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 |
|---|---|---|---|---|---|
Cohesin tethers together regions of DNA , thereby mediating higher order chromatin organization that is critical for sister chromatid cohesion , DNA repair and transcriptional regulation . Cohesin contains a heterodimeric ATP-binding Cassette ( ABC ) ATPase comprised of Smc1 and Smc3 ATPase active sites . These ATPases are required for cohesin to bind DNA . Cohesin’s DNA binding activity is also promoted by the Eco1 acetyltransferase and inhibited by Wpl1 . Recently we showed that after cohesin stably binds DNA , a second step is required for DNA tethering . This second step is also controlled by Eco1 acetylation . Here , we use genetic and biochemical analyses to show that this second DNA tethering step is regulated by cohesin ATPase . Furthermore , our results also suggest that Eco1 promotes cohesion by modulating the ATPase cycle of DNA-bound cohesin in a state that is permissive for DNA tethering and refractory to Wpl1 inhibition .
Multi-subunit protein complexes called Structural Maintenance of Chromosomes ( SMC ) mediate most aspects of higher-order chromosome organization ( Hirano , 2006 ) . Cohesin is an SMC complex that was first identified as being essential for sister chromatid cohesion , the process of holding together the sister chromatids from their synthesis in S phase until their segregation in anaphase ( Guacci et al . , 1997; Michaelis et al . , 1997 ) . Cohesin is also involved in mitotic condensation , meiotic chromosome condensation and structure , post-replicative DNA repair , and transcriptional regulation ( Onn et al . , 2008 ) . Cohesin performs these different functions by tethering together two regions of DNA , either within a single DNA molecule ( condensation and regulation of gene expression ) or between two DNA molecules ( sister chromatid cohesion and DNA repair ) . How cohesin-mediated DNA tethering is regulated is one of the critical unanswered questions in the field . The process of sister chromatid cohesion involves the binding of cohesin to chromosomes prior to DNA synthesis , and the subsequent tethering of sister chromatids by the DNA-bound cohesin to generate cohesion during S phase ( Onn et al . , 2008 ) . Cohesin appears to topologically entrap DNA through its ring-like architecture that results from the assembly of its four core subunits , Smc1 , Smc3 , Mcd1/Scc1 , and Scc3/Irr1 ( Nasmyth and Haering , 2009 ) ( Figure 1A ) . Cohesin’s loading onto DNA is controlled by its ATP Binding Cassette ( ABC ) ATPase domain ( Arumugam et al . , 2003; Heidinger-Pauli et al . , 2010b; Murayama and Uhlmann , 2014 ) . Cohesin’s ATPase domain is composed of two ATPase active sites each containing four conserved motifs: Walker A , Walker B , signature motif , and D-loop . The Smc1 ATPase active site is a hybrid site containing the Walker A and Walker B motifs encoded by Smc1 , and the D-loop and signature motifs encoded by Smc3 . Likewise , the Smc3 ATPase active site contains the Smc3-encoded Walker A and Walker B motifs , and Smc1-encoded D-loop and signature motifs ( Arumugam et al . , 2006; Haering et al . , 2004; Hopfner et al . , 2000 ) ( Figure 1A inset ) . Characterization of mutations that block ATP binding ( Walker A ) or hydrolysis ( Walker B ) suggest that ATP binding and hydrolysis by both ATPases are necessary for the DNA binding of cohesin ( Arumugam et al . , 2003; Heidinger-Pauli et al . , 2010b ) . Cohesin’s ATPase activity is also stimulated by the Scc2/Scc4 complex ( Murayama and Uhlmann , 2014 ) , which is required for the DNA binding of cohesin in vivo ( Ciosk et al . , 2000 ) . These observations led to the hypothesis that Scc2/Scc4 complex stimulates ATP hydrolysis to open the cohesin ring and allow the entry of DNA . Subsequent ATP binding closes the ring , entrapping the DNA ( Arumugam et al . , 2003 ) . Presumably , in order for cohesin to remain stably bound to DNA , its ATPase activity would have to be suppressed to prevent ring reopening and DNA escape . However , the ATPase activity of cohesin in its stable DNA-bound form had never been determined . 10 . 7554/eLife . 11315 . 003Figure 1 . Stable DNA binding does not suppress ATPase activity of cohesin . ( A ) Cartoon representation of the cohesin complex . Inset shows a model of the ABC ATPase domain of cohesin complex , based on the X-ray crystal structure of Rad50 in the presence of ATP ( Hopfner et al . , 2000 ) , and of Smc1 head domain in the presence of ATP and Mcd1 C-terminus ( Haering et al . , 2004 ) . This model has not been experimentally verified for the cohesin complex heterodimeric ATPase head domain . ( B ) Schematic showing in vitro assembly of stable cohesin-DNA bead complexes . DNA bearing CARC1 sequence was attached to dynabeads via biotin-streptavidin interaction at both ends . Cohesin was incubated with bead-bound DNA and loader in buffer containing 25 mM KCl and 25 mM NaCl , then washed in 500 mM KCl to wash off salt-sensitive cohesin . The remaining DNA-bound cohesin ( and small amount of loader ) is referred to as cohesin-DNA-beads ( CD-B ) . ( C ) Cohesin assembly on DNA-beads . S . pombe cohesin and loader complexes were purified from Y4443 and Y4483 , respectively . Purified cohesin and loader were incubated with dynabeads-DNA or dynabeads alone for 1 hour at 30°C , then cohesin was washed off as described in B . Cohesin bound DNA-beads ( DNA ) but failed to bind beads lacking DNA ( - ) . ( D ) Cohesin binding to DNA is stimulated by the loader complex . DNA binding was done as described in B & C , except loader was omitted in one sample . Percent cohesin bound was calculated by quantifying bands on Coomassie-stained SDS-PAGE . Data from two independent experiments , error bars represent standard deviation . ( E ) Effect of stable DNA binding on cohesin ATPase activity . ATPase activity of cohesin alone ( 2 ) was compared to cohesin with DNA ( 3 ) , cohesin in presence of loader complex and DNA ( 4 ) , and cohesin in stable cohesin-DNA complexes ( CD-B , 5 ) . Proteins were incubated in ATPase buffer 2 spiked with hot ATP for 1 hour at 30°C . Released Pi was calculated and plotted as described in Methods . Error bars represent standard deviation from two independent experiments . ( F ) Equal concentrations of cohesin were used in the ATPase reactions . Arrows point to S . pombe homologs of cohesin subunits , Smc1 , Smc3 ( Psm1 and Psm3 in S . pombe , ~150 kD ) , Mcd1 and Scc3 ( Rad21 and Psc3 in S . pombe , ~100 kD ) . Asterisks mark subunits of the S . pombe homolog of the loader complex , Scc2/Scc4 ( Mis4/Ssl3 in S . pombe ) . Due to the lower ability of the loader complex to bind to DNA under these conditions , there is less loader complex present in sample 5 than in sample 4 . ( G ) Stably DNA bound cohesin remained bound to DNA-beads throughout the course of the ATPase experiment . Cohesin was incubated with DNA-beads in the presence of loader and ATP as described before and washed in 0 . 5 M KCl , then resuspended in ATPase buffer . Samples were incubated for 1 hour at 30°C . Supernatant and beads were separated and visualized on SDS-PAGE . Protein gels are representative of at least 2 independent experiments . Bands in some panels were spliced from the same gel for representation purposes . Please see Figure supplements 1–4 for further characterization of cohesin’s ATPase activity while stably bound to DNA . DOI: http://dx . doi . org/10 . 7554/eLife . 11315 . 00310 . 7554/eLife . 11315 . 004Figure 1—figure supplement 1 . The ATPase activity of cohesin is stimulated by the loader and abolished by a walker-A ( K38I ) mutation in Psm3 ( S . pombe Smc3 homolog ) . Equal amounts of wild type or mutant cohesin was incubated in the presence of loader and DNA in reaction buffer spiked with hot ATP for 1 hour at 30°C . Released Pi was calculated and plotted as described in Methods . Error bars represent standard deviation from two independent experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 11315 . 00410 . 7554/eLife . 11315 . 005Figure 1—figure supplement 2 . Stably DNA-bound cohesin ( CD-B ) can be eluted off the DNA-beads by a DNase or restriction enzyme ( Mnl I ) digest . ( A ) CD-B assembled as described in Figure 1B was resuspended in CL1 buffer containing DNase . Beads were separated from the supernatant and proteins were visualized by SDS-PAGE . ( B ) CD-B assembled as described in Figure 1B was resuspended in buffer CL1 buffer containing DNase or Mnl I . Beads were separated from the supernatant and proteins were visualized by SDS-PAGE , followed by Western blotting . DOI: http://dx . doi . org/10 . 7554/eLife . 11315 . 00510 . 7554/eLife . 11315 . 006Figure 1—figure supplement 3 . Stably DNA-bound cohesin does not come off the DNA-beads after incubation with competitor DNA . CD-B assembled as described in Figure 1B was resuspended in 20 μL CL1 buffer , in the presence or absence of 0 . 5 mM ATP and 2 . 5 μg plasmid DNA ( 5x excess in mass compared to CD-B ) . Supernatant and pellets were separated at the end of 30-minute incubation at 30°C . Cohesin in supernatant and pellet fractions was visualized by Western blotting against the V5-tagged Smc3 homolog of S . pombe . DOI: http://dx . doi . org/10 . 7554/eLife . 11315 . 00610 . 7554/eLife . 11315 . 007Figure 1—figure supplement 4 . Cohesin stably assembled on DNA in the absence of loader has at least as high ATPase activity as cohesin +DNA . ( A ) Cohesin + DNA in solution ( sample 1 ) and CD-B assembled in the absence of loader ( sample 2 ) were incubated in reaction buffer spiked with hot ATP for 1 hour at 30°C . Released Pi was calculated and plotted as described in Methods . Error bars represent standard deviation from two independent experiments . ( B ) SDS-PAGE showing cohesin used in A . ( C ) Stability of cohesin on DNA , similar to Figure 1D . Cohesin that is assembled in the absence of the loader also can bind DNA stably , although the assembly is less efficient . Protein gels are representative of at least 2 independent experiments . Bands in some panels were spliced from the same gel for representation purposes . DOI: http://dx . doi . org/10 . 7554/eLife . 11315 . 007 Cohesin binding to DNA is also regulated by the antagonistic activites of Wpl1 and Eco1 . Wpl1 is thought to destabilize cohesin binding to DNA ( Chan et al . , 2012; Kueng et al . , 2006 ) . Wpl1 is antagonized by the Eco1 acetyl-transferase through its acetylation of lysines 112 and 113 of Smc3 . The key evidence supporting this view is that the deletion of WPL1 ( wpl1∆ ) suppresses the inviability of cells lacking ECO1 ( eco1∆ ) or expressing acetyl null alleles of Smc3 ( smc3-K112R , K113R ) ( Rolef Ben-Shahar et al . , 2008; Sutani et al . , 2009; Unal et al . , 2008; Zhang et al . , 2008 ) . Recent evidence suggests that Eco1 acetylation promotes cohesion by additional mechanisms besides stabilizing cohesin binding to DNA . First , while eco1∆ wpl1∆ cells are viable and have stable cohesin-DNA interaction , they have cohesion defects as severe as mutants of ECO1 or cohesin subunits ( Guacci and Koshland , 2012; Rowland et al . , 2009; Sutani et al . , 2009 ) . Second , other mutations identified in cohesin and its regulators demonstrate that stable binding of cohesin to DNA is not sufficient for cohesion ( Eng et al . , 2014; Guacci et al . , 2015 ) . Together , these data strongly argue that cohesion is a two-step process: First , cohesin associates with DNA in a stable form . Then , cohesin undergoes a second transition to tether sister chromatids together . This transition could entail conformational changes involving oligomerization ( Eng et al . , 2015 ) , or the activation of a second , independent DNA binding activity through rearrangements of the coiled coils ( Soh et al . , 2015 ) . How is cohesin-mediated DNA tethering regulated ? One hypothesis is that Eco1-mediated acetylation of Smc3 regulates this second , post-DNA binding step by modulating the cohesin ATPase ( Guacci et al . , 2015 ) . This hypothesis appears to contradict the finding that Walker A and Walker B mutations in either cohesin ATPase blocks DNA binding ( Arumugam et al . , 2003; Heidinger-Pauli et al . , 2010b ) . However , this observation does not preclude a specialized role for the Smc3 ATPase active site in regulating DNA tethering after DNA binding . Indeed , the acetylated K112 and K113 residues in Smc3 are proximal to the Smc3 ATPase active site ( Gligoris et al . , 2014; Haering et al . , 2004 ) . Moreover , a recently identified suppressor mutation located near the Smc3 ATPase active site bypasses the requirement for Smc3 acetylation in cohesion establishment ( Guacci et al . , 2015 ) . Led by these observations , we reconsider the role of the ATPase domain of cohesin as a potential regulator of the second , post-DNA binding step of cohesion establishment . Here , we present in vitro and in vivo evidence that the ATPase domain of cohesin plays a role after the initial stable DNA binding of cohesin . We provide evidence suggesting that the Smc1 ATPase active site is involved only in regulating DNA binding , whereas the Smc3 ATPase active site functions in DNA tethering as well as DNA binding . We characterize an Smc3 ATPase active site mutant in Saccharomyces cerevisiae that bypasses the requirement for Eco1 acetylation in cohesion generation , and uncouples the level of ATPase activity from cohesin’s DNA binding and tethering activities . We propose that cohesin’s ATPase has two distinct functions in regulating DNA binding and subsequent DNA tethering . We suggest that Eco1 promotes cohesion by slowing or trapping the ATPase cycle of DNA-bound cohesin to promote a conformation that is permissive for DNA tethering and refractory to Wpl1 inhibition .
Earlier models suggest that cohesin’s ATPase head domain is only involved in the initial DNA binding step , and that ATP hydrolysis releases the DNA from cohesin . These models predict that stably DNA-bound cohesin should not show ATPase activity . However , recent literature suggests that Eco1 might promote cohesion by regulating the cohesin ATPase after the stable DNA binding of cohesin . If ATPase activity is required to regulate this second step of cohesion establishment , we should be able to observe ATPase activity for purified cohesin-DNA complexes . To test this possibility , we purified Schizosaccharomyces pombe cohesin and loader complexes as described previously ( Murayama and Uhlmann , 2014 ) . Purified cohesin showed basal ATPase activity that was stimulated by the presence of the loader complex and DNA , and abolished by a Walker A mutation in Smc3 ( Figure 1—figure supplement 1 ) , similar to published results ( Murayama and Uhlmann , 2014 ) . Cohesin binding to chromosomes in vivo is enriched at centromeres and at cohesion-associated regions ( CARs ) along the chromosome arms ( Laloraya et al . , 2000; Onn et al . , 2008 ) , and is highly salt resistant ( Ciosk et al . , 2000 ) . To purify stable cohesin-DNA complexes that are physiologically relevant , we assembled cohesin on DNA molecules that contained CARC1 sequence and were coupled by both ends to dynabeads . Cohesin and its loader were incubated with DNA-beads under low salt conditions ( 25 mM KCl , 25 mM NaCl ) . The cohesin-DNA bead mix was washed with high salt ( 500 mM KCl ) to remove any free cohesin or cohesin not stably bound to DNA ( Figure 1B ) . The cohesin that remained bound to the DNA-beads was then eluted and quantified by Coomassie staining or Western blots . In the presence of the loader , 20% of the input cohesin was bound to DNA-beads after the high salt wash ( Figure 1C , D ) . In the absence of the loader , 2-fold less cohesin bound to DNA ( Figure 1D ) . Cohesin did not bind to beads that lack DNA ( Figure 1C ) . In addition , this stable population of cohesin on DNA-beads could be eluted from the beads by either a restriction enzyme digest or a DNase treatment ( Figure 1—figure supplement 2 ) . These results suggest that cohesin bound specifically to the DNA that was coupled to beads , and did so in a salt-resistant and loader-inducible manner . These DNA binding features recapitulated the properties of stable cohesin-DNA complexes assembled in vivo ( Ciosk et al . , 2000 ) . To address whether stably DNA-bound cohesin was an active ATPase , we next assessed whether cohesin in our purified , stable cohesin-DNA complexes ( CD-B , Figure 1B ) retained its ATPase activity . DNA-beads were incubated with cohesin and loader , then washed with high salt buffer to remove free and unstably bound cohesin , as described above . Cohesin-DNA beads ( CD-B ) were then resuspended in buffer containing ATP to measure ATPase activity of CD-B , compared to basal and loader/DNA-stimulated activities of cohesin . No ATPase activity was detected when cohesin was omitted from the reaction ( Figure 1E and Figure 1—figure supplement 1 ) . As shown before ( Murayama and Uhlmann , 2014 ) , cohesin’s basal ATPase activity ( sample 2 ) was stimulated about 3-fold by the presence of loader and DNA ( sample 4 ) , but only subtly increased by DNA alone ( sample 3 , Figure 1E ) . Our purified cohesin-DNA complexes on beads ( CD-B , sample 5 ) showed about 4-fold increased ATPase activity compared to the basal activity of equal amounts of cohesin ( sample 2 , Figure 1E , F ) . Furthermore , the ATPase activity of CD-B was at least as high , if not higher , as the activity of equal concentrations of loader/DNA-stimulated cohesin in solution , where additional rounds of cohesin loading were possible ( sample 4 compared to sample 5 , Figure 1E , F ) . The ATPase activity in the fraction containing purified cohesin-DNA complexes might have been derived from DNA-free cohesin that had dissociated from DNA-beads . To test this possibility , a parallel sample of CD-B was treated similarly to the ATPase assay to assess the amount of cohesin that remained bound to DNA through the course of the ATPase experiment . DNA-beads were separated from the supernatant after the 30-minute incubation with ATP and the amount of cohesin in each fraction was visualized by SDS-PAGE . Very little , if any , cohesin was found in the supernatant at the end of the incubation , suggesting that most of the ATPase activity was derived from cohesin bound to the DNA-beads ( Figure 1G ) . Even in the presence of excess competitor DNA in solution , cohesin remained stably bound to the DNA-beads ( Figure 1—figure supplement 3 ) . Thus in vitro , cohesin can have robust ATPase activity while remaining stably bound to DNA . In our preparations of purified cohesin-DNA complexes , a small amount of the loader complex was also present , due to the ability of the loader complex to bind DNA . Given that the loader complex stimulates cohesin’s ATPase activity in the presence of DNA , it was possible that the ATPase activity in our purified cohesin-DNA complexes depended upon the presence of the loader complex . To test this possibility , we repeated these experiments but omitted the loader from the binding reaction . Under these conditions , a smaller percentage of cohesin could be assembled on DNA in a salt-resistant manner ( Figure 1D and Figure 1—figure supplement 4 ) . Cohesin in these purified cohesin-DNA complexes that have no co-purifying loader complex also retained ATPase activity equivalent to that seen in the mixture of cohesin and DNA in solution ( Figure 1—figure supplement 4 ) . Thus , the persistence of ATPase activity in stable cohesin-DNA complexes was not dependent upon the loader . The presence of cohesin’s ATPase activity in the stable cohesin-DNA complex is consistent with a regulatory role for cohesin ATPase in cohesion establishment in vivo after cohesin stably binds to DNA . Cohesin exhibited robust ATPase activity after stably binding DNA , which encouraged us to re-examine the in vivo role of cohesin ATPases and Eco1 in cohesin function . Cohesin-DNA complexes in eco1∆ wpl1∆ cells are stably bound to DNA but fail to establish cohesion , indicating a post-DNA binding step is required for cohesion . We postulated that this failure to generate cohesion was due to a misregulation of the cohesin ATPase active sites . If so , a subset of mutations altering the Smc1 or Smc3 ATPase active sites might mimic the state of the ATPase upon Eco1 acetylation , and thereby might suppress the cohesion defect of eco1∆ wpl1∆ cells . To identify such suppressors , we exploited the fact that eco1∆ wpl1∆ cells remain viable in the absence of cohesion because of an unusual feature of budding yeast cell cycle that gives rise to a cohesin-independent mechanism of sister chromatid segregation ( Guacci and Koshland , 2012 ) . The reliance of eco1∆ wpl1∆ cells on this cohesin-independent segregation makes them sensitive to the microtubule depolymerizing drug benomyl . A screen for suppressors of the benomyl sensitivity of eco1∆ wpl1∆ cells should identify a subset of mutations that restore cohesion and the cohesin-dependent benomyl-resistant mechanism of segregation . Indeed , we recently reported one such benomyl-resistant suppressor allele , smc3-D1189H ( Guacci et al . , 2015 ) , that restored cohesion . smc3-D1189H is located in the ATPase head domain , near the Smc3 ATPase active site . Due to this restoration of cohesion to eco1∆ wpl1∆ cells , we termed this mutation a cohesion activator mutation . Here , we present two new SMC1 alleles from our screen , smc1-D1164E and smc1-Y1128C . Re-introduction of these mutations into the parent eco1∆ wpl1∆background generated benomyl-resistance identical to the initial suppressor eco1∆ wpl1∆ isolates and close to the resistance of wild type cells ( Figure 2A ) , demonstrating their causal role for this phenotype . The SMC1 suppressor mutations are even more proximal to the Smc3 ATPase active site than smc3-D1189H ( Figure 2B ) . smc1-Y1128C is adjacent to the signature motif , which is thought to modulate ATP binding , but this residue is not conserved . smc1-D1164E alters the invariant aspartate that is part of the conserved D-loop motif found in ABC and SMC ATPases ( Figure 2C ) . Studies from other ABC ATPases suggest that this aspartate interacts with the catalytic H loop and Walker A ( Hohl et al . , 2012; la Rosa and Nelson , 2011 ) , and mediates communications between the two ATPase active sites within the ATPase head domain ( Furman et al . , 2013 ) . Taken together , it is likely that these suppressor mutations modulate the cohesin ATPase to promote cohesion . 10 . 7554/eLife . 11315 . 008Figure 2 . smc1-D1164E and smc1-Y1128C mutations suppress the benomyl sensitivity of eco1△ wpl1△ cells and are part of the Smc3 ATPase active site . ( A ) Assessing effects of smc1-D1164E and smc1-Y1128C on eco1∆ wpl1∆ benomyl sensitivity . Haploid wild-type cells ( VG3460-2A ) , and eco1∆ wpl1∆ cells alone ( VG3502-1A ) or containing smc1-D1164E ( VG3574-5A ) , smc1-Y1128C ( VG3576-1C ) or smc3-D1189H ( VG3547-3B ) were grown to saturation in YPD at 23°C , plated as 10-fold serial dilutions on YPD alone , or containing benomyl at 12 . 5 μg/mL ( BEN ) then incubated at 23°C for 3 days . ( B ) Cartoon depicting the Smc1 and Smc3 ATPase active sites along with the position of three suppressor mutations shown in A . All three suppressor mutations are in the vicinity of the Smc3 ATPase active site . Note that the Smc1 D-loop and signature motifs form part of the Smc3 ATPase active site . ( C ) The conservation of residues around the D-loop in distant ABC ATPases . DOI: http://dx . doi . org/10 . 7554/eLife . 11315 . 008 To begin to test this hypothesis , we first asked whether smc1-D1164E or smc1-Y1128C were also cohesion activator mutations , i . e . suppressed the severe cohesion defect of eco1∆ wpl1∆ cells . Haploid wild type cells and eco1∆ wpl1∆ parent cells containing SMC1 , smc1-D1164E or smc1-Y1128C alleles were arrested in G1 , then released into media containing nocodazole to arrest them in M phase ( Figure 3A ) . To assay for cohesion , these cells contained a LacO array integrated at a CEN-proximal locus ( TRP1 ) , and a GFP-LacI fusion protein that bound to the array . In this assay , sister chromatids that have cohesion generate a single GFP spot , whereas chromatids lacking cohesion generate 2 GFP spots ( Figure 3A ) . Wild type cells had robust cohesion as shown by few M phase cells with two GFP spots ( Figure 3B and Figure 3—figure supplement 1A ) . eco1∆ wpl1∆ cells exhibited a severe cohesion defect ( more than 60% of cells had two GFP spots ) , which was reduced by half by the smc1-D1164E suppressor ( Figure 3B and Figure 3—figure supplement 1A ) . The smc1-Y1128C suppressor also reduced the eco1∆ wpl1∆ cohesion defect , but less effectively ( Figure 3—figure supplement 2 ) . Thus , both of these new suppressor mutations , like smc3-D1189H , suppressed significantly but not completely the cohesion defect of eco1∆ wpl1∆ cells . The increased CEN-proximal cohesion generated by smc1-D1164E or smc1-Y1128C was likely responsible for the increased benomyl resistance in the eco1∆ wpl1∆ background . Subsequent studies focused on smc1-D1164E , because its alters a critical and absolutely conserved D-loop aspartate , and it more robustly suppresses the cohesion defect of eco1∆ wpl1∆ cells . 10 . 7554/eLife . 11315 . 009Figure 3 . smc1-D1164E allele restores cohesion in the absence of both Eco1 and Wpl1 . ( A ) Regimen used to assess sister chromatid cohesion in cells . Mid-log phase cultures of asynchronously growing cells at 23°C were arrested in G1 with alpha factor for 3 hours , then released into media containing nocodazole for 3 hours to arrest cells in M phase . Representative images of M phase arrested cells are shown with cells being visualized by Nomarski ( NOM ) and cohesion ( GFP ) , which marks a CEN-proximal TRP1 locus . Left side shows a cell where cohesion exists ( one GFP spot . ) Right side shows a cell where sisters have separated ( 2 GFP spots ) . ( B , C ) smc1-D1164E partially restored cohesion in eco1△ wpl1△ cells at the CEN-proximal TRP1 locus . Haploid wild type ( WT , VG3460-2A ) , eco1∆ wpl1∆ ( VG3502-2A ) and smc1-D1164E eco1∆ wpl1∆ ( VG3574-5A ) cells were released from G1 and arrested in M phase using nocodazole as described in A . The percentage of cells with two GFP spots was plotted . ( B ) Cohesion loss at a CEN-proximal locus ( TRP1 ) in M phase arrested cells . DNA content of these cells is shown in Figure 3—figure supplement 1 , panel A . ( C ) Time course to assess kinetics of cohesion loss at a CEN-proximal locus ( TRP1 ) . Cell aliquots were fixed in G1 and at 20-minute intervals after release . Grey box shows S phase ( based on DNA content shown in Figure 3—figure supplement 1 , panel B ) . Please see Figure supplements 1–6 for further characterization of cohesin activator mutants . DOI: http://dx . doi . org/10 . 7554/eLife . 11315 . 00910 . 7554/eLife . 11315 . 010Figure 3—figure supplement 1 . DNA content analysis of cells from Figure 3B ( A ) and Figure 3C ( B ) . DOI: http://dx . doi . org/10 . 7554/eLife . 11315 . 01010 . 7554/eLife . 11315 . 011Figure 3—figure supplement 2 . Cohesion loss in smc1-Y1128C at CEN-proximal ( TRP1 ) locus . Haploid wild type ( WT , VG3460-2A ) , eco1∆ wpl1∆ ( VG3502-2A ) and smc1-Y1128C eco1∆ wpl1∆ ( VG3576-1C ) were arrested in G1 then released into M phase arrest as described in Figure 3A . Cohesion ( left ) and DNA content ( right ) . DOI: http://dx . doi . org/10 . 7554/eLife . 11315 . 01110 . 7554/eLife . 11315 . 012Figure 3—figure supplement 3 . Cohesion loss at a CEN-distal locus ( LYS4 ) in smc1-D1164E cells . Haploid wild type ( VG3349-1B ) , eco1∆ wpl1∆ ( VG3503-4A ) and smc1-D1164E eco1∆ wpl1∆ ( VG3575-2C ) were arrested in G1 then released into M phase arrest as described in Figure 3A . ( A ) Cohesion loss in M phase arrested cells . Cohesion ( left ) and DNA content ( right ) . ( B ) Time course to assess kinetics of cohesion loss . Cohesion ( left ) and DNA content ( right ) . Grey box shows S phase . DOI: http://dx . doi . org/10 . 7554/eLife . 11315 . 01210 . 7554/eLife . 11315 . 013Figure 3—figure supplement 4 . Effect of the smc1-D1164E on cohesin ( Mcd1 ) binding in M phase arrested eco1∆ wpl1∆ cells . Haploid eco1∆ wpl1∆ ( VG3502-1A ) and smc1-D1164E eco1∆ wpl1∆ ( VG3574-5A ) cells arrested in M phase as described in Figure 3A were fixed and processed for ChIP using Mcd1 antibodies . Mcd1p binding was assessed by qPCR . Data is presented as percentage of total DNA assayed using the same primer pairs at each site . Mcd1 ChIP at Chromosome III peri-centric CARC1 ( left side ) . Six primer pairs used to assay Mcd1 binding at loci spanning a ~2 . 6 kb region including CARC1 of chr . III . Mcd1 ChIP at chromosome XII CEN-distal CARL1 ( right side ) . Six primer pairs used to assay Mcd1p binding at loci spanning a ~4 . 5 kb region including CARL1 of chr . XII . eco1∆ wpl1∆ ( grey line , grey circles ) and smc1-D1164E eco1∆ wpl1∆ ( black line , open diamonds ) . DOI: http://dx . doi . org/10 . 7554/eLife . 11315 . 01310 . 7554/eLife . 11315 . 014Figure 3—figure supplement 5 . The Smc3 ATPase active site D-loop cohesion activator mutation smc1-D1164E cannot suppress a wpl1∆ . ( A ) Viability of smc1-D1164E in wpl1△background . Haploid wpl1△ ( VG3623-2A ) and smc1-D1164E wpl1△ ( VG3625-2C ) strains were grown then dilution plated onto YPD as described in Figure 2A . ( B ) Cohesion loss in M phase arrested smc1-D1164E in wpl1△cells . Cells were grown and arrested in M phase as described in Figure 3A . Cohesion loss at the CEN-proximal TRP1 locus ( left side ) . Haploid WT ( VG3460-2A ) , wpl1△ ( VG3623-2A ) and smc1-D1164E wpl1△ ( VG3625-2C ) . Cohesion at the CEN-distal LYS4 locus . Haploid WT ( VG3557-2A ) , wpl1△ ( VG3601-8B ) and smc1-D1164E wpl1△ ( VG3603-3D ) . DNA content analysis of cells ( bottom ) . DOI: http://dx . doi . org/10 . 7554/eLife . 11315 . 01410 . 7554/eLife . 11315 . 015Figure 3—figure supplement 6 . The Smc3 ATPase active site D-loop cohesion activator mutation smc1-D1164E partially suppresses the requirement for Eco1 . ( A ) Viability of smc1-D1164E in eco1△background . Haploid WT ( VG3596-6A ) and smc1-D1164E eco1△ ( VG3779-1E ) cells were grown and dilution plated on YPD as described in Figure 2A . ( B ) Viability of smc1-D1164E in ECO1-AID background . Haploid WT ( VG3349-1B ) , SMC1 ECO1-AID ( VG3646-1A ) and smc1-D1164E ECO1-AID ( VG3648-2C ) cells were grown as described Figure 2A then serial dilution ( 10-fold ) plated on YPD alone or YPD + Auxin and incubated 2 days at 23°C . Auxin induces depletion of ECO1-AID allowing assessment of ability to bypass the essential ECO1 function . All strains were analyzed on the same plate but rearranged for illustration purposes . ( C ) Cohesion loss at the CEN-proximal TRP1 locus in M phase arrested smc1-D1164E eco1△ cells . Strains in C were assayed for cohesion in M phase arrested cells ( top panel ) as described in Figure 3A and for DNA content ( bottom panel ) . DOI: http://dx . doi . org/10 . 7554/eLife . 11315 . 015 To better understand the impact of the smc1-D1164E suppressor on cohesion , we monitored cohesion at the CEN-proximal TRP4 locus at 20 minute intervals as cells progressed from G1 to M ( Figure 3—figure supplement 1B ) . As expected , wild type cells showed robust cohesion where very few cells had separated sisters , and eco1∆ wpl1∆ cells showed sister chromatid separation starting in S phase ( Figure 3C ) , consistent with a cohesion establishment defect that was shown previously for this strain ( Guacci and Koshland , 2012; Guacci et al . , 2015 ) . The small amount of cohesion loss in smc1-D1164E eco1∆ wpl1∆ cells also began during S phase , indicating that the residual cohesion defect in these cells resulted from a failure in cohesion establishment . However , these cells displayed fewer separated sisters throughout the time course compared to eco1∆ wpl1∆ cells . These results indicated that ( 1 ) smc1-D1164E promotes cohesion establishment in the eco1∆ wpl1∆ , albeit incompletely , and ( 2 ) once cohesion is formed in smc1-D1164E eco1∆ wpl1∆ cells , it is maintained . smc1-D1164E also partially restored cohesion establishment in eco1∆ wpl1∆ cells at the CEN-distal locus , LYS4 ( Figure 3—figure supplement 3 ) . We conclude that smc1-D1164E promotes cohesion establishment in the eco1∆ wpl1∆ , albeit incompletely . Since cohesin is already stably bound to DNA in eco1∆ wpl1∆ cells , the restoration of cohesion in smc1-D1164E eco1∆ wpl1∆ cells should occur without altering the DNA binding of cohesin . To test this prediction , we performed chromosome immunoprecipitation ( ChIP ) to compare cohesin binding to chromosomes in eco1∆ wpl1∆ cells containing SMC1 or smc1-D1164E allele . Cohesin subunits colocalize on chromosomes , making the analysis of any cohesin subunit a surrogate for cohesin binding ( Glynn et al . , 2004; Heidinger-Pauli et al . , 2010a; Lengronne et al . , 2004 ) . Here we used anti-Mcd1 antibodies as a means to assess cohesin binding to chromosomes . Cohesin binding at CARs has been shown to be reduced 2-3 fold in eco1∆ wpl1∆ cells as compared to wild type cells ( Guacci et al . , 2015; Sutani et al . , 2009 ) . In eco1∆ wpl1∆ cells containing either SMC1 or smc1-D1164E allele , there was little or no difference in Mcd1 binding at the centromere-proximal CARC1 or centromere-distal CARL1 ( Figure 3—figure supplement 4 ) . This similarity in chromosomal binding indicates that cohesion restoration in eco1∆ wpl1∆ cells by the smc1-D1164E allele was not due to an altered level of cohesin binding to chromosomes , but instead was due to a change in cohesin function at a post-DNA binding step . The restoration of cohesion in smc1-D1164E eco1∆ wpl1∆ cells indicated that the smc1-D1164E could bypass one or more aspects of cohesion regulation normally imposed by Eco1 and Wpl1 . To understand better how smc1-D1164E , and by inference the Smc3 ATPase active site , interfaced with cohesin regulators , we examined its impact on the phenotypes of wpl1∆ and eco1∆single mutants . In a wpl1∆ background , SMC1and smc1-D1164E alleles had similar viability and the same moderate cohesion defect characteristic of wpl1∆ mutants ( Figure 3—figure supplement 5 ) . The smc1-D1164E allele suppressed the essential ECO1 function , as smc1-D1164E eco1∆ cells showed good viability ( Figure 3—figure supplement 6 ) . smc1-D1164E eco1∆ cells showed a 35% cohesion defect ( Figure 3—figure supplement 6 ) , which was greatly reduced compared to 65-70% cohesion defect that was observed previously for eco1 mutant cells or cells lacking both Eco1 and Wpl1 ( Guacci et al . , 2015; Sutani et al . , 2009 ) . Taken together , these results suggest that smc1-D1164E restored cohesion by phenocopying both Eco1’s antagonism of Wpl1 and its ability to promote stably DNA-bound cohesin to tether sister chromatids . The partial restoration of cohesion in smc1-D1164E eco1∆ wpl1∆ cells could reflect the inability of smc1-D1164E to fully compensate for the loss of Eco1 or Wpl1 . Alternatively , it might reflect an inherent cost of uncoupling cohesin from its regulators . To differentiate between these possibilities , we characterized the smc1-D1164E allele when both ECO1 and WPL1 were present . In plasmid shuffle assays ( Materials and methods ) , the viability of cells containing the smc1-D1164E or SMC1 alleles was indistinguishable ( Figure 4—figure supplement 1 ) . However , otherwise wild type cells carrying the smc1-D1164E allele had a moderate cohesion defect at both CEN-proximal and CEN-distal loci , which arose around the time of replication ( Figure 4—figure supplement 2 ) . This defect was similar to that observed for the smc1-D1164E eco1∆ wpl1∆ cells . Thus , cohesin in smc1-D1164E cells had an inherent moderate cohesion establishment defect independent of cohesin regulators . Taken together , our results suggest that the smc1-D1164E allele was constitutively cohesive as it uncoupled cohesin function from its regulators , which reduced the efficiency of cohesion generation . Our two new cohesion activator mutations mapped to residues intimately involved in the Smc3 ATPase active site , and the previously characterized third cohesion activator was in close proximity ( Guacci et al . , 2015 ) . This observation suggested that the Smc3 but not the Smc1 ATPase active site might play a specialized role in activating stably DNA-bound cohesin to tether sister chromatids . To test more directly whether the two active sites played distinct roles in cohesin regulation , we compared the phenotypes of smc1-D1164E to smc3-D1161E , the analogous glutamate substitution of the D-loop aspartate of the Smc1 ATPase active site ( Figure 4A ) . We used Auxin Inducible Degron ( AID ) system to obtain conditional alleles of SMC1 and SMC3 , SMC1-AID and SMC3-AID , respectively , which can be degraded upon the addition of auxin ( Figure 4—figure supplement 3 ) . In strains containing either SMC1-AID or SMC3-AID alleles as sole source of Smc1 or Smc3 , we integrated our test alleles , smc1-D1164E and smc3-D1161E , respectively . The smc1-D1164E allele robustly suppressed the inviability associated with depletion of Smc1-AID , whereas smc3-D1161E could not support viability when Smc3-AID was depleted ( Figures 4B , C ) . Similar results in viability were seen when these alleles were tested by plasmid shuffle assays ( Figure 4—figure supplement 1 & 4 ) . This difference in viability between the smc1-D1164E and smc3-D1161E alleles suggests that Smc1 and Smc3 ATPase active sites have distinct functions . 10 . 7554/eLife . 11315 . 016Figure 4 . D-loop mutations in Smc1 and Smc3 ATPase active sites reveal that the two sites are not functionally equivalent . ( A ) Cartoon representation of cohesin’s ATPases . The Smc1-encoded D-loop is part of the Smc3 ATPase active site and the Smc3-encoded D-loop is part of the Smc1 ATPase active site . ( B ) Assessing whether the Smc3 ATPase active site D-loop mutant smc1-D1164E promotes cell viability . Haploid SMC1 SMC1-AID ( VG3764-3A ) , SMC1-AID ( VG3711-5D ) , and smc1-D1164E SMC1-AID ( VG3765-3D ) cells were grown and plated on as described in Figure 2A onto YPD alone or YPD + auxin and incubated 3 days at 23°C . SMC1-AID was depleted in media containing auxin , which allows assessment of whether smc1-D1164E promotes viability . ( C ) Assessing whether the Smc1 ATPase active site D-loop mutant smc3-D1161E promotes cell viability . Haploid SMC3 SMC3-AID ( VG3771-10C ) , SMC3-AID ( VG3651-3D ) , and smc3-D1161E SMC3-AID ( VG3773-16D ) were grown and plated dilution as described in C . ( D ) Regimen used to assess sister chromatid cohesion in cells containing AID tagged proteins . Asynchronous cells were arrested in G1 , depleted for AID tagged proteins by the addition of auxin , then released from G1 and arrested in M phase in the presence of auxin . ( E , F ) Cohesion loss at the CEN-proximal locus ( TRP1 ) in M phase cells depleted for AID tagged proteins . SMC1-AID or SMC3-AID was depleted in strains from G1 through M phase as described in D . The percentage of cells with two GFP spots ( sister separation ) is plotted . ( E ) smc1-D1164E promotes cohesion in SMC1-AID depleted cells . Haploid SMC1 SMC1-AID ( VG3794-2E ) , SMC1-AID ( VG3711-5D ) and smc1-D1164E SMC1-AID ( VG3795-2C ) assayed for cohesion . DNA content analysis of these cells can be seen in Figure 4—figure supplement 4 . ( F ) smc3-D1161E fails to promote cohesion in to promote cohesion SMC1-AID depleted cells . Haploid SMC3 SMC3-AID ( VG3797-1A ) , SMC3-AID ( VG3651-3D ) and smc3-D1161E SMC3-AID ( VG3799-3C ) strains assayed for cohesion . DNA content analysis of these cells can be seen in Figure 4—figure supplement 4 . Please see Figure supplements 1–6 for further characterization of smc1-D1164E and smc3-D1161E alleles . DOI: http://dx . doi . org/10 . 7554/eLife . 11315 . 01610 . 7554/eLife . 11315 . 017Figure 4—figure supplement 1 . Viability of smc1-D1164E allele as sole source of SMC1 . Plasmid shuffle assay showing that the smc1-D1164E allele supports viability as sole SMC1 source . Haploid SMC1 shuffle strain ( VG3568-8A ) has the endogenous SMC1 deleted but kept viable by plasmid pTH2 ( SMC1 URA3 CEN ) . This shuffle strain alone or containing SMC1 or smc1-D1164E integrated at LEU2 were grown to saturation in YPD at 23°C . Cells were plated as 10-fold serial dilutions on YPD or on media containing 5-fluoroorotic acid ( FOA ) then grown 3d at 23°C . FOA selectively kills URA3 cells so selects for Ura- cells that have lost plasmid pTH2 , thereby revealing whether integrated SMC1 or smc1-D1164E alleles can support viability as the sole SMC1 . DOI: http://dx . doi . org/10 . 7554/eLife . 11315 . 01710 . 7554/eLife . 11315 . 018Figure 4—figure supplement 2 . Time course assessing kinetics of cohesion loss in SMC1 and smc1-D1164E cells . Strains were released from G1 and arrested in M phase as described in Figure 3A . Cell aliquots were fixed in G1 and at 20-minute intervals after release then scored to assess cohesion . Cohesion ( left ) and DNA content ( right ) . Grey box shows S phase . Top panel: Cohesion loss at the CEN-proximal ( TRP1 ) locus . Haploid SMC1 ( VG3460-2A ) and smc1-D1164E ( VG3598-8A ) strains were used for this experiment . Bottom panel: Cohesion loss at the CEN-distal ( LYS4 ) locus . Haploid SMC1 ( VG3557-2A ) and smc1-D1164E ( VG3581-8B ) strains were used for this experiment . DOI: http://dx . doi . org/10 . 7554/eLife . 11315 . 01810 . 7554/eLife . 11315 . 019Figure 4—figure supplement 3 . Assessing SMC-AID depletion and Mcd1 levels after depletion . ( A ) Haploid cells expressing Smc1AID-3V5 Smc1FLAG ( VG3764-3A ) , Smc1AID-3V5 ( VG3311-5D ) , and Smc1AID-3V5 Smc1D1164E-3FLAG ( VG3565-3D ) were arrested in G1 and were depleted for Smc1AID-3V5 from G1 through M arrest as described in Figure 4D . Trichloroacetic acid extracted total cell proteins were prepared from G1 cells before auxin addition and from M phase cells under depletion of conditions then analyzed by Western Blot . Depletion of Smc1AID-3V5 protein is monitored using aV5 antibodies ( Top panel ) . Smc1FLAG and Smc1D1164E-3FLAG is monitored using αFLAG antibodies ( second panel ) . Mcd1 was monitored using αMcd1 antibodies and αTUB antibodies were used as a loading control ( bottom panel ) . Mcd1 is absent from G1 extracts due to proteolytic destruction . Mcd1 forms a trimer with Smc1 and Smc3 from S phase through M phase cells , which protects Mcd1 from destruction . When either Smc1 or Smc3 is destroyed ( Smc1AID-3V5 only in M phase + auxin ) , Mcd1 is also destroyed as it cannot be protected . Therefore , Mcd1 presence is a proxy for trimer formation . When either Smc13FLAG or Smc1D1164E-3FLAG are in Smc1AID-3V5 M phase + Auxin cells , Mcd1 is still present . Therefore , Smc1 and Smc1D1164E form the cohesin trimer ( i . e support cohesin complex assembly ) . ( B ) Haploid cells expressing Smc3AID-3V5 Smc36HA ( VG3726-6A ) , Smc3AID-3V5 ( VG3711-5D ) , and Smc3AID-3V5 Smc3D1161E-6HA ( VG3728-2A ) were arrested in G1 and were depleted for Smc3AID-3V5 from G1 through M arrest , as in ( E ) . The only difference is that αHA antibodies were used to monitor Smc36HA and Smc3D1161E-6HA . Both Smc3 and Smc3D1161E form the cohesin trimer ( support cohesin complex assembly ) as Mcd1 is present in M phase + auxin cells . DOI: http://dx . doi . org/10 . 7554/eLife . 11315 . 01910 . 7554/eLife . 11315 . 020Figure 4—figure supplement 4 . Inviability of smc3-D1161E allele as sole source of SMC3 . Plasmid shuffle assay showing smc3-D1161E cannot support viability as sole SMC3 source . Haploid SMC3 shuffle strain ( VG3464-16C ) has the endogenous SMC3 deleted but is kept viable by plasmid pEU42 ( SMC3 URA3 CEN ) . This SMC3 shuffle strain alone or with either SMC3 or smc3-D1161E integrated at LEU2 were grown and plated on YPD or FOA plates as described in Figure 4–Figure supplement 1 to assess the ability of smc3-D1161E to support viability as the sole SMC3 after pEU42 loss . DOI: http://dx . doi . org/10 . 7554/eLife . 11315 . 02010 . 7554/eLife . 11315 . 021Figure 4—figure supplement 5 . DNA content analysis of cells used in Figure 4E , F . DOI: http://dx . doi . org/10 . 7554/eLife . 11315 . 02110 . 7554/eLife . 11315 . 022Figure 4—figure supplement 6 . Cohesion assay of smc1-D1164E and smc3-D1161E at a CEN-distal locus . ( A ) smc1-D1164E has robust cohesion at the CEN-distal LYS4 locus . Haploid strains from Figure 4B were depleted from Smc1AID-3V5 with auxin , released from G1 and arrested in M phase using nocodazole in continued presence of auxin as described in Figure 4D . The percentage of cells with 2-GFP signals ( sister separation ) is plotted . The lack of G1 cells with 2-GFP spots demonstrates absence of pre-existing aneuploidy . ( B ) smc3-D1161E cells are severely defective in cohesion at the CEN-distal LYS4 locus . Haploid strains from Figure 4C were assayed for cohesion as described in A . DOI: http://dx . doi . org/10 . 7554/eLife . 11315 . 022 We next assessed cohesion at CEN-proximal and CEN-distal loci in smc1-D1164E SMC1-AID and smc3-D1161E SMC3-AID strains under conditions in which AID tagged subunits remain depleted from G1 through M phase ( Figure 4D ) . As expected , SMC1-AID and SMC3-AID cells showed severe cohesion defects that could be rescued by SMC1 and SMC3 , respectively ( Figure 4E and Figure 4—figure supplement 5A ) . smc1-D1164E SMC1-AID cells had only a modest cohesion defect , in line with our previous results for the smc1-D1164E in an otherwise wild type background . In contrast , in the presence of auxin , the smc3-D1161E SMC3-AID cells were severely compromised for cohesion , at levels comparable to cells that only have the SMC3-AID allele ( ~90% separated sisters , Figure 4F and Figure 4—figure supplement 5B ) . Similar results were obtained when cohesion was assessed at the CEN-distal locus ( Figure 4—figure supplement 6 ) . Thus , smc1-D1164E and smc3-D1161E differ in their ability to promote viability and cohesion . To understand the molecular basis for this difference in cohesion , we performed ChIP on this panel of strains to assess cohesin binding to chromosomes under conditions in which the AID-tagged proteins were depleted . The level of cohesin binding to chromosomes in smc1-D1164E SMC1-AID cells was reduced about twofold compared to SMC1 SMC1-AID ( Figure 5A and Figure 5—figure supplement 1 ) . In contrast , cohesin binding to chromosomes in smc3-D1161E SMC3-AID cells was essentially at the background levels observed in SMC3-AID-only cells ( Figure 5B and Figure 5—figure supplement 1 ) . Taken together , these data suggest that the Smc1 ATPase active site regulates cohesin ATPase to modulate cohesin’s DNA binding , whereas the Smc3 active site regulates cohesin ATPase to modulate cohesin-mediated DNA tethering after its stable binding to DNA . 10 . 7554/eLife . 11315 . 023Figure 5 . D-to-E mutations in Smc1 and Smc3 ATPase active sites uncouple the level of ATPase activity from chromosome binding . ( A , B ) ChIP of Mcd1 in M phase cells depleted for AID tagged proteins . G1 cells were depleted for AID tagged proteins then released under depletion condition and arrested in M phase as described in Figure 4D . M phase cells were fixed and processed for ChIP using Mcd1 antibodies , and the% Mcd1 binding plotted as described in Figure 3—figure supplement 4 . ( A ) ChIP of Mcd1 in smc1-D1164E cells at centromere-proximal CARC1 ( top panel ) and centromere-distal CARL1 ( bottom panel ) . Haploid M phase cells from Figure 4B , expressing Smc13FLAG Smc1AID ( SMC1; light grey line , light grey squares ) , Smc1-D1164E3FLAG Smc1AID ( smc1-D1164E SMC1-AID; red line , red circles ) , and Smc1AID alone ( SMC1-AID; black line , open triangles ) were used for ChIP under conditions in which AID-tagged proteins were degraded . ( B ) ChIP of Mcd1 in M phase in smc3-D1161E cells at centromere-proximal CARC1 ( top panel ) and centromere-distal CARL1 ( bottom panel ) . Haploid M phase cells from Figure 4C expressing Smc36HA Smc3AID ( SMC3; light grey line , light grey squares ) , Smc3-D1161E6HA Smc3AID ( smc3-D1161E SMC3-AID; orange line , orange circles ) , and Smc3AID alone ( SMC3-AID; black line , open triangles ) were used for ChIP under conditions in which AID-tagged proteins were degraded . ( C ) ATPase activity of purified S . pombe cohesin bearing D-loop mutations Psm1-D1167E or Psm3-D1132E , analogous to smc1-D1164E or smc3-D1161E , respectively . Same amount of cohesin was used in ATPase experiments ( lower panel ) . Cohesin complexes were purified from cells overexpressing wild type S . pombe cohesin ( WT ) or S . pombe cohesin with mutations analogous with Smc3 ATPase active site D-loop-E mutation ( SpSmc1DE ) or Smc1 ATPase active site D-loop mutation ( SpSmc3DE ) . ATPase assays were carried out in ATPase buffer 1 for 2 hours at 30°C . Cohesin with a K-to-I mutation in the Walker A motif of Smc3 ATPase active site ( SpSmc3WA , Psm3 K38I in S . pombe ) abolished most , if not all , ATPase activity . Coomassie-stained protein bands were spliced from the same gel for representation purposes . Please see figure supplement 1 for further characterization of chromosomal association of cohesin in smc1-D1164E or smc3-D1161E . Figure supplement 2 shows ATPase activity of Rad50 protein when the D-loop residue is mutated to an E or an A . DOI: http://dx . doi . org/10 . 7554/eLife . 11315 . 02310 . 7554/eLife . 11315 . 024Figure 5—figure supplement 1 . Chromosomal binding of cohesin in smc1-D1164E and smc3-D1161E mutants assayed by antibodies against tagged Smc1 and Smc3 subunits . ( A ) ChIP of 3Flag tagged Smc1 in M phase cells depleted for Smc1-AID . Cells from Figure 5A were processed for ChIP using anti-FLAG antibodies and% Flag tagged Smc1 binding plotted as described for Mcd1 in Suppl . Figure 2F . ChIP of Smc13FLAG and smc1-D1164E3FLAG at chr . III centromere-proximal CARC1 ( top panel ) and chr . XII centromere-distal CARL1 ( bottom panel ) . Smc13FLAG Smc1AID ( SMC1; light grey line , open squares ) , Smc1-D1164E3FLAG Smc1AID ( smc1-D1164E SMC1-AID; dark grey line , grey circles ) , and Smc1AID alone ( SMC1-AID; black line , open triangles ) . ( B ) ChIP of 6HA tagged Smc3 in M phase cells depleted for Smc3-AID . Cells from Figure 5B were processed for ChIP using anti-HA antibodies and% HA tagged Smc3 binding plotted as described for Mcd1 in Figure 3—figure supplement 4 . ChIP of Smc36HA and smc3-D1161E6HA at chr . III centromere-proximal CARC1 ( top panel ) and chr . XII centromere-distal CARL1 ( bottom panel ) . Smc36HA Smc3AID ( SMC3; light grey line , open squares ) , Smc3-D1161E6HA Smc3AID ( smc3-D1161E SMC3-AID; dark grey line , grey circles ) , and Smc3AID alone ( SMC3-AID; black line , open triangles ) . DOI: http://dx . doi . org/10 . 7554/eLife . 11315 . 02410 . 7554/eLife . 11315 . 025Figure 5—figure supplement 2 . ATPase activity of wild type and D-loop mutant Rad50 homodimer . ATPase assays were carried out with purified Rad50 homodimer from T4 bacteriophage in ATPase buffer 1 for 2 hours at 30°C . DOI: http://dx . doi . org/10 . 7554/eLife . 11315 . 025 Precedent that the D-loop mutations might impact ATPase activity come from studies of the homodimeric SMC protein Rad50 . The mutation of the D-loop aspartate in Rad50 to an alanine ( Rad50DA ) dramatically reduced its ATPase activity ( la Rosa and Nelson , 2011 ) ( Figure 5—figure supplement 2 ) . Furthermore , the substitution of the D-loop aspartate with a glutamate ( Rad50DE ) led to a 3-fold reduction in ATPase activity ( Figure 5—figure supplement 2 ) . To address how the cohesin ATPase activity would be affected by these substitutions , we purified the S . pombe cohesin complex with the analogous mutations ( Murayama and Uhlmann , 2014 ) and assayed the ATPase activity ( a sum of the ATP hydrolysis by Smc1 and Smc3 active sites ) for equivalent amounts of wild type and mutant cohesin complexes . The mutation in the S . pombe homolog of cohesin analogous to smc1-D1164E ( SpSmc1DE ) , which was competent for cohesion and viability in S . cerevisiae , reduced the ATPase activity of cohesin to levels close to the Smc3 walker A mutant ( SpSmc3WA , Figure 5C ) . In contrast , the D-loop-E mutation in the Smc1 ATPase active site ( SpSmc3DE ) , analogous to the smc3-D1161E allele that abolished chromosomal association of cohesin , led to a subtler reduction in cohesin’s total ATPase activity . These results suggested that the substitution of the D-loop aspartate with a glutamate in Smc3 and Smc1 ATPase active sites affect cohesin in unique ways that uncouple the level of ATPase activity from DNA tethering and DNA binding . The novel and distinct phenotypes of smc1-D1164E and smc3-D1161E might result from a subtle change that alters rather than abrogates D-loop function , given the chemical similarity of aspartate and glutamate . To test whether these unusual phenotypes persisted with more radical substitutions of the D-loop aspartate , we changed it to alanine in the Smc1 or Smc3 ATPase active sites . We introduced these smc1-D1164A or smc3-D1161A alleles into strains bearing SMC1-AID or SMC3-AID alleles , respectively ( Figure 6A ) . We then characterized these alleles under conditions in which the AID-tagged proteins were degraded ( Figure 6—figure supplement 1 ) . Neither smc1-D1164A nor smc3-D1161A allele was able to sustain viability on auxin plates ( Figures 6B , C ) . Similar results in viability were seen using plasmid shuffle assays ( Figure 6—figure supplement 2 ) . Moreover , the smc1-D1164A or smc3-D1161A cells were both severely compromised for cohesion , although the cohesion defect in smc3-D1161A was more severe ( Figures 6D , E , Figure 6—figure supplements 3 and 4 ) . ChIP using anti-Mcd1 antibodies showed that smc1-D1164A had 7-10 fold less cohesin binding than wild type , whereas smc3-D1161A reduced cohesin binding to chromosomes to background levels seen with the SMC3-AID alone ( Figures 7A , B ) . Similar results were obtained when tagged smc1-D1164A and smc3-D1161A proteins were assessed by ChIP ( Figure 7—figure supplement 1 ) . Therefore , the glutamate substitution in the D-loop of the Smc3 ATPase active site is unique in its ability to support cohesion suggesting it alters rather than abolishes the in vivo function of the Smc3 ATPase active site . 10 . 7554/eLife . 11315 . 026Figure 6 . D-loop-A ( DA ) mutations perturb cohesin function more severely than D-loop D to E ( DE ) mutations ( A ) Cartoon representation of the D-to-A substitution mutants in the Smc3 ATPase active site ( smc1-D1164A ) and Smc1 ATPase active site ( smc3-D1161A ) . ( B , C ) Assessing whether the DA D-loop mutants can support viability . ( B ) Smc3 ATPase active site D-loop-A mutant , smc1-D1164A , failed to promote viability . SMC1-AID smc1-D1164A ( VG3766-3C ) cells were grown and dilution plated as described in Figure 4B . SMC1-AID SMC1 ( VG3764-3A ) , SMC1-AID ( VG3311-5D ) cells were re-plated here for comparison . The smc1-D1164A SMC1-AID was moved from a different region of this same plate for clarity of presentation . ( C ) Smc1 ATPase active site D-loop-A mutant , smc3-D1161A , failed to promote viability . SMC3-AID smc3-D1161A ( VG3772-13A ) cells were grown and dilution plated . SMC3-AID SMC3 ( VG3726-6A ) and SMC3-AID ( VG3711-5D ) cells were re-plated here for comparison . The smc3-D1161A SMC3-AID was moved from a different region of this same plate for clarity of presentation . ( D , E ) Cohesion loss at the CEN-proximal locus ( TRP1 ) in M phase cells depleted for AID tagged proteins from G1 through M phase as described in Figure 4D . The percentage of cells with two GFP spots ( sister separation ) was plotted . ( D ) Cohesion loss in smc1-D1164A cells at the CEN-proximal TRP1 locus . Haploid strains SMC1 SMC1-AID ( VG3794-2E ) , SMC1-AID ( VG3711-5D ) and smc1-D1164A SMC1-AID ( VG3796-1F ) assayed for cohesion . ( E ) Cohesion loss in smc3-D1161A cells at the CEN-proximal TRP1 locus . Haploid strains SMC3 SMC3-AID ( VG3797-1A ) , SMC3-AID ( VG3651-3D ) smc3-D1161A SMC3-AID ( VG3798-2B ) assayed for cohesion . Note: smc1-D1164A SMC1-AID and smc3-D1161A SMC3-AID cells were analyzed in the same experiments as Figure 4D , E , respectively . smc-DA data was omitted from Figure 4 but presented here with controls from those experiments for clarity of presentation . DNA content analysis of cells in Figure 6D , E is shown in Figure 6—figure supplement 3 . Please see Figure supplements 1–6 for further characterization of smc1-D1164A and smc3-D1161A alleles . DOI: http://dx . doi . org/10 . 7554/eLife . 11315 . 02610 . 7554/eLife . 11315 . 027Figure 6—figure supplement 1 . Assessing SMC-AID depletion and Mcd1 levels after depletion in DA strains . ( A ) Western blotting of extracts from cells expressing Smc1AID-3V5 in G1 ( -auxin ) and M ( +auxin ) , as in Figure 4—figure supplement 2 . Cells expressing Smc1AID-3V5 Smc1D1164A-3FLAG ( VG3566-3C ) had Mcd1 in M phase after being treated with auxin to deplete Smc1AID-3V5 , indicating that cohesin complex assembly is not perturbed . ( B ) Western blotting of extracts from cells expressing Smc3AID-3V5 in G1 ( -auxin ) and M ( +auxin ) . Cells expressing Smc3AID-3V5 Smc3D1161A-6HA ( VG3731-5B ) that were treated with auxin to deplete Smc3AID-3V5 had Mcd1 in M phase , indicating that cohesin complex assembly is not perturbed . DOI: http://dx . doi . org/10 . 7554/eLife . 11315 . 02710 . 7554/eLife . 11315 . 028Figure 6—figure supplement 2 . Characterization of smc1-D1164A and smc3-D1161A mutants . ( A ) Plasmid shuffle assay showing that smc1-D1164A is severely compromised for viability as the sole SMC1 source . Haploid SMC1 shuffle strain ( VG3568-8A ) alone or with an integrated copy of either SMC1 or smc1-D1164A were grown and plated on YPD or FOA plates to assess viability in the absence of the covering plasmid ( SMC1 URA3 CEN ) as described in Figure 4—figure supplement 1 . ( B ) Plasmid shuffle assays showing that smc3-D1161A is inviable as sole source of Smc3 . Haploid SMC3 shuffle strain ( VG3464-16C ) alone or with an integrated copy of either SMC3 or smc3-D1161A were grown and plated on YPD or FOA plates to assess viability in the absence of the covering plasmid ( SMC3 URA3 CEN ) as described in Figure 4—figure supplement 3 . DOI: http://dx . doi . org/10 . 7554/eLife . 11315 . 02810 . 7554/eLife . 11315 . 029Figure 6—figure supplement 3 . DNA content analysis of cells in Figure 6D ( A ) and Figure 6E ( B ) . DOI: http://dx . doi . org/10 . 7554/eLife . 11315 . 02910 . 7554/eLife . 11315 . 030Figure 6—figure supplement 4 . Cohesion of smc1-D1164A and smc3-D1161A at the CEN-distal locus . ( A ) smc1-D1164A cells arrested in M phase + auxin have a large cohesion defect at the CEN-distal LYS4 locus . Haploid strains from Figure 6B were depleted from Smc1AID-3V5 with auxin , released from G1 and arrested in M phase + auxin as described in Figure 4D . The percentage of cells with 2-GFP signals ( sister separation ) is plotted . ( B ) smc3-D1161A cells arrested in M phase + auxin have no cohesion at the CEN-distal LYS4 locus . Haploid strains from Figure 6C were assayed for cohesion as described . Note: smc1-D1164A SMC1-AID and smc3-D1161A SMC3-AID cells were analyzed in the same experiments as Figure 4—figure supplement 5 . smc-DA data was omitted earlier but presented here with controls from those experiments for clarity of presentation . DOI: http://dx . doi . org/10 . 7554/eLife . 11315 . 03010 . 7554/eLife . 11315 . 031Figure 7 . DA mutations in Smc1 and Smc3 ATPase active sites abolish chromosome binding and ATPase activity . ( A ) ChIP of Mcd1 in M phase smc1-D1164A cells grown in auxin-containing media . Haploid M phase cells from Figure 6B were fixed and processed for ChIP using Mcd1 antibodies . % Mcd1 binding plotted as described in Figure 3—figure supplement 4 . Mcd1 ChIP at centromere-proximal CARC1 ( top panel ) and centromere-distal CARL1 ( bottom panel ) . ( B ) ChIP of Mcd1 in M phase smc3-D1161A cells grown in auxin-containing media . Haploid M phase cells from Figure 6C were fixed and processed for ChIP . Mcd1 ChIP at centromere-proximal CARC1 ( top panel ) and centromere-distal CARL1 ( bottom panel ) . ( C ) ATPase activity of purified cohesin complexes from S . pombe bearing D-loop-A mutations in Smc3 and Smc1 ATPase sctive sites . Psm1 D1167A ( SpSmc1DA ) and Psm3 D1132A ( SpSmc1DA ) , analogous to smc1-D1164A and smc3-D1161A , respectively , were purified from overexpression strains listed in Supplementary file 1 . Same amount of cohesin was used in the ATPase experiments ( lower panel ) . ATPase assays were carried out in ATPase buffer 1 for 2 hours at 30°C . ATPase activity of wild type ( WT ) and Walker A-mutant ( SpSmc3WA ) cohesin was represented here for comparison . Bands were spliced from the same gel for representation purposes . ( D ) Model for how the two ATPase active sites regulate cohesin function . Free cohesin is converted to a stable DNA-bound form by the action of the loader complex ( not shown ) and ATP binding/hydrolysis by ATPase active sites . A particular unknown nucleotide state at the Smc3 ATPase active site induces the tethering ( cohesive ) form . Upon finding the correct sister to be paired with , the Eco1-mediated acetylation of Smc3 leads to the stabilization of this cohesive state . Absent this stabilization , the Smc3 ATPase active site destabilizes the tethering form or induces cohesin dissociation from chromosomes . Wpl1 could either promote this Smc3 ATPase active site function or destabilize the cohesin bound to DNA in the non-tethering form . Figure supplement 1 shows cohesin binding to DNA in smc1-D1164E and smc3-D1161E strains . DOI: http://dx . doi . org/10 . 7554/eLife . 11315 . 03110 . 7554/eLife . 11315 . 032Figure 7—figure supplement 1 . Chromosomal binding of smc1-D1164A and smc3-D1161A mutants assayed with antibodies against tagged Smc proteins . ( A ) ChIP of Smc13FLAG in M phase smc1-D1164A cells at centromere-proximal CARC1 and centromere-distal CARL1 . Haploid M phase cells from Figure 6B were fixed and processed for ChIP using anti-FLAG antibodies . ( B ) ChIP of Smc36HA in M phase smc3-D1161A cells at centromere-proximal CARC1 and centromere-distal CARL1 . Haploid M phase cells from Figure 6C were fixed and processed for ChIP using anti-HA antibodies . DOI: http://dx . doi . org/10 . 7554/eLife . 11315 . 032 Finally , we asked whether analogous aspartate-to-alanine substitution mutations altered the ATPase activity of purified S . pombe cohesin . Both mutations ( SpSmc1DA and SpSmc3DA ) reduced the ATPase activity of cohesin to levels comparable to the walker A mutation ( SpSmc3WA , Figure 7C ) . The similar severe defects in ATPase activities for the S . pombe Smc1DE and Smc1DA cohesin complexes was striking given the dramatic differences in the analogous complexes to promote viability , cohesion and cohesin binding to DNA in S . cerevisiae ( Figures 4–7 ) . This functional difference reinforces the conclusion that the cohesin’s DNA binding and DNA tethering activities can be uncoupled from the level of cohesin ATPase activity .
Until this study , there was no evidence for a role for the ATPase regulating post-DNA binding steps . Here , we show that cohesin retains its ATPase activity after stably binding to DNA in vitro , implying ATP hydrolysis plays an additional role ( s ) after cohesin binds DNA . We also show that cohesion activator mutations in key residues of the Smc3 ATPase active site , smc1-D1164E and smc1-Y1128C , suppress the severe cohesion defect of eco1∆ wpl1∆ cells , similar to the previously described smc3-D1189H allele ( Guacci et al . , 2015 ) . These cohesion activators restore cohesion in eco1∆ wpl1∆ cells without increasing the amount of cohesin stably bound to DNA ( this study , Guacci et al . , 2015 ) . Lastly , a mutation analogous to one of these activator mutations , smc1-D1164E , abolishes most , if not all , cohesin ATPase activity when introduced in purified S . pombe cohesin complex . Taken together , our results suggest the robust ATPase activity of DNA-bound cohesin is physiologically relevant , likely acting as an inhibitor of the conversion of stably bound cohesin to a form capable of tethering sister chromatids . We also show that the cohesion activator phenotype of the smc1-D1164E allele is a unique feature of the D-loop of the Smc3 ATPase active site . The analogous substitution in the D-loop of the Smc1 ATPase active site , smc3-D1161E , not only fails to act as a cohesion activator but also is unable to support viability , fails to load cohesin to DNA , and cannot generate cohesion . These results suggest that the Smc1 and Smc3 ATPase active sites differentially affect cohesin ATPase function and that the Smc3 ATPase active site has a distinct function in the generation of cohesion after stable DNA binding of cohesin . This distinct function provides an explanation for the evolutionary diversion of the SMC subunits . Indeed , use of asymmetrical ATPases for regulation purposes is common to ABC transporters ( Antony and Hingorani , 2004; Furman et al . , 2013; Yang et al . , 2003 ) . We suspect the differential roles of cohesin ATPase active sites and a specialized role for the Smc3 active site in DNA tethering was missed previously , in part due to their composite nature and the particular mutations that have been analyzed to date . How the D-loop of the Smc3 ATPase active site executes its specialized function in regulating the cohesin ATPase remains unclear . The D-loop may impact the nucleotide state only in the Smc3 ATPase active site . Alternatively , it could alter the nucleotide state of both active sites . Support for the latter comes from studies on the ABC ATPase domains in ABC transporters , which suggest that ATP hydrolysis by the two sites is cooperative ( Holland and Blight , 1999 ) , and that D-loops are involved in mediating the communicaton between the two active sites within the ATPase ( Furman et al . , 2013; Hohl et al . , 2012 , 2014 ) . The phenotypes of the D-loop mutants in the Smc1 and Smc3 active sites did not correlate with the ATPase activities of the corresponding purified S . pombe cohesin complexes . For example , cohesin complexes with aspartate-to-glutamate substitution in the D-loop of the Smc1 active site had significant ATPase activity in vitro , but were severely perturbed for DNA binding in S . cerevisiae . In addition , cohesin complexes with aspartate-to-glutamate substitution in the D-loop of the Smc3 ATPase had low levels of ATPase activity indistinguishable from the walker A mutant , but could promote DNA binding and cohesion in S . cerevisiae . While it is possible that this lack of correlation could be explained by our use of the ATPase activity of the S . pombe wild type and mutant cohesins as an approximation of the S . cerevisiae cohesin , this explanaition seems unlikely . The aspartate of the D-loop is absolutely conserved in all SMC complexes and highly conserved among very diverse ABC ATPase modules . Our analysis of analogous substitutions in the homodimeric Rad50 protein shows similar reductions in the ATPase activity . Finally , many published biochemical and structural studies have used cohesin complexes from different organisms as a proxy for S . cerevisiae cohesin to provide physiologically relevant insights . Intriguingly , the uncoupling of ATPase activity and function by D-loop mutations has been observed for several other ABC ATPases ( Furman et al . , 2013; Grossmann et al . , 2014 ) . Therefore , we favor a model in which cohesin’s functions and conformations could be coordinated by the different states in the hydrolysis cycle of ATPase active sites . For example , the tethering-competent state of cohesin could require that the Smc3 active site be bound with ADP+Pi . The smc1-D1164E mutation may mimic the ADP-Pi state , whereas the smc1-D1164A mutant could be blocked elsewhere in the ATPase cycle ( ATP bound or nucleotide free ) . Both the smc1-D1164E and smc1-D1164A would dramatically reduce overall ATPase activity but only the smc1-D1164E would promote cohesin function . A similar case of trapping the ADP+Pi state is hypothesized for the MRP1 ABC transporter , for a glutamate substitution of an aspartate residue ( Qin and Cao , 2008 ) . Elucidating the exact relationship between the particular nucleotide states of the Smc1 and Smc3 ATPase active sites and cohesin functions will likely require in vitro reconstitution of cohesin’s various functions and detailed structural studies . The phenotypes of smc1-D1164E cells suggest that Eco1 modulation of cohesin’s ATPase impacts two distinct modes of cohesin regulation . First , smc1-D1164E suppresses the inviability of eco1∆ cells , which normally die because they are unable to inhibit Wpl1 ( Rolef Ben-Shahar et al . , 2008; Rowland et al . , 2009; Sutani et al . , 2009 ) . This result suggests that Eco1 acetylation alters the Smc3 ATPase active site to make cohesin resistant to Wpl1 . Second , smc1-D1164E also suppresses the cohesion defect in eco1△ wpl1△ cells , suggesting that it phenocopies the second function of Eco1 acetylation , the activation of DNA-bound cohesin to tether sister chromatids . A clue to how Eco1 modulates cohesin ATPase to promote DNA tethering comes from our in vitro and in vivo characterization of smc1-D1164E . Since smc1-D1164E suppresses the eco1∆ cohesion defect , we propose that smc1-D1164E mimics the Eco1 acetylation of Smc3 by blocking the cohesin ATPase at a particular stage in its hydrolysis cycle , promoting DNA tethering . In the absence of acetylation , continued cycles of ATP hydrolysis prevent cohesin from maintaining the conformation necessary for tethering ( Figure 7D ) . The completion of the ATPase cycle may render cohesin in a conformation capable of being recognized by and acted upon Wpl1 , which leads to cohesin dissociation from DNA . This model predicts that Eco1-mediated acetylation of the Smc3 may have an impact on the ATPase activity of cohesin . However , previous studies on in vitro acetylated cohesin or acetyl-mimic mutants of cohesin suggest that Eco1 acetylation does not impact cohesin’s total ATPase activity ( Ladurner et al . , 2014 ) . This could be due to ( 1 ) the inherent inefficiency of acetylation both in vitro and in vivo , resulting in only a small subset of cohesin being acetylated , ( 2 ) the inability of glutamine substitutions of K112 and K113 in Smc3 to mimic the acetylated state , ( 3 ) the failure to assay cohesin ATPase activity on DNA . Alternatively , Eco1-mediated acetylation may stabilize an intermediate state of cohesin in a way that makes subsequent cycles of ATP hydrolysis unable to alter the tethering conformation . Either way , the result is that acetylation traps cohesin in its tethering form to stabilize cohesion during or soon after establishment . Testing whether Eco1-mediated acetylation directly downregulates the cohesin ATPase cycle when bound to DNA and how the ATPase cycle affects cohesin conformation will require better reconstitution of cohesin function in vitro . Our model of transient conformational changes that must be stabilized provides a plasticity to cohesin-mediated DNA tethering . Cohesin that is bound to a locus could form transient tethers with a number of different DNA loci and at different times . The spatial and temporal control of Eco1 would ensure that only the proper tethers are stabilized , allowing a very tight regulation for cohesion establishment in sister chromatid cohesion . This plasticity of cohesin tethering and its Eco1-dependent regulation may be critical for other aspects of cohesin function , such as condensation , DNA repair and transcriptional regulation .
S . pombe cohesin complex was over-expressed and purified from S . cerevisiae as described ( Murayama and Uhlmann , 2014 ) with minor modifications . Briefly , over-expressing cells were harvested and lysed with a cryomill . For cohesin purifications , 50 mL of cell powder was thawed in the presence of 100 mL CLH300 buffer ( 50 mM HEPES , pH 7 . 5 , 1 mM DTT , 300 mM NaCl , 20% glycerol , and protease inhibitor cocktail from Roche ) and the clarified by centrifugation . The clarified supernatant was bound to 2 mL IgG beads ( Invitrogen ) overnight in the presence of RNase A ( 10 μg/mL ) . The beads were washed 300 mL in H300 ( 300 mM NaCl , 25 mM HEPES pH 7 . 5 , 0 . 5 mM TCEP , 10% glycerol ) . To elute protein off the beads , 5 mL of elution buffer ( H300 with 2 mM MgCl2 , 5 U/mL Prescission protease , 10 μg/mL RNase A ) was added . Eluted protein was diluted to achieve 100 mM NaCl and then bound to heparin column . The column was washed in 4xColumn Volume ( CV ) H100M ( 25 mM HEPES pH 7 . 5 , 0 . 5 mM TCEP , 10% glycerol , 100 mM NaCl and 2 mM MgCl2 ) , 4xCV H300 with 2 mM MgCl2 . Cohesin was eluted from the column at 600 mM NaCl . Cohesin containing fractions were pooled , concentrated and frozen in same buffer containing 200 mM NaCl . Mutants of cohesin were generated by site-directed mutagenesis and over-expression strains were generated as described ( Murayama and Uhlmann , 2014 ) . Mutant proteins were purified using the same protocol as described above . The loader complex from S . pombe was expressed and purified as described ( Murayama and Uhlmann , 2014 ) . Rad50 protein from T4 bacteriophage was purified as described ( la Rosa and Nelson , 2011 ) . CARC1 DNA substrates were prepared as described ( Onn and Koshland , 2011 ) . Briefly , for each binding reaction , 500 ng biotin-labeled DNA was assembled on 20 μL streptavidin-conjugated dynabeads . Beads were washed 6x 30 μL in CL1 buffer ( 35 mM Tris pH 7 . 5 , 1 mM TCEP , 25 mM NaCl , 25 mM KCl , 1 mM MgCl2 , 15% glycerol and 0 . 003% Tween 20 ) and resuspended in 60 μL CL1 . 80 nM cohesin complex and 0 . 5 mM ATP ( with or without 80 nM loader ) was added to beads and incubated at 30°C for 1 hour in a reaction volume of 100 μL . Assembled cohesin-DNA complexes were washed in 100 μL buffer once in CL1 , three times in CL1 with 500 mM KCl , and once more with CL1 . Resuspending beads in 30 μL SDS sample buffer eluted bead-bound cohesin-DNA complexes . Samples were analyzed by SDS-PAGE followed by Coomassie staining or western blotting . Stable cohesin-DNA complexes assembled as described above were resuspended in CL1 buffer in the presence of 2U DNase or 5U Mnl I at 30°C for 30 minutes . Eluted proteins ( supernatant ) and beads were subjected to SDS-PAGE and proteins were visualized by Coomassie staining or Western blotting . For basal ATPase activity of cohesin , 150 nM cohesin was diluted in 100 μL ATPase reaction buffer 1 ( 25 mM HEPES pH 7 . 5 , 100 mM NaCl , 10% glycerol , 1 mM TCEP , 5 mM MgCl2 ) . 10 μL of this sample was run on SDS-PAGE to visualize cohesin used in ATPase experiments . The ATPase reaction was started by the addition of 20 μL of 5x ATP hot mix ( 2 . 5 mM ATP , 1 μL of 10 μCi/ μL ATP-γ-P32 in 96 . 5 μL 5xATP buffer ) to 80 μL cohesin mix . Thus , in our ATPase reactions , the final concentration of cohesin was 120 nM , and ATP was 0 . 5 mM . At appropriate time points , 20 μL of this reaction mix was taken out and mixed with 380 μL of 0 . 25 mg/mL BSA . 300 μL of STOP buffer ( 100 μL KPO4 1 M , 100 μL 1N HCl , 100 μL 20% Norit ) was immediately added to the samples . Samples were then spun at 4°C 10000 rpm for 3 minutes . 500 μL of the supernatant was taken out in a new tube and spun again at 4°C 10000 rpm for 3 minutes . 350 μL of the supernatant of this second spin was counted in 5 mL of scintillation cocktail using a scintillation counter . Reads at time zero were counted as background and subtracted from later time points . To get% ATP hydrolysis , 10 μL of 1:5 diluted 5xATP buffer was counted to represent 100% ATP hydrolysis . For loader- and DNA-stimulated ATPase activity of cohesin , 80 nM cohesin , 80 nM loader , and 2 . 5 μg plasmid DNA containing the sequence for CARC1 ( pIO2 ) were incubated in 50 μL ATPase reaction buffer 2 ( 25 mM HEPES pH 7 . 5 , 25 mM NaCl , 25 mM KCl , 10% glycerol , 1 mM TCEP , 5 mM MgCl2 ) . To measure the activity of stably DNA-bound cohesin complexes ( CD-B ) , 80 nM cohesin , 80 nM loader and 0 . 5 mM ATP was incubated with 100 μL dynabeads coupled to 2 . 5 μg DNA in 500 μL total volume of CL1 at 30°C for 1 hour . Unbound/unstably bound cohesin was washed off using the following washes: 1x500 μL CL1 , 3 x 500 μL in CL1 with 500 mM KCl , then 1x500 μL CL1 . CD-B were then resuspended in 50 μL ATPase buffer 2 . This sample of CD-B contained about% 20 of the initial input cohesin , which was approximately the same amount of cohesin as the other samples in this experiment ( Figure 1E ) , as could be judged by the protein gel . The reaction was started by the addition of 10 μL 5xATP hot mix to 40 μL protein/DNA mix for 2 hours at 30°C and samples were processed as described above . The ATPase activity of T4 Rad50 and mutants was measured in ATPase buffer 1 for 2 hours at 30°C as described above , except 1 μM Rad50 was used . Error bars represent standard error of data from at least two independent experiments . Yeast strains used in this study are A364A background , and their genotypes are listed in Supplementary file 1 . SC minimal and YPD media were prepared as described ( Guacci et al . , 1997 ) . Benomyl ( a gift from Dupont ) and camptothecin ( Sigma ) plates used to assess drug sensitivity were prepared as previously described ( Guacci and Koshland , 2012 ) . Preparation of auxin ( Sigma ) containing media for depletion of AID tagged proteins was as previously described ( Eng et al . , 2014 ) . Cells were grown to saturation in YPD media at 23°C then plated in 10-fold serial dilutions . Cells were incubated on plates at relevant temperatures or containing drugs as described . For plasmid shuffle assays , cells were grown to saturation in YPD media to allow loss of covering plasmid , then plated in 10-fold serial dilutions on YPD or FOA media . Cohesion was monitored using the LacO-LacI system where cells contained a GFP-LacI fusion and tandem LacO repreats integrated at one chromosomal locus , which recruits the GFP-LacI ( Straight et al . , 1996 ) . CEN-distal cohesion was monitored by integrating LacO repeats at LYS4 , located 470 kb from CEN4 . CEN-proximal cohesion is monitored by integrating LacO at TRP1 , located 10 kb from CEN4 . Cells were fixed , and processed to allow the number of GFP signals in each cell to be scored and the percentage of cells with 2-GFP spots determined as previously described ( Guacci and Koshland , 2012 ) . Data is from 2 independent experiments and 200-300 cells scored for each data point in each experiment . Site directed mutagenesis using the Stratagene Quick-change kit was employed to generate all mutants described . Mutations were confirmed by sequencing the entire ORF to ensure it was the only change . Details about the Auxin mediated destruction of AID tagged proteins in yeast was previously described ( Eng et al . , 2014 ) . Briefly , the TIR1 E3-ubiquiting ligase placed under control of the GPD promoter and marked by C . Albicans TRP1 replaced the TRP1 gene on chromosome IV . SMC1 , SMC3 and ECO1 were internally tagged with 3V5-AID2 sequences and transformed into yeast strains bearing TIR1 to replace SMC1 , SMC3 and ECO1 at the endogenous locus . PCR screening and auxin mediated sensitivity were used to identify clones containing AID tagged genes . Genetic screen of eco1∆ wpl1∆ cells for cohesion activator suppressors was done as previously described ( Guacci et al . , 2015 ) . Chromatin Immunoprecipitation ( ChIP ) was performed as previously described ( Eng et al . , 2014; Wahba et al . , 2013 ) . Microscopy . Images were acquired with a Zeiss Axioplan2 microscope ( 100X objective , NA=1 . 40 ) equipped with a Quantix CCD camera ( Photometrics ) . Flow cytometry analysis was performed as previously described ( Eng et al . , 2014 ) . | The bulk of the genetic material in cells from yeast to humans is organized into chromosomes . These chromosomes must be duplicated and the copies need to be segregated every time cells divide . Cohesin is a protein complex that helps to organize the structure of chromosomes by tethering together two regions of DNA , either within a chromosome or between chromosomes . Problems with cohesin have been linked to cancer and birth defects , but it is not clear how cohesin binds DNA and how it makes a tether between two DNA regions . It is also unclear how cohesin’s activity is coordinated with the series of events that allow cells to divide ( known as the cell cycle ) . Cohesin has two active sites that can break down molecules of ATP . Previous research had suggested that these active sites ( called ATPases ) controlled cohesin’s activity by regulating whether or not it could bind to DNA . However , Çamdere et al . now reveal that cohesin’s ATPases do not simply provide an ‘on/off switch’ for DNA binding . The experiments , which involved a combination of genetic , cell biology and biochemical techniques in budding yeast , instead revealed that one of cohesin’s ATPases regulates structural rearrangements in cohesin that is already bound to DNA . These structural rearrangements fine-tune the complex’s ability to tether two regions of DNA . Further experiments then revealed that two cohesin regulators ( namely Eco1 and Wpl1 ) altered this ATPase active site to control cohesin’s DNA tethering and DNA binding activities . These findings provide a molecular explanation for how these regulators control cohesin’s activity to make sure that the chromosomes have the correct structure during cell division . The next challenge is to identify the structural changes in cohesin that are triggered by cohesin’s two ATPases and to understand how these structural changes promote DNA binding followed by DNA tethering . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"biochemistry",
"and",
"chemical",
"biology",
"cell",
"biology"
] | 2015 | The ATPases of cohesin interface with regulators to modulate cohesin-mediated DNA tethering |
Type I neurofibromatosis ( NF1 ) is caused by mutations in the NF1 gene encoding neurofibromin . Neurofibromin exhibits Ras GTPase activating protein ( Ras-GAP ) activity that is thought to mediate cellular functions relevant to disease phenotypes . Loss of murine Nf1 results in embryonic lethality due to heart defects , while mice with monoallelic loss of function mutations or with tissue-specific inactivation have been used to model NF1 . Here , we characterize previously unappreciated phenotypes in Nf1-/- embryos , which are inhibition of hemogenic endothelial specification in the dorsal aorta , enhanced yolk sac hematopoiesis , and exuberant cardiac blood island formation . We show that a missense mutation engineered into the active site of the Ras-GAP domain is sufficient to reproduce ectopic blood island formation , cardiac defects , and overgrowth of neural crest-derived structures seen in Nf1-/-embryos . These findings demonstrate a role for Ras-GAP activity in suppressing the hemogenic potential of the heart and restricting growth of neural crest-derived tissues .
NF1 is a common human disorder characterized by benign and malignant tumors of neural crest origin , pigmentation defects , learning disorders , cardiovascular abnormalities and a wide spectrum of other abnormalities including a predilection for leukemia ( especially juvenile myelomonocytic leukemia , [JMML] ) and vascular defects ( Cichowski and Jacks , 2001; Friedman et al . , 2002 ) . Some of these phenotypes , including JMML and vascular defects , are shared by patients with related disorders associated with activation of the Ras signaling pathway , which together have been termed the 'RAS-opathies' ( Rauen et al . , 2010 ) . Neurofibromin contains a protein domain termed the GAP-related domain ( GRD ) that is homologous to yeast IRA proteins . The NF1 GRD is able to complement yeast IRA mutants and hydrolyze GTP bound to active Ras , thereby down-regulating Ras signaling ( Ballester et al . , 1990; Xu et al . , 1990 ) . Interestingly , however , missense mutations in humans with NF1 have been identified that alter amino acids throughout the protein , suggesting functional domains outside of the GRD ( Mattocks et al . , 2004 ) . Additional cellular functions for neurofibromin have also been identified , including modulation of protein kinase A ( PKA ) and cyclic adenosine monophosphate ( cAMP ) pathways ( Brown et al . , 2010; The et al . , 1997; Wolman et al . , 2014 ) . A C-terminal region of neurofibromin has also been shown to interact with a major class of heparan sulfate proteoglycans ( Hsueh et al . , 2001 ) while full-length neurofibromin can bind to the scaffolding domain of caveolin-1 ( Boyanapalli et al . , 2006 ) . Therapeutic strategies for the treatment of NF1 have focused on modulation of the Ras pathway , but the degree to which Ras dysregulation accounts for the diverse aspects of the human disease , or for the equally diverse features of various animal models of NF1 , remains a critical question in the field . Mouse models of NF1 have demonstrated critical developmental functions for neurofibromin in multiple tissues , including neural crest , endothelium , and hematopoietic stem and progenitor cells ( HSPCs ) ( Brossier and Carroll , 2012; Costa and Silva , 2003; Gitler et al . , 2003; Bollag et al . , 1996; Zhang et al . , 1998 ) . HSPCs arise during midgestation from a transient population of endothelial cells called hemogenic endothelium ( HE ) located in the yolk sac , the dorsal aorta , vitelline and umbilical arteries ( Bertrand et al . , 2010; Boisset et al . , 2010; Chen et al . , 2009; Kissa and Herbomel , 2010; Lam et al . , 2010; Oberlin et al . , 2010; Zovein et al . , 2008 ) . HE gives rise to HSPCs through a direct transition of endothelial cells into hematopoietic cells , independent of cell division ( Kissa and Herbomel , 2010; Eilken et al . , 2009 ) . This endothelial to hematopoietic transition ( EHT ) was thought to occur only in the major arteries of the embryo , the placenta , and the yolk sac , but recent studies have identified the heart and the head as sites of de novo hematopoiesis ( Dzierzak and Speck , 2008; Nakano et al . , 2013; Li et al . , 2012 ) . In the heart , hemogenic endocardial cells are integrated into the outflow cushion and atria and undergo EHT on embryonic day ( E ) 9 . 5 . Unlike arterial HE cells that give rise to the full repertoire of hematopoietic cells , hemogenic endocardial cells produce a transient population of hematopoietic cells restricted to the erythroid/myeloid lineage , similar in potential to an early wave of erythroid/myeloid progenitors ( EMPs ) that emerge starting at E8 . 5 in the yolk sac ( Nakano et al . , 2013; Palis et al . , 1999 ) . Later in gestation , the heart is associated with a less-defined second wave of hematopoiesis characterized by aggregates of endothelial and hematopoietic cells called blood islands . Cardiac blood island formation is a prevalent physiological process that has been identified in embryonic mice , chicks , quails and humans , but surprisingly , little is known about the formation of these structures ( Hiruma and Hirakow , 1989; Hutchins et al . , 1988; Kattan et al . , 2004; Ratajska et al . , 2006; Red-Horse et al . , 2010; Wu et al . , 2013; Jankowska-Steifer et al . , 2015 ) . What is known about cardiac blood islands comes primarily from histological studies . Blood islands form in the subepicardial space near the interventricular sulci between E11 and E14 and consist primarily of erythroblasts , but have also been associated with megakaryocytes , platelets , and leukocytes ( Ratajska et al . , 2009; Red-Horse et al . , 2010 ) . Clonal and histological analysis suggests that blood islands emerge from the endocardium , protruding into the myocardium where they pinch off , forming blood-filled spheres or tubules that join the coronary plexus ( Red-Horse et al . , 2010; Jankowska-Steifer et al . , 2015 ) . It has been suggested that hematopoietic cells enter cardiac blood islands by diapedesis , but other routes such as circulation or the de novo generation of hematopoietic cells from the endocardium in situ have not been ruled out ( Ratajska et al . , 2006 ) . Cardiac blood island formation was found to be more robust in Tbx18 null mouse embryos , and thought to be an indirect consequence of aberrant signaling ( Wu et al . , 2013 ) . Here we show that hyperactive Ras signaling increases cardiac blood island formation , and that endocardial cells of the blood islands have functional characteristics of HE and express Runx1 , a marker of HE .
E 12 . 5–13 . 5 Nf1-deficient fetuses were reported to have increased numbers of committed hematopoietic progenitors in the fetal liver ( Largaespada et al . , 1996; Bollag et al . , 1996; Zhang et al . , 1998 ) . Since many fetal liver progenitors in the midgestation embryo originate in the yolk sac ( Lux et al . , 2008 ) , we examined the number of EMPs in the yolk sac of E10 . 5 Nf1-deficient embryos . Nf1-/- yolk sacs ( Figure 1A ) contained significantly more EMPs , specifically due to an increased number of erythroid progenitors ( Figure 1B ) , suggesting that Ras signaling positively regulates EMP numbers . We next examined the impact of Nf1 deficiency on hematopoiesis in the major arteries . The majority of HE cells in the major arteries ( dorsal aorta , vitelline and umbilical ) undergo EHT between E9 . 0-–10 . 5 , resulting in the formation of Kit+ CD31+ Runx1+ hematopoietic cells that remain briefly attached as clusters to the luminal wall of the arteries . In contrast to the increase in EMPs observed in the yolk sac , CD31+ Kit+ Runx1+ hematopoietic cluster cells were decreased in the dorsal aortas of E10 . 5 Nf1-deficient embryos ( Figure 1C , D ) . The decrease in CD31+ Kit+ Runx1+ hematopoietic cluster cells appears to be due to decreased de novo generation , as fewer Runx1+ CD31+ Kit--/low HE cells were present in the dorsal aortas ( Figure 1C , D ) . These data suggest that disruption of neurofibromin function augments the formation of EMPs in the yolk sac , but inhibits the specification of hemogenic endothelium in the dorsal aorta . 10 . 7554/eLife . 07780 . 003Figure 1 . Nf1 deficiency increases yolk sac hematopoiesis but decreases specification of hemogenic endothelium in the dorsal aorta at E10 . 5 . ( A ) Quantification of erythroid and myeloid progenitors ( EMPs ) in the yolk sacs of E10 . 5 Nf1+/+ , Nf1+/- and Nf1-/- conceptuses ( Nf1+/+ n = 7; Nf1+/- n = 8; Nf1-/- n = 3 ) . One-way ANOVA and Bonferroni’s multiple comparison test was applied to determine significance , error bars represent the standard deviation ( SD ) ( B ) Percent of EMP colony type . Mk: megakaryocyte; Mix: granulocyte-erythroid-monocyte-megokaryocyte; BFU-E: burst forming unit-erythroid; G/M: granulocyte-macrophage colonies . There were significantly more BFU-E progenitors in the yolk sacs of Nf1-/- compared to Nf1+/+ and Nf1+/- littermates , p≤0 . 0001 . ( C ) Confocal Z-projections ( Z intervals = 2 μm ) of Nf1+/+ , Nf1+/- and Nf1-/- dorsal aortas at E10 . 5 , immunostained for CD31 ( red ) Runx1 ( green ) and Kit ( cyan ) . Scale bars = 100 μm . Aortas are oriented with the ventral aspect on the left . ( D ) Quantification of CD31+ Runx1+ Kit+ hematopoietic cluster cells and CD31+ Runx1+ Kit-/low hemogenic endothelial cells within the dorsal aorta at E10 . 5 , One-way ANOVA and Bonferroni’s multiple comparison test applied to determine significance , error bars represent the SD and n = 3 for all genotypes . ** indicates that p≤0 . 01 . DOI: http://dx . doi . org/10 . 7554/eLife . 07780 . 003 Nf1 deficiency results in embryonic lethality by midgestation ( approximately E13 ) due to cardiac defects . These defects include enlarged endocardial cushions and a malformed outflow tract ( Brannan et al . , 1994; Jacks et al . , 1994; Lakkis and Epstein , 1998 ) . Despite midgestation lethality , E11 . 5 Nf1-/- embryos appeared grossly normal ( Figure 2A ) . However , blood -filled protrusions were often visible on the ventricles of Nf1-/- embryos ( Figure 2B arrowheads ) . Whole-mount confocal analyses revealed that the protrusions are blood island-like structures budding from the ventricular endocardium , as they express CD31 and the hematopoietic markers CD41 and Runx1 ( Figure 2C arrowheads ) . The blood islands are concentrated laterally on both ventricles of Nf1-/- embryos ( Figure 2C ) , in contrast to wild-type embryos , in which it was reported that blood islands are generally located on the dorsal surface in the interventricular sulcus ( Jankowska-Steifer et al . , 2015 ) . Cardiac blood island formation was more robust in Nf1-deficient embryos compared to Nf1+/+ and Nf1+/- littermates; an average of 63 . 7 ± 7 . 6 blood islands could be identified via confocal microscopy on the ventricles of E11 . 5 Nf1-/- embryos , whereas Nf1+/+ and Nf1+/- littermates averaged 0 . 3 ± 0 . 6 and 1 . 2 ± 1 . 3 blood islands , respectively ( Figure 2D ) . At E10 . 5 , only 25% ( 1/4 ) of Nf1-deficient embryos displayed small budding cardiac blood islands , whereas 92% ( 11/12 ) of E11 . 5 embryos had robust blood island formation , indicating that cardiac blood islands arise between E10 . 5 and 11 . 5 in Nf1-/- embryos . 10 . 7554/eLife . 07780 . 004Figure 2 . Ectopic formation of cardiac blood islands in Nf1-/- embryos . ( A ) Gross view of E11 . 5 Nf1+/+ , Nf1+/- and Nf1-/- embryos ( B ) Isolated hearts from embryos in ( A ) Black arrowheads point to two examples of blood-filled protrusions . ( C ) Confocal Z-projections ( Z interval = 5 μm ) of CD31 ( red ) , CD41 ( cyan ) and Runx1 ( green ) immunostained Nf1+/+ , Nf1+/- and Nf1-/- E11 . 5 embryos . Blood island-like structures ( arrowheads ) are visible on the ventricle of the Nf1-/- embryo . Scale bars = 500 μm . ( D ) Quantification of blood islands on the ventricles of E11 . 5 embryos , One-way ANOVA and Bonferroni’s multiple comparison test applied to determine significance , error bars represent SD . ( E ) Number of erythroid and myeloid progenitors per flushed E11 . 5 ventricles . One-way ANOVA and Bonferroni’s multiple comparison test applied to determine significance , error bars represent the standard SD . Nf1+/+ n = 10 , Nf1+/- n = 33 , and Nf1-/- n = 6 . *** indicates that p≤0 . 001 . CFU-C: colony-forming units-culture; V: ventricle; A: atrium; FL: fetal liver; Mk: megakaryocyte; Mix: granulocyte-erythroid-monocyte-megakaryocyte; BFU-E: burst forming unit-erythroid; G/M: granulocyte-macrophage colonies . DOI: http://dx . doi . org/10 . 7554/eLife . 07780 . 004 To determine if the ectopic cardiac blood islands harbored functional hematopoietic progenitors we performed colony-forming assays . To eliminate circulating blood cells , the atrium was removed and circulating blood flushed from the ventricles before the ventricles were dissociated and plated in methylcellulose supplemented with cytokines . Nf1-/- ventricles contained significantly more EMPs than their Nf1+/+ and Nf1+/- littermates ( Figure 2E ) . This suggests that the phenotypic hematopoietic cells in the blood islands are functional erythroid and myeloid progenitors . We used CD31 , Runx1 , Kit and CD41 whole-mount immunofluorescence and confocal microscopy to examine the structure of cardiac blood islands in Nf1-deficient ventricles at E11 . 5 . Single optical projections through the blood islands indicate that they are cystic structures that consist of a layer of CD31+ endocardial cells that is continuous with the endocardium lining the ventricular trabeculae ( Figure 3A , C ) . A layer of 3–4 CD31 bright cells with morphology between a flat endocardial cell and a rounded hematopoietic cell lined the base of most blood islands ( Figure 3A–D ) . Some of these cells express the hemogenic endothelial marker Runx1 but they do not express high levels of the early hematopoietic markers CD41 and Kit , suggesting that they are hemogenic endocardial cells that have not yet initiated the transition into hematopoietic cells ( Figure 3B , D , arrows ) . Within the cystic structure of most blood islands , there are also rounded cells that are CD31+ Kit+ Runx1+ or CD31+ CD41+ Runx1+; these cells are phenotypic and morphological HSPCs ( Figure 3B , D , arrowheads ) . These data suggest that blood islands are derived from the endocardium of the ventricle and that the endocardium of blood islands has a latent HE potential that is held in check by Ras-GAP activity . 10 . 7554/eLife . 07780 . 005Figure 3 . Nf1-/- cardiac blood islands . ( A ) Single optical projection through the ventricle of an E11 . 5 Nf1-/- embryo immunostained for CD31 ( red ) , CD41 ( cyan ) and Runx1 ( green ) . Blood islands ( arrowheads ) are visible sprouting from the ventricles of Nf1-/- embryos . ( B ) Single optical projection through blood islands on the ventricles of E11 . 5 Nf1-/- embryos . Runx1+ endocardial cells are visible in the blood islands ( arrows ) . Arrowheads indicate examples of CD31+ CD41+ Runx1+ hematopoietic cells . ( C ) Single optical projection through the ventricle of an E11 . 5 Nf1-/- embryo immunostained for CD31 ( red ) , Kit ( cyan ) and Runx1 ( green ) . ( D ) Single optical projection through blood islands . Arrows indicate Runx1+ endocardial cells . Arrowheads indicate examples of CD31+Kit+ Runx1+ hematopoietic cells . Scale bars = 50 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 07780 . 005 In addition to robust cardiac blood island formation , E11 . 5 Nf1-/- embryos have enlarged fetal livers populated by Runx1+ and CD41+ hematopoietic cells ( Figure 2C ) , consistent with previous studies that found significantly higher numbers of fetal liver clonogenic progenitors ( Zhang et al . , 1998; Largaespada et al . , 1996; Bollag et al . , 1996 ) . Furthermore , competitive repopulation assays comparing Sca1+lin–/dim cells isolated from the fetal livers of Nf1-/- and Nf1+/+ embryos demonstrated that Nf1-/- cells have a growth advantage , particularly in the myeloid compartment ( Bollag et al . , 1996 ) . Thus , the enlargement of the fetal liver may be due to elevated proliferation of Nf1-/- hematopoietic cells . In order to determine if the increase in cardiac blood islands seen in Nf1-/- embryos is due specifically to loss of the Ras-GAP activity of neurofibromin , we engineered a missense mutation within the GRD . Arginine 1276 has been shown to be the 'arginine finger' of the GRD and is critical for catalytic activity . Mutation of this residue to proline was identified in a family with NF1 , and crystal structures of related GAP domains were consistent with empiric studies showing loss of GAP activity following R1276P mutagenesis ( Ahmadian et al . , 1997; Scheffzek et al . , 1997; Klose et al . , 1998; Hiatt et al . , 2004 ) . We generated 'knockin' mice in which arginine 1276 was mutated to proline ( R1276P ) and designated these mice Nf1GRD/+ ( Figure 4—figure supplement 1 ) . We generated an additional line of engineered mice in order to control for minor changes to intronic genomic sequences necessitated by the gene targeting and selection strategy ( see Materials and methods and Figure 4—figure supplement 1 ) . For these control mice , designated Nf1GRDCTL/+ , we utilized the identical targeting strategy but arginine 1276 was left intact . Appropriate targeting in several ES cell clones for each of the Nf1GRD or Nf1GRDCTL constructs was demonstrated by Southern blotting ( Figure 4—figure supplement 1 ) . These were used to generate chimeric animals that were then bred for germ line transmission . Heterozygous Nf1GRD/+ mice appeared normal and were able to breed , but heterozygous intercrosses failed to produce any viable homozygous Nf1GRD/GRD pups ( Table 1 ) . One out of 61 embryos genotyped at E12 . 5 was Nf1GRD/GRD , and 11of 63 ( 17 . 5% ) were Nf1GRD/GRD at E11 . 5 ( Table 1 ) . Hence , homozygous R1276P mutation of Nf1 causes midgestation embryonic lethality with most embryos succumbing by E12 . 5 . 10 . 7554/eLife . 07780 . 006Table 1 . Genotypes from Nf1GRD/+ X Nf1GRD/+ intercrosses . DOI: http://dx . doi . org/10 . 7554/eLife . 07780 . 006AgeTotal+/+Nf1GRD/+Nf1GRD/GRDE11 . 563133911E12 . 56119411P06223390 Total cell lysates from Nf1GRD/+ and Nf1GRD/GRD embryos exhibited similar levels of neurofibromin protein of expected apparent molecular weight of 250–280 kDa ( Figure 4A ) . The relative neurofibromin protein expression was similar to that of wild-type embryos and was increased relative to Nf1+/- embryos ( Figure 4B ) . 10 . 7554/eLife . 07780 . 007Figure 4 . Neurofibromin protein expression and activity from the Nf1 alleles . ( A ) Total cell lysates from E10 . 5 Nf1+/+ , Nf1+/- , Nf1-/- , Nf1GRD/+ , and Nf1GRD/GRD embryos were analyzed by SDS-PAGE followed by immunoblotting with either anti-neurofibromin ( top panel ) or anti-beta tubulin ( bottom panel ) antibodies as indicated . ( B ) Band intensities from 5 immunoblots as in ( A ) were quantified by ImageJ . The relative neurofibromin expression for each genotype compared to wild-type is indicated . All data are represented as the mean ± S . E . ** , p<0 . 05; *** , p<0 . 001; NS = not significant ( p<0 . 001 , one-way ANOVA between groups , post hoc multiple comparisons , Tukey’s test ) . ( C ) A cross-section of a peripheral nerve ( demarcated in white and indicated by an arrow ) from each of Nf1GRD/flox and Wnt1-Cre; Nf1GRD/flox P0 animals shows elevated expression of pERK , a downstream indicator of Ras activity , in Wnt1-Cre; Nf1GRD/flox animals ( right panel ) . An adjacent blood vessel ( BV ) is indicated . ( D ) Adrenal medullary tissue within an adrenal gland from either a Nf1GRD/flox or Wnt1-Cre; Nf1GRD/flox animal shows increased pERK expression in a hyperplastic area from the Wnt1-Cre; Nf1GRD/flox animal ( right panel ) . pERK-positive cells are marked by arrowheads . Background fluorescence from non-neural-crest-derived adrenal cortical and blood cells is evident in the Nf1GRD/flox sample . ( E ) Cardiac cushions from E11 . 5 embryos show elevated pERK staining in Nf1GRD/GRD embryos compared to Nf1+/+ animals as indicated within the dashed oval . Scale bars = 50 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 07780 . 00710 . 7554/eLife . 07780 . 008Figure 4—figure supplement 1 . Generation of Nf1GRDand Nf1GRDCTL mouse lines . ( A ) Schematic diagram outlining the targeting strategy to develop the Nf1GRD mouse line by modifying the endogenous mouse Nf1 locus with a mutation corresponding to the human NF1 R1276P missense allele . This mutation abrogates neurofibromin GAP activity without impairing secondary or tertiary protein structure or reducing cellular levels of neurofibromin ( Klose , et al . , 1998 ) . ( B ) Strategy to develop the Nf1GRDCTLknock-in 'control' mouse by targeting the endogenous mouse Nf1 locus with a construct identical to that used to target the NF1 R1276P mutation in ( A ) with the exception that no mutation is introduced . Knock-in Nf1GRDCTLmice generated from this construct are a stringent control for Nf1GRDanimals . For both ( A ) and ( B ) asterisks denote regions where additional DNA sequences are identically introduced into introns as part of the targeting process . The 'an' cassette imparts G418 resistance and is self-excised in the male germ line . N = NcoI restriction endonuclease site . ( C ) Southern blots of genomic DNA from embryonic stem ( ES ) cell clones targeted with either the Nf1GRDor Nf1GRDCTLallele display a 13 kb wild type ( WT ) band as well as a 5 . 5 kb mutant ( MT ) band . Five and three positive clones were isolated with genotype Nf1GRDor Nf1GRDCTL , respectively , as shown . ( D ) DNA products from PCR reactions performed with primers specific for the Nf1 wildtype , Nf1 knockout ( KO ) , or Nf1 GRD alleles using template DNA isolated from amniotic sacs of E10 . 5 embryos . DOI: http://dx . doi . org/10 . 7554/eLife . 07780 . 00810 . 7554/eLife . 07780 . 009Figure 4—figure supplement 2 . Increased pERK staining in neural crest derivatives of Nf1GRD/flox newborn animals following deletion by Wnt1-Cre . ( A ) pERK staining was observed in neural crest-derived enteric ganglia within the intestinal wall that was more evident in Wnt1-Cre; Nf1GRD/flox animals ( right panel ) . Arrowheads indicate cells exhibiting positive staining . ( B ) Both axons ( arrows ) and nerve cell bodies ( arrowheads ) were readily visualized in Wnt1-Cre; Nf1GRD/flox newborns but not in control animals ( data not shown ) . DOI: http://dx . doi . org/10 . 7554/eLife . 07780 . 009 To assess whether the introduced R1276P mutation within the GAP domain of neurofibromin disrupts Ras-GAP activity in vivo , tissues were examined for elevated phosphorylated extracellular-signal regulated kinase ( pERK ) , a downstream effector of Ras , as evidence of up-regulated Ras pathway activity . Nf1GRD/flox newborns in which Nf1 was deleted by Wnt1-Cre , displayed elevated pERK staining in neural crest-derived tissues such as peripheral nerves ( Figure 4C ) , within hyperplastic adrenal medullary tissue ( Figure 4D ) , and in enteric ganglia ( Figure 4—figure supplement 2 ) . Wnt1-Cre; Nf1GRD/flox newborns showed prominent pERK staining in the axons and cell bodies of peripheral nerves ( Figure 4—figure supplement 2 ) that was not observed in control animals . Elevated pERK staining was also seen in the enlarged cardiac cushions of Nf1GRD/GRD embryos , indicating the R1276 mutation is sufficient to elevate pERK levels ( Figure 4E ) . Multiple reports showed that mutation of the conserved 'arginine finger' within the GAP domain decreases neurofibromin GAP function while leaving the domain structurally intact ( Ahmadian et al . , 1997; Scheffzek et al . , 1997; Klose et al . , 1998; Hiatt et al . , 2004 ) . These observations indicate that inactivation of neurofibromin GAP activity elevates the phosphorylation of the Ras pathway effector ERK in vivo . Nf1GRDCTL mice either heterozygous or homozygous for the control allele in which arginine 1276 was left intact , appeared normal . Intercrosses of Nf1GRDCTL/+ mice produced 6 of 26 Nf1GRDCTL/GRDCTL offspring ( 23% ) . These control mice were not examined further , and we conclude that embryonic lethality observed in Nf1GRD/GRD embryos is due specifically to the R1276P mutation . Histologic analysis of E11 . 5 Nf1GRD/GRD embryos revealed abnormal cardiac outflow tract morphology and enlarged endocardial cushions , similar to those seen in Nf1-/- embryos ( Figure 5A ) , which have been previously described in detail ( Brannan et al . , 1994; Jacks et al . , 1994; Lakkis and Epstein , 1998 ) . Atrioventricular endocardial cushions were also enlarged and ventricular septal defects were present , similar to the phenotype seen in Nf1-/- embryos ( Figure 5B ) . Sympathetic ganglia , derived from neural crest , were enlarged in both Nf1GRD/GRD and Nf1-/- embryos ( Figure 5C ) . Enlargement of sympathetic ganglia in Nf1GRD/GRD mutants was confirmed by immunofluorescence staining for neurofilament and tyrosine hydroxylase ( Figure 6A , B ) . 10 . 7554/eLife . 07780 . 010Figure 5 . Inactivation of Nf1 GRD function affects heart and sympathetic ganglia development . ( A ) Sections of hearts from E12 . 5-13 . 5 Nf1+/+ , Nf1GRD/GRD , and Nf1-/- embryos . The enlarged endocardial cushions in hearts from Nf1GRD/GRD embryos ( arrowheads ) are similar to the oversized cushions of Nf1-/- embryos . ( B ) Enlarged atrioventricular endocardial cushions ( arrowheads ) and ventricular septa defects ( arrows ) in Nf1GRD/GRD and Nf1-/- embryos . ( C ) Sympathetic ganglia ( arrowheads ) are similarly enlarged in Nf1GRD/GRD and Nf1-/- embryos . Scale bars = 500 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 07780 . 01010 . 7554/eLife . 07780 . 011Figure 6 . Enlarged sympathetic ganglia in E11 . 5 Nf1GRD/GRD embryos . ( A ) Transverse sections of E11 . 5 Nf1+/+ , Nf1GRD/+ , and Nf1GRD/GRD embryos stained with antibodies against neurofilament . Arrowheads indicate sympathetic ganglia . ( B ) Transverse sections of E11 . 5 Nf1+/+ , Nf1GRD/+ , and Nf1GRD/GRD embryos stained with antibodies against tyrosine hydroxylase . Scale bars = 100 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 07780 . 011 Tissue-specific loss of Nf1 in neural crest results in late gestation lethality , bypassing the midgestation cardiac defects seen in Nf1 null mutants ( Gitler et al . , 2003 ) . In order to examine in more detail if the Ras-GAP function of neurofibromin is necessary in developing neural crest in embryos surviving past midgestation , we crossed Wnt1-Cre; Nf1GRD/+ mice with Nf1flox/flox mice to generate Wnt1-Cre; Nf1GRD/flox offspring . At E18 . 5-P0 , no viable Wnt1-Cre; Nf1GRD/flox pups were identified out of 80 genotyped , although 12 non-viable pups ( 15% ) were stillborn or died shortly after birth ( Table 2 ) . Live Wnt1-Cre; Nf1GRD/flox embryos were recovered between E12 . 5 and 16 . 5 at the expected frequency ( Table 2 ) . 10 . 7554/eLife . 07780 . 012Table 2 . Genotypes from Wnt1-Cre; Nf1GRD/+ X Nf1flox/flox crosses . DOI: http://dx . doi . org/10 . 7554/eLife . 07780 . 012AgeTotalNf1flox/+Nf1GRD/floxWnt1-Cre;Nf1flox/+Wnt1-Cre;Nf1GRD/floxE12 . 5-–16 . 52710537 *E18 . 5-P0803218180 ***2 non-viable Wnt1-Cre; Nf1GRD/flox embryos were recovered at E12 . 5-–16 . 5**12 non-viable Wnt1-Cre; Nf1GRD/flox pups were recovered at E18 . 5-P Histologic examination of Wnt1-Cre; Nf1GRD/flox embryos revealed overgrowth of the adrenal medulla when compared to control Nf1GRD/flox embryos that phenocopied adrenal medullary defects seen in Wnt1-Cre; Nf1flox/flox embryos ( Figure 7A ) , described previously ( Gitler and Epstein , 2003 ) . Massive enlargement of paraspinal neural crest-derived ganglia was also noted in both Wnt1-Cre; Nf1GRD/flox and Wnt1-Cre; Nf1flox/flox embryos ( Figure 7B , C ) . These findings suggest that loss of neurofibromin Ras-GAP function in neural crest is sufficient to reproduce the late-gestation lethality and tissue overgrowth that results from by tissue-specific deletion of Nf1 in neural crest . 10 . 7554/eLife . 07780 . 013Figure 7 . Hyperplasia of neural crest derivatives is similar in Nf1GRD/flox and Nf1flox/flox newborn animals in which Nf1 is deleted in neural crest cells with Wnt1-Cre . ( A ) Adrenal medullary tissue ( demarcated in white and indicated with an arrowhead ) contained within an adrenal gland of P0 wild-type , P0 Wnt1-Cre; Nf1GRD/flox , or E16 . 5 Wnt1-Cre; Nf1flox/flox animals . The tissue is similarly overgrown in Wnt1-Cre; Nf1GRD/flox and Wnt1-Cre; Nf1flox/flox newborns/fetuses . Scale bars = 100 μm . The arrow indicates a tumor-like medullary protrusion . ( B ) Sagittal sections showing peripheral ganglia ( arrowheads ) in Nf1GRD/flox newborn pups , and abnormally enlarged ganglia and tumor-like overgrowth of nerve tissue adjacent to the lumbar spine in a Wnt1-Cre; Nf1GRD/flox newborn pup and an E16 . 5 Wnt1-Cre; Nf1flox/flox fetuse . Scale bars = 500 μm . ( C ) Magnifications of images in ( B ) , with hyperplastic tissue demarcated in black and marked by arrowheads . Lu , lung; Li , liver; Scale bars = 500 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 07780 . 013 We examined E11 . 5 Nf1GRD/GRD embryos for evidence of cardiac blood island formation to determine if this results from the loss of Ras-GAP activity . Nf1GRD/GRD embryos appeared grossly normal at E11 . 5 ( Figure 8A ) , but blood-filled protrusions were often visible on the ventricles ( Figure 8B , arrowheads ) . Whole-mount immunofluorescence revealed that the blood-filled protrusions were phenotypically identical to the ectopic cardiac blood islands that formed on the ventricles of Nf1-/- embryos ( Figure 8C , arrowheads ) . Furthermore , Nf1GRD/GRD embryos had enlarged fetal livers populated with Runx1+ CD41+ hematopoietic cells , similar to Nf1-/- embryos ( Figure 8C ) . Ventricular blood islands were evident in histologic sections after hematoxylin and eosin ( H and E ) staining of Nf1GRD/GRD and Nf1-/- embryos and had similar structural characteristics ( Figure 8D , arrowheads ) . Single optical projections through Nf1GRD/GRD blood islands confirm that they were associated with CD31+ CD41+ Runx1+ phenotypic hematopoietic cells ( Figure 8E ) . An average of 26 . 3 ± 9 . 2 blood islands could be identified via confocal microscopy on the ventricles of E11 . 5 Nf1GRD/GRD embryos , whereas Nf1+/+ and Nf1GRD/+ ventricles contained no cardiac blood islands ( Figure 8F ) . However , Nf1-deficient E11 . 5 embryos had , on average , >2 fold more morphological cardiac blood islands as compared to Nf1GRD/GRD embryos ( compare Figure 8F and 2D , p≤0 . 022 ) , suggesting that the Nf1GRD is a hypomorphic Nf1 allele , at least in regard to cardiac blood island formation . Flushed Nf1GRD/GRD ventricles contained significantly more EMPs than Nf1+/+ and Nf1GRD/+ littermates ( Figure 8G ) , but there was a trend towards fewer progenitors than in Nf1-/- embryos . 10 . 7554/eLife . 07780 . 014Figure 8 . E11 . 5 Nf1GRD/GRD embryos form ectopic cardiac blood islands . ( A ) Gross view of E11 . 5 Nf1+/+ , Nf1GRD/+ and Nf1GRD/GRD littermates . ( B ) Isolated hearts from E11 . 5 Nf1+/+ , Nf1GRD/+ and Nf1GRD/GRD embryos . Black arrowheads indicate blood-filled protrusions on the ventricle of the Nf1GRD/GRD heart . ( C ) Confocal Z-projections ( Z interval = 5 μm ) of CD31 ( red ) , CD41 ( cyan ) and Runx1 ( green ) immunostained E11 . 5 Nf1+/+ , Nf1GRD/+ and Nf1GRD/GRD embryos . Arrowheads point to blood islands on the ventricle of the Nf1GRD/GRDembryo . Scale bars = 500 μm . ( D ) Cell aggregates resembling blood islands ( arrowheads ) in hearts of E12 . 5 Nf1+/+ , Nf1-/- and Nf1GRD/GRD embryos . Lower panels , are magnifications of images in top panels . Scale bars = 100 μm . ( E ) Single optical projection through the cardiac blood islands of an E11 . 5 Nf1GRD/GRD embryo immunostained for CD31 ( red ) , Runx1 ( green ) and CD41 ( cyan ) . Scale bars = 50 μm . ( F ) Quantification of blood islands on the ventricles of E11 . 5 embryos . One-way ANOVA and Bonferroni’s multiple comparison test applied to determine significance , error bars represent the SD . ( G ) Number of erythroid and myeloid progenitors per flushed E11 . 5 ventricle . One-way ANOVA and Bonferroni’s multiple comparison test applied to determine significance; error bars represent SD . Nf1+/+ n = 21 , Nf1GRD/+ n = 22 and Nf1GRD/GRD n = 8 . * indicates that p≤0 . 05 and *** indicates that p≤0 . 001 . V: ventricle; A: atrium; FL: fetal liver; Mk: megakaryocyte; Mix: granulocyte-erythroid-monocyte-megakaryocyte; BFU-E: burst forming unit-erythroid; G/M: granulocyte-macrophage colonies . DOI: http://dx . doi . org/10 . 7554/eLife . 07780 . 014
In this study we have identifed ectopic cardiac blood island formation as a novel phenotype that arises in Nf1-deficient embryos . Furthermore , using a mouse that expresses a mutant form of neurofibromin with decreased Ras-GAP activity , we demonstrated that the phenotype is a direct result of dysregulation of the Ras signaling pathway . We also showed that some endocardial cells in the ectopic blood islands express Runx1 , a master regulator of hematopoiesis and a marker of HE . This , in addition to the enrichment of both phenotypic and functional hematopoietic progenitors in the ventricles of E11 . 5 Nf1-deficient embryos , suggests that the endocardial cells are producing hematopoietic cells de novo . We also observed dysregulation of de novo hematopoietic progenitor formation in Nf1-deficient embryos in normal sites of hematopoiesis . A previous study in zebrafish embryos found that the downstream effector of the Ras signaling pathway , pERK , has a biphasic role in blood cell formation from endothelium ( Zhang et al . , 2014 ) . When zebrafish embryos were treated with an ERK signaling inhibitor prior to artery-—vein specification , runx1 and myb expression in the dorsal aorta decreased; however , when treated with an ERK signaling inhibitor after artery-—vein specification but before EHT , runx1 and myb expression increased ( Zhang et al . , 2014 ) . Thus , early in development pERK is necessary for de novo generation of hematopoietic cells , but after artery-vein specification , pERK inhibits the specification of HE cells ( Zhang et al . , 2014 ) . Consistent with a role for ERK signaling in HE specification , increased signaling through the fibroblast growth factor ( FGF ) receptor , which is upstream of ERK and regulated by Ras-GAP , decreases runx1 expression in the dorsal aorta of zebrafish ( Pouget et al . , 2014 ) . The mechanism by which increased FGF signaling decreases runx1 expression in the HE involves inhibition of bone morphogenic protein signaling , which is required for runx1 expression ( Pouget et al . , 2014; Wilkinson et al . , 2009; Pimanda et al . , 2007 ) . Consistent with these findings , we show that loss of Nf1 , which is associated with activation of the Ras-pERK pathway , results in fewer Runx1+ HE cells in the dorsal aorta at E10 . 5 , as well as fewer CD31+ Kit+ Runx1+ hematopoietic cluster cells . In contrast , the yolk sac of Nf1-/- embryos produced more EMPs when compared to littermate controls , consistent with the positive role for FGF signaling in regulating erythropoiesis and myelopoiesis at a similar earlier stage in the zebrafish embryo ( Yamauchi et al . , 2006; Walmsley et al . , 2008 ) . Thus the level of Ras activity must be carefully titrated , as elevating Ras signaling in the dorsal aorta limits HE specification , but enhances EMP formation in the yolk sac , and unleashes the hematopoietic potential of the endocardium . It was previously reported that hematopoietic cells derived from the fetal livers of Nf1-deficient mice are hyperproliferative and cause a JMML-like myeloid proliferative disorder when transplanted into irradiated recipients ( Birnbaum et al . , 2000; Largaespada et al . , 1996; Zhang et al . , 2001; Zhang et al . , 1998 ) . Based on immunoflourescence , it appears that hematopoiesis is elevated as early as E11 . 5 in fetal livers of both Nf1-/- and Nf1GRD/GRD embryos compared to controls , thus implicating activated Ras in hyperproliferation of hematopoietic cells ( at E11 . 5 , primarily EMPs ) that populate the fetal liver . Likewise , our results implicate the loss of neurofibromin Ras-GAP function within neural crest cells as sufficient to result in overgrowth of sympathetic and dorsal root ganglia and of the adrenal medulla . The ability of the Nf1 gene product to act as a Ras GAP has been known for a quarter of a century ( Ballester et al . , 1990; Xu et al . , 1990 ) , but the degree to which this function accounts for some or all Nf1 phenotypes has been an ongoing topic of research with relevance for therapeutic strategies . We and others have provided evidence for the ability of neurofibromin to affect alternate signaling pathways , including PKA and cAMP ( Guo et al . , 1997; The et al . , 1997; Hegedus et al . , 2007; Brown et al . , 2010; Wolman et al . , 2014 ) . Prior work has suggested that midgestation embryonic lethality resulting from loss of Nf1 can be rescued by transgenic expression of the isolated neurofibromin GRD , but this was not sufficient for rescue of neural crest overgrowth ( Ismat et al . , 2006 ) . Failure to rescue neural crest overgrowth could have been the result of inadequate transgenic expression of GRD in this tissue , or because of the necessity for an additional function of neurofibromin outside of the GRD . The findings reported here for Nf1GRD/GRD embryos do not rule out the existence of critical non-GRD functions of neurofibromin in the neural crest or other tissues . In fact , the Nf1GRD has characteristics of a hypomorphic allele that could be explained by non-GRD related functions . Rather , we demonstrate the necessity of GRD function for normal embryonic development . The development of the Nf1GRD/+ mouse line described here will allow researchers to determine the necessity of GRD function across the spectrum of developmental and tumor phenotypes observed in mouse models of neurofibromatosis .
Embryos were removed from the uterus and dissected in phosphate buffered saline ( PBS ) with 20% fetal bovine serum and antibiotics . The yolk sacs were removed and the hearts were dissected , the atrium was cut away and the ventricles were flushed with PBS using an insulin needle and syringe to remove circulating blood cells . The ventricles and yolk sacs were then dissociated in 0 . 125% collagenase Type I ( Sigma , St Louis , MO ) for 20–30 min at 37°C , triturated , washed , and filtered to obtain a single cell suspension . Single cell solutions of embryonic ventricles or yolk sacs were plated in methylcellulose ( MethocultM3434; Stem Cell Technologies , Vancouver , BC ) and colonies were counted 7–8 days after plating . Embryos were prepared as described previously ( Yokomizo et al . , 2012 ) . The following primary antibodies were used; rat anti-mouse CD31 ( Mec 13 . 3 , BD Pharmingen , San Diego , CA ) , rat anti-mouse CD117 ( 2B8 , eBiosciences , San Diego , CA ) , rat anti-mouse CD41 ( MWReg30 , BDBiosciences , Franklin Lakes , NJ ) and rabbit anti-human/mouse Runx ( EPR3099 , Abcam , Cambridge , MA ) . Secondary antibodies used were goat anti-rat Alexa Fluor 647 ( Invitrogen , Carlsbad , CA ) , goat-anti rat Alexa Fluor 555 ( Abcam ) and goat anti-rabbit Alexa Fluor 488 ( Invitrogen ) . Images were acquired on a Zeiss LSM 710 AxioObserver inverted microscope with ZEN 2011 software and processed with Fiji software ( Schindelin , et al . , 2012 ) . Hematopoietic cells in the dorsal aorta were counted using the cell counter plugin ( version February 29 , 2008 , Kurt De Vos; http://rsb . info . nih . gov/ij/plugins/cell-counter . html ) . E10 . 5 embryos were dissected free of the amniotic sac , frozen in liquid nitrogen , thawed , and disrupted by pipetting in Hank’s Balanced Salt Solution containing 5 mM ethylenediaminetetraacetic acid ( EDTA ) . Total cell lysates were prepared by heating samples in boiling Laemli buffer ( 66 mM Tris–HCl , pH 6 . 8 , 2% ( w/v ) SDS , 10 mM EDTA ) . The samples were subjected to sodium dodecyl sulfate polyacrylamide gel electrophoresis ( SDS-PAGE ) and immunoblotting analysis using anti-neurofibromin antibody ab17963 ( Abcam ) . Immunoreactive bands were visualized by chemiluminescence . Quantification of individual band intensities was performed using ImageJ . One-way analysis of variance ( ANOVA ) was used to assess statistical differences between band intensities . Significant ANOVA results were analyzed post hoc by the Tukey-Kramer multiple comparisons test . Whole mouse embryos or dissected hearts were fixed in 2 or 4% paraformaldehyde , dehydrated in ethanol , and embedded in paraffin for sectioning . Tissues were visualized with H and E stain or by immunofluorescent detection of marker proteins according to standard practices . Detailed protocols are available at http://www . pennmedicine . org/heart/ . Antibodies used for immunofluorescence include rabbit polyclonal anti-tyrosine hydroxylase ( AB152 , EMD Millipore/Chemicon , Billerica , MA ) , rabbit polyclonal anti-pERK ( #9101 , Cell Signaling Technology , Inc . , Danvers , MA ) and mouse monoclonal anti-neurofilament ( 2H3 , Developmental Studies Hybridoma Bank , Department of Biology , University of Iowa , Iowa City , IA ) . Images were adjusted using Adobe Photoshop using settings applied across the entirety of each image . All mouse manipulations were performed in accordance with protocols approved by the Institutional Animal Care and Use Committee ( IACUC ) of the University of Pennsylvania following guidelines described in the US National Institutes of Health Guide for the Care and Use of Laboratory Animals . Nf1-/- , Nf1flox/+ and Wnt1-Cre mice have been described previously ( Brannan et al . , 1994; Jacks et al . , 1994; Zhu et al . , 2001; Danielian et al . , 1998; Jiang et al . , 2000 ) . Nf1GRD/+ and Nf1GRDCTL/+ mice were produced by targeting C57BL/6 ES cells ( Genoway , Lyon , France ) with a targeting vector designed to replace arginine 1276 with proline ( R1276P ) or , in the case of Nf1GRDCTL/+ to leave arginine 1276 as arginine . The selection strategy ( Figure 4—figure supplement 1 ) included a self-excising floxed neomycin resistance cassette that , after excision , leaves a single loxP site within intron 27 . The Nf1GRDCTL/+ mice were created in order to control for possible unpredicted effects related to the introduction of small changes in genomic sequence , other than those encoding the R1276P missense mutation , necessitated by the targeting strategy . Nf1GRD/+ and Nf1GRDCTL/+ mice were genotyped using polymerase chain reaction primers listed below , which produce a 175 bp wild-type band and a 248 bp mutant band ( Figure 4—figure supplement 1 ) . All mice were maintained on a C57BL/6 background . GRDF: 5’- GAGGGGAGATGTCAAAGATGTATTGTGTAACTAC-3’ GRDR: 5’- CAACCTTCAAACAGTACTAAAGTCCATCATGG-3’ | Messages are carried from the surface of a cell to the cell’s nucleus in order to regulate various processes such as how often the cell will divide . The Ras-signaling pathway carries some of these messages . A gene called Nf1 encodes a protein in this pathway that deactivates another protein called Ras when the message is no longer required . If a mutation in Nf1 prevents it from deactivating Ras , the pathway becomes hyperactivated . In humans , this results in a disorder called Neurofibromatosis type I , which is characterized by tumors that affect many parts of the body . When the expression of Nf1 is turned off in mice , the mice die as embryos because of cardiac defects . But a mouse in which Nf1 has been turned off in specific organs or tissues other than the heart can survive , and these mice are used to model Neurofibromatosis type I and to help to identify potential treatments . Yzaguirre et al . have now identified new roles for Nf1 during embryonic development . In the embryo , blood cells originate from the cells lining the blood vessels . The experiments revealed that , when the Nf1 gene was mutated in mice , fewer blood cells formed from the lining of the major blood vessel that leaves the embryonic heart . In contrast , these mutant mice formed more structures called cardiac blood islands than a normal mouse . These structures line the heart , and have the potential to generate new blood cells for the heart to pump . These results shed new light on how blood is originally formed from the lining of the heart and blood vessels , and show that Ras signaling must be tightly regulated to maintain normal blood development in the embryo . Furthermore , Yzaguirre et al . demonstrated that this excessive formation of cardiac blood islands resulted specifically from the loss of Nf1’s role in the Ras-signaling pathway . This was achieved by using gene targeting to generate a mouse that expresses Nf1 with a minor change that affects only the protein’s interaction with Ras . In the future , this new strain of mouse will be a useful tool in determining if specific aspects of Neurofibromatosis type I can be attributed to loss of Nf1’s role in Ras-signaling and could therefore be treated by medicines that target this pathway . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"developmental",
"biology",
"cell",
"biology"
] | 2015 | Loss of neurofibromin Ras-GAP activity enhances the formation of cardiac blood islands in murine embryos |
Armor plate changes in sticklebacks are a classic example of repeated adaptive evolution . Previous studies identified ectodysplasin ( EDA ) gene as the major locus controlling recurrent plate loss in freshwater fish , though the causative DNA alterations were not known . Here we show that freshwater EDA alleles have cis-acting regulatory changes that reduce expression in developing plates and spines . An identical T → G base pair change is found in EDA enhancers of divergent low-plated fish . Recreation of the T → G change in a marine enhancer strongly reduces expression in posterior armor plates . Bead implantation and cell culture experiments show that Wnt signaling strongly activates the marine EDA enhancer , and the freshwater T → G change reduces Wnt responsiveness . Thus parallel evolution of low-plated sticklebacks has occurred through a shared DNA regulatory change , which reduces the sensitivity of an EDA enhancer to Wnt signaling , and alters expression in developing armor plates while preserving expression in other tissues .
The repeated evolution of similar adaptive phenotypic traits in multiple populations is a fascinating evolutionary phenomenon observed in many organisms ( Wood et al . , 2005; Elmer and Meyer , 2011; Conte et al . , 2012; Martin and Orgogozo , 2013; Stern , 2013 ) . The threespine stickleback ( Gasterosteus aculeatus ) is a particularly favorable species to characterize the molecular mechanisms underlying repeated evolution of adaptive phenotypic traits in nature , because many populations have evolved similar morphological and skeletal traits following widespread colonization of new freshwater environments by migratory marine ancestors at the end of the last ice age ( Bell and Foster , 1994 ) . One of the most striking and ubiquitous morphological changes seen in sticklebacks is repeated alteration in bony armor seen along the sides of fish . Marine sticklebacks are typically covered from head to tail with 32 ( or more ) bony lateral plates . In contrast , freshwater fish characteristically lack most plates , typically retaining only 0–7 plates in the anterior flank region . This dramatic difference in anterior-posterior patterning of armor plates was used to assign different species names to marine and freshwater sticklebacks in the 1800s ( Cuvier and Valenciennes , 1829 ) . Subsequent studies have shown that different armor patterns are highly heritable , and are likely controlled by a relatively simple genetic system ( Münzing , 1959; Hagen and Gilbertson , 1973; Avise , 1976; Ziuganov , 1983; Banbura , 1994 ) . More recently , genome-wide linkage mapping in crosses between divergent sticklebacks identified a major locus on stickleback chromosome IV that controls over 75% of the variance in armor plate number in F2 offspring ( Colosimo et al . , 2004; Cresko et al . , 2004 ) , as well as several unlinked modifier genes that each control 5–10% of the variance in plate numbers ( Colosimo et al . , 2004 ) . High-resolution mapping , chromosome walking , and transgenic rescue experiments showed that the major armor plate locus corresponds to the ectodysplasin ( EDA ) gene on stickleback chromosome IV ( Colosimo et al . , 2005 ) . The EDA gene encodes a secreted protein in the tumor necrosis factor ( TNF ) family that plays a key role in cell signaling during the development of multiple neural crest and ectodermal tissues , including skin , hair , and teeth ( Mikkola and Thesleff , 2003; Cui and Schlessinger , 2006 ) . Humans with null mutations in EDA have defects in multiple ectoderm and neural crest derived tissues , including sparse hair , absent sweat glands , dental abnormalities , and dermal bone changes in the skull ( Mikkola and Thesleff , 2003; Cui and Schlessinger , 2006; Yavuz et al . , 2006; Clauss et al . , 2008; Lesot et al . , 2009 ) . Zebrafish and medaka mutants with perturbations in the EDA pathway display severe skeletal abnormalities , such as loss of fins and scales , missing and abnormally shaped teeth , and abnormal craniofacial morphology ( Harris et al . , 2008; Iida et al . , 2014 ) . While both high-resolution mapping and transgenic rescue experiments confirm that EDA is the major locus controlling armor plates in sticklebacks , the molecular difference between marine and freshwater fish is still unclear . Most freshwater populations share four amino acid differences in the EDA gene , as well as numerous non-coding changes that together make up a characteristic freshwater haplotype ( Colosimo et al . , 2005 ) . However , the four amino acid changes occur at positions that also vary among other species , so these coding changes are unlikely to be the basis of major changes in EDA function . In addition , there exists at least one low-plated stickleback population that has the identical EDA protein-coding sequence as marine fish ( Colosimo et al . , 2005 ) . This key population from Nakagawa Creek in Gifu , Japan ( NAKA ) is a low-plated stream population with a predominately marine-like sequence in both coding and non-coding regions . NAKA fails to complement armor plate changes when crossed with a typical Canadian low-plated population ( Schluter et al . , 2004 ) , suggesting that NAKA and other low-plated fish share a modification in the same major locus . Based on the absence of amino acid changes in NAKA , and the deleterious nature of coding region changes in human patients , Colosimo et al . ( 2005 ) proposed that an unknown regulatory change at the stickleback EDA locus is the most likely basis of the common EDA variants found in freshwater fish . Here we further investigate the EDA locus in order to study the causative base pair changes that underlie repeated evolution of low-plated sticklebacks .
In order to test if EDA is differentially expressed in marine and freshwater fish due to cis-regulatory differences , we performed allele-specific expression in F1 hybrid fish made by crossing marine and freshwater sticklebacks . The F1 hybrids are heterozygous for both the marine and freshwater haplotypes at the EDA locus , and therefore express both alleles in an identical trans-acting environment . We then isolated RNA from 10 different developing tissues , and determined whether the freshwater and marine EDA transcripts were expressed at the same or different levels using pyrosequencing ( Figure 1 , see ‘Materials and methods’ ) . No significant expression differences between marine and freshwater EDA alleles were observed in the fins or the lower jaw . However , the freshwater EDA allele was expressed almost fourfold lower than the marine allele in the developing anterior and posterior flanks ( corresponding to sites where armor plates had already appeared , or were not forming yet; respectively ) , and in the dorsal and pelvic spines ( p < 0 . 01 , Student's t test ) , as well as twofold lower in the premaxilla ( p < 0 . 05 , Student's t test ) . These data suggest that the marine and freshwater haplotypes at the EDA locus have cis-acting regulatory changes that reduce expression of the freshwater allele in particular tissues , including the flank regions where armor plates normally form . 10 . 7554/eLife . 05290 . 003Figure 1 . EDA shows allele-specific expression differences in several tissues , indicating cis-regulatory divergence . Allele-specific expression in F1 freshwater-marine heterozygous larvae reveals significant differential expression of the marine and freshwater alleles in dorsal spines 1 and 2 , the pelvic spine , the premaxilla , and the presumptive armor plates ( anterior and posterior flanks ) . In all of these bony tissues the marine allele of EDA is expressed more highly than the freshwater allele , suggesting that there are differences in the cis-regulatory sequences controlling EDA expression . Several other tissues , however , do not show significant allelic imbalance in EDA expression; their allelic ratios are close to 1 ( dashed line ) . The control shows results from a 1:1 mixture of plasmids containing the freshwater and marine alleles . Red-shaded structures and bars indicate tissues with significant allelic-imbalance compared to control ( ***p < 0 . 001 , **p < 0 . 01 , *p < 0 . 05 by two-tailed t-test ) . DOI: http://dx . doi . org/10 . 7554/eLife . 05290 . 003 Previous studies narrowed the minimal candidate interval controlling armor plates to a 16 kb interval containing EDA and flanking regions ( Colosimo et al . , 2005 ) . To look for possible shared molecular changes that might account for the regulatory difference between marine and freshwater sticklebacks , we amplified and sequenced the EDA candidate interval from low-plated Japanese NAKA fish , and compared it to other high- and low-plated stickleback populations ( Figure 2 and ‘Materials and methods’ ) . This analysis identified a single T → G nucleotide change , located at position chrIV:12811481 ( gasAcu1 assembly , Jones et al . , 2012 ) in the intergenic region downstream of EDA , that was shared between NAKA and all other low-plated sticklebacks examined . 10 . 7554/eLife . 05290 . 004Figure 2 . All low-plated populations share a single base pair change in the genetic region controlling armor plates . Genome-wide comparisons of low- and high-plated fish reveal a T → G base pair change ( black box ) that is shared between all low-plated populations tested , including the low-plated Japanese NAKA fish that otherwise shows a primarily marine-like haplotype in the EDA region . Geographic population codes and DNA sequences from marine high-plated populations and freshwater low-plated populations are shown in red and blue , respectively , along with representative Alizarin Red stained fish showing typical armor plate patterns in different fish . The 16 kb candidate interval controlling armor plate number ( blue bar , Colosimo et al . , 2005 ) is shown beneath predicated genes in the region . Also shown are the numbered positions ( 4–16 ) of previously identified SNPs that differentiate most low- and high-plated sticklebacks other than NAKA ( Colosimo et al . , 2005 ) . These numbered SNPs correspond to positions chrIV: 12800508 , 12808303 , 12808630 , 12811933 , 12813328 , 12813394 , 12815024 , 12815027 , 12816201 , 12816202 , 12816360 , 12816402 , and 12816464 in the stickleback genome assembly ( Jones et al . , 2012 ) . Blank positions represent occasional gaps in sequence coverage for individual fish from large population surveys ( Colosimo et al . , 2005; Jones et al . , 2012 ) . The position of the shared T → G change ( chrIV:12811481 ) is indicated with a short black vertical line in the overall genomic interval , and in a 3 . 2 kb region that was used to test for possible regulatory enhancers in the EDA region ( orange bar , chrIV:12808949–12812120 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 05290 . 004 Given the known role of EDA in plate formation , we hypothesized that this intergenic base pair change ( Figure 2 ) lies in a developmental enhancer that modulates EDA gene expression during armor plate development . Therefore , we cloned a 3 . 2 kb region surrounding the SNP ( orange bar , Figure 2 ) from high-plated marine fish and tested for enhancer activity using a GFP reporter construct ( p3 . 2mar-GFP , see ‘Materials and methods’ ) . In two-month-old transgenic fish , this 3 . 2 kb region drives consistent GFP expression at multiple sites , including the anterior and posterior armor plates , the junction between the pelvic spine and girdle , the upper edge of the pelvic girdle , the base of the pectoral fin , the cranial ganglia surrounding the eyes and lips , and the premaxilla and jaw ( Figure 3 , Figure 4A , B ) . Comparison to the endogenous pattern of EDA expression using in situ hybridization suggests that the GFP construct recapitulates typical EDA patterns in cranial ganglia , premaxilla , jaw , pectoral fin base , armor plates , and pelvic girdle base ( Figure 3 ) . However , some domains of endogenous EDA expression are not accounted for by the enhancer region , including the dorsal and pelvic spines , suggesting that this construct contains some but not all of the regulatory information controlling EDA expression during normal development . 10 . 7554/eLife . 05290 . 005Figure 3 . Reporter expression driven by an EDA enhancer matches several regions of endogenous EDA expression . ( A , D , G , J ) Negative control DapB RNAscope in situ staining shows no positive brown signal appearing around the face ( A and D ) , the plates ( G ) , or the pelvic junction ( J ) . The slight brown color in the pelvic spine is due to natural pigmentation at this site . ( B , E , H , K ) Endogenous EDA expression is localized to the premaxilla , lips , lower jaw , cranial ganglia , gill and pectoral fin base ( B and E ) ; armor plates ( H ) ; and the junction between the pelvic spine and the pelvic girdle ( K ) . ( C , F , I , L ) The p3 . 2mar-GFP construct drives reporter expression at several corresponding sites , including the lips , premaxilla , lower jaw and cranial ganglia surrounding the eyes ( C and F ) ; in the armor plates; ( I ) and at the pelvic junction ( L ) . Anatomical abbreviations as in other figures , including: lips ( L ) , premaxilla ( PM ) , lower jaw ( J ) , cranial ganglia ( CG ) , gills ( G ) , pectoral fin base ( PF ) , anterior plates ( AP ) , and pelvic spine junction ( PSJ ) . Scale bars are 1 mm long . DOI: http://dx . doi . org/10 . 7554/eLife . 05290 . 00510 . 7554/eLife . 05290 . 006Figure 4 . Enhancer expression in plates and other structures is reduced by a single base pair change . ( A , B ) A 3 . 2 kb enhancer region from high-plated fish drives GFP expression in all armor plates ( AP ) of 2-month-old ( 20 mm long ) marine stickleback larvae , with expression preceding plate ossification , and stronger expression in the first 7 armor plates . The p3 . 2mar-GFP construct also drives expression in the lips ( L ) , premaxilla ( PM ) , lower jaw ( J ) , cranial ganglia ( CG ) , the base of the pectoral fins ( PF ) , and the pelvic spine-girdle junction ( PSJ ) . Panel B is a higher magnification view of the area boxed in panel A . ( C , D ) The single base pair change in the p3 . 2mar ( T → G ) -GFP construct results in greatly reduced enhancer activity in the posterior plates , and reduced but detectable expression in plates 4–7 ( D ) . This stable line also retains expression in the cranial ganglia and lips , reduced expression in the pelvic junction and the pectoral fin base , and novel strong expression in the spinal cord . Panel D is a higher magnification view of the area boxed in panel B . The hsp70 promoter in the GFP vector drives strong expression in the lens ( LN ) of all transgenic fish , helping to identify carriers following microinjection experiments ( Chan et al . , 2010 ) . Scale bars are 1 mm long . DOI: http://dx . doi . org/10 . 7554/eLife . 05290 . 00610 . 7554/eLife . 05290 . 007Figure 4—figure supplement 1 . Plate enhancer activity is altered by a single base pair change ( additional examples from independent transgenic fish ) . ( A , B ) Examples of transient transgenics with mosaic GFP expression under the control of the marine high-plated marine EDA enhancer ( p3 . 2mar-GFP ) . Multiple transgenic founders share expression in the cranial ganglia ( CG ) surrounding the eyes and the lips ( L ) , the premaxilla ( PM ) , under the jaw ( J ) , and in armor plates ( AP ) . ( C , D ) Site-directed mutagenesis of the p3 . 2mar-GFP construct generating p3 . 2mar ( T → G ) -GFP results in loss of armor plate expression in transient transgenics . However , expression in the cranial ganglia ( CG ) around the eyes and lips ( L ) , as well as some expression surrounding the base of the pelvic spine-girdle junction ( PSJ ) remains in several fish . Copy number , integration sites , and mosaicism can vary in injected sticklebacks , giving rise to a range of expression levels . Despite this variability , consistent expression patterns can still be detected by comparing results from multiple injected fish . Overall , posterior plate expression was seen in 9 of 20 transgenic larvae with green eyes following injection of p3 . 2mar-GFP , vs of 0 of 27 transgenic larvae following injection of p3 . 2mar ( T → G ) -GFP . Scale bar in D is 2 mm long . DOI: http://dx . doi . org/10 . 7554/eLife . 05290 . 007 We next performed site-directed mutagenesis to change the T found in high-plated fish to the G found in all sequenced low-plated fish , while maintaining the sequence of the high-plated marine haplotype throughout the rest of the enhancer construct . The p3 . 2mar ( T → G ) -GFP plasmid still drove detectable expression in the anterior plates , cranial ganglia , jaws , and pectoral fin base , but showed greatly reduced GFP expression in the posterior armor plates and pelvic girdle junction ( Figure 4C , D , Figure 4—figure supplement 1 ) . Thus , the single base pair change shared by all low-plated sticklebacks produces striking but localized differences in gene expression , with prominent reduction occurring in the flank region where plates normally develop in marine fish . Previous studies have shown that Wnt signaling acts upstream of EDA in the early proliferation and specification of tissues in many vertebrates ( Laurikkala et al . , 2002; Cui and Schlessinger , 2006; Häärä et al . , 2011; Arte et al . , 2013 ) . To test whether Wnt also acts upstream of plate development in sticklebacks , we tested whether implants of either Wnt-3a or Dkk-1 ( an inhibitor of Wnt signaling , Glinka et al . , 1998 ) altered normal patterns of armor plate formation . Beads soaked in PBS , Wnt-3a , or Dkk-1 proteins were surgically implanted into the mid-flank of 2-month-old marine fish , and fish were then aged to 6 months to test for effects on plate size and number . Control bead implantation had no significant effect on overall plate morphology ( Figure 5A , B ) . In contrast , exposure to ectopic Wnt signaling at the juvenile stage induced hypermorphic plate development , characterized by adult fish with larger plates and plate fusions surrounding the sites of Wnt-3a bead implantation ( Figure 5C ) . Conversely , the addition of the Wnt inhibitor Dkk-1 resulted in a hypomorphic phenotype marked by the absence of plates surrounding the bead implantation site ( Figure 5D ) , suggesting that Wnt signaling plays an important role in normal plate development . 10 . 7554/eLife . 05290 . 008Figure 5 . Wnt signaling regulates armor plate development . Live Calcein staining of 6-month-old fish marks newly ossified bones in green . ( A ) Armor plates in an untreated high-plated adult marine fish . The normal morphologies of two individual plates are outlined with dashed lines . ( B ) Control beads soaked in PBS were implanted between the two outlined plates at two months of age . After bead implantation , fish continued to develop a full set of armor plates , with minimal changes in plate morphology ( n = 8 ) . ( C ) Implantation of Wnt-3a beads results in hypermorphic growth and armor plate fusion in the regions surrounding the exogenous Wnt-3a signal ( n = 11 ) . ( D ) Conversely , beads soaked in the Wnt inhibitor Dkk-1 inhibit plate formation surrounding the site of bead implantation ( n = 10 ) . Scale bar in D is 2 mm long . DOI: http://dx . doi . org/10 . 7554/eLife . 05290 . 008 To examine whether ectopic Wnt signaling also causes changes in EDA expression , we placed Wnt-3a protein beads into marine fish and used in situ hybridization to visualize EDA expression 48 hr later ( Figure 6A , B ) . These experiments revealed a strong ring of induced EDA expression surrounding the site of Wnt-3a bead implantation ( Figure 6B ) . We then implanted Wnt-3a beads into a stable transgenic line carrying the p3 . 2mar-GFP reporter construct described above . Implanted Wnt-3a beads , but not control beads , induced a strong ring of GFP expression directly around the site of bead implantation ( Figure 6C , D ) . In contrast , implantation of Wnt-3a protein beads failed to produce a similar strong ring of GFP expression in transgenic fish carrying the mutated p3 . 2mar ( T → G ) -GFP construct ( Figure 6E , F ) . Unexpectedly , the p3 . 2mar ( T → G ) -GFP construct did show a novel GFP response to the cyanoacrylate glue used in the implantation procedure , which was not seen in fish carrying p3 . 2mar-GFP . This expression was also observed in control manipulations with PBS beads ( Figure 6E ) or cyanoacrylate glue alone ( data not shown ) , and was therefore distinct from the strong Wnt-3a response observed only with the fish carrying the p3 . 2mar-GFP construct . 10 . 7554/eLife . 05290 . 009Figure 6 . Single point mutation alters Wnt responsiveness of the EDA plate enhancer . Beads soaked in either PBS or Wnt-3a protein were implanted in the flanks of 2-month-old ( 24 mm long ) marine fish . All images were taken at 48 hr post bead implantation . ( A , B ) RNAscope in situ hybridization for EDA expression induced by control bead placement ( A ) or Wnt-3a protein ( B ) . The addition of Wnt-3a beads induces a ring of EDA expression ( brown color in B ) directly surrounding the implantation site . ( C , D ) Bead implantation into the stable p3 . 2mar-GFP transgenic fish line . Control beads fail to induce GFP activity ( C ) , whereas Wnt-3a beads induce a strong GFP response , seen in a ring surrounding the bead implantation site ( D ) . ( E , F ) Bead implantation into the stable p3 . 2mar ( T → G ) -GFP line of transgenic fish . A ring of GFP expression is only seen at a distance from the implantation site of either control ( E ) or Wnt-3a ( F ) beads , corresponding to the location where cyanoacrylate glue was placed following implantation . Strong expression immediately surrounding the bead is not seen with Wnt-3a beads , in contrast to the result seen with p3 . 2mar-GFP transgenic fish ( compare panels F and D ) . Scale bar in F is 1 mm long . ( G ) In vitro analysis of enhancer response to Wnt signaling via β-catenin co-transfection shows a strong induction of p3 . 2mar-Luc ( green squares ) with 50 ng or more of β-catenin in human HaCaT keratinocyte cells . The β-catenin-responsiveness of the p3 . 2mar ( T → G ) -Luc is significantly lower ( black triangles ) . Combined p-values were calculated using Meta-P ( ***p < 0 . 001 , **p < 0 . 01 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 05290 . 009 Canonical Wnt signaling normally activates gene expression through changes in the activity of β-catenin ( Logan and Nusse , 2004 ) . Cotransfection of a β-catenin expression construct ( pRK5-sk-βcatΔGSK ) with the marine EDA enhancer driving a luciferase reporter ( p3 . 2mar-Luc ) produced a significant , dose-dependent increase in luciferase expression in cultured human keratinocytes in vitro ( Figure 6G ) . Engineering the single SNP change in the marine enhancer ( p3 . 2mar ( T → G ) -Luc ) reduced but did not eliminate response to β-catenin in the heterologous system ( 28% lower expression with 50 ng of β-catenin , p < 0 . 001 , n = 4 ) . Our combined experiments show that Wnt signaling can alter armor plate development and EDA expression in sticklebacks . The EDA enhancer region from high-plated sticklebacks also responds to Wnt signaling , while the single base pair mutation shared between NAKA and other low-plated sticklebacks significantly reduces Wnt responsiveness both in vivo and in vitro .
Previous work has shown that repeated armor plate reduction in sticklebacks is due in large part to genetic changes in the EDA region , though the causative molecular lesion ( s ) remained unknown ( Colosimo et al . , 2005; Jones et al . , 2012 ) . Our allele-specific expression experiments show that the freshwater allele of EDA is expressed at lower levels than the marine allele in F1 hybrids , confirming prior suggestions that there were likely to be cis-acting regulatory differences between marine and freshwater EDA variants ( Colosimo et al . , 2005 ) . In addition , we have now identified a specific enhancer region in the key armor plates region , shown that the marine version of this enhancer normally drives expression in developing armor plates , and identified a specific T → G base pair change within the enhancer that is shared by all sequenced low-plated freshwater fish . Experimental recreation of the T → G base pair change reduces both armor plate expression and Wnt responsiveness of the enhancer , suggesting that this specific DNA change is the likely causative regulatory lesion in the EDA locus that leads to low-plated phenotypes in sticklebacks . Like other genes found to underlie major morphological differences between marine and freshwater fish ( Shapiro et al . , 2006; Miller et al . , 2007; Chan et al . , 2010 ) , the EDA gene is a key developmental control gene that is essential for formation of multiple tissues . Mutations in the coding region of EDA in both zebrafish and medaka cause deleterious phenotypes at multiple body sites , including complete loss of scales , partial loss of fins and teeth , and multiple craniofacial abnormalities ( Harris et al . , 2008; Iida et al . , 2014 ) . In contrast , the T → G regulatory change we have identified in an EDA enhancer leads to partial loss of EDA expression , particularly in the posterior flank region ( Figure 4 ) . This regulatory change thus alters EDA expression at the same body site where freshwater fish lack body armor , while preserving important functions of EDA in other tissues . These results provide a new example of a specific regulatory change linked to morphological evolution in natural populations ( Martin and Orgogozo , 2013 ) , and add to growing evidence that regulatory changes are a predominant mechanism underlying adaptive evolution in sticklebacks ( Jones et al . , 2012 ) and other organisms ( Wray , 2007; Carroll , 2008 ) . Our results also provide new insight into genomic mechanisms contributing to repeated evolution . Previous analyses identified a shared low-plated EDA haplotype that has been fixed in most low-plated Pacific and Atlantic freshwater populations , and that is also present at very low frequency in the heterozygous state in marine populations ( Colosimo et al . , 2005; Barrett et al . , 2008; Bell et al . , 2010 ) . Thus , EDA has become a classic example of rapid parallel evolution based on a preexisting genetic variant that increases in frequency when marine populations colonize new freshwater environments ( Stern , 2013 ) . The current results suggest that repeated evolution of low-plated phenotypes might also result from independent mutations occurring in the EDA locus in different populations . In previous surveys , Japanese NAKA fish were the only low-plated freshwater population that did not share the same EDA haplotype as other freshwater populations ( Colosimo et al . , 2005 ) . Our experiments show that NAKA and other freshwater sticklebacks share an identical T → G non-coding regulatory mutation that reduces expression of EDA specifically in developing posterior armor plates . Characteristic flanking SNPs are not shared between NAKA and other low-plated populations , suggesting that the same T → G mutation has likely occurred independently on two very different haplotypes . Recurrent mutations can be due to a particular DNA sequence that has a high intrinsic mutation rate . For example , previous studies of pelvic reduction in sticklebacks suggest that a key pelvic enhancer repeatedly deleted in freshwater populations has sequence features shared with fragile sites in human chromosomes ( Chan et al . , 2010 ) . Individual base pairs can also be prone to particular mutations . For example , C → T transition mutations are particularly common at CpG dinucleotides in mammalian genomes , due to a high rate of spontaneous deamination of methylated C residues ( Mancini et al . , 1997; Xia et al . , 2012 ) . In contrast , the recurrent regulatory mutation we have identified at the stickleback EDA locus is a T → G transversion substitution , one of the least common types of changes seen in large scale studies of spontaneous germ-line mutations in humans ( Kong et al . , 2012; Genome of the Netherlands Consortium , 2014 ) as well as in flies , worms , and yeast ( Lynch , 2010 ) . It is possible that the shared T → G change arose not by independent mutation , but by extensive recombination or gene conversion from the typical freshwater EDA haplotype . Migratory marine populations include rare individuals that are heterozygous for both marine and freshwater haplotypes , which likely arise by repeated rounds of introgression of freshwater alleles into marine populations ( Colosimo et al . , 2005; Schluter and Conte , 2009 ) . Sequence studies suggest that recombination can occur between typical marine and freshwater haplotypes , producing smaller and smaller blocks of sequence shared among most low-plated populations ( Colosimo et al . , 2005 ) . In previous studies , the minimal shared freshwater region was approximately 16 kb , consisting of regions of both the EDA gene and two flanking genes involved in immune functions ( Colosimo et al . , 2005 ) . However further recombination between marine and freshwater haplotypes could narrow this region further , conceivably approaching the size of a single base pair . For example , we have recently surveyed 263 migratory marine sticklebacks from Alaska and identified 12 completely plated individuals that are heterozygous for the T → G change in the EDA enhancer ( minor allele frequency 2 . 3% ) . Analysis of flanking SNPs suggests one of these carriers is heterozygous for a larger characteristic freshwater haplotype , three are heterozygous for a much shorter freshwater haplotype , and eight are heterozygous at the T → G position but are marine-like at other characteristic flanking SNPs tested ( Supplementary file 1 ) . These data show that migratory marine populations can carry freshwater haplotypes of different sizes , including much smaller regions surrounding the key T → G regulatory change . Although most low-plated populations have clearly fixed a multi-kilobase haplotype surrounding EDA , the large size of this haplotype may reflect co-selection for additional phenotypes controlled by the closely linked genes ( Colosimo et al . , 2005 ) . The geographically distant NAKA population is low-plated but shares only the T → G change , either because of an independent mutation , or because of fixation of a tiny fragment of the typical EDA haplotype . The NAKA population may be useful in the future for distinguishing the phenotypic effects of the isolated T → G regulatory change versus the larger EDA haplotype typically found in most low-plated sticklebacks . The absence of a greater range of armor plate mutations at the EDA locus could be due to the relatively high frequency of a preexisting freshwater haplotype , whose frequency in migratory populations exceeds the rate of many spontaneous mutations . Alternatively , the T → G change could represent one of very few possible ways of producing a major change in armor plate patterns while still preserving other functions of the EDA gene . A constrained spectrum of mutations has been observed in other contexts involving very specific phenotypes . For example , nearly all patients with classic achondroplasia contain the same Gly380Arg ( G → C ) substitution in FGFR3 ( Horton et al . , 2007 ) . This Gly380Arg substitution leads to a constitutively active FGF receptor that is thought to confer a selective advantage to spermatogonial cells ( Tiemann-Boege et al . , 2002; Choi et al . , 2008 ) . Identical amino acid substitutions in particular genes also underlie several examples of repeated evolution including insecticide resistance in insects ( GABA ) , tetrodotoxin resistance in snakes ( NaK-ATPase ) , C4 photosynthesis in plants ( PEPC ) , and dark pigmentation in mice and birds ( MC1R ) ( Stern , 2013 ) . These and other well-studied cases typically involve particular amino acid changes that alter protein activity in specific ways , rather than completely ablating protein function . In contrast , few examples are known of identical recurrent base pair mutations in non-coding regulatory sequences ( Martin and Orgogozo , 2013 ) , though multiple cases are now being uncovered in large-scale sequencing surveys of replicate microbial evolution ( Tenaillon et al . , 2012; Blank et al . , 2014 ) . In a recent large-scale study of parallel temperature adaptation over 2000 generations , recurrent use of particular genes was at least 10 times more common than recurrent use of the same base pair changes within those genes ( Tenaillon et al . , 2012 ) . Of the relatively rare recurrent base pair changes , those affecting protein-coding sequence also outnumbered those affecting non-coding intergenic sequence by nearly threefold . The T → G base pair change we have identified near EDA provides a rare example in vertebrates of a particular non-coding base pair change contributing to repeated adaptive evolution . Our experiments also show that Wnt signaling acts upstream of EDA control sequences in armor plate patterning , and that the low-plated SNP reduces Wnt responsiveness of the EDA enhancer ( Figures 5 , 6 ) . Although canonical Wnt signaling typically acts through the β-catenin and Lef transcription factors , the particular T → G base pair change we have identified does not alter a canonical Lef binding sequence . However , Wnt signaling is known to interact with multiple additional signaling and transcription factor pathways , and the T → G change does alter a predicted binding site for c-Jun in the marine sequence ( Newburger and Bulyk , 2009 ) , which can act in collaboration with Wnt signaling in chondrocyte dedifferentiation ( Hwang et al . , 2005 ) , osteopontin promoter activation in mammary cells ( El-Tanani et al . , 2004 ) , and complexes with β-catenin to bind the promoters of Wnt target genes both in mammalian cells and in zebrafish ( Gan et al . , 2008 ) . There are seven base pair positions in the predicted marine c-Jun binding site , and 21 corresponding single bp mutations that could alter one of these bases . 19 of these potential mutations are predicted to eliminate c-Jun binding ( Messeguer et al . , 2002 ) . Of these 19 mutations , the T → G change found in low- plated sticklebacks is the only mutation that is also predicted to create a new overlapping binding site for AP-2α in the low-plated sequence . AP-2α has been shown to inhibit Wnt signaling by complexing with APC/β-catenin ( Li and Dashwood , 2004; Li et al . , 2009b ) . A new binding site for AP-2α could contribute to the reduced Wnt responsiveness of the freshwater EDA gene , or may contribute to other novel expression patterns that are not yet understood ( such as the enhanced cyanoacrylate response we have observed with the T → G mutated enhancer ) . Future experiments are needed to test whether c-Jun , AP-2α or other factors interact directly with the EDA enhancer of either marine or freshwater sticklebacks . However , the simultaneous loss and gain of specific binding sites is a good example of the type of dual molecular constraints that could limit the range of possible base pair substitutions found underlying adaptive regulatory evolution at the EDA locus . Our findings that connect Wnt signaling , plate development , and EDA signaling in sticklebacks also suggest new candidates for trans-acting genetic factors that may modify armor plate number in evolving populations . Previous genetic studies have shown that while the majority of the variance ( >75% ) in armor plate number in stickleback crosses maps to the EDA locus , the remainder of the variance can be explained by multiple plate modifier loci located on other chromosomes ( Colosimo et al . , 2004 ) . Interestingly , two of the three previously mapped armor plate modifier regions contain genes for members of the Wnt pathway: WNT11 on chromosome VII and β-catenin ( CTTNB1 ) on chromosome X . Given the dramatic effects of Wnt signaling on armor plate development and EDA regulation ( Figures 5 , 6 ) , these or other components of the Wnt signaling pathway are strong candidates for additional loci that may contribute to the adaptive fine-tuning of armor plate numbers that is known to occur in many low-plated populations ( Hagen and Gilbertson , 1972 , 1973; Moodie , 1972; Moodie et al . , 1973; Bell and Haglund , 1978 ) .
Allele-specific expression differences were detected using pyrosequencing analysis of F1 hybrid fish as previously described ( Wang and Elbein , 2007 ) . In brief , a marine female from Rabbit Slough , AK was crossed to a freshwater benthic male fish from Paxton Lake , British Columbia to generate F1 hybrids that were heterozygous for a SNP in the EDA gene . Hybrid fish were raised to 13 mm standard length , a stage where the first few armor plates are forming in anterior tissues , but posterior plates have not yet formed . Multiple tissues were dissected , including: first dorsal spine , second dorsal spine , pelvic spines , pectoral fins , caudal fin , dorsal fin , anal fin , premaxilla with oral teeth , lower jaw ( approximately the articular and dentary with oral teeth ) , left anterior flank skin between the second dorsal spine and pelvic spine ( where anterior plates are forming ) , and left posterior flank skin between the dorsal fin and anal fin ( where posterior plates will later form ) . RNA was prepared from dissected tissues using the TRI Reagent Protocol ( Life Technologies , Carlsbad , CA ) . cDNA was synthesized using the Superscript III Supermix ( Life Technologies ) with random hexamer primers . A 183 bp product from the EDA gene was amplified using a biotinylated forward primer 5′-TCCACCAGAAGCGGGATACA-3′ and the reverse primer 5′-TTATGCCCCGGTTATCCTGTG-3′ . Amplified products were sequenced using the primer 5′-TCTCCTCATGACCCTCT-3′ , and the percentage of the two SNP alleles was calculated by EpigenDx , Inc . ( Hopkinton , MA ) . The 16 kb EDA candidate interval from NAKA fish was amplified as several long PCR products and assembled using Sanger sequencing ( GenBank entry KP164994 ) . Alignment of the NAKA sequence with the complete sequence of the EDA region from Salmon River ( SALR ) marine and Paxton Benthic ( PAXB ) freshwater BAC clones ( Colosimo et al . , 2005 ) ; and the Bear Paw Lake ( BEPA ) reference genome ( Jones et al . , 2012 ) ; identified 13 positions where low-plated NAKA , PAXB , and BEPA differed from high-plated SALR fish . Reexamination of these positions in sequence reads from 21 marine and freshwater genomes ( Jones et al . , 2012 ) placed with SAMtools ( Li et al . , 2009a ) against the BEPA reference genome , and resequencing of additional fish , identified the chrIV:12811481 position as shared among all low-plated sticklebacks examined . Population codes and source locations are as previously described ( Colosimo et al . , 2005; Jones et al . , 2012 ) . The region surrounding the T → G base pair change was subsequently amplified from 263 fully plated migratory sticklebacks collected from Rabbit Slough , AK ( RABS ) , using 5′-TTGACAAGTGATGTTCTCTGTGGCC-3′ and 5′-ATGTTGGAGCTGGCAGGAGGAGG-3′ . All heterozygous carrier fish were then tested for the characteristic flanking SNPs previously found to distinguish most high-plated and low-plated haplotypes in previous studies ( Colosimo et al . , 2005 ) . SNPs 5 and 6 at positions 12808303 and 12808630 were determined by amplifying and sequencing a genomic region using 5′-CAGAGGAGGTGAAACCGCACTTACA-3′ and 5′-TGGGAACGCGTCGACATTTGGGA-3′ . SNP 7 at position 12811933 was called from the same genomic amplification used to recover the T → G regulatory change . SNPs 8 and 9 at positions 12813328 and 12813394 were determined by amplifying and sequencing a genomic region using 5′-GTGCCCAGGAGCTCTAGACTTGGC-3′ and 5′-TCTCACATCCGGCAGCGACAAGC-3′ . The plate enhancer region was amplified from genomic DNA of a marine fish from Salmon River , British Columbia using 5′-ATGTGGCCAGATAGGCCACATGGTGTGGGAGAGCAGTGATCG-3′ and 5′-ATGTGGCCTATCTGGCCATGTTGGAGCTGGCAGGAGGAGG-3′ primers that each contain SfiI linkers . The 3 . 2 kb amplified fragment was cloned into the SfiI site of the pT2HE GFP reporter vector ( modified from Kawakami , 2007 ) to generate p3 . 2mar-GFP . Site directed mutagenesis was performed on the p3 . 2mar-GFP plasmid to induce a single freshwater base pair change using two 40 bp overlapping primers 5′-AATTAGTTCCATCTTGAGAGGCAGAGAGAAGATGGTTCCT-3′ and 5′-AGGAACCATCTTCTCTCTGCCTCTCAAGATGGAACTAATT-3′ . A 15-cycle PCR amplification using 50 ng of plasmid , 125 ng of primers , and Phusion polymerase was performed to induce the base pair change ( Zheng et al . , 2004 ) . The resulting plasmid , p3 . 2mar ( T → G ) -GFP , was verified by DNA sequencing . For cell culture experiments , the enhancer inserts from p3 . 2mar-GFP and p3 . 2mar ( T → G ) -GFP were excised from the pT2He plasmid using SfiI and cloned into the XhoI site of the pTA-Luc vector ( Clontech Laboratories , Inc . , Mountain View , CA ) , to generate p3 . 2mar-Luc and p3 . 2mar ( T → G ) -Luc . The β-catenin expression plasmid pRK5-sk-βcatΔGSK was a gift from the Nusse Lab . Transgenic sticklebacks were generated by microinjection of freshly fertilized eggs as previously described ( Chan et al . , 2010 ) . Plasmids were co-injected with Tol2 transposase mRNA as described ( Fisher et al . , 2006; Wada et al . , 2010 ) . Mature Tol2 mRNA was synthesized by in vitro transcription using the mMessage mMachine SP6 kit ( Life Technologies ) . All enhancer assays were performed on high-plated fish derived from Little Campbell River ( British Columbia ) , Bodega Bay ( California ) , or Rabbit Slough ( Alaska ) . Microscopic observation for GFP expression was conducted with a MZFLIII fluorescent microscope ( Leica Microsystems , Bannockburn , IL ) using GFP2 filters and a ProgResCF camera ( Jenoptik AG , Jena , Germany ) to distinguish GFP expression from autofluorescence in pigmented fish . We generated stable lines by making crosses from mosaic founder animals . Two-month-old fish ( 20–24 mm standard length ) were fixed in 4% paraformaldehyde overnight at 4°C , washed , and stored in methanol at −20°C for up to 6 months prior to in situ hybridization . Fish were rehydrated through a series of methanol/water washes ( 90% , 75% , 50% , 25% , 0 ) , bleached in 6% hydrogen peroxide rocking at room temperature for up to 3 hr , and treated with 10 μg/ml Proteinase K in water rocking for 7 . 5 min in order to detect EDA signal . From this point , the RNAscope Brown Protocol was followed with an EDA probe designed by Advanced Cell Diagnostics ( Hayward , CA ) with two procedural modifications: Pretreatment 2 was performed at 40°C and the hybridization step with EDA probe was allowed to proceed overnight ( Wang et al . , 2012; Gross-Thebing et al . , 2014 ) . Affi-Gel Blue Gel beads ( BioRad Laboratories , Inc . , Hercules , CA ) were soaked overnight in PBS , 1 . 2 μg of recombinant human Wnt-3a ( R&D Systems , Minneapolis , MN ) , or 1 . 2 μg of recombinant mouse Dkk-1 ( R&D Systems ) . Marine-derived fish were raised to 20–24 mm standard length ( first four armor plates present ) , anesthetized with Tricaine ( 0 . 017% wt/vol ) , and an average of 12 beads were placed into the flank of each fish , posterior to the apparent plates . Cyanoacrylate glue ( Loctite Super Glue ) was used to close the skin surrounding the implantation site . Fish were allowed to recover for 48 hr before further experimentation or to continue developing into adulthood . The beads' effects on overall plate development were analyzed in live adult fish using 0 . 2% Calcein in aquarium water to mark newly ossified bones as previously described ( Kimmel et al . , 2003; Wada et al . , 2010 ) . HaCaT human keratinocyte cells ( Boukamp et al . , 1988 ) were cultured in DMEM supplemented with 10% FBS , 2 mM L-glutamine and 1% penicillin-streptomycin . Cells were seeded into 24-well plates at a density of 1 × 105 cells/well and transfected after 24 hr . 300 ng of p3 . 2mar-Luc or p3 . 2mar ( T → G ) -Luc plasmids were cotransfected together with 0–100 ng of pRK5-sk-βcatΔGSK using Lipofectamine 2000 ( Life Technologies ) according to manufacturer's protocol . After 6 hr of transfection , cell culture medium was replaced with standard medium supplemented with 2 . 8 mM calcium chloride ( Sigma , St . Louis , MO ) . Cell lysates were collected after 48 hr and assayed using the Dual-Luciferase Reporter Assay System ( Promega , Madison , WI ) according to the manufacturer's instructions . | Stickleback fish develop bony plates on their surface to protect themselves from predators . The extent and pattern of their bony armor depends on their habitat: marine sticklebacks are typically covered from head to tail with bony plates , but freshwater sticklebacks retain only a few plates on their sides . One gene that promotes the formation of the bony plates is called ectodysplasin ( EDA ) . This encodes a signaling protein that is important for the development of the skeleton , skin and many other tissues . Variations in the sequence of this gene are shared among different stickleback populations worldwide . However , it has not been clear which genetic changes can explain how lightly armored freshwater sticklebacks could have evolved from their well-armored marine ancestors on several separate occasions . Here , O'Brown et al . studied EDA in marine and groups of freshwater sticklebacks that have evolved in different locations around the world . The experiments show that the level of expression of EDA in the developing plates and spines is lower in the freshwater fish . O'Brown et al . thought this could be due to genetic changes in regions of EDA that lie outside the region that encodes the protein , so called ‘regulatory elements’ . Indeed , further experiments found that all freshwater fish have a small change in the DNA of a regulatory element that switches on the gene in plate-forming regions of the body . When this change was introduced into marine sticklebacks , the fish had lower levels of gene expression in these plate-forming regions . These findings demonstrate that lightly armored sticklebacks have evolved multiple times from their well-armored marine ancestors through the same small change in their DNA that alters the expression of the EDA gene . The next challenge will be to understand why this particular small change in DNA appears to be favored over all the other changes that could occur in the regulatory element , and to see if factors that act through this regulatory switch also modify armor structures in natural populations . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"chromosomes",
"and",
"gene",
"expression",
"evolutionary",
"biology"
] | 2015 | A recurrent regulatory change underlying altered expression and Wnt response of the stickleback armor plates gene EDA |
Changes in cell proliferation define transitions from tissue growth to physiological homeostasis . In tendons , a highly organized extracellular matrix undergoes significant postnatal expansion to drive growth , but once formed , it appears to undergo little turnover . However , tendon cell activity during growth and homeostatic maintenance is less well defined . Using complementary methods of genetic H2B-GFP pulse-chase labeling and BrdU incorporation in mice , we show significant postnatal tendon cell proliferation , correlating with longitudinal Achilles tendon growth . Around day 21 , there is a transition in cell turnover with a significant decline in proliferation . After this time , we find low amounts of homeostatic tendon cell proliferation from 3 to 20 months . These results demonstrate that tendons harbor significant postnatal mitotic activity , and limited , but detectable activity in adult and aged stages . It also points towards the possibility that the adult tendon harbors resident tendon progenitor populations , which would have important therapeutic implications .
Development , growth , and homeostasis rely on the precise regulation of cell proliferation and differentiation to generate and maintain a functioning organism . Frequent cell divisions grow tissues to the proper size , as do modifications to non-cellular tissue properties such as to the extent of extracellular matrix . However , once size is achieved , each tissue maintains its physiological functionality either through stem cell-mediated mechanisms as in the intestine ( Simons and Clevers , 2011 ) , the duplication of specialized cell types as in the liver ( Miyajima et al . , 2014 ) , or in the virtual absence of cell division as in the heart ( Senyo et al . , 2014 ) . In some cases , the transition from active proliferating to terminally differentiated cells has been attributed to a change in regenerative potential as observed in neonate verses adult mouse hearts ( Senyo et al . , 2014 ) . Therefore , understanding transitions in cell turnover are important for setting the framework for more deeply understanding proliferative-driven growth stages and distinguishing between specific homeostatic renewal mechanisms in the adult . This knowledge is significant in considering therapies for tendon injuries , which can be challenging to treat due to their imperfect healing and propensity for re-injury ( Thomopoulos et al . , 2015 ) . Tendons begin as aggregations of cells that secrete and organize a highly ordered matrix to connect the musculoskeletal system and enable movement . Therefore , tendon growth and maintenance must not only involve its matrix but also the cells that generate and eventually reside within it , making knowledge of transitions in cell division important for understanding these processes . In the adult , the tendon matrix contains organized type I collagen fibrils and tenocytes , which are mature tendon cells possessing cellular extensions that project into the matrix ( Kalson et al . , 2015; Kannus , 2000 ) . This mature stellate morphology differs greatly from that of the rounded shape of embryonic and neonatal tenoblasts . During embryogenesis , limb bud mesenchymal cells express the transcription factor , Scleraxis ( Scx ) , and coalesce into tendon primordia , which organize to connect muscle and bone ( Schweitzer et al . , 2001 ) . Through the transgenic labeling of cell cycle state using the Fluorescent ubiquitination-based cell cycle indicator ( Fucci ) , robust numbers of mitotic tendon cells have been detected prior to birth ( Esteves de Lima et al . , 2014 ) . Specific segments of the limb display more proliferative activity than others ( Huang et al . , 2015 ) , suggesting that there are localized effects on tendon cell proliferation during embryogenesis . In addition to cell growth , these embryonic stages are marked by an increase in the number of collagen fibrils deposited in the matrix ( Kalson et al . , 2015 ) . These collagen fibrils grow in length and diameter to grow the tissue ( Ezura et al . , 2000 ) . Scanning electron microscopy at postnatal stages ( P0 and P42 ) has shown an increase in the diameter of the collagen fibrils rather than an increase in collagen fibril or cell number in the tail tendons of mice ( Kalson et al . , 2015 ) . These observations have led to a model whereby tendon postnatal growth is primarily driven by expansion of the extracellular matrix ( ECM ) , which results in a reduction in cell density across the whole tissue in growth and aging ( Dunkman et al . , 2013; Kalson et al . , 2015 ) . However , a direct analysis of cell turnover and the transition in proliferative activity from birth to adult and aged stages has not been performed . Although adult tendons display limited proliferation , some cell division has been detected in vitro and in vivo , especially in the context of injury . Tendon-derived stem/progenitor cells were characterized based on their ex vivo abilities to proliferate , clonally expand , and undergo serial transplantation ( Bi et al . , 2007 ) . However , the identity and in vivo activity of the resident cell population remains unknown . Other studies have reported proliferation in adult tendons during homeostasis and repair ( Lindsay and Birch , 1964; Runesson et al . , 2013; Tan et al . , 2013 ) . Because these studies were performed without genetic lineage tracing tools , the origin of the cells proliferating in response to injury was unclear . Recent lineage tracing experiments have shown that in multiple cell-lineages , including alpha-Smooth muscle actin ( Acta2 ) , Scx , and S100a4 lineage cells , can contribute to the healing tissue depending on the tendon and injury type , suggesting these populations may retain proliferative abilities in adult tendons ( Best and Loiselle , 2019; Dyment et al . , 2014; Dyment et al . , 2013; Howell et al . , 2017; Sakabe et al . , 2018 ) . Interestingly , in aging , tendon cell number per unit area decreases , suggesting declining proliferative abilities with age ( Dunkman et al . , 2013 ) . Consistent with this interpretation , tendon-derived cells from aged human tendons have reduced proliferative abilities and increased markers of senescence compared with adult-derived tendon cells in culture ( Kohler et al . , 2013 ) . Together , these studies indirectly indicate that adult and aged tendon cells have reduced proliferative activity , yet sub-populations of tendon cells may divide in adult tendons under specific injury conditions . Nevertheless , it remains unclear to what extent , if any , there is physiological cell turnover in the adult tendon without injury and how this may differ from cell turnover during periods of active tendon growth . Therefore , we sought to examine cell turnover rates in limb tendons during growth , adulthood , and aging . Using complementary methods of genetic pulse-chase labeling to trace the cell division history and BrdU/EdU incorporation to detect proliferation , we were able to identify changes in tendon cell turnover from birth to the early juvenile period ( beginning around 3–4 weeks ) , with comparisons to adult and aged stages ( ≥3 months and ≥18 months , respectively ) . We detect relatively high levels of proliferation during the neonatal period ( P0-P7 ) and a rapid decline by P21 . Although proliferation was significantly reduced after one month of age , surprisingly we were able to identify a small population of tendon cells that continued to proliferate in adult and aged mice , albeit at a very low rate . Understanding which cell populations can continue to divide in adults and the mechanisms driving the switch from proliferative to more quiescent stages would greatly benefit clinical approaches to tendon injuries .
To characterize cell proliferation in the tendon , we used the doxycycline ( Dox ) inducible Histone 2B-green fluorescent protein reporter mouse model ( Col1a1-tetO-H2B-GFP; ROSA-rtTA , henceforth referred to as H2B-GFP ) , which has been used to quantify cell proliferation and identify slowly cycling label-retaining cell populations based on the stability and dilution of H2B-GFP protein in each cell ( Chakkalakal et al . , 2014; Foudi et al . , 2009 ) . After H2B-GFP expression is induced by Dox addition , Dox is removed for the chase period and H2B-GFP protein becomes diluted in proportion with each subsequent cell division ( Figure 1A ) . Therefore , cells cycling more frequently will dilute H2B-GFP protein more quickly and will appear unlabeled earlier in the chase period; more slowly cycling cells will retain H2B-GFP protein longer during the chase . To verify the H2B-GFP system worked efficiently in the tendon , we pulsed mice with Dox from embryonic stage ( E ) 10 to birth and examined H2B-GFP expression on postnatal day ( P ) 0 . Two photon microscopy images of histological sections of pulsed P0 Achilles ( Figure 1B–B’’ ) tendons showed widespread expression of H2B-GFP throughout the Hoechst+ tendon nuclei . We next confirmed by flow cytometry that more than 90% of the tendon cells were positive for H2B-GFP at P0 ( Figure 1D ) , indicating efficient labeling of all tendon cell populations examined . To ensure we were enriching for tendon cells , we only analyzed cells from dissected tendon tissues that were negative for CD45 and CD31 to remove blood and endothelial cells , respectively ( Figure 1—figure supplement 1A ) . We also found the background level of H2B-GFP expression without the addition of Dox was very low ( <1%; Figure 1—figure supplement 1B ) . At all stages analyzed , tendon cells were isolated from extensor , deep and superficial flexor , and Achilles tendons in the hindlimbs and extensor , deep and superficial flexor tendons in the forelimbs . To determine the total cumulative proliferation of tendon cells from birth to aged mice , we next examined H2B-GFP expression in tendons that had been pulsed with Dox at embryonic stages and allowed to chase without Dox for over 18 months using section and FACS analysis . Tendons in section imaged using 2-photon microscopy appeared to have reduced H2B-GFP+ expression in Hoechst+ nuclei ( Figure 1C–C’’ ) compared to P0 tendons ( Figure 1B–B’’ ) . We found that H2B-GFP+ cells had shifted in the intensity of GFP ( Figure 1E ) with only 20 . 1 ± 1 . 4% of the cells H2B-GFP+ at 645 days ( Figure 2A ) . Previous studies calculate that 7-8 divisions are needed for a cell to fall below the GFP detection threshold ( Foudi et al . , 2009 ) . This would indicate that the H2B-GFP+ population at 645 days proliferated less than 7-8 times , while the H2B-GFP– population proliferated at a minimum of 7-8 times since birth . Together , these data show that all tendon cells proliferate after birth , but that a subpopulation of the cells display limited proliferative activity . To more deeply assess the dynamic changes of tendon cell proliferation after birth , we analyzed Dox pulsed mice at multiple stages of chase from P0 to P80 . We observed a decrease in the total percentage of H2B-GFP+ cells from 93 . 6 ± 1 . 2% at P0 to 76 . 7 ± 9 . 6% at P7% and 52 . 0% at P14 ( Figure 2A ) . However , the percentage of H2B-GFP+ tendon cells remained relatively constant between P14 and P80 with no significant differences among any pair of time points ( Figure 2A ) , suggesting limited cell division occurred from P21-P80 . Interestingly , the percentage of H2B-GFP+ cells was further reduced at 645 days to 20 . 1 ± 1 . 4% indicating low but detectable amounts of cell proliferation continue in adult and aged mice ( Figure 2A , B ) . As significant changes in the percentage of positive and negative H2B-GFP cells between P14-P80 were not observed , we next examined alterations in H2B-GFP intensity , as this would reveal more subtle changes in cell division that occur . We noted a marked shift in the H2B-GFP+ intensity from 105 at P0 to 103 after 645 days ( Figures 1E and 2C ) . Using a logarithmic decay equation to define the dilution of GFP signal mathematically ( see Materials and methods ) , we also observed increased proliferation at early stages ( Figure 2D ) . Our calculations show that tendon cells were dividing at a rate of 19 ± 4 . 2% per day from P0 to P7 and 9 ± 4 . 25% per day from P7 to P14 ( Figure 2D ) . Proliferation rates decreased to 3 . 85 ± 0 . 07% per day between P14 and P21 and 1 . 75 ± 0 . 64% per day from P21 to P80 . After P80 , the rate of tendon cell proliferation was markedly decreased to 0 . 1 ± 0 . 13% per day by P600 ( Figure 2D ) . Together , these proliferation rates derived from mathematical modeling of H2B-GFP decay and the absolute loss of H2B-GFP intensity over time from our flow cytometry analysis indicate that there are relatively high levels of proliferation at the early postnatal stages . In addition , this proliferative activity is greatly diminished after one month of age , but not extinguished in adult or aged tendons . To complement our mathematical model of H2B-GFP decay , we used flow cytometry to quantify the percentage of tendon cells that had incorporated Bromodeoxyuridine ( BrdU ) , a thymidine analog that incorporates into replicating DNA , for different BrdU administration lengths and stages . We performed intraperitoneal ( IP ) injection of BrdU and harvested tendons to determine the number of BrdU+ cells after 24 hours . For flow cytometry analysis , highly proliferative organs ( gastrocnemius muscle ) were used as positive controls , tendon tissues from mice that had not received BrdU treatment were used as negative controls , and Scleraxis ( Scx ) -Cre;Rosa-LSL-TdTomato+ ( abbreviated Scx-Cre;TdTom ) mice were used to analyze Scx-descendent tendon cells ( Blitz et al . , 2009 ) . We found that BrdU injection at P0 resulted in 76 ± 13 . 8% BrdU+ tendon cells at P1 , while at P8 and P22 , 26 ± 6 . 5% and 8 . 4 ± 3 . 4% of the tendon cells were BrdU+ positive , respectively ( Figure 3A , B ) . In adult mice , we observed that less than 1% of the cells were BrdU+ ( P60 = 0 . 4 ± 0 . 2% , P370 = 0 . 5 ± 0 . 1% ) . To verify these findings in tissue sections , we injected EdU at P1 and P59 and examined Scx-Cre;TdTom+ and EdU+ tendon cells in section one day later . Consistent with our BrdU and H2B-GFP results , we observed more Scx-Cre;TdTom+/EdU+ cells in the Achilles tendon at P2 compared with P60 mice ( Figure 3C ) . Interestingly , we also observed noticeable doublets of EdU+ cells in rows along on the longitudinal axis of the Achilles tendon ( Figure 3C , B’ ) . This indicates that the cells divided and retained their relative position in channels along the long axis of the tendon and is consistent with prior work noting an increase in cell number along the longitudinal axis at postnatal stages ( Kalson et al . , 2015 ) . These results show a high rate of proliferation immediately following birth , and a decrease in the first weeks of postnatal life , specifically after P21 , which is consistent with our H2B-GFP mathematical model . However , the low percentage of BrdU+ cells at P60 and P370 suggests minimal turnover in adult tendons . To more accurately quantify the amount of cell proliferation in adults , we administered BrdU continuously in the drinking water of Scx-GFP;Scx-Cre;TdTom mice for 90 to 100 days . We found that after long periods of BrdU administration , 4 month old mice had incorporated BrdU into 2 . 35 ± 1 . 2% of the Scx-Cre;TdTom+ cells and 2 . 75 ± 2 . 9% of the Scx-GFP+ cells ( Figure 3D , E; Figure 3—figure supplement 1 ) , using flow cytometry . Quantification of BrdU stained tendon sections at 3 and 6 months showed a similar percentage of BrdU+ cells ( Figure 3F ) , further supporting a low , but detectable rate of turnover in adult mouse tendons . We also observed a significantly greater number of cells incorporating BrdU in the outer compared with internal tendon regions ( Figure 3F ) . In mice older than 1 year of age , 90 days administration of BrdU yielded 0 . 48 ± 0 . 26% of BrdU+ tendon cells ( Figure 3D ) , however , this decrease was not statistically significant between 4 , 6 and 13 month stages . Since we have determined that there is a transition in cell division rate during the first postnatal month , we also predict that there are dynamic gene expression changes occurring during this period , especially for genes important for proliferation and matrix production . We performed RT-qPCR assays on RNA isolated from whole distal limb tendon homogenate for a small set of transcripts . These assays provide further information about cell proliferation ( Mki67 ) , tendon cell identity and differentiation ( Scx , Mkx ) , and matrix production and assembly ( Col1a2 , Col3a1 , Fmod ) , during tendon growth . An analysis of variance ( ANOVA ) on ∆CT values for each gene demonstrated a significant change in expression of all genes across the developmental range ( p<0 . 05 ) . Tukey’s Honestly Significant Difference ( HSD ) post hoc tests revealed the specific pairs of time points for which relative expression is significantly different ( see Supplementary file 1 ) . For many of the genes , relative expression levels decreased during the first month of age . Although KI-67 protein expression is commonly used as a marker of proliferating cells , Mki67 mRNA expression has been shown to correlate with protein levels and the number of KI-67 positive cells seen in histological sections ( Prihantono et al . , 2017 . ; Schleifman et al . , 2014 ) . Based on this , we examined Mki67 transcript levels as another independent way to assess the number of mitotically active cells . Mki67 gene expression was highest during the first week after birth ( P0 to P7 ) , and no significant differences were observed between P0 , P7 , and P14 ( all p>0 . 8; Figure 4; Supplementary file 1 ) . By P21 , however , the relative amount of Mki67 mRNA present in the tendon became significantly reduced compared to earlier timepoints ( P0 , P7 , and P14 , all p<0 . 05; Supplementary file 1 ) and remained low throughout the rest of the time series . By P35 , Mki67 expression levels approached the lower limit of detection for our RT-qPCR assays ( CT values ~35 ) . Therefore , these results suggest that the number of proliferating cells is highest during the first week after birth , but by P35 most tendon cells are no longer mitotically active . The expression of Scx , Mkx , Col1a2 , and Col3a1 measured via RT-qPCR also decreased by P35 compared with P0 , while Scx alone shows significantly increased expression at P14 relative to birth and later stages ( Figure 4; Supplementary file 1 ) . Fmod expression follows a different pattern , however , with higher transcript measurements at all timepoints from P7 to P28 compared to P0; however , none of these differences in Fmod expression achieved statistical significance during post hoc testing ( Figure 4; Supplementary file 1 ) . To understand how tendon cell number changes relative to matrix expansion during growth , we also quantified tendon cell density during the first postnatal month . Using 2-photon microscopy and second harmonic generation ( SHG ) imaging to generate 3D images of Scx-GFP+ Achilles tendons , we counted Hoechst+ cells and examined collagen organization at P0 , P7 , P14 , and P28 ( Figure 5A–D’’’; Videos 1–5 ) . As has been previously reported ( Kalson et al . , 2015 ) , we observed a decrease in cell density per unit area , with 42 . 3 ± 11 . 44 , 26 ± 4 . 6 , 21 . 8 ± 3 . 3 , and 9 . 6 ± 1 . 9 cells per 50 mm x 50 mm at P0 , P7 , P14 , and P28 , respectively ( Figure 5E ) . These results suggest that matrix expansion outpaces cell proliferation , at least for the cross-sectional area of the tendon . Consistent with this observation , using the same imaging conditions , we observed an increase in SHG signal intensity from P0 to P28 , suggesting an increase in collagen density at these stages ( Figure 5A’’’–D’’’ ) . We also noticed a larger variability in cell density at P0 compared to P28 , which could indicate natural variability in growth rates during early stages . To understand how cell division compares with longitudinal tendon growth , we measured the Achilles tendon length from the enthesis to its connection with the gastrocnemius muscle at postnatal and adult stages . Strikingly , we observed rapid growth in the early postnatal stages with the Achilles tendon increasing from 0 . 127 ± 0 . 019 cm at P0 to 0 . 35 ± 0 . 017 cm at P14 , and to 0 . 436 ± 0 . 018 cm at P21 . However , the length of the Achilles tendon did not change significantly between P21 to P28 ( Figure 5B ) , and only increased modestly from 0 . 496 ± 0 . 01 cm at P28 to 0 . 54 ± 0 . 007 cm after P30 ( P30-P270 ) ( Figure 5B ) . Overall , the time periods where we observed significant increases in Achilles tendon length correspond directly with our observations of periods of active tendon cell proliferation . This suggests the interesting possibility that , in parallel with matrix expansion , cell proliferation during the first two weeks after birth may in some way contribute to longitudinal growth or result from mechanical or chemical changes that occur during this dynamic longitudinal growth period .
Defining the transition from developmental growth to adult homeostasis is important for understanding functional tissue physiology . Adult tissues range from high self-renewal activity driven by stem cell populations , such as in the blood and intestine , to low or even no self-renewal as has been reported for the liver and heart , respectively . The tendon presents an intriguing case as growth and maintenance involve both its highly organized matrix and the cells that reside within it . Many studies have highlighted the changes that tendon matrix undergoes in growth , adulthood , and aging . However , the activity of the cells as the matrix transitions from growth to maturation is less well understood . Previous work has suggested that cell proliferation in adult tendons is limited ( Runesson et al . , 2013 ) , but it is unclear when and to what extent this decline in cell division occurs . Identifying the shift from proliferative growth to homeostasis is also important for properly defining cellular growth periods and understanding self-renewal mechanisms in the adult tendon . During postnatal development , the tendon ECM undergoes increases in collagen fibril diameter , collagen content , and mechanical properties ( Ansorge et al . , 2011 ) . In our study , we sought to define the changes that occur to the cells within the tendon during the same periods of growth and homeostasis . Using the H2B-GFP system and BrdU/EdU labeling , we detected significant cell proliferation prior to one month of age . In addition to loss of H2B-GFP , the intensity of H2B-GFP expression decreased demonstrating that all tendon cells divide at least once during postnatal life . The decrease in H2B-GFP intensity across the time series is best described by a logarithmic decay model , which yields proliferation rates similar to those measured via BrdU labeling . Although there were some discrepancies between BrdU labeling and our H2B-GFP mathematical model at P21 , these differences were modest and could be attributed to differences in BrdU incorporation into the tendon , a low level ( <1% ) leakiness of the H2B-GFP system ( Figure 1—figure supplement 1B ) , or a potential difference in cell populations as we examined Scx-lineage ( Scx-Cre;TdTom+ ) cells with BrdU and total tendon cells negatively sorted for CD31/CD45 with H2B-GFP . Despite the potential drawbacks from each method , we obtained similar results from these complementary approaches further strengthening our conclusions . Future analysis focusing on specific sub-populations of cells residing in the tendon –such as macrophages , the sheath or epitenon cells , tendon-derived stem/progenitor cells ( Bi et al . , 2007 ) , the S100a4-expressing population ( Best and Loiselle , 2019 ) , the αSMA-expressing cells ( Dyment et al . , 2014 ) , or through the use of extracellular matrix reporters ( reviewed in Delgado Caceres et al . , 2018 ) – would further refine this analysis and determine if the mitotic potential in the growing and adult tendon is restricted to specific populations of tendon cells . In relation to the growth of the tissue and consistent with others ( Dunkman et al . , 2013; Kalson et al . , 2015 ) , we have observed decreased cell density and an increase in collagen organization as observed by second harmonic generation signal in postnatal Achilles tendons as the mice mature from P0 to P28 . This corresponds with rapid elongation of the Achilles tendon in the early postnatal stages with little change occurring from P30 to adulthood . Taken together , our analyses show there is significant proliferation even as the tendon cells are reduced in density . Although this indicates that matrix expansion outpaces cell growth , it also points towards a possible co-regulation of proliferation and matrix expansion during early postnatal stages , which could have interesting implications for how cells regulate , or respond to , ECM expansion and changes in biochemical and mechanical signals . Our gene expression analysis also demonstrates interesting changes in the first month of this transition from growth to homeostasis . Relative expression of Mki67 is significantly downregulated by P21 compared to the earlier time points . In later stages of the time series Mki67 transcripts are reduced to nearly undetectable levels ( CT ~35 ) . Concurrently , the relative expression patterns of tendon transcription factors ( Scx and Mkx ) , and pro-collagen genes ( Col1a2 and Col3a1 ) largely match that of Mki67 . Both Scx ( Murchison et al . , 2007; Schweitzer et al . , 2001; Shukunami et al . , 2006 ) and Mkx ( Ito et al . , 2010; Liu et al . , 2014 ) are involved in tenocyte differentiation , as well as matrix organization via interactions with Smad3 ( Berthet et al . , 2013 ) . Our findings on the coordinated downregulation of Scx , Mkx , Col1a2 , and Col3a1 after P14 fits within this established framework and suggests that the period from P0 to P14 is a key window of postnatal tendon development . The persistence of Fmod expression is concordant with previous studies of ECM proteins during the postnatal period in mice ( Ezura et al . , 2000 ) , indicating that collagen fibril formation slows early , but fibril growth , mediated by Fmod , continues into the juvenile period ( >1 month ) . Although our gene expression analysis is consistent with previous studies , this work was limited to a handful of tendon genes and an unbiased next generation sequencing method such as RNA-seq would provide a more comprehensive analysis of gene expression during this postnatal transition . The tendon has also been shown to undergo regenerative healing during fetal and early postnatal periods ( Ansorge et al . , 2011; Favata et al . , 2006; Howell et al . , 2017 ) . The timing in these studies is reminiscent of the other organ systems such as the heart ( Bassat et al . , 2017 ) , which demonstrate more regenerative potential at neonatal compared to adult stages . In mice , tendons injured prior to one week of life undergo regenerative healing , with mechanical properties of the healed tendon nearly matching those of the uninjured controls; injured tendons of mice older than 3 weeks of age healed imperfectly through scar formation ( Ansorge et al . , 2011; Howell et al . , 2017 ) . Previous work has also demonstrated that the regenerative abilities of injured fetal sheep tendons are not affected by transplantation into an adult environment ( Favata et al . , 2006 ) , suggesting that the regenerative properties of developing tendons are intrinsic . Interestingly , neonatal cardiac regeneration has been attributed to the ability of cardiomyocytes to proliferate during the first 1–2 weeks of postnatal life ( Bassat et al . , 2017 ) . It is interesting to speculate that the swift decline in tendon cell proliferation that we observed at 3 weeks of age may also underlie the shift in regenerative to reparative healing in the tendon . In addition to defining distinct postnatal periods of cell proliferation , our work also establishes the presence of cell division in tendon cells at adult and aged stages . One caveat of our study is that although mice over 18 months are considered "aged" , age-related tendon phenotypes are not observed until after 22 months ( Ackerman et al . , 2017 ) . It would be interesting to perform our long BrdU administrations at these stages to determine if significant differences in mitotic activity are observed . Despite the low levels of proliferation , we detected BrdU incorporation in both Scx-lineage and Scx-GFP+ cells in adults . Although our current understanding of the self-renewal mechanisms in the tendon are limited , studies have shown that tendon-derived stem/progenitor cells divide readily and are multipotent when isolated and expanded in culture ( Bi et al . , 2007 ) . These cells can also form tendon-like tissues upon transplantation ( Bi et al . , 2007 ) , but how this activity reflects that of resident cells in their native environment is unclear . In the context of injury , recent studies have shown contributions to the healing tissue from Scx-GFP-negative cells originating from tendon sheath regions ( Dyment et al . , 2014; Wang et al . , 2017 ) . These results are consistent with earlier studies showing that external tendon cell populations exhibited increased proliferative capacity compared with internal tendon cells in culture ( Banes et al . , 1988 ) . Other studies have shown contributions from Scx- or S100a4-lineage cells to the bridging tendon tissue ( Best and Loiselle , 2019 ) , which would suggest that tendon cells within the tendon body are also capable of proliferating . However , it is unknown if these previously identified cells are responsible for the homeostatic proliferation detected in adults . A combined study using genetic lineage tracing of sub-populations of tendon cells with the H2B-GFP or BrdU labeling system would address this question . In examining human tendon tissue , studies have used Carbon-14 ( C14 ) isotope analysis to infer human tendon turnover rates based on of known changes in atmospheric C14 levels originating from atomic bomb tests . These studies show that the majority of the tendon core mass is formed by adolescence ( Heinemeier et al . , 2013 ) . Consistent with this previous study , our results show that most cell division in the tendon occurs prior to the juvenile stage . However , our work also indicates continued low levels of proliferative activity in adults . As the previous C14 studies were performed with tissue samples , which are predominantly matrix , it is unclear , as the authors also note , if they could detect low rates of turnover by a small population of cells . Therefore , even though the C14 results indicate very little tissue turnover after adolescence , they do not exclude the possibility of a slowly cycling tendon cell population in humans . Interestingly , further C-14 analysis of collagen isolated from tendinopathy samples showed evidence of collagen turnover after adolescent periods ( Heinemeier et al . , 2018 ) . Although it is unclear if the collagen turnover is a cause or effect of the tendinopathy , these results suggest that adult tendon cells can be actived at adult stages to remodel their matrix significantly . In summary , by using complementary genetic and chemical labeling methods , we have gained a comprehensive understanding of the dynamic cell proliferation rates in the tendon from birth to aging . We show that limb tendon cells remain proliferative throughout early postnatal stages ( P0-P21 ) and that mitotic activity declines significantly in juvenile periods with a small population of cells continuing to divide from one month and 1–2 years of life . The timing of these changes in cell turnover appears to be correlated with the timing at which the tendon matrix is undergoing expansion and maturation ( Figure 6 ) , as well as when the tendon cells are changing morphology from rounded to stellate . These changes in cell division also correlate with the transition from regenerative to reparative healing that has been documented in murine tendons . These findings are important to consider in studying tendon growth , maturation and self-renewal mechanisms , and have implications for identifying and characterizing self-renewal mechanisms of a tissue .
We thank Andrew Brack ( UCSF ) and Konrad Hochedlinger ( MGH ) for the Doxycycline ( Dox ) inducible H2B-GFP ( Col1a1:tetO-H2B-GFP; ROSA:rtTA ) heterozygous mice used in these studies . To induce transgene expression , Dox ( Sigma D9891 , 2 mg/ml , supplemented with sucrose at 10 mg/ml ) was added to the drinking water of timed pregnant females at E10-birth as described ( Foudi et al . , 2009 ) . Scx-GFP and Scx-Cre mice were provided by the Schweitzer lab ( Blitz et al . , 2009; Schweitzer et al . , 2001 ) . Gt ( ROSA ) 26Sortm9 ( CAG-tdTomato ) Hze ( Ai9 ) were obtained from Jackson Laboratory ( Jax cat# 007909 ) . All experiments were performed according to our protocol approved by the Massachusetts General Hospital Institutional Animal Care and Use Committee ( IACUC: 2013N000062 ) . The tendon cells were isolated from the distal forelimb and hindlimb tendon tissue ( Achilles , extensor , deep and superficial flexor tendons ) from mice at time points between P0 and 2 years . Limb tendons were enzymatically dissociated in a solution containing 0 . 2% collagenase II ( Worthington Cat# LS004176 ) in DMEM ( Gibco Cat#11956–092 ) with 1% P/S ( Corning Cat#30002 CL ) and 1% Hepes ( Gibco Cat#15630–80 ) for 2 hr at 37°C . Subsequently , a secondary digestion solution containing 0 . 2% Collagenase I ( Gibco Cat# 17100–017 ) and 0 . 4% Dispase ( Gibco Cat# 1710541 ) was added and the samples were incubated for an additional for 30 min at 37°C . The digested cells were filtered with 30 µm filters ( MACS Cat# 130041407 ) and washed . For the H2B-GFP+ studies , we enriched for tendon cells from H2B-GFP+ mice by excluding for CD31+ and CD45+ cells using FACS prior to analysis ( BD Cat#551262 , Cat# 557659 ) . For the BrdU analysis , cells were stained with anti-BrdU following tendon tissue dissociation ( Biolegend , Cat# 339808 ) , and tendons from Scx-Cre;TdTom+; or Scx-Cre;TdTom+;Scx-GFP+ mice were used to analyze TdTom+ or GFP+ tendon cells . Flow cytometry was performed using 5 ml tubes ( BD Biosciences Cat# 352235 ) on a FACSAria II ( BD Biosciences ) . For each independent experiment , gates were defined by positive and negative control tendon cells from TdTom+/TdTom- and GFP+/GFP- cells . For the negative controls for BrdU analysis , BrdU antibody staining was performed on tendon cells isolated from mice that were not administered BrdU . To ensure reproducibility of H2B-GFP emission intensity between different samples and sorting times , the voltage of the photomultiplier receiving signal from the 488 nm laser was normalized using Green Flow Cytometry Reference Beads prior to every sort ( Molecular Probes Cat# C16508 ) . BrdU was injected at a concentration of 150 mg/kg ( Sigma Cat#B5002 ) as described ( Magavi and Macklis , 2008 ) . Flow cytometry analysis was performed as described previously . For BrdU immunostaining , sections underwent antigen retrieval and immunostaining using anti-BrdU ( 1:100; Abcam Cat# 6326 ) . EdU was administered at 20 mg/kg as described ( Salic and Mitchison , 2008 ) and tendon sections were stained using the Click-iT EdU kit ( Invitrogen Cat# C10337 ) . For histological sections , tendons were fixed overnight in 4% PFA , followed by 5% sucrose for 1 hr , and 30% sucrose overnight before being mounted in OCT . A Leica cryostat ( CM3050S ) was used to obtain 8–10 μm sections . Pictures were taken with Zeiss AxioImager D2 with ( 10X and 20X magnification ) and prepared using Adobe Photoshop and Illustrator . For 2-photon imaging of H2B-GFP expression in mouse Achilles tendons ( Figure 1 ) , we chose 2-photon microscopy due to the longer wavelengths used to image the sample and the high collagen content of the tendon . The longer wavelengths penetrated the tendon tissue more efficiently and produced a better signal to noise ratio that was not affected by the collagen fibers , resulting in a more unified GFP signal . Achilles tendon samples from at least three mice were sectioned and analyzed at P0 ( end of pulse ) and 680 days ( end of chase ) . For the P0-P28 data ( Figure 5 ) , at least 3 Achilles tendons from different mice were analyzed . For all 2-photon imaging , we stained the nucleus with Hoechst 33258 ( ThermoFisher Cat# H3569 ) at 1:100 , 000 dilution . To standardize signal detection between sections and samples , the laser power was adjusted to ‘Bright Z’ mode . The images were analyzed using FIJI ( Schindelin et al . , 2012 ) . For P28 transverse Scx-GFP Achilles tendon Video 5 , the ‘reslice’ feature in the FIJI software was used to convert the image from the sagittal to transverse view ( Schindelin et al . , 2012 ) . The images were taken with optical slices every 0 . 4 µm with a 25X wet lens ( XLPlan N 25X WMP ) on an Olympus 2P microscope FLOVIEW FVMPE-RS . Tendons from at least three mice were measured per stage from the calcaneus to the gastrocnemius muscle , and the data were analyzed using Prism software ( Graphpad ) . For BrdU incorporation quantification ( Figure 3F ) , we examined 3–4 tendon transverse sections in 3–4 regions along the tendon per mouse from at least three mice . In these transverse sections , we counted BrdU in the outer tendon regions within the outer 20 µm of the tendon and the inner regions comprised the regions internal from this area . The data points in Figure 3F represent the ratio of BrdU+/TdTom+ cells in the inner or outer regions of 3–4 sections from one region along the tendon . For cell density counting , we counted cells in three 50 µm x 50 µm squares in 7–9 optical transverse sections per Achilles tendon per mouse with a gap of 20 microns between sections and at least three mice were analyzed per stage . The 50 µm x 50 µm square was created using FIJI ( Schindelin et al . , 2012 ) , and in each optical section the nucleus number was counted in at least three different locations . To define the dilution of the GFP signal mathematically , we modeled the change in signal intensity using a logarithmic decay equation:P ( t ) =P ( 0 ) e−kt In this formula , we assume that GFP signal intensity decreases through dilution by cell proliferation . We calculated the constant between populations at different times ( k ) , by comparing the populations’ median GFP intensity at particular times ( P ( t ) ) . Assuming that the increase in tendon cell number could be measured by the decrease of the GFP intensity ( Figure 2B and D ) , we calculated the dilution of GFP between each time point from P0 to 645 days ( Figure 2C ) . Fresh , whole limb tendons ( pooled forelimb and hindlimb from a single individual; n = 3 mice per time point ) were dissected from mice euthanized via CO2 and immediately placed in cold TRIzol ( Invitrogen 15596026 ) . Tendons were roughly chopped with clean microdissection scissors in TRIzol and frozen at −80C until RNA extraction via TRIzol-chloroform and a proprietary kit . Briefly , the homogenate in TRIzol was thawed on ice , vortexed , and transferred to a clean microcentrifuge tube to remove tissue debris . The traditional TRIzol-chloroform extraction protocol was followed until phase separation . An equal volume of ethanol was added to the upper aqueous phase and the mixture was transferred to a Zymo IIC spin column ( Zymo Research C1011 ) for purification and DNase I treatment using the Zymo Direct-Zol system ( Zymo Research R2050 , R2060 ) following the manufacturer’s guidelines . RNA quality was examined using spectrophotometry ( NanoDrop 2000c , Thermo Scientific ) and capillary electrophoresis ( 2100 Bioanalyzer , Agilent ) , and concentration was measured via fluorometric quantitation ( Qubit HS RNA assay , Invitrogen Q32852 ) . The final RNA product was stored at −80C . Total RNA was reverse transcribed using the SuperScript IV first strand synthesis system ( Thermo Fisher 18091050 ) . 100 ng total RNA for each sample ( n = 3 per time point ) was converted to cDNA using oligo ( dT ) 20 primers . SYBR green assays ( Applied Biosystems 4367659 ) were run in technical triplicate with 1 ng of cDNA template in each 12 . 5 µl reaction . Samples were amplified for 40 cycles using the LightCycler 480 II real time PCR system ( Roche Diagnostics ) . All targets were normalized to Gapdh ( see Supplementary file 2 for primer sequences ) . Relative expression values were calculated for visualization using the ∆CT method ( Livak and Schmittgen , 2001 ) ; statistics were performed on ∆CT values ( Supplementary file 1 ) . All self-designed primers were designed using PrimerBLAST ( Ye et al . , 2012 ) . For the RT-qPCR assays , statistical differences among the six timepoints were investigated via ANOVA and post hoc pairwise comparisons were computed using Tukey’s Honestly Significant Difference test on the ∆CT values ( n = 3 biological replicates per time point; alpha = 0 . 05 ) . R statistical software ( R Development Core Team , 2018 ) was used for all RT-qPCR calculations and visualizations . Data analysis in R was facilitated using R packages included in the Tidyverse collection ( Wickham , 2017 ) and statistical analysis was performed using the implementations of ANOVA and Tukey’s HSD in 'stats' version 3 . 5 . 1 ( R Development Core Team , 2018 ) . For each stage analyzed by flow cytometry , least three mice were used per group . Statistical differences between time points for all flow cytometry analysis were calculated using a Welch’s t-test . One-way ANOVA was used to calculate statistical differences in cell density . A two-tailed t-test was used to calculate significance between BrdU incorporation in outer and inner tendon regions and Achilles tendon length measurements between different stages . | Muscles are anchored to bones via fibrous structures called tendons , which are made from specialized cells and large proteins called collagens . Tendons in newborns have round cells that can easily divide to make new cells , allowing the tissue to grow . Compared to newborns , tendon cells in adults have a star-like shape and appear to have stopped dividing . Tendon cells in adults are less abundant than in newborns , with more of the tendon made up of collagen . Adult tendons also do not heal as well as young tendons following an injury . However , it was previously unknown when the characteristics of young tendons are lost . Additionally , it was unclear whether adult tendon cells completely stop dividing or simply do so more slowly , or if these changes in cell division are what causes adult tendons to heal less easily . To investigate this , Grinstein et al . used microscopy and cellular and molecular tools to examine tendon cell division in mice of different ages . The results indicated that tendon cells continue to divide after birth , but they do so more slowly as mice age . The researchers saw the most significant changes in the mice after they reached two weeks of age , which is also when the structure of the tendons begins to transition to that of an adult . Young mice , which are growing and learning to move , have tendons that produce new cells faster than adult tendons . This may be necessary for young tendons to grow and adapt to the strains of movement . As tendons age , most of their cells lose the ability to divide , which seems to prevent the tendons from healing fully after injury . Notably , Grinstein et al . found that some cells are still able to divide slowly in adult tendons , which may be important for healing . Further research is needed to examine the mechanisms involved in these changes and the factors that drive them , but a deeper understanding of tendon biology could lead to new therapies for treating tendon injuries . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"stem",
"cells",
"and",
"regenerative",
"medicine",
"developmental",
"biology"
] | 2019 | A distinct transition from cell growth to physiological homeostasis in the tendon |
Recent longitudinal neuroimaging studies in patients with electroconvulsive therapy ( ECT ) suggest local effects of electric stimulation ( lateralized ) occur in tandem with global seizure activity ( generalized ) . We used electric field ( EF ) modeling in 151 ECT treated patients with depression to determine the regional relationships between EF , unbiased longitudinal volume change , and antidepressant response across 85 brain regions . The majority of regional volumes increased significantly , and volumetric changes correlated with regional electric field ( t = 3 . 77 , df = 83 , r = 0 . 38 , p=0 . 0003 ) . After controlling for nuisance variables ( age , treatment number , and study site ) , we identified two regions ( left amygdala and left hippocampus ) with a strong relationship between EF and volume change ( FDR corrected p<0 . 01 ) . However , neither structural volume changes nor electric field was associated with antidepressant response . In summary , we showed that high electrical fields are strongly associated with robust volume changes in a dose-dependent fashion .
Electroconvulsive therapy ( ECT ) remains the most effective approach for treatment resistant depressive episodes , as well as the most established neuromodulation technique ( UK ECT Review Group , 2003; Fink and Taylor , 2007 ) . Despite intensive research , however , the mechanism of action for ECT remains unknown , but does involve at least two potentially therapeutic components: electric perturbation and/or seizure activity . One common element across various neuromodulation techniques is the application of different intensities of electric field ( EF ) on the human brain . Understanding how ECT-induced EF interacts with the cortex and subcortical structures can both advance our mechanistic understanding of ECT and enrich our understanding of other neuromodulation approaches such as magnetic seizure therapy ( MST ) , transcranial magnetic stimulation ( TMS ) , transcranial direct current stimulation ( tDCS ) , and deep brain stimulation ( DBS ) . A recent longitudinal ECT-imaging study from the Global ECT-MRI Collaboration ( GEMRIC ) ( Oltedal et al . , 2018 ) assessed hippocampal volume changes in a large cohort of subjects ( N = 268 ) receiving right unilateral ( RUL ) or bilateral ( BL ) electrode placements . The results demonstrated that the volume of the hippocampus increased over the course of ECT treatment and correlated with the number of ECT sessions administered during the ECT series . In addition , the subjects receiving RUL electrode placement had a significantly larger volume change ipsilateral to the side of stimulation , consistent with previous ECT-neuroimaging observations ( Abbott et al . , 2014; Dukart et al . , 2014; Pirnia et al . , 2016; Bouckaert et al . , 2016; Sartorius et al . , 2016; Cano et al . , 2018 ) . Our most recent study of 331 subjects with longitudinal MRI scanning pre- and post-ECT showed brain volume increases across several subcortical and cortical regions with strong lateralization of the effects if the electrode placement was RUL ( Ousdal et al . , 2019 ) . Contrary to a priori expectations ( Joshi et al . , 2016; Cano et al . , 2017 ) , increased volume in these key areas did not translate to better clinical outcome . While the association between the number of ECT sessions and volume change and the laterality of the volume changes all implied a dose–response causative relationship , the role of ECT-mediated neuroplasticity and the underlying mechanism for antidepressant response remains elusive . Furthermore , given the naturalistic design of the studies included for mega-analysis ( e . g . , non-responders had a longer ECT course and were frequently switched to bilateral treatment at varying intervals ) , both the number of ECT sessions and electrode placement varied depending on the clinical response , further confounding the dose-response association and its interpretation . Recent research has challenged the notion that a primary purpose of electric stimulation in treating depression is to generate widespread seizure activity ( Sackeim , 2015; Regenold et al . , 2015 ) . Alternatively , electric stimulation may be a therapeutic component of ECT and similar to other non-convulsive neuromodulation treatments . Finite-element simulation was developed to estimate the spatial distribution of the electric field on a voxel-wise basis ( Lee et al . , 2012; Bikson et al . , 2012 ) . The technique was recently validated in humans ( Huang et al . , 2017 ) . Preliminary computational analyses based on three realistic head models suggested that the ECT electric field distribution had a direct association with clinical and cognitive outcomes , explaining the rationale behind different electrode placement strategies in ECT treatment ( Bai et al . , 2017 ) . This finding is in agreement with our previous observation where RUL treatment induced higher volumetric changes in the right hippocampus compared to the left ( Oltedal et al . , 2018 ) , implying that more lateralized electric stimulation rather than a global generalized seizure , may be responsible at least for part of the antidepressant effects of ECT . However , to date , no study has demonstrated the relationship between ECT electric field distribution and treatment response . In this study , we used the large Global ECT-MRI Research Collaboration ( GEMRIC ) ECT-imaging data set to explicitly determine the relationships between regional 1 ) electric field strength and volume changes , 2 ) volume changes and antidepressant response , and 3 ) electric field and antidepressant response . For the purpose of our primary research question and in contrast to previous GEMRIC investigations , we limited the analyses to subjects that only received right unilateral electrode placement .
Subjects showed an average MADRS improvement of 61 . 3%±33 . 9% following ECT ( pre-ECT MADRS 33 . 9 ( range: 14–54 ) , post-ECT MADRS 12 . 9 ( range: 0–51 ) . Highly significant correlations between age and clinical response ( t = 5 . 75 , df = 149 , r = 0 . 43 , p<10−7 , older patients responded better ) , as well as age and total brain volume ( t = −7 . 32 , df = 149 , r = −0 . 51 , p<10−10 ) were also observed . ECT was associated with increased volume across all brain regions except the brain stem and bilateral cerebellum cortex ( Supplementary file 1 ) . In the majority of the regions , right hemisphere regions had greater volumetric change with respect to the corresponding left hemisphere region; no left hemipshere regions had greater volumetric changes compared to the corresponding right-sided region ( Supplementary file 2 , Figure 1 ) . Average EF strongly correlated with ∆Vol across the ROIs ( Figure 1 , t = 3 . 77 , df = 83 , r = 0 . 38 , p=0 . 0003 ) . To show that this correlation was not simply due to a general effect of the hemisphere ( right side had higher EF and volume change while left side had lower values ) , we calculated laterality indices in both EF and volume change . The correlation between laterality indices for EF and ∆Vol also had a positive relationship ( Figure 2 , t = 2 . 13 , df = 40 , r = 0 . 32 , p=0 . 04 ) across 42 regions ( brain stem is missing , since it is not a bilateral structure ) . In a multiple regression analysis controlled for age , number of ECT sessions and site , we found that left hippocampus and left amygdala had a strong relationship with EF in these regions ( FDR corrected p<0 . 01 , Table 1 ) . Post hoc analyses of the hippocampus ( Figure 3 ) and amygdala ( Figure 4 ) illustrate that the relationship between EF and ∆Vol was dose-dependent and scaled across the hemispheres ( hippocampus: t = 5 . 97 , df = 300 , r = 0 . 3259 , p<0 . 0001; amygdala: t = 11 . 3538 , df = 300 , r = 0 . 5482 , p<0 . 0001 ) . Age was a necessary covariate since it was a confound in our model: both the spatial distribution of EF and volume changes correlate with age ( Deng et al . , 2015 ) . We add number of ECT as a covariate to the model to be able to compare the relative influence of EF and number of ECT on volume change . In both left hippocampus and amygdala the effect size of EF was the largest ( hippocampus: tEF = 4 . 5 , tAge = −2 . 7 , tECTnum = 3 . 3 , amygdala: tEF = 3 . 9 , tAge = −1 . 1 , tECTnum = 2 . 1; Table 1 ) . We also investigated the spatial specificity of these correlations . First , we permutated the regional labels in the volumetric changes across all possible ROIs and calculated the correlations between the EF and ∆Vol . The correlation between EF and the corresponding ∆Vol ( Figure 3—figure supplement 1 , Figure 4—figure supplement 1 , left panels , indicated with red dot ) was always in the top 5% among all possible correlations . Second , we permutated the region labels in the EF across all possible ROIs and calculated the correlations between the EF and ∆Vol ( Figure 3—figure supplement 1 , Figure 4—figure supplement 1 right panels ) . Overall these results indicate a strong spatial selectivity in the relationship between EF and ∆Vol . We further investigated if EF directly or indirectly ( mediated via volume change ) leads to clinical response . In a multiple regression analysis , we tested if volumetric changes controlled for age , number of ECT sessions , and site had an effect on clinical response measured by MADRS changes . Results indicated that none of the volume changes across the 85 ROIs had a significant relationship with clinical response ( Supplementary file 3 , hippocampus: tΔVOL = 0 . 2 , tAge = 5 . 4 , tECTnum = −2 . 7 , amygdala: tΔVOL = 0 . 1 , tAge = 5 . 6 , tECTnum = −3 . 0 ) . These results therefore contradicted the hypothesis that EF by increasing brain volume indirectly exerts its effect on clinical response , given the negative results between the volume change ( mediator ) and MADRS change ( outcome ) . Testing the direct effect of the EF , we failed to find significant correlations between EF and clinical response ( Supplementary file 4 , hippocampus: tEF = 1 . 2 , tAge = 5 . 7 , tECTnum = −3 . 0 , amygdala: tEF = 1 . 1 , tAge = 5 . 7 , tECTnum = −3 . 0 ) . Similar to earlier studies , age strongly correlated with both clinical response ( Haq et al . , 2015; O'Connor et al . , 2001 ) , also see Clinical Results ) and EF distribution ( Deng et al . , 2015 ) , therefore we controlled for age in our model . The rationale for including the number of ECT treatments as covariate needs more explanation . Due to the naturalistic nature of the design , where most sites followed the patient until response or site-determined criteria for ECT discontinuation , we observed a negative relationship between clinical response and the number of ECT treatments . Not controlling for this variable could lead to spurious correlation between volume change and clinical response ( for more on this see Oltedal et al . , 2018 ) . In a post-hoc analysis , we also examined the interaction between EF and volume change in relation to clinical outcome ( excluding age as a covariate ) , but we again failed to find significant effects for any region . To explore further , we investigated if changing age to baseline volume in the mixed model would modify results , but we did not find significant effects ( age and baseline volume correlates strongly across almost all regions – Supplementary file 5 ) .
This investigation is the first demonstration that the ECT-induced electric field is related to increases in cortical and subcortical structures . These results support that the electric field , independent or synergistic with seizure activity and other stimulation parameters , can have a profound effect on the biology of the human brain . The electric path originates from the ECT electrode handle , which delivers a constant stimulus current from the scalp . From the scalp , the electric path travels through skin , skull , cerebral spinal fluid , and brain . Each tissue type has different conductive properties and abundant individual variability ( Deng et al . , 2015 ) . This variability creates different electric field doses despite the similar current at the scalp . These differences in current may lead to both differences in volume changes as well as clinical outcomes . In our investigation , the electric field-induced volume change in the bilateral amygdala and the left hippocampus suggests regional specificity , but the association of these volumetric changes with clinical outcomes remains elusive . Better controlled prospective trials are needed to answer if these robust volume changes and corresponding electric field distributions are associated with any clinical or cognitive consequences .
GEMRIC is a multi-site consortium focused on improving and individualizing ECT by researching the still elusive mechanisms of action and response-related biomarkers ( Oltedal et al . , 2017 ) . Patients in the GEMRIC database participated in clinical and imaging assessments pre- and post-ECT series . To control for the differential effects of electrode placement on electric fields , we only included subjects who received high-dose ( six times the seizure threshold ) right unilateral electrode placement throughout the ECT series . We screened 281 patients from 10 sites ( Oltedal et al . , 2018 ) , and data were included from 7 GEMRIC sites with the RUL only criteria ( n = 151 , 92 F , age: 57 . 5 ± 17 . 1 , 12 with bipolar depression , 139 with major depression , demographic summary is in Table 2A and B ) . Depression severity was assessed with the Montgomery–Åsberg Depression Rating Scale ( MADRS ) ( Montgomery and Asberg , 1979 ) or 17- or 24-item Hamilton Depression Rating Scale ( HAM-D ) ( Hamilton , 1960 ) . For sites collecting only the 17- or 24-item HAM-D , a validated equation was used to convert the 17-item HAM-D to a MADRS score ( Heo et al . , 2007 ) . Clinical response was estimated as the percentage change of the MADRS scores ( ΔMADRS = ( MADRSTP1-MADRSTP2 ) /MADRSTP1 ) . Although more conservative than absolute change or post-ECT depression outcomes ( Vickers , 2001 ) , the rationale for the use of the proportional change score was to control for the variability of the pre-ECT MADRS . The range of the number of sessions for the ECT series was between 7 and 20 . Half of the subjects were medication free during the ECT series ( n = 69 ) . Concurrent pharmacotherapy for the remaining subjects was as follows: selective serotonin reuptake inhibitors ( SSRI , n = 28 ) , serotonin norepinephrine reuptake inhibitors ( SNRI , n = 37 ) , tricyclic antidepressants ( TCA , n = 10 ) , and no record of concurrent medications ( n = 6 ) . Only five subjects received medication changes during the ECT series ( two medication free subjects started SSRI and TCA , one subject switched from SNRI to TCA , one from SNRI to TCA and one from SSRI to SNRI ) . The results did not change if we used medication status or diagnosis ( bipolar or unipolar depression ) as a nuisance variable in the linear models of this study . All sites’ contributing data ( Table 2B ) received approval by their local ethical committees or institutional review board , and the centralized mega-analysis was approved by the Regional Ethics Committee South-East in Norway ( 2013/1032 ECT and Neuroradiology , June 1 , 2015 ) . The image processing methods have been detailed previously ( Oltedal et al . , 2018; Oltedal et al . , 2017 ) . In brief , the sites provided longitudinal 3T T1-weighted MRI images ( at baseline and after the end of the course of ECT ) with a minimal resolution of 1 . 3 mm in any direction ( detailed parameters in Supplementary file 6 ) . The raw DICOM images were uploaded and analyzed on a common server at the University of Bergen , Norway . To guarantee reproducibility , in addition to the common platform , the processing pipelines were implemented in a docker environment ( Merkel , 2014 ) . First , images were corrected for scanner-specific gradient-nonlinearity ( Jovicich et al . , 2006 ) . Further processing was performed with FreeSurfer version 5 . 3 , which includes segmentation of subcortical structures ( Fischl et al . , 2002 ) and automated parcellation of the cortex ( Desikan et al . , 2006 ) . In addition to brainstem and bilateral cerebellum , this automated process identified 33 cortical and eight subcortical regions in each hemisphere . Altogether this resulted in 85 regions of interest ( ROIs ) ( Supplementary file 1 ) . Next Quarc ( Holland et al . , 2011 ) was used for unbiased , within-subject assessment of estimation of longitudinal volume change ( ΔVol - % ) ( Figure 5 ) . In summary , we calculated bias-free estimation of volumetric change from 85 brain regions across the timespan of an ECT course in 151 individuals who received between 4 to 20 ECT sessions ( 1 ½ week to 2 month ) . We estimated ECT-induced electric fields with Realistic Volumetric-Approach to Stimulate Transcranial Electric Stimulation ( ROAST v1 . 1 ) ( Huang et al . , 2017 ) . After segmentation of the structural MRI T1-weighted images , ROAST builds a three-dimensional tetrahedral mesh model of the head . The segmentation identifies five tissue types: white and gray matter of the brain , cerebrospinal fluid , skull , and scalp , and assigns them different conductivity values: 0 . 126 S/m , 0 . 276 S/m , 1 . 65 S/m , 0 . 01 S/m , and 0 . 465 S/m respectively . ECT electrodes of 5 cm diameter were placed over the C2 and FT8 EEG ( 10–20 system ) sites . Study sites from the GEMRIC database used either the Thymatron ( Somatics , Venice , Florida , six sites , N = 121 ) or spECTrum ( MECTA Corp . , Tualatin , Oregon , one site , N = 30 ) devices . The electric field was solved using the finite-element method with unit current on the electrodes and , subsequently , it was scaled to the current amplitude of the specific devices ( Thymatron 900 mA , spECTrum 800 mA ) . These procedures resulted in a voxel-wise electric field distribution map in each individual . We calculated the average electric field across the 85 three-dimensional ROIs in every individual ( Figure 5 ) based on the Freesurfer parcellations and segmentations . The voxel values with the top and lowest one percentile in each ROI were omitted during calculations to reduce boundary effects . We assessed the relationship between EF and ∆ Vol with the following linear mixed effect model in all 85 regions: 1 ) ΔVol ~EF + Age + number of ECT sessions + site ( where EF , age , and number of ECT sessions were fixed effects , and site was random effect , while the dependent variable was volume change ) . Age , number of ECT sessions , and site , considered as nuisance variables , were included based on our prior observations of an inverse relationship between ECT session number and response ( Oltedal et al . , 2018 ) . Further , age is also shown to impact clinical response ( older patients have increased probability of response , in our sample: t = 5 . 75 , df = 149 , r = 0 . 43 , p<10−7 ) and age-related changes on brain structure are related to EF ( Deng et al . , 2015 ) . We used Benjamini and Hochberg false discovery rate ( FDR ) correction method ( Benjamini and Hochberg , 1995 ) to control for multiple comparisons across 85 ROIs , where a conservative FDR -corrected p<0 . 01 was chosen as the statistical threshold of significance . We assessed the relationship between ∆MADRS and EF and ∆ Vol with the following linear mixed effect model in all 85 regions: 1 ) ΔMADRS ~ ΔVol + Age + number of ECT sessions + site; and 2 ) ΔMADRS ~EF + Age + number of ECT sessions + site ( site as random effect ) . We used the same Benjamini and Hochberg FDR correction for multiple comparison corrections . In addition to analyzing the percentage change of the clinical response , we also evaluated the same models with absolute changes , using baseline MADRS as a covariate . We provided the results of these analyses in the second half of the corresponding Supplementary Files . | Electroconvulsive therapy , or ECT for short , can be an effective treatment for severe depression . Many patients who do not respond to medication find that their symptoms improve after ECT . During an ECT session , the patient is placed under general anesthesia and two electrodes are attached to the scalp to produce an electric field that generates currents within the brain . These currents activate neurons and make them fire , causing a seizure , but it remains unclear how this reduces symptoms of depression . For many years , researchers thought that the induced seizure must be key to the beneficial effects of ECT , but recent studies have cast doubt on this idea . They show that increasing the strength of the electric field alters the clinical effects of ECT , without affecting the seizure . This suggests that the benefits of ECT depend on the electric field itself . Argyelan et al . now show that electric fields affect the brain by making a part of the brain known as the gray matter expand . In a large multinational study , 151 patients with severe depression underwent brain scans before and after a course of ECT . The scans revealed that the gray matter of the patients’ brains expanded during the treatment . The patients who experienced the strongest electric fields showed the largest increase in brain volume , and individual brain areas expanded if the electric field within them exceeded a certain threshold . This effect was particularly striking in two areas , the hippocampus and the amygdala . Both of these areas are critical for mood and memory . Further studies are needed to determine why the brain expands after ECT , and how long the effect lasts . Another puzzle is why the improvements in depression that the patients reported after their treatment did not correlate with changes in brain volume . Disentangling the relationships between ECT , brain volume and depression will ultimately help develop more robust treatments for this disabling condition . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"neuroscience"
] | 2019 | Electric field causes volumetric changes in the human brain |
Plants produce phylogenetically and spatially restricted , as well as structurally diverse specialized metabolites via multistep metabolic pathways . Hallmarks of specialized metabolic evolution include enzymatic promiscuity and recruitment of primary metabolic enzymes and examples of genomic clustering of pathway genes . Solanaceae glandular trichomes produce defensive acylsugars , with sidechains that vary in length across the family . We describe a tomato gene cluster on chromosome 7 involved in medium chain acylsugar accumulation due to trichome specific acyl-CoA synthetase and enoyl-CoA hydratase genes . This cluster co-localizes with a tomato steroidal alkaloid gene cluster and is syntenic to a chromosome 12 region containing another acylsugar pathway gene . We reconstructed the evolutionary events leading to this gene cluster and found that its phylogenetic distribution correlates with medium chain acylsugar accumulation across the Solanaceae . This work reveals insights into the dynamics behind gene cluster evolution and cell-type specific metabolite diversity .
Despite the enormous structural diversity of plant specialized metabolites , they are derived from a relatively small number of primary metabolites , such as sugars , amino acids , nucleotides , and fatty acids ( Maeda , 2019 ) . These lineage- , tissue- or cell- type specific specialized metabolites mediate environmental interactions , such as herbivore and pathogen deterrence or pollinator and symbiont attraction ( Mithöfer and Boland , 2012; Pichersky and Lewinsohn , 2011 ) . Specialized metabolism evolution is primarily driven by gene duplication ( Moghe and Last , 2015; Panchy et al . , 2016 ) , and relaxed selection of the resulting gene pairs allows modification of cell- and tissue-specific gene expression and changes in enzymatic activity . This results in expanded substrate recognition and/or diversified product formation ( Khersonsky and Tawfik , 2010; Leong and Last , 2017 ) . The neofunctionalized enzymes can prime the origin and diversification of specialized metabolic pathways ( Schenck and Last , 2020; Weng et al . , 2012; Weng , 2014 ) . There are many examples of mechanisms that lead to novel enzymatic activities in specialized cell- or tissue-types , however , the principles that govern assembly of multi-enzyme specialized metabolic pathways are less well established . One appealing hypothesis involves the stepwise recruitment of pathway enzymes ( Noda-Garcia et al . , 2018 ) . In rare cases , non-homologous specialized metabolic enzyme genes occur in proximity to each other in a genomic region , forming a biosynthetic gene cluster ( Nützmann et al . , 2016; Nützmann and Osbourn , 2014; Rokas et al . , 2018 ) . In recent years , an increasing number of specialized metabolic gene clusters ( SMGCs ) were experimentally identified or bioinformatically predicted in plants ( Boutanaev et al . , 2015; Castillo et al . , 2013; Schläpfer et al . , 2017 ) . However , although most experimentally characterized plant SMGCs are co-expressed , the majority of the bioinformatically predicted ones do not show coexpression under global network analysis ( Wisecaver et al . , 2017 ) . While examples of SMGCs are still relatively rare in plants , experimentally validated cases were reported for a surprisingly diverse group of pathways . These include terpenes ( Chae et al . , 2014; Prisic et al . , 2004; Qi et al . , 2004; Wilderman et al . , 2004 ) , cyclic hydroxamic acids ( Frey et al . , 1997 ) , biosynthetically unrelated alkaloids ( Itkin et al . , 2013; Winzer et al . , 2012 ) , polyketides ( Schneider et al . , 2016 ) , cyanogenic glucosides ( Takos et al . , 2011 ) , and modified fatty acids ( Jeon et al . , 2020 ) . However , whereas each cluster encodes multiple non-homologous enzymes of a biosynthetic pathway , evolution of their assembly is not well understood . Acylsugars are a group of insecticidal ( Leckie et al . , 2016 ) and anti-inflammatory ( Herrera-Salgado et al . , 2005 ) chemicals mainly observed in glandular trichomes of Solanaceae species ( Fan et al . , 2019; Schuurink and Tissier , 2020 ) . These specialized metabolites are sugar aliphatic esters with three levels of structural diversity across the Solanaceae family: acyl chain length , acylation position , and sugar core ( Fan et al . , 2019 ) . The primary metabolites sucrose and aliphatic acyl-CoAs are the biosynthetic precursors of acylsucroses in plants as evolutionarily divergent as the cultivated tomato Solanum lycopersicum ( Fan et al . , 2016; Figure 1 ) , Petunia axillaris ( Nadakuduti et al . , 2017 ) and Salpiglossis sinuata ( Moghe et al . , 2017 ) . The core tomato acylsucrose biosynthetic pathway involves four BAHD [BEAT , AHCT , HCBT , DAT ( D'Auria , 2006 ) family acylsucrose acyltransferases ( Sl-ASAT1 through Sl-ASAT4 ) , which are specifically expressed in the type I/IV trichome tip cells ( Fan et al . , 2016; Schilmiller et al . , 2015; Schilmiller et al . , 2012 ) . These enzymes catalyze consecutive reactions utilizing sucrose and acyl-CoA substrates to produce the full set of cultivated tomato acylsucroses in vitro ( Fan et al . , 2016 ) . Co-option of primary metabolic enzymes contributed to the evolution of acylsugar biosynthesis and led to interspecific structural diversification across the Solanum tomato clade . One example is an invertase-like enzyme originating from carbohydrate metabolism that generates acylglucoses in the wild tomato S . pennellii through cleavage of the acylsucrose glycosidic bond ( Leong et al . , 2019 ) . In another case , allelic variation of a truncated isopropylmalate synthase-like enzyme ( IPMS3 ) – from branched chain amino acid metabolism – leads to acylsugar iC4/iC5 ( 2-methylpropanoic/3-methylbutanoic acid ) acyl chain diversity in S . pennellii and S . lycopersicum ( Ning et al . , 2015 ) . Acylsugar structural diversity is even more striking across the family . Previous studies revealed variation in acyl chain length ( Ghosh et al . , 2014; Liu et al . , 2017; Moghe et al . , 2017 ) : Nicotiana , Petunia and Salpiglossis species were reported to accumulate acylsugars containing only short acyl chains ( carbon number , C ≤ 8 ) . In contrast , some species in Solanum and other closely related genera produce acylsugars with medium acyl chains ( C ≥ 10 ) . These results are consistent with the hypothesis that the capability to produce medium chain acylsugars varies across the Solanaceae family . In this study , we identify a metabolic gene cluster on tomato chromosome 7 containing two non-homologous genes – acylsugar acyl-CoA synthetase ( AACS ) and acylsugar enoyl-CoA hydratase ( AECH ) – affecting medium chain acylsugar biosynthesis . Genetic and biochemical results show that the trichome enriched AACS and AECH are involved in generating medium chain acyl-CoAs , which are donor substrates for acylsugar biosynthesis . Genomic analysis revealed a syntenic region on chromosome 12 , where the acylsucrose biosynthetic Sl-ASAT1 is located ( Fan et al . , 2016 ) . Phylogenetic analysis of the syntenic regions in Solanaceae and beyond led to evolutionary reconstruction of the origin of the acylsugar gene cluster . We infer that sequential gene insertion facilitated emergence of this gene cluster in tomato . These results provide insights into specialized metabolic evolution through emergence of cell-type specific gene expression , the formation of metabolic gene clusters and illuminates additional examples of primary metabolic enzymes being co-opted into specialized metabolism .
S . pennellii natural accessions ( Mandal et al . , 2020 ) , as well as the S . lycopersicum M82 ×S . pennellii LA0716 chromosomal substitution introgression lines ( ILs ) ( Eshed and Zamir , 1995 ) , offer convenient resources to investigate interspecific genetic variation that affects acylsugar metabolic diversity ( Mandal et al . , 2020; Schilmiller et al . , 2010 ) . In a rescreen of ILs for S . pennellii genetic regions that alter trichome acylsugar profiles ( Schilmiller et al . , 2010 ) , IL7-4 was found to accumulate increased C10 medium chain containing acylsugars compared with M82 ( Figure 2 , A and B ) . The genetic locus that contributes to the acylsugar phenotype was narrowed down to a 685 kb region through screening selected backcross inbred lines ( BILs ) ( Ofner et al . , 2016 ) that have recombination breakpoints on chromosome 7 ( Figure 2C ) . Because tomato acylsucrose biosynthesis occurs in trichomes , candidate genes in this region were filtered based on their trichome-specific expression patterns . This analysis identified a locus containing multiple tandemly duplicated genes of three families – an acyl-CoA synthetase ( ACS ) , enoyl-CoA hydratase ( ECH ) , and BAHD acyltransferase . Our analysis ( Moore et al . , 2020 ) revealed co-expression of four Sl-ASATs ( Fan et al . , 2019 ) and three genes at the locus – Solyc07g043630 , Solyc07g043660 , and Solyc07g043680 ( Supplementary file 1 and Figure 2—figure supplement 1 ) . Expression of these three genes was trichome enriched ( Figure 2D ) , and thus they were selected for further analysis . The three candidate genes were tested for involvement in tomato acylsugar biosynthesis by making loss of function mutations using the CRISPR-Cas9 gene editing system . Two guide RNAs ( gRNAs ) were designed to target one or two exons of each gene to assist site-specific DNA cleavage by hCas ( Brooks et al . , 2014; Figure 3—figure supplement 1 , A–C ) . In the self-crossed T1 progeny of stably transformed M82 plants , at least two homozygous mutants were obtained in Solyc07g043630 , Solyc07g043660 , and Solyc07g043680 ( Figure 3—figure supplement 1 , A–C ) , and these were analyzed for leaf trichome acylsugar changes . Altered acylsugar profiles were observed in the ACS-annotated Solyc07g043630 or ECH-annotated Solyc07g043680 mutants ( Figure 3 , A and B ) , but not in the ACS-annotated Solyc07g043660 mutant ( Figure 3—figure supplement 1D ) . Despite carrying mutations in distinctly annotated genes ( ACS or ECH ) , the two mutants exhibited the same phenotype – no detectable medium acyl chain ( C10 or C12 ) containing acylsugars ( Figure 3 , A and B ) . We renamed Solyc07g043630 as acylsugar acyl-CoA synthetase 1 ( Sl-AACS1 ) and Solyc07g043680 as acylsugar enoyl-CoA hydratase 1 ( Sl-AECH1 ) based on this analysis . Further genomic analysis revealed that Sl-AACS1 and Sl-AECH1 belong to a syntenic region shared with a locus on chromosome 12 , where Sl-ASAT1 is located ( Figure 3C and Figure 3—figure supplement 2 ) . Sl-ASAT1 is specifically expressed in trichome tip cells and encodes the enzyme catalyzing the first step of tomato acylsucrose biosynthesis ( Fan et al . , 2016 ) . This led us to test the cell-type expression pattern of Sl-AACS1 and Sl-AECH1 . Like Sl-ASAT1 , the promoters of both genes drove GFP expression in the trichome tip cells of stably transformed M82 plants ( Figure 3C ) . This supports our hypothesis that Sl-AACS1 and Sl-AECH1 are involved in tomato trichome acylsugar biosynthesis . Taken together , we identified a metabolic gene cluster involved in medium chain acylsugar biosynthesis , which is composed of two cell-type specific genes . ACS and ECH are established to function in multiple cell compartments for the metabolism of acyl-CoA ( Buchanan et al . , 2015 ) , the acyl donor substrates for ASAT enzymes . We sought to understand the organelle targeting of Sl-AACS1 and Sl-AECH1 , to advance our knowledge of acylsugar machinery at the subcellular level . We constructed expression cassettes of Sl-AACS1 , Sl-AECH1 and Solyc07g043660 with C-terminal cyan fluorescent protein ( CFP ) , hypothesizing that the targeting peptides reside at the N-terminus of precursor proteins . When co-expressed in tobacco leaf epidermal cells , three CFP-tagged recombinant proteins co-localized with the mitochondrial marker MT-RFP ( Nelson et al . , 2007; Figure 4A and Figure 4—figure supplement 1A ) . To rule out the possibility of peroxisomal localization , we fused Sl-AACS1 , Sl-AECH1 , or Solyc07g043660 with N-terminus fused yellow fluorescent protein ( YFP ) , considering that potential peroxisomal targeting peptides are usually located on the C-terminus ( Brocard and Hartig , 2006 ) . The expressed YFP-recombinant proteins were not co-localized with the peroxisomal marker RFP-PTS ( Nelson et al . , 2007; Figure 4—figure supplement 1B ) . Instead , they appeared distributed in the cytosol ( Figure 4—figure supplement 1B ) , presumably because the N-terminal YFP blocked the mitochondria targeting signal . Taken together , protein expression and co-localization analyses suggest that Sl-AACS1 , Sl-AECH1 , and Solyc07g043660 encode enzymes targeted to mitochondria . Sl-AACS1 belongs to a group of enzymes that activate diverse carboxylic acid substrates to produce acyl-CoAs . We hypothesized that Sl-AACS1 uses medium chain fatty acids as substrates , because ablation of Sl-AACS1 eliminated acylsugars with medium acyl chains . To characterize the in vitro activity of Sl-AACS1 , we purified recombinant His-tagged proteins from Escherichia coli . Enzyme assays were performed by supplying fatty acid substrates with even carbon numbers from C2 through C18 ( Figure 4B ) . The results showed that Sl-AACS1 utilized fatty acid substrates with lengths ranging from C6 to C12 , including those with a terminal branched carbon ( iC10:0 ) or an unsaturated bond ( trans-2-decenoic acid , C10:1 ) ( Figure 4 , B and C ) . However , no activity was observed with the 3-hydroxylated C12 and C14 fatty acids as substrates ( Figure 4B ) . These results support our hypothesis that Sl-AACS1 produces medium chain acyl-CoAs , which are in vivo substrates for acylsugar biosynthesis . To test whether Sl-AACS1 and Sl-AECH1 can produce medium chain acyl-CoAs in planta , we transiently expressed these genes in Nicotiana benthamiana leaves using Agrobacterium-mediated infiltration ( Sainsbury et al . , 2009 ) . It is challenging to directly measure plant acyl-CoAs , due to their low concentration and separate organellar pools . We used an alternative approach and characterized membrane lipids , which are produced from acyl-CoA intermediates . We took advantage of the observation that N . benthamiana membrane lipids do not accumulate detectable acyl chains of 12 carbons or shorter . N . benthamiana leaves were infiltrated with constructs containing Sl-AACS1 or Sl-AECH1 individually , or together ( Figure 4D ) . In contrast to the empty vector control , infiltration of Sl-AECH1 led to detectable levels of C12 acyl chains in the leaf membrane lipid phosphatidylcholine ( PC ) ( Figure 4D ) . We also observed increased C14 acyl chains in PC , phosphatidylglycerol ( PG ) , sulfoquinovosyl diacylglycerol ( SQDG ) , and digalactosyldiacylglycerol ( DGDG ) in Sl-AECH1 infiltrated plants ( Figure 4D and Figure 4—figure supplement 1C ) . These results suggest that Sl-AECH1 participates in generation of medium chain acyl-CoAs in planta , which are channeled into lipid biosynthesis . No medium chain acylsugars were detected , presumably due to the lack of core acylsugar biosynthetic machinery in N . benthamiana mesophyll cells . We asked whether the closest known homologs of Sl-AECH1 from Solanum species can generate medium chain lipids when transiently expressed in N . benthamiana . Two SQDGs with C12 chains were monitored by LC/MS as peaks diagnostic of lipids containing medium chain fatty acids ( Figure 4—figure supplement 2 , A and B ) . The results showed that only the putative Sl-AECH1 orthologs Sopen07g023250 ( Sp-AECH1 ) and Sq_c37194 ( Sq-AECH1 ) – from S . pennellii and S . quitoense respectively – generated medium chain lipids in the infiltrated leaves ( Figure 4—figure supplement 2C ) . This confirms that not all ECHs can produce medium chain lipids and suggests that the function of Sl-AECH1 evolved recently , presumably as a result of neofunctionalization after gene duplication ( Figure 4—figure supplement 2C ) . Medium chain acylsugars were documented in Solanum species besides cultivated tomato , including S . pennellii ( Leong et al . , 2019 ) , S . nigrum ( Moghe et al . , 2017 ) , as well as the more distantly related S . quitoense ( Leong et al . , 2020; Hurney , 2018 ) and S . lanceolatum ( Herrera-Salgado et al . , 2005 ) . We hypothesized that evolution of AACS1 and AECH1 contributed to medium chain acylsugar biosynthesis in Solanum . As a test , we analyzed the genomes of Solanum species other than cultivated tomato . Indeed , the acylsugar related synteny containing ACS and ECH was found in both S . pennellii and S . melongena ( eggplant ) , suggesting that the cluster assembly evolved before divergence of the tomato and eggplant lineage ( Figure 5A ) . We applied gene expression and genetic approaches to test the in vivo functions of ACS and ECH in selected Solanum species . To explore the expression pattern of S . pennellii ACS and ECH cluster genes , we performed RNA-seq analysis on trichomes and shaved stems to identify acylsugar biosynthetic candidates ( Supplementary file 2 ) . The expression pattern of S . pennellii cluster genes is strikingly similar to S . lycopersicum: one ECH and two ACS genes are highly enriched in trichomes , including the orthologs of Sl-AACS1 and Sl-AECH1 . Sp-AACS1 function ( Sopen07g023200 ) was first tested by asking whether it can reverse the cultivated tomato sl-aacs1 mutant acylsugar phenotype . Indeed , Sp-AACS1 restored C12 containing acylsugars in the stably transformed sl-aacs1 plants ( Figure 5—figure supplement 1 ) . To directly test Sp-AACS1 and Sp-AECH1 function , we used CRISPR-Cas9 to make single mutants in S . pennellii LA0716 . No medium chain acylsugars were detected in T0 generation mutants with edits for each gene ( Figure 5B and Figure 5—figure supplement 2 , A and C ) . Similar to the ACS-annotated Solyc07g043660 cultivated tomato mutant ( Figure 3—figure supplement 1D ) , deletion of S . pennellii ortholog Sopen07g023220 has no observed effects on S . pennellii trichome acylsugars ( Figure 5—figure supplement 2D ) . The medium chain acylsugar producer S . quitoense ( Hurney , 2018 ) was used for AACS1 and AECH1 functional analysis because of its phylogenetic distance from the tomato clade - it is in the Solanum Leptostemonum clade ( including eggplant ) - and the fact that it produces medium chain acylsugars . We found trichome-enriched putative orthologs of AACS1 and AECH1 in the transcriptome dataset of S . quitoense ( Moghe et al . , 2017 ) , and tested their in vivo function through virus-induced gene silencing ( VIGS ) ( Figure 5—figure supplement 2E ) . Silencing either gene led to decreased total acylsugars ( Figure 5C and Figure 5—figure supplement 2F ) , which correlated with the degree of expression reduction in each sample ( Figure 5D ) . These results are consistent with the hypothesis that Sq-AACS1 and Sq-AECH1 are involved in medium chain acylsugar biosynthesis , because all acylsugars in S . quitoense carry two medium chains ( Leong et al . , 2020; Hurney , 2018 ) . The importance of AACS1 and AECH1 in medium chain acylsugar biosynthesis in distinct Solanum clades inspired us to explore the evolutionary origins of the gene cluster . We sought to understand how the acylsugar gene cluster evolved and whether it correlates with the distribution of medium chain acylsugars across the Solanaceae family . Taking advantage of the available genome sequences of 13 species from Solanaceae and sister families , we analyzed the regions that are syntenic with the tomato acylsugar gene cluster ( Figure 6—figure supplement 1 ) . This synteny was found in all these plants , including the most distantly related species analyzed , Coffea canephora ( coffee , Rubiaceae ) ( Figure 6—figure supplement 1 ) . BAHD acyltransferases were the only genes observed in the syntenic regions both inside and outside the Solanaceae , in contrast to ECH and ACS , which are restricted to the family ( Figure 6A and Figure 6—figure supplement 1 ) . Within the syntenic regions of the species analyzed , ECH homologs , including pseudogenes , are present in all Solanaceae except for Capsicum species , while ACS is more phylogenetically restricted , being found only in Nicotiana and Solanum ( Figure 6A and Figure 6—figure supplement 1 ) . We then performed phylogenetic analysis to reconstruct the evolutionary history of the ACS , ECH , and BAHD acyltransferase genes in the syntenic region ( Figure 6 ) . This analysis revealed a model for the temporal order of emergence for the three types of genes , leading to their presence in the syntenic regions in extant Solanaceae plants ( Figure 6B ) . We propose that the BAHD acyltransferase was the first of three genes that emerged before the divergence between Solanaceae and Rubiaceae , and was likely lost in Convolvulaceae . This hypothesis is based on the discovery of a BAHD acyltransferase pseudogene in the syntenic region of C . canephora ( Figure 6A and Figure 6—figure supplement 1 ) , which is one of the closest Coffea sequences sister to the ASAT clade ( Figure 6—figure supplement 2 and Figure 6—figure supplement 5 ) . In our model , ECH was likely inserted into the syntenic region before the Solanaceae-specific whole genome duplication ( WGD ) event ( Figure 6B and Figure 6—figure supplement 3 ) . We propose that ACS was inserted into the synteny through segmental duplication ( Bailey et al . , 2002; Figure 6—figure supplement 4 ) . However , whether ACS insertion happened before or after the Solanaceae-specific WGD event cannot be resolved by the phylogenetic analysis ( Figure 6—figure supplement 4 ) . If the insertion happened before WGD , one ACS gene loss on chromosome 12 in Solanum – as well as two independent gene losses on chromosomes 7 and 12 in both Petunia ( Figure 6—figure supplement 4 ) and in Salpiglossis sinuata ( Figure 6—figure supplement 6 ) – should have happened . However , if the insertion happened after WGD , then only one gene loss in Petunia and Salpiglossis needs to be invoked ( Figure 6—figure supplement 6 ) . The latter scenario is more likely based on the principle of parsimony . Our ancestral state reconstruction inference supports the notion that the medium chain acylsugars co-emerged with the ACS/ECH genes in the syntenic regions in the common ancestor of Solanum ( Figure 6—figure supplement 7 ) . This leads us to propose that the emergence of both ACS and ECH genes in the synteny was a prerequisite for the rise of medium chain acylsugars in Solanaceae species ( Figure 6B ) . Consistent with the hypothesis , only short chain acylsugars were observed in Petunia ( Liu et al . , 2017; Figure 6—figure supplement 8 ) , which correlates with the absence of ACS homolog ( Figure 6B ) . In contrast , medium chain acylsugars were detected throughout Solanum ( Figure 6—figure supplement 8 ) , supported by the observation that both ACS and ECH homologs are present in extant Solanum species ( Figure 6B ) . Interestingly , although Nicotiana species collectively have both ACS and ECH genes ( Figure 6B ) , they do not produce medium chain acylsugars ( Figure 6—figure supplement 8 ) presumably due to gene losses . For example , the ECH homologs are pseudogenes in N . benthamiana and N . tomentosiformis ( Figure 6A ) . These results show that the presence of both functional ACS and ECH genes are associated with the accumulation of medium chain acylsugars , supporting our hypothesis above . Although no medium chain acylsugars were detected in Nicotiana species examined , the ACS and ECH genes may have been present in the syntenic region prior to divergence of Solanum and Nicotiana . This suggests that one or more species that diverged from the common ancestor of Solanum and Nicotiana could have medium chain acylsugars . To test this hypothesis , we extended the phenotypic analysis to six such Solanaceae genera ( Figure 6—figure supplement 8 ) . Indeed , we found that species in Jaltomata , Physalis , Iochroma , Atropa , and Hyoscyamus , which diverged from the common ancestor with Nicotiana but before Solanum , have medium chain acylsugars ( Figure 6—figure supplement 8 ) .
This study identified a S . lycopersicum chromosome 7 synteny of ACS , ECH , and BAHD acyltransferase genes including two involved in medium chain acylsugar biosynthesis . The discovery of this locus was prompted by our observation of increased C10 containing acylsugars in tomato recombinant lines carrying this region from the wild tomato S . pennellii chromosome 7 . In vitro biochemistry revealed that Sl-AACS1 produces acyl-CoAs using C6-C12 fatty acids as substrates . The function of AACS1 and AECH1 in cultivated and wild tomato medium chain acylsugar biosynthesis was confirmed by genome editing , and extended to the phylogenetically distant S . quitoense using VIGS . The trichome tip cell-specific expression of these genes is similar to that of previously characterized acylsugar pathway genes ( Fan et al . , 2019 ) . There are increasing examples of plant specialized metabolic innovation evolving from gene duplication and neofunctionalization of primary metabolic enzymes ( Maeda , 2019; Milo and Last , 2012; Moghe and Last , 2015; Zi et al . , 2014 ) . Recruitment of Sl-AACS1 and Sl-AECH1 from fatty acid metabolism provides new examples of ‘hijacking’ primary metabolic genes into acylsugar biosynthesis , in addition to an isopropylmalate synthase ( Sl-IPMS3 ) and an invertase ( Sp-ASFF1 ) ( Leong et al . , 2019; Ning et al . , 2015 ) . We hypothesize that both AACS1 and AECH1 participate in generation of medium chain acyl-CoAs , the acyl donor substrates for ASAT-catalyzed acylsugar biosynthesis . Indeed , Sl-AACS1 exhibits in vitro function consistent with this hypothesis , efficiently utilizing medium chain fatty acids as substrates to synthesize acyl-CoAs . Strikingly , Sl-AECH1 perturbs membrane lipid composition when transiently expressed in N . benthamiana leaves , generating unusual C12-chain membrane lipids . These results suggest that evolution of trichome tip cell-specific gene expression potentiated the co-option of AACS1 and AECH1 in medium chain acylsugar biosynthesis . Analogous to trichomes producing medium chain acylsugars , seeds of phylogenetically diverse plants accumulate medium chain fatty acid storage lipids ( Ohlrogge et al . , 2018 ) . In contrast , fatty acids with unusual structures , including those of medium chain lengths , are rarely found in membrane lipids , presumably because these would perturb membrane bilayer structure and function ( Millar et al . , 2000 ) . For example , seed embryo-specific expression of three neofunctionalized enzyme variant genes in Cuphea species – an acyl-ACP thioesterase ( Dehesh et al . , 1996 ) , a 3-ketoacyl-ACP synthase ( Dehesh et al . , 1998 ) , and a diacylglycerol acyltransferase ( Iskandarov et al . , 2017 ) – lead to production of medium chain seed storage lipids ( Voelker and Kinney , 2001 ) . Trichome tip cell restricted expression of AACS1 and AECH1 represents an analogous example of diverting neofunctionalized fatty acid enzymes from general metabolism into cell-specific specialized metabolism . It is notable that we obtained evidence that Sl-AACS1 and Sl-AECH1 are targeted to mitochondria . Because the other characterized acylsugar biosynthetic enzymes – ASATs and Sl-IPMS3 – appear to be cytoplasmic , these results suggest that medium chain acylsugar substrates are intracellularly transported within the trichome tip cell . It is worth noting that Sl-AACS1 seems to show higher activity with C8 fatty acid than with C10 or C12 ( Figure 4 , B and C ) , while no C8 containing acylsugars were described in tomato trichomes ( Ghosh et al . , 2014 ) . This suggests that C8 fatty acids are not synthesized in trichomes . Beyond employing functional approaches , this study demonstrates the value of a combined comparative genomic and metabolomic analysis in reconstructing the evolutionary history of a gene cluster: in this case over tens of millions of years . We propose that the acylsugar gene cluster started with a ‘founder’ BAHD acyltransferase gene , followed by sequential insertion of ECH and ACS ( Figure 6B ) . This de novo assembly process is analogous to evolution of the antimicrobial triterpenoid avenacin cluster in oat ( Qi et al . , 2006; Qi et al . , 2004 ) . There are two noteworthy features of our approach . First , reconstructing acylsugar gene cluster evolution in a phylogenetic context allows us to deduce cluster composition in extant species ( Figure 6B ) . Second , it links cluster genotype with acylsugar phenotype and allows inference of acylsugar diversity across the Solanaceae ( Figure 6 and Figure 6—figure supplement 8 ) . The current architecture of the Solanaceae acylsugar synteny merely represents a snapshot of a genomic neighborhood that is dynamic , which echoes a recent study of triterpene biosynthetic gene clusters in the Brassicaceae ( Liu et al . , 2019; Peters , 2020 ) . De novo assembly of the gene cluster was accompanied by gene duplication , transposition , pseudogenization , and deletion in different genera . In the case of non-acylsugar producer Capsicum , although phylogenetic analysis revealed putative Sl-AACS1 and Sl-AECH1 orthologous genes , they are not harbored in the syntenic region , probably due to translocation or assembly quality issues ( Figure 6A and Figure 6—figure supplement 1 ) . In Nicotiana , the ECH genes became pseudogenized ( Figure 6B ) , which is associated with lack of detectable plant medium chain acylsugars ( Figure 6—figure supplement 8 ) . In tomatoes , the trichome expressed Solyc07g043660 derives from a recent tandem duplication ( Figure 6—figure supplement 4 ) , yet its deletion has no effect on trichome acylsugars ( Figure 3—figure supplement 1D ) . A parsimonious explanation is that Solyc07g043660 is experiencing functional divergence , which may eventually lead to pseudogenization as observed for other genes in the syntenic region . In this study , we identified an acylsugar SMGC in the context of a multiple chromosome syntenic region . This synteny resulted from WGD , and the acylsugar-related genes are co-expressed , and involved in the same metabolic pathway . This resembles the tomato steroidal alkaloid gene cluster consisting of eight genes that are dispersed into two syntenic chromosome regions ( Itkin et al . , 2013 ) . In fact , this tomato alkaloid SMGC is located next to the acylsugar cluster ( Figure 3—figure supplement 2 ) , which is reminiscent of two physically adjacent SMGCs in the fungus Aspergillus ( Wiemann et al . , 2013 ) . Tomato steroidal alkaloids and acylsugars both serve defensive roles in plants , but are biosynthetically and structurally distinct and are stored in different tissues . This raises intriguing questions . Did the separation of acylsugar and alkaloid SMGCs into two chromosomes occur contemporaneously and by the same mechanism ? Did this colocalization confer selective advantage through additive or synergistic effects of multiple classes of defensive metabolites ? Answering these questions requires continued mining and functional validation of metabolic gene clusters across broader plant species and analysis of impacts of clustering in evolutionary and ecological contexts .
The seeds of cultivated tomato Solanum lycopersicum M82 were obtained from the C . M . Rick Tomato Genetic Resource Center ( https://tgrc . ucdavis . edu/ ) , RRID:SCR_014954 . Tomato introgression lines ( ILs ) and tomato backcross inbred lines ( BILs ) were from Dr . Dani Zamir ( Hebrew University of Jerusalem ) . The tomato seeds were treated with ½ strength bleach for 30 min and washed with de-ionized water three or more times before placing on wet filter paper in a Petri dish . After germination , the seedlings were transferred to peat-based propagation mix ( Sun Gro Horticulture ) and transferred to a growth chamber for two or three weeks under 16 hr photoperiod , 28°C day and 22°C night temperatures , 50% relative humidity , and 300 μmol m−2 s−1 photosynthetic photon flux density . The youngest fully developed leaf was submerged in 1 mL extraction solution in a 1 . 5 mL screw cap tube and agitated gently for 2 min . The extraction solution contains acetonitrile/isopropanol/water ( 3:3:2 ) with 0 . 1% formic acid and 10 μM propyl-4-hydroxybenzoate as internal standard . The interactive protocol of acylsugar extraction is available in Protocols . io at http://dx . doi . org/10 . 17504/protocols . io . xj2fkqe . Assembly of the CRISPR-Cas9 constructs was as described ( Brooks et al . , 2014; Leong et al . , 2019 ) . Two guide RNAs ( gRNAs ) were designed targeting one or two exons of each gene to be knocked out by the CRISPR-Cas9 system . The gRNAs were obtained from gene blocks ( gBlocks ) synthesized by IDT ( Integrated DNA Technologies , location ) ( Supplementary file 3 ) . For each CRISPR construct , two gBlocks and four plasmids from Addgene , pICH47742::2 × 35 S-5′UTR-hCas9 ( STOP ) -NOST ( Addgene no . 49771 ) , pICH41780 ( Addgene no . 48019 ) , pAGM4723 ( Addgene no . 48015 ) , pICSL11024 ( Addgene no . 51144 ) , were mixed for DNA assembly using the Golden Gate assembly kit ( NEB ) . For in planta tissue specific reporter gene analysis , 1 . 8 kb upstream of the annotated translational start site of Sl-AACS1 and Sl-AECH were amplified using the primer pairs SlAACS1-pro_F/R and SlAECH1-pro_F/R ( Supplementary file 3 ) . The amplicon was inserted into the entry vector pENTR/D-TOPO , followed by cloning into the GATEWAY vector pKGWFS7 . For ectopically expressing Sp-AACS1 in the cultivated tomato CRISPR mutant sl-aacs1 , Sp-AACS1 gene including 1 . 8 kb upstream of the translational start site of Sp-AACS1 was amplified using the primer pair SpAACS1-pro-gene_F/R ( Supplementary file 3 ) . The amplicon was inserted into the entry vector pENTR/D-TOPO , followed by cloning into the GATEWAY vector pK7WG . The plant transformation of S . lycopersicum M82 and S . pennellii LA0716 was performed using the Agrobacterium tumefaciens strain AGL0 following published protocols ( Leong et al . , 2019; McCormick , 1997 ) . The primers used for genotyping the S . lycopersicum M82 transgenic plants harboring pK7WG or pKGWFS7 construct are listed in Supplementary file 3 . For genotyping the S . lycopersicum M82 CRISPR mutants in the T1 generation , the sequencing primers listed in Supplementary file 3 were used to amplify the genomic regions harboring the gRNAs and the resultant PCR products were sent for Sanger sequencing . For genotyping the S . pennellii LA0716 CRISPR mutants in the T0 generation , the sequencing primers listed in Supplementary file 3 were used to amplify the genomic regions harboring the gRNAs . The resulting PCR products were cloned into the pGEM-T easy vector ( Promega ) and transformed into E . coli . Plasmids from at least six individual E . coli colonies containing the amplified products were extracted and verified by Sanger sequencing . For protein subcellular targeting analysis , the open reading frame ( ORF ) of Sl-AACS1 , Sl-AECH1 , and Solyc07g043660 were amplified using the primers listed in Supplementary file 3 . These amplicons were inserted into pENTR/D-TOPO respectively , followed by subcloning into the GATEWAY vectors pEarleyGate102 ( no . CD3-684 ) and pEarleyGate104 ( no . CD3-686 ) , which were obtained from Arabidopsis Biological Resource Center ( ABRC ) . For the pEarleyGate102 constructs , the CFP was fused to the C-terminal of the tested proteins . For the pEarleyGate104 constructs , the YFP was fused to the protein N-terminus . Transient expressing the tested proteins was performed following an established protocol ( Batoko et al . , 2000 ) with minor modifications . In brief , cultures of A . tumefaciens ( strain GV3101 ) harboring the expression vectors were washed and resuspended with the infiltration buffer ( 20 mM acetosyringone , 50 mM MES pH 5 . 7 , 0 . 5% glucose [w/v] and 2 mM Na3PO4 ) to reach OD600nm = 0 . 05 . Four-week-old tobacco ( Nicotiana tabacum cv . Petit Havana ) plants grew in 21°C and 8 hr short-day conditions were infiltrated , and then maintained in the same growth condition for three days before being sampled for imaging . The GV3101 cultures containing the mitochondria marker MT-RFP ( Nelson et al . , 2007 ) were co-infiltrated to provide the control signals for mitochondrial targeting . In separate experiments , the GV3101 cultures containing the peroxisome marker RFP-PTS ( Nelson et al . , 2007 ) were co-infiltrated to provide the control signals for peroxisomal targeting . A Nikon A1Rsi laser scanning confocal microscope and Nikon NIS-Elements Advanced Research software were used for image acquisition and handling . For visualizing GFP fluorescence in trichomes of the tomato transformants , the excitation wavelength at 488 nm and a 505- to 525 nm emission filter were used for the acquisition . For visualizing signals of fluorescence proteins in the tobacco mesophyll cells , CFP , YFP and RFP , respectively , were detected by excitation lasers of 443 nm , 513 nm , 561 nm and emission filters of 467–502 nm , 521–554 nm , 580–630 nm . For N . benthamiana transient expression of Sl-AACS1 , Sl-AECH1 , and homologs of AECH1 , the ORFs of these genes were amplified using primers listed in Supplementary file 3 , followed by subcloning into pEAQ-HT vector using the Gibson assembly kit ( NEB ) . Linearization of pEAQ-HT vector was performed by XhoI and AgeI restriction enzyme double digestion . A . tumefaciens ( strain GV3101 ) harboring the pEAQ-HT constructs were grown in LB medium containing 50 µg/mL kanamycin , 30 µg/mL gentamicin , and 50 µg/ml rifampicin at 30 °C . The cells were collected by centrifugation at 5000 g for 5 min and washed once with the resuspension buffer ( 10 mM MES buffer pH 5 . 6 , 10 mM MgCl2 , 150 µM acetosyringone ) . The cell pellet was resuspended in the resuspension buffer to reach OD600nm = 0 . 5 for each strain and was incubated at room temperature for 1 hr prior to infiltration . Leaves of 4 to 5 week-old N . benthamiana grown under 16 hr photoperiod were used for infiltration . Five days post infiltration , the infiltrated leaves were harvested , ground in liquid nitrogen , and stored at −80 °C for later analysis . The membrane lipid analysis was performed as previously described ( Wang and Benning , 2011 ) . In brief , the N . benthamiana leaf polar lipids were extracted in the organic solvent containing methanol , chloroform , and formic acid ( 20:10:1 , v/v/v ) , separated by thin layer chromatography ( TLC ) , converted to fatty acyl methylesters ( FAMEs ) , and analyzed by gas-liquid chromatography ( GLC ) coupled with flame ionization . The TLC plates ( TLC Silica gel 60 , EMD Chemical ) were activated by ammonium sulfate before being used for lipid separation . Iodine vapor was applied to TLC plates after lipid separation for brief reversible staining . Different lipid groups on the TLC plates were marked with a pencil and were scraped for analysis . For LC/MS analysis , lipids were extracted using the buffer containing acetonitrile/isopropanol/water ( 3:3:2 ) with 0 . 1% formic acid and 10 μM propyl-4-hydroxybenzoate as the internal standard . To express His-tagged recombinant protein Sl-AACS1 , the full-length Sl-AACS ORF sequence was amplified using the primer pair SlAACS1-pET28_F/R ( Supplementary file 3 ) and was cloned into pET28b ( EMD Millipore ) using the Gibson assembly kit ( NEB ) . The pET28b vector was linearized by digesting with BamHI and XhoI to create overhangs compatible for Gibson assembly . The pET28b constructs were transformed into BL21 Rosetta cells ( EMD Millipore ) . The protein expression was induced by adding 0 . 05 mM isopropyl β-D-1-thiogalactopyranoside to the cultures when the OD600nm = 0 . 5 . The E . coli cultures were further grown overnight at 16 °C , 120 rpm . The His-tagged proteins were purified by Ni-affinity gravity-flow chromatography using the Ni-NTA agarose ( Qiagen ) following the product manual . Measurement of acyl-CoA synthetase activity was performed using minor modifications of the coupled enzyme assay described by Schneider et al . , 2005 . A multimode plate reader ( PerkinElmer , mode EnVision 2104 ) compatible with the 96-well UV microplate was used for the assays . The fatty acid substrates were dissolved in 5% Triton X-100 ( v/v ) to make 5 mM stock solutions . The enzyme assay premix was prepared containing 0 . 1 M Tris-HCl ( pH 7 . 5 ) , 2 mM dithiothreitol , 5 mM ATP , 10 mM MgCl2 , 0 . 5 mM CoA , 0 . 8 mM NADH , 250 µM fatty acid substrate , 1 mM phosphoenolpyruvate , 20 units myokinase ( Sigma-Aldrich , catalog no . M3003 ) , 10 units pyruvate kinase ( Roche , 10128155001 ) , 10 units lactate dehydrogenase ( Roche , 10127230001 ) , and was aliquoted 95 µL each to the 96-well microplate . The reaction was started by adding 5 µL ( 1–2 µg ) proteins . The chamber of the plate reader was set to 30 °C and the OD at 340 nm was recorded every 5 min for 40 min . Oxidation of NADH , which is monitored by the decrease of OD340nm , was calculated using the NADH extinction coefficient 6 . 22 cm2 µmol−1 . Every two moles of oxidized NADH is equivalent to one mole of acyl-CoA product generated in the reaction . To measure the parameters of enzyme kinetics , the fatty acid substrate concentration was varied from 0 to 500 µM , with NADH set at 1 mM . The fatty acid substrates , sodium acetate ( C2:0 ) , sodium butyrate ( C4:0 ) , sodium hexanoate ( C6:0 ) , sodium octanoate ( C8:0 ) , sodium decanoate ( C10:0 ) , sodium laurate ( C12:0 ) , sodium myristate ( C14:0 ) , sodium palmitate ( C16:0 ) , and sodium stearate ( C18:0 ) , were purchased from Sigma-Aldrich . Trans-2-decenoic acid ( C10:1 ) , 8-methylnonanoic acid ( iC10:0 ) , 3-hydroxy lauric acid ( C12:OH ) , and 3-hydroxy myristic acid ( C14:OH ) were purchased from Cayman Chemical . Total RNA was extracted from trichomes isolated from stems and shaved stems of 7-week-old S . pennellii LA0716 plants using the RNAeasy Plant Mini kit ( Qiagen ) and digested with DNase I . A total of four RNA samples extracted from two tissues with two replicates were used for RNA sequencing . The sequencing libraries were prepared using the KAPA Stranded RNA-Seq Library Preparation Kit . Libraries went through quality control and quantification using a combination of Qubit dsDNA high sensitivity ( HS ) , Applied Analytical Fragment Analyzer HS DNA and Kapa Illumina Library Quantification qPCR assays . The libraries were pooled and loaded onto one lane of an Illumina HiSeq 4000 flow cell . Sequencing was done in a 2 × 150 bp paired end format using HiSeq 4000 SBS reagents . Base calling was done by Illumina Real Time Analysis ( RTA ) v2 . 7 . 6 and output of RTA was demultiplexed and converted to FastQ format with Illumina Bcl2fastq v2 . 19 . 1 . The paired end reads were filtered and trimmed using Trimmomatic v0 . 32 ( Bolger et al . , 2014a ) with the setting ( LLUMINACLIP: TruSeq3-PE . fa:2:30:10 LEADING:3 TRAILING:3 SLIDINGWINDOW:4:30 ) , and then mapped to the S . pennellii LA0716 genome v2 . 0 ( Bolger et al . , 2014a ) using TopHat v1 . 4 ( Trapnell et al . , 2009 ) with the following parameters: -p ( threads ) 8 , -i ( minimum intron length ) 50 , -I ( maximum intron length ) 5000 , and -g ( maximum hits ) 20 . The FPKM ( Fragments Per Kilobase of transcript per Million mapped reads ) values for the genes were analyzed via Cufflinks v2 . 2 ( Trapnell et al . , 2010 ) . For differential expression analysis , the HTseq package ( Anders et al . , 2015 ) in Python was used to get raw read counts , then Edge R version 3 . 22 . 5 ( McCarthy et al . , 2012 ) was used to compare read counts between trichome-only RNA and shaved stem RNA using a generalized linear model ( glmQLFit ) . For VIGS analysis of Sq-AACS1 and Sq-AECH1 in S . quitoense , the fragments of these two genes , as well as the phytoene desaturase ( PDS ) gene fragment , were amplified using the primers listed in Supplementary file 3 , cloned into pTRV2-LIC ( ABRC no . CD3-1044 ) using the ligation-independent cloning method ( Dong et al . , 2007 ) , and transformed into A . tumefaciens ( strain GV3101 ) . The VIGS experiments were performed as described ( Leong et al . , 2020 ) . In brief , the Agrobacterium strains harboring pTRV2 constructs , the empty pTRV2 , or pTRV1 were grown overnight in separate LB cultures containing 50 µg/mL kanamycin , 10 µg/mL gentamicin , and 50 µg/ml rifampicin at 30 °C . The cultures were re-inoculated in the induction media ( 50 mM MES pH5 . 6 , 0 . 5% glucose [w/v] , 2 mM NaH2PO4 , 200 µM acetosyringone ) for overnight growth . The cells were harvested , washed , and resuspended in the buffer containing 10 mM MES , pH 5 . 6 , 10 mM MgCl2 , and 200 µM acetosyringone with the OD600nm = 1 . Different cultures containing pTRV2 constructs were mixed with equal volume of pTRV1 cultures prior to infiltration . The 2- to 3-week-old young S . quitoense seedlings grown under 16 hr photoperiod at 24°C were used for infiltration: the two fully expanded cotyledons were infiltrated . Approximately three weeks post inoculation , the fourth true leaf of each infiltrated plant was cut in half for acylsugar quantification and gene expression analysis , respectively . The onset of the albino phenotype of the control group infiltrated with the PDS construct was used as a visual marker to determine the harvest time and leaf selection for the experimental groups . At least fourteen plants were analyzed for each construct . The trichome acylsugars were extracted using the solution containing acetonitrile/isopropanol/water ( 3:3:2 ) with 0 . 1% formic acid and 1 μM telmisartan as internal standard , following the protocol at http://dx . doi . org/10 . 17504/protocols . io . xj2fkqe . The leaf RNA was extracted using RNeasy Plant Mini kits ( Qiagen ) and digested with DNase I . The first-strand cDNA was synthesized by Superscript II ( Thermofisher Scientific ) using total RNA as templates . Quantitative real-time PCR was performed to analyze the Sq-AACS1 or Sq-AECH1 mRNA in S . quitoense leaves using the primers listed in Supplementary file 3 . EF1α was used as a control gene . A QuantStudio 7 Flex Real-Time PCR System with Fast SYBR Green Master Mix ( Applied Biosystems ) was used for the analysis . The relative quantification method ( 2-ΔΔCt ) was used to evaluate the relative transcripts levels . Trichome acylsugars extracted from tomato IL and BILs were analyzed using a Shimadzu LC-20AD HPLC system connected to a Waters LCT Premier ToF mass spectrometer . Ten microliter samples were injected into a fused core Ascentis Express C18 column ( 2 . 1 mm ×10 cm , 2 . 7 μm particle size; Sigma-Aldrich ) for reverse-phase separation with column temperature set at 40°C . The starting condition was 90% solvent A ( 0 . 15% formic acid in water ) and 10% solvent B ( acetonitrile ) with flow rate set to 0 . 4 mL/min . A 7 min linear elution gradient was used: ramp to 40% B at 1 min , then to 100% B at 5 min , hold at 100% B to 6 min , return to 90% A at 6 . 01 min and hold until 7 min . For analyzing trichome acylsugars extracted from S . pennellii transgenic plants and membrane lipids from N . benthamiana , a Shimadzu LC-20AD HPLC system connected to a Waters Xevo G2-XS QToF mass spectrometer was used . The starting conditions were 95% solvent A ( 10 mM ammonium formate , pH 2 . 8 ) and 5% solvent B ( acetonitrile ) with flow rate set to 0 . 3 mL/min . A 7 min linear elution gradient used for acylsugar analysis was: ramp to 40% B at 1 min , then to 100% B at 5 min , hold at 100% B to 6 min , return to 95% A at 6 . 01 min and hold until 7 min . A 12 min linear elution gradient used for the lipid analysis was: ramp to 40% B at 1 min , then to 100% B at 5 min , hold at 100% B to 11 min , return to 95% A at 11 . 01 min and hold until 12 min . For analyzing trichome acylsugars extracted from other plants , a Waters Acquity UPLC was coupled to a Waters Xevo G2-XS QToF mass spectrometer . The starting condition was 95% solvent A ( 10 mM ammonium formate , pH 2 . 8 ) and 5% solvent B ( acetonitrile ) with flow rate set to 0 . 3 mL/min . A 7 min linear elution gradient was: ramp to 40% B at 1 min , then to 100% B at 5 min , hold at 100% B to 6 min , return to 95% A at 6 . 01 min and held until 7 min . A 14 min linear elution gradient was: ramp to 35% B at 1 min , then to 85% B at 12 min , then to 100% B at 12 . 01 min , hold at 100% B to 13 min , return to 95% A at 13 . 01 min and held until 14 min . For Waters LCT Premier ToF mass spectrometer , the MS settings of electrospray ionization in negative mode were: 2 . 5 kV capillary voltage , 100°C source temperature , 350°C desolvation temperature , 350 liters/h desolvation nitrogen gas flow rate , 10 V cone voltage , and mass range m/z 50 to 1500 with spectra accumulated at 0 . 1 s/function . Three collision energies ( 10 , 40 , and 80 eV ) were used in separate acquisition functions to generate both molecular ion adducts and fragments . For Waters Xevo G2-XS QToF mass spectrometer , the MS settings of the negative ion-mode electrospray ionization were as follows: 2 . 00 kV capillary voltage , 100°C source temperature , 350°C desolvation temperature , 600 liters/h desolvation nitrogen gas flow rate , 35 V cone voltage , mass range of m/z 50 to 1000 with spectra accumulated at 0 . 1 s/function . Three collision energies ( 0 , 15 , and 35 eV ) were used in separate acquisition functions . The MS settings for positive ion-mode electrospray ionization were: 3 . 00 kV capillary voltage , 100°C source temperature , 350°C desolvation temperature , 600 liters/h desolvation nitrogen gas flow rate , 35 V cone voltage , mass range of m/z 50 to 1000 with spectra accumulated at 0 . 1 s/function . Three collision energies ( 0 , 15 , and 45 eV ) were used in separate acquisition functions . The Waters QuanLynx software was used to integrate peak areas of the selected ion relative to the internal standard . For quantification purpose , data collected with the lowest collision energy was used in the analysis . The publicly available tomato RNA-seq datasets and the methods used for normalizing FPKM , gene expression correlation analysis were described in a recent study ( Moore et al . , 2020 ) . 926 RNA-seq Sequence Read Archive ( SRA ) files for tomato from 47 studies were downloaded from National Center for Biotechnology Information ( NCBI; https://www . ncbi . nlm . nih . gov/ ) ( Table S6 in Moore et al . , 2020 ) . Reads were filtered using Trimmomatic ( Bolger et al . , 2014b ) based on the sequence quality with default settings , and mapped to the tomato NCBI S . lycopersicum genome 2 . 5 using TopHat ( Trapnell et al . , 2009 ) . Read files with <70% mapped reads were discarded . Fragments per kilobase of transcript per million mapped reads ( FPKM ) were calculated using Cufflinks ( Trapnell et al . , 2010 ) . The pipeline for FPKM calling used in this study was put in https://github . com/ShiuLab/RNAseq_pipeline ( Uygun et al . , 2020; copy archived at https://github . com/elifesciences-publications/RNAseq_pipeline ) . The median FPKM of multiple replicates was used for each sample , resulting in FPKM values in 372 samples . To draw the heatmap of gene expression profiles , FPKM values of a gene across all the samples were scaled , where the maximum FPKM was scaled to 1 , while the minimum value was 0 . Protein sequences of annotated genes and the corresponding annotation files in General Feature Format ( GFF ) of 11 Solanaceae species , Ipomoea trifida , and Coffea canephora were downloaded from National Center for Biotechnology Information ( NCBI , https://www . ncbi . nlm . nih . gov/genome/ ) or Solanaceae Genomics Network ( SGN , https://solgenomics . net/ ) . The GFF files contain the coordinates of annotated genes on assembled chromosomes or scaffolds . The sources and version numbers of sequences and GFF files used are: S . lycopersicum ITAG3 . 2 ( SGN ) and V2 . 5 ( NCBI ) , S . pennellii SPENNV200 ( NCBI ) and v2 . 0 ( SGN ) , S . tuberosum V3 . 4 ( SGN ) , S . melongena r2 . 5 . 1 ( SGN ) , Capsicum annuum L . zunla-1 V2 . 0 ( SGN ) , C . annuum_var . glabriusculum V2 . 0 ( SGN ) , Nicotiana attenuata NIATTr2 ( SGN ) , N . tomentosiformis V01 ( NCBI ) , N . benthamiana V1 . 0 . 1 ( SGN ) , Petunia axillaris V1 . 6 . 2 ( SGN ) , P . inflata V1 . 0 . 1 ( SGN ) , I . trifida V1 . 0 ( NCBI ) , and C . canephora Vx ( SGN ) . To hypothesize the evolutionary history of genes in the acylsugar gene cluster , putative pseudogenes , which are homologs to protein-coding genes but with predicted premature stops/frameshifts and/or protein sequence truncation , were also identified for each species as described ( Wang et al . , 2018 ) . Protein sequences from Arabidopsis thaliana , Oryza sativa , and S . lycopersicum were used as queries in the searches against the genomic regions of target species using TBLASTN ( Altschul et al . , 1990 ) . The intergenic genomic sequences were identified as potential pseudogenes using the pipeline from as previously described ( Campbell et al . , 2014; Zou et al . , 2009 ) . If one of the six-frame translated sequences of the intergenic genomic sequences had significant similarity to annotated protein sequences , and had premature stops/frameshifts and/or were truncated ( <30% of functional paralogs ) , the gene was defined as a pseudogene . Genome-wide syntenic analysis was conducted using annotated protein-coding genes and putative pseudogenes from all the species with MCScanX-transposed ( Wang et al . , 2013 ) as described ( Wang et al . , 2018 ) . The MCScanX-based analysis did not lead to a syntenic block of acylsugar gene cluster on chromosome 7 of S . melongena r2 . 5 . 1 , which can be due to true absence , issues with genome assembly , or lack of coverage . To verify this , protein sequences of S . lycopersicum genes in genomic blocks on chromosome 7 were searched against an updated S . melongena genome from The Eggplant Genome Project ( http://ddlab . dbt . univr . it/eggplant/ ) that led to the identification of the synteny . Homologous genes of Sl-AACS1 ( ACS ) , Sl-AECH1 ( ECH ) , and Solyc07g043670 ( BAHD acyltransferase ) were obtained through BLAST ( Altschul et al . , 1990 ) search from the genomes of 11 Solanaceae species , Ipomoea trifida , and Coffea canephora with an Expect value threshold of 1e-5 . To simply the phylogenetic tree , sequences which are distantly related to the target genes were removed , and the remained sequences were used to rebuild the phylogenetic trees . The amino acid sequences were aligned using MUSCLE ( Edgar , 2004 ) with the default parameters . The phylogenetic trees were built using the maximum likelihood method with 1000 bootstrap replicates . The trees were generated using RAxML/8 . 0 . 6 ( Stamatakis , 2014 ) with the following parameters: -f a -x 12345 p 12345 -# 1000 m PROTGAMMAAUTO --auto-prot=bic , and were shown with midpoint rooting . The final sequence alignments used to generate the phylogenetic trees were provided in Supplementary file 5 . Ancestral trait state reconstruction was conducted using the maximum likelihood model Mk1 in Mesquite 3 . 6 ( Massidon and Maddison , 2018 ) . Four traits were inferred for their ancestral states . They are the presence of medium chain acylsugars , presence of ACS genes in the synteny , presence of ECH genes in the synteny , and presence of both ACS and ECH genes in the synteny . The phylogeny of Solanaceae species was based on a previous study ( Särkinen et al . , 2013 ) . Acyl chains were characterized from the corresponding fatty acid ethyl esters following transesterification of acylsugar extractions as previously reported ( Ning et al . , 2015 ) . Plants were grown for 4–8 weeks and approximately ten leaves were extracted for 3 min in 10 mL of 1:1 isopropanol:acetonitrile with 0 . 01% formic acid . Extractions were dried to completeness using a rotary evaporator and then 300 µL of 21% ( v/v ) sodium ethoxide in ethanol ( Sigma ) was added and incubated for 30 min with gentle rocking and vortexing every five minutes and 400 µL hexane was added and vortexed for 30 s . To the hexane layer , 500 µL of saturated sodium chloride in water was added and vortexed to facilitate a phase separation . After phase separation , the top hexane layer was transferred to a new tube . The phase separation by addition of 500 µL hexane was repeated twice , with the final hexane layer transferred to a 2 mL glass vial with a glass insert . The fatty acid ethyl esters were analyzed using an Agilent 5975 single quadrupole GC-MS equipped with a 30 m , 0 . 25 mm internal diameter fused silica column with a 0 . 25 µm film thickness VF5 stationary phase ( Agilent ) . Injection of 1 µL of each hexane extract was performed using splitless mode . The gas chromatography program was as follows: inlet temperature , 250°C; initial column temperature , 70°C held for 2 min; ramped at 20 °C/min until 270°C , then held at 270°C 3 min . The helium carrier gas was used with 70 eV electron ionization . Acyl chain type was determined through NIST Version 2 . 3 library matches of the mass spectra of the corresponding ethyl ester and relative abundances were determined through integrating the corresponding peak area over the total acyl chain peak area . | Plants produce a vast variety of different molecules known as secondary or specialized metabolites to attract pollinating insects , such as bees , or protect themselves against herbivores and pests . The secondary metabolites are made from simple building blocks that are readily available in plants , including amino acids , fatty acids and sugars . Different species of plant , and even different parts of the same plant , produce their own sets of secondary metabolites . For example , the hairs on the surface of tomatoes and other members of the nightshade family of plants make metabolites known as acylsugars . These chemicals deter herbivores and pests from damaging the plants . To make acylsugars , the plants attach long chains known as fatty acyl groups to molecules of sugar , such as sucrose . Some members of the nightshade family produce acylsugars with longer chains than others . In particular , acylsugars with long chains are only found in tomatoes and other closely-related species . It remained unclear how the nightshade family evolved to produce acylsugars with chains of different lengths . To address this question , Fan et al . used genetic and biochemical approaches to study tomato plants and other members of the nightshade family . The experiments identified two genes known as AACS and AECH in tomatoes that produce acylsugars with long chains . These two genes originated from the genes of older enzymes that metabolize fatty acids – the building blocks of fats – in plant cells . Unlike the older genes , AACS and AECH were only active at the tips of the hairs on the plant’s surface . Fan et al . then investigated the evolutionary relationship between 11 members of the nightshade family and two other plant species . This revealed that AACS and AECH emerged in the nightshade family around the same time that longer chains of acylsugars started appearing . These findings provide insights into how plants evolved to be able to produce a variety of secondary metabolites that may protect them from a broader range of pests . The gene cluster identified in this work could be used to engineer other species of crop plants to start producing acylsugars as natural pesticides . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"plant",
"biology",
"biochemistry",
"and",
"chemical",
"biology"
] | 2020 | Evolution of a plant gene cluster in Solanaceae and emergence of metabolic diversity |
Traditional cultivation approaches in microbiology are labor-intensive , low-throughput , and yield biased sampling of environmental microbes due to ecological and evolutionary factors . New strategies are needed for ample representation of rare taxa and slow-growers that are often outcompeted by fast-growers in cultivation experiments . Here we describe a microfluidic platform that anaerobically isolates and cultivates microbial cells in millions of picoliter droplets and automatically sorts them based on colony density to enhance slow-growing organisms . We applied our strategy to a fecal microbiota transplant ( FMT ) donor stool using multiple growth media , and found significant increase in taxonomic richness and larger representation of rare and clinically relevant taxa among droplet-grown cells compared to conventional plates . Furthermore , screening the FMT donor stool for antibiotic resistance revealed 21 populations that evaded detection in plate-based assessment of antibiotic resistance . Our method improves cultivation-based surveys of diverse microbiomes to gain deeper insights into microbial functioning and lifestyles .
Culture-independent surveys of naturally occurring microbial populations through marker gene amplicons and shotgun metagenomes have revealed intriguing associations between the gut microbial communities and human health ( Knight et al . , 2017; Lynch and Pedersen , 2016 ) . However , inferring the taxonomic composition or functional potential of complex gut microbiomes does not reveal mechanistic underpinnings of observed associations ( Surana and Kasper , 2017; Ni et al . , 2017; Schmidt et al . , 2018 ) . One of the essential steps to address such shortcomings is microbial cultivation , which enables the recovery of complete reference genomes ( Mukherjee et al . , 2017 ) , accurate identification of taxonomy and functional potential of new strains ( Forster et al . , 2019; Zou et al . , 2019 ) , and validation of causality through perturbation experiments ( Schmidt et al . , 2018 ) . Microbial cultivation is currently experiencing a pronounced revival ( Forster et al . , 2019; Zou et al . , 2019; Browne et al . , 2016; Lagier et al . , 2016; Villa et al . , 2020 ) , yet the majority of cultivation efforts that rely on traditional cultivation strategies require arduous manual picking of thousands of colonies , impeding the efforts to harmonize discoveries that emerge from ‘omics strategies with downstream mechanistic investigations in the rapidly advancing field of microbiome research . Recent years have witnessed numerous new cultivation strategies that increase the throughput in isolating and studying gut-associated bacteria . For example , a recent well plate-based growth experiment screened 96 phylogenetically diverse human gut-associated bacterial strains across 19 media and determined their nutritional preferences and biosynthetic capabilities ( Tramontano et al . , 2018 ) . Another study that relied on ‘SlipChip’ ( Du et al . , 2009 ) , a microfluidic device that can isolate hundreds of microbial cells and enable targeted cultivation , successfully recovered an organism that was a member of the genus Oscillibacter ( Ma et al . , 2014 ) , which had been one of the ‘most wanted taxa’ from the human gut ( Fodor et al . , 2012 ) , a list of uncultivated yet highly prevalent taxa from the Human Microbiome Project ( Huttenhower et al . , 2012 ) . Biomimetic devices represent another active area of research ( Bhatia and Ingber , 2014 ) . For instance , the ‘gut-on-a-chip’ offers a controlled microfluidics platform which mimics the physical and functional features of the intestinal environment and enables complex in vitro chemical gradients and multicellular interactions ( Kim et al . , 2012; Kim et al . , 2016 ) that can establish stable co-culturing of complex bacterial populations ( Jalili-Firoozinezhad et al . , 2019 ) . Although these techniques increase the throughput in isolating and manipulating gut organisms as compared to plate-based culture , their throughput is insufficient for isolating rare organisms among the thousands of gut-associated species or performing large-scale perturbation experiments . Droplet microfluidics offers a promising alternative for high-throughput anaerobic cultivation . The aqueous droplets , with typical volumes ranging from picoliters to nanoliters , are generated and manipulated with an oil phase in microfluidic channels . An extensive arsenal of droplet microfluidic tools has been developed for use in standard aerobic lab environments , where oxygen is present ( Kaminski et al . , 2016 ) . For instance , droplets can be generated and sorted at rates exceeding 10 kHz ( Sciambi and Abate , 2015 ) , reagents can be added by pico-injection or droplet merging ( Abate et al . , 2010; Niu et al . , 2008 ) , and the droplets can further be stored in a regular array and retrieved for downstream applications ( Jiang et al . , 2016; Cole et al . , 2017 ) . Droplets also eliminate a major bottleneck of conventional broth and plate-based culture: the overgrowth of fast-growing populations over slow-growers . In particular , the stochastic isolation of individual bacterial cells in discrete droplets prior to cultivation eliminates the competition that favors fast-growers and yields more accurate representation of the distribution of microbial cells from the input sample ( Jiang et al . , 2016; Zengler et al . , 2002 ) . For instance , Jiang et al . , 2016 isolated environmental soil-associated bacteria in droplets and found an increase in the diversity of taxa with an increased representation of rare organisms ( Jiang et al . , 2016 ) . Villa et al . , 2020 recently cultivated human gut microbes in thousands of nanoliter droplets to characterize metabolic variation in polysaccharide-degrading gut bacteria and analyzed their growth kinetics ( Villa et al . , 2020 ) . Due to small droplet volume , bacteria confined within droplets can reach a critical threshold concentration of quorum sensing molecules faster than would occur in bulk culture , which can lead to improved growth in certain culture medium ( Boedicker et al . , 2009 ) . These key advantages afforded by droplet microfluidics thus present an ideal technology to improve the speed and efficiency of traditional strategies used for anaerobic cultivation . Here , we present an end-to-end platform for high-throughput automated isolation , cultivation , and sorting of anaerobic bacteria in microfluidic droplets . The technology is comprised of droplet microfluidic devices operated inside of an anaerobic chamber and an automated rapid image processing system . We characterized our technology’s ability using a stool sample from a human subject and three different growth media . The droplet-based anaerobic isolation ( i ) achieved a larger representation of microbes in the original stool sample compared to traditional cultivation strategies regardless of the growth medium , ( ii ) promoted the growth of a larger fraction of rare and slow-growing taxa in the original sample , and ( iii ) detected significantly more antibiotic resistant strains from the stool sample than could be detected through traditional plate-based cultivation . Overall , droplet-based cultivation has the potential to increase the throughput and accuracy of cultivating pure strains from anaerobic environments . As fecal microbiome transplant is becoming an increasingly powerful approach for the treatment of several gut conditions such as Clostridium difficile colitis , there is great need for rapidly and affordably screening of these complex microbial populations . Our technology enables rapid and efficient screening for antibiotic resistant microbes in donor stool samples and improves the safety of fecal microbial transplant treatments .
We developed an array of droplet microfluidic technologies for the high-throughput cultivation and manipulation of anaerobic microbial communities ( Figure 1 ) . The microfluidic devices are housed within an anaerobic chamber along with a microscope , syringe pumps , a high frame rate camera , electrodes , and an incubator ( Figure 1a ) . A computer external to the anaerobic chamber controls the camera , syringe pumps , and electrodes ( via a voltage amplifier also external to the chamber ) . We generated droplets at a flow focusing junction from liquid culture medium into oil ( Figure 1b ) . Our strategy stochastically encapsulated microbial cells in the droplets ( ~65–115 pL ) by diluting fecal cell suspensions or liquid broth cultures in the medium so that ~2% to 12% of droplets initially contain only one live bacterial cell according to Poisson statistics . We then placed the droplet emulsion in an incubator at 37°C and the isolated viable strains can clonally replicate within a droplet provided the strain can grow within the environmental conditions ( Figure 1c ) . Since each colony is isolated within a confined droplet , the slow-growing populations avoid the competitive overgrowth of fast-growing populations – which often occurs in traditional broth or Petri agar cultivation . We validated the anaerobic isolation and culture method using extremely oxygen sensitive anaerobes from the model mouse gut microbiota community , the Altered Schaedler’s Flora ( Wymore Brand et al . , 2015; Figure 1—figure supplement 1 ) . The droplets remained stable for several days in culture , although by 4 days the arid atmosphere of the anaerobic chamber led to a significant reduction in droplet volume due to evaporation ( Figure 1—figure supplement 2 ) . We also developed an image-based sorting algorithm and microfluidic control system for sorting bacterial colonies in droplets based on the colony density ( Figure 1d ) . Importantly , image-based droplet sorting does not require fluorescent strains or reporters and therefore has broad applicability in processing environmental samples ( Zang et al . , 2013 ) . The high-frame rate camera along with a custom LabView code automatically detects the droplet approaching the sorting junction and performs a wavelet-based image analysis of an optical density-like measurement , which we termed the Wavelet OD ( see Materials and methods ) . If the Wavelet OD satisfies an empirically-defined thresholding criteria , the computer will actuate the electrodes via a voltage amplifier and deflect the droplet bacteria colony into the ‘keep’ path ( Figure 1d ) . The Wavelet OD value varies between 0 ( empty droplets ) and 1 ( droplets with a very dense colony ) . Droplets were sorted at a rate of ~30 Hz . We retrieved slow-growing colonies by sorting droplets with a Wavelet OD within an empirically defined lower and upper threshold value ( Figure 1e–f ) . To explore the growth potential of human gut bacteria in microfluidic droplets , we used a single human fecal sample from a previously characterized FMT donor ( Lee et al . , 2017 ) and cultivated microorganisms in droplets composed of three rich media up to 2 days ( Figure 1c , Figure 2—figure supplement 1 , and Video 1 ) . In parallel , we used the same set of media on plates for cultivation from the same sample for up to seven days . Our three rich media included Brain Heart Infusion Supplemented ( BHIS ) , Gut Microbiota Medium ( GMM ) , and Yeast Casitones Fatty Acid ( YCFA ) . After cultivation , we extracted the genomic DNA from plate scrapings or by breaking the pooled droplet emulsion . The samples were sequenced using paired-end 16S rRNA gene sequencing on the Illumina platform with primers targeting the V4 region . To infer highly resolved microbial community structures in our amplicon data , we used Minimum Entropy Decomposition ( MED ) ( Eren et al . , 2015a ) , which uses Shannon entropy to identify highly variable nucleotide positions among amplicon sequences and iteratively decomposes a given sequencing dataset into oligotypes , or ‘amplicon sequence variants’ ( ASVs ) , in which the entropy is minimal . The single-nucleotide resolution afforded by this strategy allows the identification of closely related but distinct taxa , better explaining micro-diversity patterns that may remain hidden otherwise ( Eren et al . , 2014; Needham et al . , 2017 ) . Reads were filtered to ensure only organisms grown during the culture period are presented in our data ( see Materials and methods and Figure 2—figure supplement 2 ) . Our data revealed no significant variation in the community composition between biological replicates or cultivation time for a given culture method ( droplets or plates ) and media ( Figure 2—figure supplement 3 ) . Across media and cultivation time , the community richness ( number of detected ASVs ) in droplets was larger than that on plates ( Figure 2a–b , p<0 . 005 , Mann-Whitney U test ) . In particular , droplets enabled an increase in richness between 15% ( BHIS ) and 410% ( YCFA ) . The community diversity , measured by the Shannon index , increased in BHIS and YCFA droplets over their corresponding plates , but not in GMM ( Figure 2c ) . The intra-phyla richness of across plates was non-normally distributed , with a notable lack of representation of Bacteroidetes on YCFA and Proteobacteria on GMM and YCFA ( Figure 2—figure supplement 4 ) . The most abundant ASV in all droplet samples , except 2 day GMM droplets , belonged within the closely related genera Hafnia and Obesumbacterium and had a mean abundance of 24% averaged across media and cultivation time ( Figure 2—figure supplement 5 ) . The droplets also featured a similar taxonomic composition at the family level across the three media , whereas the plate-based cultures more drastically differed from each other and from the input sample ( Figure 2d and Figure 2—figure supplement 6 ) . One of the general bottlenecks of plate-based cultivation efforts is to detect and isolate organisms that are rare in the input sample , because abundant taxa are often over-represented in plates ( Zou et al . , 2019 ) . Our data showed that droplets were able to grow a larger number of organisms that were low-abundance ( <1% ) in the original stool sample based on 16S rRNA gene amplicons ( Figure 3 , p<0 . 005 , Mann-Whitney U test ) . In particular , out of ASVs that were <1% abundant in the original stool sample , 105 ASVs from droplets and 57 ASVs from plates , averaged over cultivation time and media , were detected . Across droplet and plate experiments , 41 ASVs were detected which were not detected in the raw stool and would require an increased read count to resolve their true abundance . Next , we investigated how the composition of closely related taxa that resolve to the same taxonomic group in our cultivation efforts compared to their composition in the stool . For this , we performed an oligotyping analysis on all sequencing reads that matched to a single taxon , Bacteroides , an abundant genus in our dataset and one of the most variable genera across individuals ( Arumugam et al . , 2011 ) . Hierarchical clustering of our samples based on the distribution of Bacteroides oligotypes revealed that the composition of Bacteroides populations measured by the 16S rRNA gene amplicons in droplet-based cultures were more similar to those in raw stool than plate-based cultures , as they clustered closer to the stool sample ( Figure 4 , p<0 . 005 , multiscale bootstrap resampling ) . This indicates the influence of growth biases associated with plate-based cultivation was lessened in droplet-based cultures regardless of the medium , and a larger fraction of Bacteroides populations were accessible through droplets ( Figure 4 ) . In summary , droplet-based cultivation increases the richness and representation of rare taxa broadly across gut-associated phyla . Relatively slow growth rate is one possible explanation for the apparent low abundance of any given taxon within a sample . To investigate whether we could increase the relative abundance of ASVs which were <1% abundant in the raw stool sample , we automatically sorted droplets based on the colony density ( Video 2 ) . We performed two independent sorting experiments using human stool samples grown in BHIS droplets to keep only low-density colonies . The false positive sorting rate was low , with at most 8% of droplets incorrectly sent into the keep path . Sorted droplet cultures resulted in a shift in community composition ( p<0 . 005 , Kolmogorov-Smirnov test ) , with a noticeable change in the abundance of the top 20 ranked ASVs ( Figure 5a ) . Next , we investigated which ASVs were amplified from <1% in raw stool to >1% in unsorted and sorted droplets . The average number of ASVs amplified from the <1% to >1% condition in unsorted BHIS droplets was 6 . 5 out of the 158 ASVs detected in total , while sorting increased the average number of amplified ASVs to 12 . 5 out of 158 ( Figure 5b ) , thereby indicating that droplet sorting based on optical density can provide some preference in amplifying low abundance taxa . As a case in point , plate-based dilution cultures would require at least 65 standard culture plates to have a 90% chance at isolating the 6 Alistipes populations enhanced by the two droplet sorting experiments ( see Figure 5—figure supplement 1 for detailed explanation ) . Additionally , we note that sorted samples amplified taxa across a wider range of the phylogenetic tree than unsorted samples . Finally , to ensure that bacteria remain viable after droplet cultivation , we streaked sorted droplets from one experiment onto an agar plate and cultured the bacteria on the plate for 2 days . To test whether growing colonies on the plate represented distinct taxa , we randomly picked 24 of them . Sanger sequencing of their 16S rRNA genes resolved to genera Hafnia ( 12/24 ) , Enterococcus ( 8/24 ) , and Bacteroides ( 4/24 ) ( Figure 5—figure supplement 2 ) . In total , sorting slow-growing organisms in droplets leads to a further enhancement in the representation of rare taxa beyond that achieved by droplet isolation and cultivation alone . Rapid emergence and spread of antibiotic resistance is a major global public health problem , threatening prevention and treatment options for bacterial infections . Antibiotic resistance also poses a major risk for the development of new treatment methods for gastrointestinal tract diseases . For instance , Clostridium difficile is known to be resistant to multiple antibiotics , and FMT emerged as a major treatment option for recurrent C . difficile colitis ( Wang et al . , 2016 ) . Therefore , antibiotic resistance screening of stool samples is essential for determination of healthy donors and minimizing health risks for large scale FMT applications . In plate-based antibiotic screening , bacterial antibiotic resistance may be unobservable because the plate environment introduces an artificial bias that prevents that organism from growing . Since our droplet technology reduced biases associated with plates , and in particular increased the richness and representation of rare taxa , we hypothesized that our droplet technology could uncover antibiotic resistant members from the FMT donor stool which could not be determined using plate-based cultivation . We chose three widely used antibiotics – ampicillin ( 100 μg/mL ) , ciprofloxacin ( 5 μg/mL ) , and vancomycin ( 10 μg/mL ) , and screened for resistant members in plates and droplets ( Figure 6 and Figure 6—figure supplement 1 ) . The concentrations were chosen to be commensurate with commercially available plates . In the case of droplet cultures , droplets without a growing colony were removed by sorting in order to remove dead or nonviable cells – the reads therefore represent only organisms which grew in the presence of the antibiotics . Overall , droplets detected a much larger number of resistant organisms than plate-based cultures ( Figure 6—figure supplement 2 ) . In Figure 6 , we show ASVs with the most significant difference between droplet and plate screens by filtering for ASVs which are at least 0 . 5% abundant and 10x greater than their counterpart . Overall , the droplet culture identified 21 new organisms from the human stool sample that exhibited strong antibiotic resistance . Additionally , ASVs which are present at greater than 1% abundance in both droplets and plates are depicted . Both droplets and plates recovered known resistance patterns including Bacteroides spp . and Parabacteroides spp . resistance to ampicillin ( Boente et al . , 2010 ) , Enterococcus sp . and Enterobacteriaceae gen . sp . resistance to ciprofloxacin ( Hooper , 2002; Paterson , 2006 ) , and Enterobacteriaceae gen . spp . resistance to vancomycin ( Citron et al . , 2012 ) . At the filtering criteria listed above , growth on plates exceeded droplets only for five populations across all three antibiotics , including a Bacteroides sp . on ampicillin plates and 4 Clostridium spp . on ciprofloxacin plates . Potentially , this preferred growth reflects an improved base fitness on antibiotic-free plates versus droplets for these organisms ( Figure 2a ) . Droplet-based screening detected a number of important antibiotic resistant organisms ( 21 unique organisms ) which could not be determined using plates . For example , ampicillin and vancomycin droplet-based screening detected growth of an organism within the genera Pseudomonas , which contains many opportunistic pathogens , and also within the genera Shewenella , which is a progenitor of antibiotic resistance genes in humans ( Yousfi et al . , 2017 ) . Screening for ciprofloxacin resistance within droplets better displayed known resistances of Bacteroides spp . and Parabacteroides spp . to fluoroquinolone class antibiotics ( Snydman et al . , 2017 ) as well as the reduced activity against gram-positive bacteria ( Poole , 2000 ) . In summary , our droplet platform increases the detection of antibiotic resistant organisms present in human gut samples .
Here , we developed a droplet-based microfluidic platform for isolating , cultivating , and sorting human gut-associated anaerobic bacteria . We cultivated the bacteria across three different rich media using our microfluidic platform and compared the growth to that on conventional plates . The droplet cultivation featured several advantages including an increase in community richness over plates by 15% up to 410% , depending on the medium , and an enhancement in the cultivation of low abundance ( <1% ) strains in raw stool . We also found a reduction in the variability of community composition across different media , and the droplets enabled cultivation of strains from the clinically important genus , Bacteroides , several of which did not grow on plates within the experiments conducted here . Further , sorting droplets based on colony density led to a further enhancement of low abundance strains in the sorted fraction . Finally , we applied our technology towards the detection of antibiotic resistant organisms present in FMT donor stool . Cultivation of the FMT sample in droplets led to the detection of many antibiotic resistant organisms which could not be detected using traditional plate cultures . The droplet-based culture provides several key benefits over traditional plate-based cultivation . First , droplets offer a high-throughput platform for culturing , manipulating , and monitoring bacterial colonies . In aerobic systems , droplet microfluidics has enabled significant progress in fields ranging from pathogen detection , to antibiotic susceptibility testing , to strain engineering ( Kaminski et al . , 2016 ) . Here , we demonstrated that droplet generation , cultivation , and sorting can be extended to anaerobic systems , potentially paving the way towards improved throughput in studying gut microbiota . Second , parasitism , amensalism , and competition are eliminated between strains since each colony is isolated in its own droplet . Our data showed a broad representation of taxa across phyla in droplets ( indicating that the base nutrient requirements are sufficient for growth ) , but a heterogeneous representation on plates . For instance , the lack of Bacteroidetes and Proteobacteria on YCFA plates suggests many of the Firmicutes members have a competitive advantage over the other phyla on YCFA . Although we are uncertain as to the exact cause , a comparison between experiments employing picking of thousands of colonies on YCFA medium versus 70 separate media finds a similar diversity of Firmicutes between the two studies even though only YCFA medium was used , suggesting a strong fitness of many Firmicutes on YCFA ( Forster et al . , 2019; Lagier et al . , 2016 ) . Competitive interactions on the YCFA plates may therefore have led to dominance of many Firmicutes , while the lack of competition in YCFA droplets enabled representation from the other phyla . This elimination of competition can also favor slow-growing organisms . For instance , our droplet sorting for slow-growing organisms enriched for six separate Alistipes populations ( Rikenellaceae family , Figure 5 ) . These Alistipes were either not detected or in very low abundance in plate cultures ( Figure 2 ) . Together , this suggests that these Alistipes are slow growing for all three tested media and that isolation in droplets prevents competitive overgrowth . Finally , the concentration of quorum sensing molecules can increase in droplets faster than bulk culture due to the inherently small volume of each droplet ( Boedicker et al . , 2009 ) . Interestingly , the distribution and use of quorum sensing signaling genes in gut ecosystems may be wide spread – a recent study identified ~38% of genomes from the rumen microbiota possess quorum sensing-related genes ( Won et al . , 2020 ) . Although we do not know the total extent to which quorum sensing is utilized by droplet-grown bacteria , one likely candidate is Hafnia spp . which were the predominant population in droplets across all three rich media and the growth dynamics of at least some strains of Hafnia alvei are modulated by quorum sensing ( Hou et al . , 2017 ) . Combined , these droplet benefits – high-throughput , reduction of competition , and increase in quorum sensing – likely led to the improvement in droplet-grown representation . However , we note that disentangling the interplay of these benefits to determine the exact mechanism which allowed any given organism to grow would likely require mechanistic studies on a case-by-case basis . As a case study , we applied our technology towards detecting antibiotic resistant members present in the FMT donor stool . Recently , transplantation of an FMT donor stool sample led to two recipient patients developing bacteremia , with one patient dying ( DeFilipp et al . , 2019 ) . The stool sample was found to contain a rare extended-spectrum beta-lactamase producing ( ESBL ) Escherichia coli only after bacteremia occurred in the recipient patients . This rare ESBL E . coli went undetected during the initial safety screening of the donor stool . Here , we demonstrated a high-throughput method to detect rare antibiotic resistant organisms that might otherwise remain undetected by traditional cultivation . Our technology thus serves as an effective and efficient tool for screening FMT donor samples and their safer use in transplantation . We envision that several further refinements to our droplet-based cultivation strategy could address limitations of the current study . First , in order for anaerobic droplet cultivation to be widely utilized among microbiologists , the droplet generating and sorting technologies must be easily integrated into standard microbiology workflows . Since microfluidic droplet generating devices are now commercially available ( Dolomite , 2020 ) , methods presented in our study are likely immediately transferable for isolation and cultivation of anaerobic bacteria in droplets . However , anaerobic droplet sorting is technologically more involved and will likely require commercial development of an anaerobic droplet sorter before it is widely adopted . Second , in this study , we isolated single living bacterial cells into droplets in order to prevent interspecies competition . However , isolation inhibits the growth of organisms that rely on other microbial or host cells , such as Saccharibacteria ( formerly TM7 ) ( He et al . , 2015 ) , or obligate endosymbionts , such as Wolbachia ( Hosokawa et al . , 2010 ) . In droplets , co-encapsulation of two cross-feeding auxotrophic strains into a single droplet can induce growth , whereas growth will not occur when only one auxotroph is present within a droplet ( Park et al . , 2011; Hsu et al . , 2019 ) . Future studies could stochastically co-encapsulate multiple gut bacteria into droplets ( by increasing the loading cell density during droplet generation ) and investigate the resulting growth dynamics . Third , the arid environment of the anaerobic chamber led to a reduction in droplet volume over 4 days due to evaporation , limiting the extent of longitudinal studies . This issue could likely be resolved through humidity control inside the incubator . Finally , here we loaded droplets with three different rich media in order to broadly enrich the cultivated community representation across taxa , since previous surveys found the majority of gut bacteria which grow in defined nutrient limited media also grow in rich media ( Tramontano et al . , 2018 ) . However , some bacteria require minimal medium with specific carbohydrates , vitamins , or trace elements ( Tramontano et al . , 2018; Oberhardt et al . , 2015 ) while others utilize surface features such as hydrophobicity , roughness , and surface chemistry to form biofilms and proliferate ( Tuson and Weibel , 2013 ) . Further enrichment of anaerobic organisms within our droplet platform could be achieved by incorporating droplet generation with defined medium ( Villa et al . , 2020 ) , combinatorially generated gradients of medium ( Churski et al . , 2012 ) , or varying the droplet surfactant chemistries to improve biofilm formation ( Chang et al . , 2015 ) . Our approach for isolation , cultivation , and sorting of gut microbiota in droplets enriched the representation of taxa across bacterial phyla , including organisms which are rare and/or slow growing . The improved representation of taxa afforded by droplet cultivation enabled the detection of antibiotic resistant organisms in an FMT donor stool sample which were not detected by traditional plate-based cultivation . Additionally , bacteria remained viable throughout the droplet cultivation and sorting processes suggesting that our anaerobic droplet technology is compatible with traditional downstream microbiology techniques . Going forward , our technology could facilitate overcoming difficulties in traditional plate-based cultivation and pave the way for rapid recovery and detection of novel strains in complex systems such as the human gut microbiome .
The microfluidic droplet generation and droplet sorting devices were fabricated using soft lithography techniques . The device architectures were adapted from Mazutis et al . , 2013 . Briefly , we first fabricated molds from the negative photoresist , SU-8 3050 , on 4" silicon wafers . The height of both the droplet generation and sorting devices was 50 μm . We then poured a 10:1 ratio of PDMS ( RTV 615 ) parts A to B onto the mold , degassed the PDMS , and cured at 80°C for at least 1 hr . Next , we removed the cured PDMS from the mold , punched holes for the inlets , outlets , and electrodes , and plasma bonded the PDMS to either a glass slide ( for droplet generating devices ) or a glass slide with a conductive indium-tin oxide on the rear side ( for droplet sorting devices , Delta Technologies - Part No . CG-811N-S207 ) . To increase the microchannel hydrophobicity , we coated the microchannels with Aquapel ( Pittsburgh Glass Works ) followed by Fluorinert FC-40 ( Sigma ) . Finally , for droplet sorting devices , we created the electrodes by placing the microfluidic device on a 90°C hotplate , flowing a low melting temperature solder ( Indium Corporation of America , 51% In/32 . 5% Bi/16 . 5% Sn ) into the electrode holes , and connecting the solder to standard wires . The syringe pumps , high-frame rate camera , and electrodes are controlled through custom written LabView code . We generated droplets using two syringe pumps ( Harvard Apparatus Pump 11 Pico Plus Elite ) which controlled the liquid and oil ( Bio-Rad Droplet Generation Oil for EvaGreen ) flow rates . We used two separate droplet generating devices here ( see Supplementary file 1 ) with flow rates specified in the Source Data 1 , Experiment Info . The droplet volumes ranged from ~65–115 pL . In a typical experiment , approximately 0 . 5–1 mL of droplets were generated in approximately 20 min – 1 hr , depending on the droplet generating device . For droplet sorting , the droplet reinjection flow rate was set to 20 μL/hr and the oil phase for droplet spacing was set to 180 μL/hr . The microfluidic devices were monitored using an inverted microscope ( Nikon Ts2R ) under 4x and 10x magnification . A high-frame rate camera ( Basler acA640-750um ) captured 672 × 360 pixel images at a rate of 925 Hz and the exposure time was set to 59 μs per frame . Our LabView code automatically analyzed each droplet near the sorting junction and made a sorting decision based off the wavelet OD ( see droplet image analysis ) . Droplets which satisfied the sorting conditions were sent into the keep path by actuating the electrodes ( Figure 1d ) . The remaining droplets flowed down the waste path . The sorting rate was ~30 Hz . The electrodes were actuated by outputting a true decision to an NI-DAQ 6211 which set the analog out to the desired voltage . The analog output voltage is then amplified ( TREK Model 2220 ) at 200 V/V . The electrode actuations used a 10 kHz , 800 V p-p , 30% duty cycle square wave which was activated for 10 ms . An upper limit on the false positive sorting error , ε , is given by ε ≤ [f un ( 1 - f s ) ] / [f s ( 1 - f un ) ] , where f un is the fraction of slow-growing colonies in the unsorted sample and f s is the fraction of slow-growing colonies in the sorted sample ( i . e . , the droplets sent into the keep path ) . The estimate on ε is an equality when the false negative rate is zero . However , we did not count the slow-growing fraction in the waste stream and therefore the actual false positive rate is likely lower than the upper limit . We sorted bacterial colonies in droplets based on an optical density-like measurement , which we termed the Wavelet OD . Custom LabView code first located each droplet as it approached the sorting junction by detecting the droplet edges along the center of the channel ( i . e . , the red dots in Figure 1d ) . The interior region of each droplet ( ~60×80 pixels ) was then analyzed using a discrete wavelet frame decomposition from the LabView function IMAQ Extract Texture Feature VI ( Unser , 1995 ) . We optimized the wavelet parameters for speed and accuracy . In particular , we used biorthogonal 3 . 1 wavelets with the Low Low High subband , a 15 × 15 pixel window with a step size of 5 pixels , and the co-occurrence matrix quantized into 15 gray levels with a 3 × 3 pixel displacement distance . The number of non-zero elements in the wavelet feature vector was then normalized to one to obtain the Wavelet OD . Droplets with a Wavelet OD between 0 . 3 and 0 . 7 were empirically identified as slow-growers and sorted into the keep channel . Droplets with a wavelet OD of less than 0 . 3 were typically empty droplets while droplets with a wavelet OD of greater than 0 . 7 contained a dense bacteria colony . Bacteria cultivation , droplet generation , and droplet sorting were all carried out inside a vinyl anaerobic chamber ( Coy Laboratory Products ) supplied with an 86% N2/10% O2/4% H2 gas mixture . The O2 and H2 concentrations were monitored using an anaerobic monitor ( Coy CAM-12 ) . The H2 concentration was maintained between 1 . 5–2 . 5% and the O2 concentration was typically less than one ppm . A hydrogen sulfide reducing column ( Coy ) was placed inside the chamber to prevent corrosion of the electronic components from H2S buildup . The stool sample we used was previously collected from a fecal microbiota transplant donor ( Lee et al . , 2017 ) . University of Chicago Ethics Committee and the University of Chicago Institutional Review Board ( IRB 132–0212 ) approved the sample collection , and we obtained written and informed consent from the single stool donor . We aliquoted the sample by spinning down 50 μL of stool diluted in 100 μL of PBS , carrying forward the supernatant . The supernatant was stored at −80°C . The live cell and dead cell densities in our aliquots were measured to be ~1 . 5 ± 1 . 0 x 108/mL and ~21 ± 3 x 108/mL , respectively , using the Live/Dead BacLight kit ( ThermoFisher ) . Because the fluorescent dyes in the Live/Dead BacLight kit require oxygenation of the surrounding medium to fluoresce , we exposed the aliquot to air , which may impact the true live/dead measurement . For plate cultures , 100 μL of the aliquoted bacteria suspension was spread onto the plates followed by scraping the plate after cultivation to collect DNA . For both plate and droplet cultures , the DNA was extracted using the Qiagen DNeasy Blood and Tissue Kit according to the manufacturer’s directions . We chose a high plating density of ~1 cell/500 μm2 – which resulted in ‘bacterial lawns’ after cultivation on rich medium . A high plating density was chosen to ensure rare cells are plated , as compared to limiting dilution where distinct colonies can be grown at the expense of not plating rare populations . For droplets , aliquots were diluted 200x , so that according to Poisson statistics , the mean percentage of droplets that will contain one living bacteria is 2% to 12% , accounting for the uncertainty in cell density and the different droplet volumes . In a typical experiment , we generated and cultured a ~0 . 5–1 mL emulsion of ~65–115 pL droplets ( i . e . , ~4–15 million droplets ) . The droplets were stored and cultivated in a snap-cap 15 mL culture tube ( Fisherbrand - 149569C ) . After cultivation , we pipetted out 100 μL of the droplet emulsion ( with an exception for sorted droplets described below ) , mixed it with 100 μL of 1H , 1H , 2H , 2H-Perfluoro-1-octanol ( PFO ) , vortexed and centrifuged the solution , and removed the oil and PFO . Because our sorting rate was limited to ~30 Hz , the total volume of the droplets which were sorted into the ‘keep’ path was significantly less than 100 μL . In the two sorting experiments conducted here , the sorted droplets only formed a very thin layer on top of the oil . Roughly , the estimated total volume observed by pipetting , was 1–10 μL . To extract the DNA , we therefore first added 150 μL DNA-free water and 150 μL PFO to the sorted droplets followed by the same vortex and centrifugation steps . For both plate and droplet cultures , we note that dead cells present in the initial inoculum may generate a small uncertainty in the detected cultivated organisms . In order to verify that the droplet cultivation platform is also compatible with traditional microbiology workflows , we further cultured bacteria grown in droplets on a plate after sorting . In one experiment , we cultured human stool bacteria in BHIS droplets for 1 day , sorted the droplets , and then streaked ~10 μL of the sorted droplet emulsion and oil onto a BHIS plate . We then cultured the plate for 2 days at 37°C and then randomly picked 24 colonies . Each colony was placed into 2 mL of BHIS broth , mixed , and 1 mL was extracted for Sanger sequencing . DNA was isolated using the Qiagen DNeasy Blood and Tissue Kit according to the manufacturer’s instructions . We amplified the 16S rRNA using the primers 27F ( AGAGTTTGATCMTGGCTCAG ) and 1492R ( GGTTACCTTGTTACGACTT ) by mixing 250 nM of the 27F and 1492R primers , 10 μL GoTaq Green Master Mix ( Promega ) , 2 μL extracted DNA , and 7 μL DNA free water followed by running PCR amplification with ( i ) 3 min at 95°C , ( ii ) 30 cycles of 30 s at 95°C , 30 s at 54°C , 1 min at 72°C , and ( iii ) 10 min at 72°C . Each sample was then run through a 2% agarose gel , the band was excised from the gel , and the DNA purified using the QIAquick Gel Extraction Kit ( Qiagen ) according to the manufacturer’s instructions . The DNA was then Sanger sequenced with the 27F and 1492R primers at the University of Chicago Comprehensive Cancer Center DNA Sequencing and Genotyping Facility . Finally , the forward and reverse reads were aligned in Benchling ( https://benchling . com ) using the MAFFT algorithm with default parameters and the consensus sequence was searched using the Standard Nucleotide BLAST for the closest taxonomic match . Antibiotic resistance screening was performed by adding ampicillin ( 100 μg/mL ) , ciprofloxacin ( 5 μg/mL ) , or vancomycin ( 10 μg/mL ) to BHIS plates or BHIS droplets . For antibiotic plate cultures , the raw stool inoculum was diluted 10 , 000x before plating ( plating density of ~0 . 2 cell/mm2 ) . The droplet loading density was the same as in the rich media experiments . Plate screenings were cultured for 3 days and the DNA was then collected through a plate scraping . Droplet screenings were first cultured for 1 day . Next , because the addition of the antibiotics leads to more droplets containing nonviable single cells , inclusion of these nonviable cells would decrease our signal-to-noise . To prevent this , we therefore sorted droplets after the 1 day culture period using our optical density-based droplet sorter to remove drops without a grown colony . After sorting , the DNA was prepared for sequencing as described above . The Environmental Sample Preparation and Sequencing Facility at Argonne National Laboratory ( Argonne , IL , USA ) performed library preparation and sequencing of our DNA isolates following their established protocol developed through the Earth Microbiome Project ( https://earthmicrobiome . org/protocols-and-standards/16s/ ) . Briefly , 35 cycles of amplification were performed using the primer set described previously ( Caporaso et al . , 2012; Caporaso et al . , 2011 ) that target the V4 region of the 16S rRNA gene to generate our amplicons from purified DNA , and Illumina MiSeq paired-end sequencing ( 2 × 151 ) was used to sequence our amplicon libraries . Although here we sequenced the V4 region , we note that sequencing of the V4-V5 region can improve taxonomic resolution ( Nelson et al . , 2014 ) . We analyzed the raw sequencing reads using illumine-utils ( Eren et al . , 2013a ) to ( 1 ) de-multiplexed raw sequencing reads into samples , ( 2 ) join paired-end sequences , and ( 3 ) remove low-quality sequences by requiring a minimum overlap size of 45 nucleotides between the two reads in each pair and removing any read that contained more than two mismatches in the overlapped region ( mismatches in sequences survive these criteria were resolved with the use of the higher quality base ) . Finally , we inferred amplicon sequence variants ( ASVs ) in our dataset using Minimum Entropy Decomposition ( MED ) ( Eren et al . , 2015a ) through the oligotyping pipeline v2 . 1 ( Eren et al . , 2013b ) , and taxonomy was assigned to each ASV using the SILVA database ( Quast et al . , 2013 ) . For both droplet and plate cultures , DNA from dead or nonviable cells in the initial inoculum can be carried over into the collected DNA post culture . Therefore , to ensure that the measured ASVs represent organisms which grew during the culture period , we applied a conservative filtering threshold ( Figure 2—figure supplement 2 ) . We first created a threshold for each sample by fitting a percentage of dead or nonviable DNA , pn . v , from the raw stool initial inoculum to sample ASVs which were at least 10 times less than the corresponding raw stool ASVs . Thus , pn . v represents the fraction of dead or nonviable DNA in the post culture DNA . Next , a 90% confidence interval for each ASV was determined using maximum likelihood estimation ( MLE ) as follows . Consider a given ASV , k , with a true proportional abundance , pk , and measured proportional abundance , p^k=x/n , where x is the number of ASV reads and n is the total number of reads in the sample . Let pk be reparametrized by θk=lnpk / ( 1-pk ) . The 90% confidence interval for θk is given by θ^k ±1 . 645/−l″ ( θ^k;x ) , where lθk; x=constant+xθk-n ln ( 1+eθk ) , is the reparametrized log likelihood estimator for the binomial process of random read sampling , and the derivatives are taken with respect to θk , and evaluated at θ^k . The 90% confidence interval on the number of counts for ASV , k , is then given by transforming back to the lower and upper estimate of pk and multiplying by the number of sample counts . For each ASV , if the lower limit of the 90% confidence interval is below the fitted threshold , the read is discarded . We characterized each sample using standard ecological metrics including richness ( R ) , Shannon’s diversity index ( H’ ) , and rank-abundance curves . ASVs were first normalized by proportion so that for each sample , Σipi = 1 , where pi is the proportional abundance of ASV i . The richness is the total count of ASVs detected within a sample and Shannon’s index , H’ = -Σi pi log ( pi ) . Rank-abundance curves were obtained by ordering the ASVs by decreasing pi for each sample . Additionally , we also counted the richness of ASVs for each sample which were <1% abundant in the raw stool sample ( Rlow ) in order to investigate if droplets can enhance the cultivation of rare species . We statistically tested the metrics R , H’ , and Rlow using the nonparametric Mann-Whitney U test under the null hypothesis that the difference between the means of the metrics on plates and droplets , irrespective of culture media , antibiotic , and cultivation time , is zero . The statistical testing was implemented in the R package using the function wilcox . test . We also statistically tested if there was any difference in the rank-abundance distributions between samples with the same cultivation condition ( i . e . , droplet or plate ) and the same medium . In particular , we applied the Kolmogorov-Smirnov test ( R function ks . test ) under the null hypothesis that two samples can be generated from the same distribution . Next , hierarchical clustering was applied to infer associations between samples . Samples were clustered at the family level by calculating the Bray-Curtis dissimilarity index in the R function vegdist in the package vegan and then plotting the dendrogram . Finally , we statistically tested that Bacteroides oligotypes cluster closer to raw stool using hierarchical cluster analysis with multiscale bootstrap resampling ( R function pvclust ) ( Suzuki and Shimodaira , 2006 ) . We also tested if sorting slow-growing colonies could amplify the relative abundance of rare ASVs . Rank-abundance curves for the two independent sorting experiments were first generated and the combined sorted and unsorted distributions were statistically compared using the two-sample Kolmogorov-Smirnov test , which tests if two sample distributions can be drawn from a common distribution . Next , we investigated the ability of the sorted droplets to increase the abundance of low-abundant ASVs from the raw stool . We arbitrarily set the limit to 1%; ASVs which were <1% abundant in raw stool were considered amplified if the ASV’s relative abundance was >1% in the sorted or unsorted droplets . The ASVs which satisfied this condition across two separate sorting experiments in BHIS droplets were pooled together for phylogenetic analysis . The phylogenetic tree was generated by performing multiple sequence alignment using the default settings on Clustal Omega followed by calculating a DNA Neighbour Joining tree in Jalview . Finally , the increase in proportional abundance for the ASVs which satisfied the above condition in unsorted and sorted droplets was calculated . We used anvi’o v5 . 5 ( Eren et al . , 2015b ) to visualize heat map visualizations of ASV percent relative abundances and clustering dendrograms , and used the open-source vector graphics editor Inkscape ( available from https://inkscape . org ) to finalize them for publication . | The human gut is inhabited with hundreds of billions of bacterial cells from a wide range of families . This complex mixture of bacteria is part of the gut microbiome , along with other lifeforms such as viruses , archaea and fungi . As well as interacting with each other , the bacteria in the microbiome interact with our cells and available nutrients . Studying these interactions can help us understand how this community of bacteria influence health and disease . One way to study the diversity of the microbiome is to take a sample , such as a section of stool , and perform DNA sequencing to determine which types of bacteria are present . This can reveal how the composition of the gut microbiome relates to our health , but cannot confirm whether these bacteria are the cause or the effect of most diseases . To overcome this problem , researchers need to be able to grow pure strains of these bacteria in order to unravel their underlying mechanisms . For over a century , the conventional way to cultivate bacteria has been to grow them in a Petri dish . However , this method promotes the growth of more abundant , fast-growing bacterial strains . This results in a huge disconnect between the bacteria grown in a Petri dish and the diversity within the human gut , which is hindering our understanding of gut health and disease . Now , Watterson et al . have built a machine that improves the speed and number of cultivated bacterial organisms , thus paving the way for more detailed investigations of the human gut microbiome . This new system works by growing bacteria in millions of miniscule droplets which can be physically separated to help the expansion of slower growing species . Watterson et al . cultivated bacterial cells from a stool sample from a single donor using the droplet system and compared this to traditional culturing methods . The droplet technology increased the number of different organisms that were able to grow by up to four times , including those that were rare or slow-growing . Bacteria in the donor stool were then screened for populations that were resistant to antibiotics . This identified 21 antibiotic resistant bacteria which only grew in the droplets and not in Petri dishes . This droplet-based technology will make it possible to study bacterial strains that were previously difficult to grow . Furthermore , this method could help identify whether stool from a donor contains any antibiotic resistant strains , which can lead to clinical complications once transplanted . In future , this new technology could be used in laboratories or hospitals to study the role of the gut microbiome in health and disease . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"tools",
"and",
"resources",
"microbiology",
"and",
"infectious",
"disease"
] | 2020 | Droplet-based high-throughput cultivation for accurate screening of antibiotic resistant gut microbes |
During development eukaryotic gene expression is coordinated by dynamic changes in chromatin structure . Measurements of accessible chromatin are used extensively to identify genomic regulatory elements . Whilst chromatin landscapes of pluripotent stem cells are well characterised , chromatin accessibility changes in the development of somatic lineages are not well defined . Here we show that cell-specific chromatin accessibility data can be produced via ectopic expression of E . coli Dam methylase in vivo , without the requirement for cell-sorting ( CATaDa ) . We have profiled chromatin accessibility in individual cell-types of Drosophila neural and midgut lineages . Functional cell-type-specific enhancers were identified , as well as novel motifs enriched at different stages of development . Finally , we show global changes in the accessibility of chromatin between stem-cells and their differentiated progeny . Our results demonstrate the dynamic nature of chromatin accessibility in somatic tissues during stem cell differentiation and provide a novel approach to understanding gene regulatory mechanisms underlying development .
During the development of a multicellular organism , gene expression is tightly regulated in response to spatially and temporally restricted signals . Changes to gene expression are accompanied by concomitant changes to chromatin structure and composition . Therefore chromatin states vary widely across developmental stages and cell types . Functional regions of a genome , including promoters and enhancers , can be identified by their relative lack of nucleosomes . These regions of ‘open chromatin’ can be assayed by their accessibility to extrinsic factors . Consequently , chromatin accessibility profiling techniques are commonly used to investigate chromatin states ( reviewed in [Tsompana and Buck , 2014] ) . Chromatin is highly accessible in pluripotent cell types such as embryonic stem ( ES ) cells , but is compacted following differentiation ( Meshorer and Misteli , 2006 ) . It has been suggested that this open chromatin represents a permissive state to which multiple programmes of gene regulation may be rapidly applied upon differentiation ( Gaspar-Maia et al . , 2011 ) . The nature of chromatin accessibility across different developmental stages in vivo is less well understood . Imaging studies have been used to demonstrate gross changes to chromatin structure , for example changes to the distribution of heterochromatin have been observed in post-mitotic cells ( Francastel et al . , 2000; Le Gros et al . , 2016 ) . Molecular studies investigating chromatin states in vivo during development have tended to utilise heterogeneous tissues due to the fact that profiling the epigenome of individual cell types frequently requires physical isolation of cells or nuclei , which can be laborious and prone to human error ( McClure and Southall , 2015 ) . Therefore , there is a lack of information regarding cell-type-specific changes to chromatin states in in vivo models . Whilst recently developed methods such as ATAC-seq have become popular and address many of the limitations inherent to earlier techniques such as DNAse-seq ( i . e . requires fewer cells and increased assay speed ) , these techniques still require the physical separation of cells and isolation of genomic DNA before chromatin accessibility is assayed ( Buenrostro et al . , 2013 ) . It has been suggested that ectopic expression of untethered DNA adenine methyltransferase ( Dam ) results in specific methylation of open chromatin regions whilst nucleosome bound DNA is protected ( Wines et al . , 1996; Bulanenkova et al . , 2007; Boivin and Dura , 1998; Singh and Klar , 1992 ) . However , the efficacy of using Dam methylation for chromatin accessibility profiling on a genomic scale is not clear . Furthermore , expression of Dam in a cell-type-specific manner , at levels low enough to avoid toxicity and oversaturated signal , has not been possible until now . Transgenic expression of fusions of Dam to DNA-binding proteins is a well-established method used to assess transcription factor occupancy ( DNA adenine methyltransferase identification - DamID ) ( van Steensel and Henikoff , 2000 ) . Recently , it was demonstrated that DamID could be adapted to profile DNA-protein interactions in a cell-type-specific manner by utilising ribosome re-initiation to attenuate transgene expression ( Marshall et al . , 2016; Aughey and Southall , 2016; Southall et al . , 2013 ) . This technique is referred to as Targeted DamID ( TaDa ) . Here , we take advantage of TaDa to express untethered Dam in specific cell types to produce chromatin accessibility profiles in vivo , without the requirement for cell separation . We show that Chromatin Accessibility profiling using Targeted DamID ( CATaDa ) yields comparable results to both FAIRE and ATAC-seq methods , indicating that it is a reliable and reproducible method for investigating chromatin states . By assaying multiple cell types within a tissue , we show that chromatin accessibility is dynamic throughout the development of Drosophila central nervous system ( CNS ) and midgut lineages . These data have also enabled us to identify enriched motifs from regulatory elements that dynamically change their accessibility during differentiation , as well as to identify functional cell-type-specific enhancers . Finally , we show that compared to their differentiated progeny , somatic stem cell Dam-methylation signals are more widely distributed across the genome , indicating a greater level of global chromatin accessibility .
We reasoned that low-level expression of transgenic E . coli Dam , using tissue-specific GAL4 drivers in Drosophila , would specifically methylate regions of accessible chromatin exclusively in a cell-type of interest . Detection of these methylated sequences could yield chromatin accessibility profiles for defined cell populations in vivo ( Figure 1 ) . To determine if CATaDa produces an accurate reflection of chromatin accessibility , we compared data acquired using this approach with commonly used alternative techniques . A recent study generated ATAC and FAIRE-seq data from Drosophila imaginal eye discs ( Davie et al . , 2015 ) . Using CATaDa , we expressed E . coli Dam in the eye disc of Drosophila third instar larvae so that we could compare Dam methylation profiles to these previously collected data . Chromatin accessibility profiles produced with CATaDa in the eye disc were highly reproducible between replicates ( r2 = 0 . 947 ) ( Figure 2—figure supplement 1 ) . CATaDa profiles showed good agreement with data produced with ATAC-seq and FAIRE-seq . Visual inspection of the data showed that many regions of accessible chromatin identified by ATAC and FAIRE are also represented by CATaDa , whilst condensed regions are reliably inaccessible ( Figure 2A , B ) . We also observe that CATaDa profiles exhibited features consistent with chromatin accessibility . For example , open chromatin is enriched at transcriptional start sites ( TSS ) ( Figure 2C ) . We observe that CATaDa signal frequency increases dramatically towards the centre of ATAC or FAIRE peaks ( Figure 2D ) . The overlap of Dam identified peaks with ATAC and FAIRE peaks is 48 . 6% and 49 . 4% , respectively ( In comparison , 55 . 9% of ATAC peaks are also identified in FAIRE data – Figure 2E ) . A Monte Carlo simulation determined that this is a highly significant overlap ( p<1 × 10−5 ) and peak heights at shared ATAC and CATaDa peaks show significant correlation ( p<1 × 10-16 , r2 = 0 . 138 ) ( Figure 2—figure supplement 2A ) . We found that increasing the stringency of our peak calling notably decreased the number of peaks identified that coincided with ATAC peaks , but had relatively little impact on unique CATaDa peak discovery ( Figure 2—figure supplement 2B ) . Given these data , we suggest that the majority of these peaks are not false positives , but are genuinely accessible sites that are not detected by ATAC-seq . Further examination of these unique peaks indicates that they are significantly smaller than the shared peaks ( Figure 2—figure supplement 2C ) . We also observe that for peaks identified in either ATAC or FAIRE data that are not present in CATaDa , there is a relative lack of GATC motifs , which suggests that there may be cases in which false negatives are observed due to the limitations of the resolution achievable by Dam methylation ( Figure 2—figure supplement 3A–B ) . To further investigate the differences between CATaDa and ATAC or FAIRE-seq , we investigated the detection of peaks at different genomic features . We found that whilst CATaDa identified fewer peaks than FAIRE in regions proximal to gene promoters ( when compared with ATAC ) , CATaDa was notably better at identification of non-promoter adjacent accessible sites in ATAC data compared to FAIRE-seq ( Figure 2E ) . The lack of promoter peaks identified can again be explained by the relative depletion of GATC sites upstream of TSS ( Figure 2—figure supplement 3C ) . It was previously shown that ATAC-seq and FAIRE-seq data demonstrated high chromatin accessibility at experimentally validated eye-antennal enhancers ( Davie et al . , 2015 ) . CATaDa profiles similarly showed increased open chromatin at these regions ( Figure 3A , B ) . We found that for 57 . 9% of FlyLight eye enhancers , a corresponding peak was called in CATaDa profiles ( 333 of 575 enhancers ) . CATaDa was comparable to FAIRE-seq and ATAC-seq which identified 48% and 68 . 7% respectively , of validated FlyLight enhancers as peaks ( Figure 3C ) . In Drosophila , neurons are derived from asymmetrically dividing neural stem cells ( NSCs ) . NSC divisions produce one self-renewing daughter NSC and a ganglion mother cell ( GMC ) , which divides once more to produce neurons or glia ( Homem and Knoblich , 2012 ) . To further test the technique and investigate how local and global chromatin accessibility changes during the process of nervous system differentiation , we expressed Dam in specific cells with GAL4 drivers that cover four different developmental stages within the lineage . These include NSCs ( worniu- GAL4 ) , GMCs and newly born neurons ( R71C09-GAL4 [Figure 4—figure supplement 1B , Li et al . , 2014] ) , differentiated larval neurons ( nSyb-GAL4 ) , and also mature adult neurons ( nSyb-GAL4 ) ( Figure 4A ) . By examining candidate genes differentially expressed during neural development , we observed that chromatin accessibility relates to gene expression in an expected manner . For example , intronic open chromatin peaks can be seen at the bruchpilot ( brp ) locus , in both third instar ( L3 ) and adult neurons , whilst these peaks are reduced or absent in the progenitor cell types ( Figure 4B ) . This corresponds with the expression of brp , which is specifically transcribed in neurons and has an important role in synapse function ( Wagh et al . , 2006 ) . In contrast , the adjacent gene to brp , Wnt2 , displays peaks which are most apparent in the NSC and intermediate cell types . Wnt signalling is known to be important for the control of stem cell populations , therefore , these results are also expected ( Ring et al . , 2014 ) . Similar patterns are observed at a number of other loci . At the asense ( ase ) locus , ( a NSC-specific transcription factor ) , chromatin is highly accessible at the promoter and upstream intergenic region in NSCs ( Figure 4—figure supplement 2B ) . This signal is considerably reduced in fully differentiated neurons in which ase is not expressed . Interestingly , open chromatin is still detectable in these regions in the GMCs/newly born neurons . This pattern is also observed with other NSC expressed factors such as deadpan ( dpn ) , CyclinE ( CycE ) and prospero ( pros ) ( Figure 4—figure supplement 2 ) . Furthermore , GMC/newly born neuron profiles frequently show intermediate signal at these loci , indicating that functional elements required for regulation of NSC gene expression are not immediately rendered inaccessible following differentiation ( Figure 4B and Figure 4—figure supplement 2 ) . It is to be expected that many of the functional elements marked by accessible chromatin that are important for regulating gene expression in a given neural cell type would show dynamic accessibility across the lineage ( i . e . stem cell-specific enhancers would not be expected to be open in mature neurons ) . We examined regions of differential chromatin accessibility to determine the extent to which chromatin accessibility is changed during development of the nervous system . Hierarchical clustering of regions of chromatin with differential accessibility between cell types reveals two major clusters in which chromatin is either open in stem cells but inaccessible in neurons , or vice versa ( Figure 4C ) . Intriguingly , there are other clusters where maximal chromatin accessibility is observed in either GMCs/early neurons or larval neurons . Therefore , it is not as simple as NSC accessible regions progressively closing during differentiation and neuronal regions gradually opening . There are a large number of loci that are inaccessible in NSCs , then open in the intermediate GMCs/newly born neurons stage before being rendered inaccessible again in terminally differentiated neurons ( Figure 4C ) . In addition , a cluster enriched in larval neurons demonstrates that the chromatin accessibility landscape of larval neurons , although similar , is distinct from adult neurons . Regions of open chromatin are thought to identify functional regulatory elements such as enhancers . Therefore , it is to be expected that these regions will be enriched for motifs belonging to transcription factors involved in neurogenesis . Identification of enriched motifs in sequences that were accessible in NSCs showed that expected transcription factor binding sites were highly enriched . For example , the E-box motif – CAGCNG – which is bound by the NSC proneural factor ase ( Figure 4D ) ( Southall and Brand , 2009; Jarman et al . , 1993 ) . Regions in which open chromatin was specifically enriched in mature neurons yielded a sequence motif corresponding to the transcription factor , Ci . In all groups , sequence motifs were also identified for which no known binding partner could be identified ( Figure 4—figure supplement 3 ) . Analysis of further subdivision of these clusters revealed yet more novel motifs for the individual cell types examined , as well as indicating that the ase-like motif is specifically enriched for sequences which are accessible solely in the NSCs , and not their progeny ( Figure 4—figure supplement 4 ) . Gene ontology ( GO ) analysis of genes at which enriched chromatin accessibility was observed yielded expected biological process terms for each of the cell types examined ( Figure 4—figure supplement 5 ) . For example , terms such as ‘neuroblast fate determination’ and ‘chromosome segregation’ were more highly enriched in stem cells relative to neurons , whilst ‘regulation of behaviour’ and ‘synaptic vesicle docking during exocytosis’ were enriched for differentiated neurons but not NSCs ( Figure 4E ) . Having observed chromatin accessibility changes in the cells of the developing CNS , we asked whether similar patterns would be observed in adult somatic stem cell lineages . The Drosophila midgut contains a pool of cycling intestinal stem cells ( ISCs ) that persists in the adult to maintain a population of terminally differentiated cells which mediate the absorptive and secretory functions of the organ ( Jiang and Edgar , 2011; Nászai et al . , 2015 ) . In contrast to neurogenesis , a single committed immature progenitor cell ( enteroblast – EB ) is produced from stem cell divisions , which then differentiates without further divisions to produce the mature epithelial cells of the midgut ( Ohlstein and Spradling , 2007 ) . To examine chromatin accessibility in the cells of the adult midgut , we expressed Dam in the ISCs and EBs , as well as in the terminally differentiated absorptive cells , the enterocytes ( ECs ) ( Figure 5A ) . As with the CNS data , we observed predictable changes in chromatin accessibility at loci for genes with variable expression in the lineage . For example , escargot ( esg ) a transcription factor required for ISC self-renewal ( Korzelius et al . , 2014 ) , displays multiple peaks of accessible chromatin at the gene body and surrounding region in ISCs and EBs , whilst little signal is observed in the ECs ( Figure 5B ) . In contrast the nubbin locus ( encoding EC marker – Pdm1 ) , displays peaks predominantly in the EC data , with relatively closed chromatin in the progenitor cell types ( Figure 5C ) . As observed in the CNS , hierarchical clustering revealed two major groups in which accessible chromatin was enriched in either in the stem cells ( ISCs ) or differentiated cell ( ECs ) ( Figure 5D ) . Smaller clusters were again evident in which accessible chromatin was up or downregulated exclusively in the intermediate EBs . However , this was much less pronounced than the changes observed in GMCs/early neurons of the developing CNS . This indicates that , similar to the CNS lineages , the majority of chromatin accessibility changes involved in specifying the fully differentiated cells do not occur until after EB maturation . As with the cells of the CNS , we were able to identify motifs specifically enriched in each of these groups ( Figure 5—figure supplement 1 ) . ISCs and NSCs fulfil similar roles in their respective organs in the production of highly specialised functional cells . However , whilst NSCs exist for a short amount of time during fly development to produce relatively long-lived neurons that persist in the adult CNS for the animal’s lifetime , the ISCs act post-developmentally to constantly replenish ECs in the adult gut . By comparing the chromatin accessibility of these two cell types , it is apparent that there are similarities in their chromatin states . At loci involved in growth or cell division , we see similar accessibility profiles across differentiation between the two tissue types ( Figure 5—figure supplement 1 ) . Given the similarities that we observed for individual loci between CNS and midgut lineages , we queried whether it was possible to observe trends between the cells in the two lineages on a global scale . Principal component analysis reveals two distinct clusters in which >80% of the variance is explained in the first two principal components ( Figure 5E ) . These clusters represent the two distinct tissue types , ( CNS and midgut ) rather than immature and differentiated cells . By examining the overall correlation between all cell types we observed a number of interesting features . Firstly , as expected all cell types correlated most closely with either their direct progeny or progenitor cell ( Figure 5F ) . Therefore by clustering the data we were able to recapitulate the familial relationship between the cell types of the two lineages . The greatest similarities were observed between the intermediate progenitors and their cognate stem cells ( R2 = 0 . 94/0 . 98 for CNS and midgut respectively ) . Interestingly , the greatest correlation outside of a lineage was between the two stem cell types ( R2 = 0 . 76 ) , whilst differentiated cells exhibited only weak correlation ( ISCs vs NSC , R2 = 0 . 51 ) . This indicates that somatic stem cell types may utilise a broadly similar chromatin landscape for the maintenance of multipotency , whilst lineage-specific variation is relatively small . Enhancer activity is closely linked to gene expression , therefore , many tissue-specific enhancers are required to orchestrate correct spatial and temporal transcription ( Pennacchio et al . , 2013 ) . However , identification of functional enhancers can be challenging . Chromatin accessibility data have previously been used to identify novel enhancers ( Davie et al . , 2015; Crawford et al . , 2006 ) . We reasoned that it would be possible to identify genomic regions corresponding to cell-type- specific enhancers by comparing dynamically accessible regions between cell types . In support of this , we observed that the sequence covered by the 71C09-GAL4 line used in this study to profile GMCs/newly born neurons , displayed a higher peak specifically at this region than in either the stem cell or differentiated neuron data ( Figure 4—figure supplement 1 ) . Interestingly , a clear peak can still be observed in the NSC data , without concomitant reporter expression . Therefore , an enrichment of accessible chromatin does not necessarily correspond to an active enhancer in a given cell type . This is consistent with previous observations that DNase hypersensitive regions are often not active enhancers ( Zhou et al . , 2017; Thurman et al . , 2012 ) . We selected accessible regions with large differences between at least two cell types in the lineage , which satisfied various criteria for us to designate them as putative enhancers ( see Materials and methods ) . We then identified available reporter lines from the Vienna tiles ( VT ) ( Kvon et al . , 2014 ) and FlyLight ( Jenett et al . , 2012 ) collections of GAL4 reporter lines that contained sequences encompassing our predicted enhancers upstream of a GAL4 reporter , and verified their expression in the tissues of interest . We identified enhancer-GAL4 lines in which reporter expression matched our predictions for enhancer activity . In the CNS Vienna line VT017417 and FlyLight line GMR56E07 both showed expression in the early part of the lineage in the CNS , with GFP reporter expression detectable predominantly in NSCs and GMCs ( Figure 6A , B ) . This is consistent with accessible chromatin readings from our CATaDa data for these cell types in which progenitor cells displayed prominent peaks , whereas differentiated neurons did not . Similarly , we were able to detect functional cell-type-specific enhancers in the midgut . The Vienna line , VT004241 , showed reporter expression predominantly in Delta positive ISCs ( Figure 6C ) . Therefore , it is possible to use CATaDa data to identify novel cell-type-specific enhancers in multiple tissues . Overall 17 of 30 lines ( 57% ) tested showed GFP expression in tissues and cell types that closely matched our predictions based on CATaDa data . This is in line with the rate of enhancer prediction by previous methods ( Kvon et al . , 2014 ) . It is commonly accepted that the global accessibility of chromatin in a given cell type correlates broadly with potency . Whilst there are multiple lines of evidence showing that pluripotent cells have high levels of open chromatin , little data exist to support this idea in somatic multipotent stem cells and their progeny . We reasoned that by examining the distribution of normalised mapped sequencing reads ( for each GATC fragment ) we would be able to determine the nature of open chromatin in a given cell type . Initial examination of read distributions indicated that differentiated neurons had fewer GATC fragments with lower read counts ( Figure 7—figure supplement 1A ) . However , having a limited number of replicates prevented a thorough statistical analysis of this relationship . To increase the significance of our analysis of read distributions , we decided to incorporate further neuronal and NSC datasets into our analysis , which were available to us as control data from existing DamID studies ( [Marshall et al . , 2016] and unpublished data ) . Including our previously described neuronal data , we examined the distribution of twelve adult and larval neuronal Dam accessibility datasets ( derived from individual post-mitotic neuronal subtypes , cholinergic , GABAergic , and glutamatergic ) and compared to data derived from NSCs . From these distributions it is apparent that majority of GATC fragments in the genome have very few corresponding mapped reads in all samples , whilst fragments having over ~10 reads per million ( rpm ) were relatively infrequent . In other words , most of the genome is inaccessible or accessible at very low levels , whilst hyper-accessible regions are comparatively rare in all cell types ( Figure 7A ) . The greatest difference between the distributions of neurons and NSCs was apparent at very low read numbers . We observed that there was a significantly greater proportion of GATC fragments with low read counts but without being completely inaccessible ( ~1–3 rpm ) , in the NSC data compared to neurons ( Figure 7A ) . This abundance of genomic regions in the stem cells with low-level chromatin accessibility indicates that open chromatin is more broadly distributed than in NSCs than neurons . We find that this trend is also apparent in the intermediate progenitor cell types , having intermediate amounts of GATC fragments mapping to low read counts ( Figure 7B , Figure 7—figure supplement 1A ) . Conversely , we observed a trend towards greater number of GATC fragments to which zero reads were mapped as differentiation progresses ( Figure 7—figure supplement 1B ) . These data demonstrate that differentiated cells are more likely to have regions of chromatin that are completely inaccessible to Dam , indicating a globally lower amount of accessible chromatin in neurons . These trends are also observed in the midgut cells , implying that global changes to accessible chromatin are a common feature of somatic stem cell lineages in vivo ( Figure 7—figure supplement 1C ) .
Recent studies have provided insights into chromatin accessibility of individual cell-types using accessibility assays coupled with cell sorting ( Pearson et al . , 2016 ) . Whilst these strategies have been proven to produce meaningful biological data , they suffer from being technically challenging , particularly with regards to cell isolation . We have demonstrated that CATaDa yields chromatin accessibility profiles for defined cell types in vivo without the need for cell isolation , fixation or the extraction of naked chromatin . In addition to its ease of use , CATaDa also has the advantage that the marking of accessible DNA occurs in vivo . Due to this , there are no artefacts associated with chemical fixation or washing of the chromatin prior to the assay ( Baranello et al . , 2016 ) . Furthermore , widely used tissue dissociation protocols have recently been shown to result in substantial gene expression artefacts , a risk that is also circumvented by labelling the DNA in vivo ( van den Brink et al . , 2017 ) . In addition , as Dam is expressed in vivo over several hours , the profiles produced will reflect dynamic changes to chromatin structure over the entire time period during which Dam is expressed . CATaDa is limited by its resolution , which is restricted by the frequency of GATC sites in the genome ( median spacing of ~200 bp in Drosophila ) . However , this can be increased by using a modified Dam in conjunction with immunoprecipitation ( Dam-IP ) ( Xiao et al . , 2010; Xiao and Moore , 2011 ) . Due to the dependence of Dam for methylation of GATC sequences , biases may be observed at loci which are depleted for GATC . It is worth noting that extensive sequence biases have also been reported for DNAse and ATAC-seq ( Madrigal , 2015 ) . Although single cell protocols have recently been developed for chromatin accessibility techniques ( Jin et al . , 2015a; Buenrostro et al . , 2015 ) , for routine experiments it is more usual to require a relatively large number of cells . For example , for FAIRE-seq it is recommended to have a minimum of 1 × 106 cells ( Tsompana and Buck , 2014; Simon et al . , 2013 ) , whilst DNase-seq typically requires 1 × 107 cells ( Tsompana and Buck , 2014; Song and Crawford , 2010 ) . In contrast , DamID experiments can be performed with as few as 1000 cells ( Tosti , 2017 ) , therefore CATaDa is likely to also be effective with low cell numbers , making it competitive with ATAC-seq ( 500–50 , 000 cells ) ( Buenrostro et al . , 2013; Marshall et al . , 2016 ) . Furthermore , single cell DamID has also recently been demonstrated , indicating that the minimum number of cells required for CATaDa is one ( Kind et al . , 2015 ) . Although we see instances of false negatives in our peak identification with CATaDa due to lack of available GATC sites for methylation , we also observe a number of peaks unique to CATaDa . Whilst it is possible that these signals are the result of an experimental artefact , it seems likely that they are genuine accessible loci . This could be accounted for by the fact that Dam methylase is a relatively small protein compared to Tn5 transposase ( utilised in ATAC-seq – 55 and 32 KDa respectively [Naumann and Reznikoff , 2002; Boye et al . , 1992] ) , Therefore , Dam may be able to methylate sites which are inaccessible to Tn5 . Alternatively , Dam may be able to methylate some nucleosome bound DNA in which the GATC site is exposed to the nucleoplasm , which may be insufficient for transposition . Whilst the presence of some potential false positives and negatives in the data may be a problem for some applications , overall , CATaDa produces similar overall results to ATAC and FAIRE-seq . As there is already a significant amount of disagreement between ATAC and FAIRE-seq , it is impossible to say whether a unique peak called by any method is a true peak or not , and as these techniques work via different principles , it is likely that these differences reflect biases unique to each approach . Due to the technical differences between CATaDa and currently favoured alternatives ( e . g . ATAC-seq ) , it is clear that the choice of which of these techniques is most appropriate is dependent on the application in question . The limited resolution of CATaDa means that it is not well suited for identifying precise limits of enhancer regions or nucleosome positions when compared to ATAC-seq . However , we have demonstrated that CATaDa resolution is sufficient to investigate broad differences in chromatin accessibility between samples , to identify enriched sequence motifs , and even to identify individual cell-type-specific enhancers . Therefore , the technique is suited to answering questions regarding the biology of cell-types of interest . The requirement for expression of Dam may represent difficulties for non-genetically-tractable model systems , or complicate experimental design if a mutant background is desired , ( although it should be noted that in the latter case , a genetically encoded cell-type-specific marker may have to be included anyway to facilitate cell-separation for ATAC-seq or other approaches – therefore a complicated genetic background may be unavoidable ) . On the other hand , the advantages of being able to assay individual cell types residing within a complex tissue , are clear . If the driver used is specific enough , this may even be achieved with minimal or no dissection . Targeted DamID is rapidly being embraced by the Drosophila community with , at present , over 135 laboratories having requested the reagents and a number of papers already published ( Southall et al . , 2013; Dinges et al . , 2017; Marshall and Brand , 2017; Cheetham and Brand , 2018; Jin et al . , 2015b; Spéder and Brand , 2018 ) . Progress is also being made in adapting it for use in vertebrate models ( Tosti , 2017 ) . Whenever the binding of a protein of interest ( POI ) is investigated with this technique , Dam-only data ( representing chromatin accessibility ) for the cell type being assayed is also generated , as it is the control for which the Dam-POI is normalised . Therefore , researchers performing DamID experiments can now take advantage of this data , getting a ‘2-for-1 deal’ whenever they use TaDa to profile the binding of a POI . Furthermore , much of these data are already available from published studies that could be readily analysed to provide novel biological insights . The sequence of events which lead to repression of open chromatin in the transition between stem cells and their progeny is not well defined . Through identification of functional elements of the genome at various stages in this process , we can begin to understand how the dynamic chromatin landscape impacts the regulation and maintenance of differentiated cell states . Interestingly , we observe that chromatin accessibility in intermediate cell types is broadly more similar to their stem cell precursors than their differentiated progeny in CNS and midgut lineages ( Figure 4B , C , Figure 5B , C , F ) . This indicates that many stem-cell-specific regulatory regions remain accessible in intermediate cell types and are not fully repressed until terminal differentiation . This seems particularly surprising in the case of the EBs in the midgut considering that these cells do not undergo further mitotic divisions and are committed to a particular cell fate before their genesis by Notch signalling in the ISC ( Ohlstein and Spradling , 2007 ) . Furthermore , in the intermediate cells of the larval CNS ( GMCs and immature neurons ) , a relatively high proportion of cells profiled are neurons , which express markers thought to be associated with fully differentiated cells , suggesting that these regulatory regions may be open prior to terminal fate specification . The reason for these regions of chromatin remaining accessible in stem cell progeny is unclear , however there are several plausible explanations . Firstly , regions that are bound by transcription factors that activate transcription may be replaced by transcriptional repressors . Such repressive factors are known to have detectable open chromatin ‘footprints’ , similarly to activating factors ( Mall et al . , 2017; Dinges et al . , 2017 ) . Alternatively , the same factors that bind in the stem cells may recruit new binding partners that alter their activity to initiate repression rather than activation of gene expression . This explanation would require that further modifications occur to the chromatin following repressor activity as open chromatin regions are lost in fully differentiated cells , suggesting that repressors may no longer be bound . Retention of open chromatin regions in intermediate cell types may also reflect increased plasticity , indicating that cell fate has not yet been fully determined and that lineage reversion or dedifferentiation is possible given the introduction of the correct combination of factors . This idea is supported by the fact that immature post-mitotic neurons have been experimentally induced to dedifferentiate by interventions that are ineffective in fully differentiated adult cells ( Southall et al . , 2014; Marshall and Brand , 2017 ) . It has been suggested that in some physiological contexts differentiated cells may revert to replenish stem cell pools ( Kai and Spradling , 2004; Yan et al . , 2017; Cheetham and Brand , 2018; Jin et al . , 2015b ) . This idea may help to explain retention of plasticity in these cell types . It is widely asserted that terminally differentiated cells have limited accessible chromatin whilst their progenitors maintain a broadly open chromatin landscape . However , there are few studies that investigate this phenomenon in vivo . Furthermore , the chromatin state in lineage committed intermediate progenitor cell types has been little studied . With CATaDa we have acquired evidence to indicate that NSCs in the developing Drosophila CNS appear to have more broadly open chromatin landscapes than their fully differentiated progeny . These differences are predominantly in the range of 1 to 3 rpm , which reflects chromatin with very low accessibility . Furthermore , the intermediate or immature progenitors , in both the CNS and midgut , retain a relatively open chromatin state , similar to that of their stem cell precursors ( Figure 4B–C , Figure 5B , D , F ) . Together , these data support the model of stem cells containing more chromatin that has the potential to be accessed , whereas in differentiated cells , this flexibility is reduced and more regions are rendered completely inaccessible . In pluripotent stem cells , the more highly accessible chromatin landscape is thought to promote rapid initiation of multiple gene expression programmes . The fact that this is also a feature of somatic stem cells suggests that these cells may retain an unexpected level of plasticity even after their terminal cell fate has been specified . In conclusion , we have shown that cell-type-specific chromatin accessibility profiles can be obtained through tightly controlled expression of Dam methylase . These data can be used to predict cell-type-specific enhancers , as well as gaining insights into the global regulation of chromatin . Of particular interest are the dynamic changes in accessibility as cells progress towards terminal differentiation ( e . g . the unique open regions observed in GMCs/early neurons ) and the delayed compaction of stem cell gene associated chromatin . Also , analysis of the genome-wide distribution of chromatin accessibility supports a model of gradual compaction of large regions of low accessibility chromatin during differentiation . Overall , our results from profiling developing cell types illuminate the dynamic nature of chromatin accessibility in differentiation , and hint at organising principles which may apply to all somatic stem-cell lineages .
tub-GAL80ts; UAS-LT3-NDam ( Southall et al . , 2013 ) was used to allow cell-specific expression of Dam . The following GAL4 driver lines were used to drive Dam expression in the CNS: wor-GAL4 ( Albertson et al . , 2004 ) for neuroblasts , GMR71C09-GAL4 ( Bloomington #39575 ) for GMCs and newly born neurons and nSyb-GAL4 ( Bloomington #51941 ) for mature larval and adult neurons . For expression of Dam in the gut the following lines were used for ISC , EB , and EC expression , respectively: esg-GAL4 , UAS-2xEYFP/Cyo; Su ( H ) GBE-GAL80/TM3 Sb , Su ( H ) GBE-GAL4 , UAS-CD8GFP/Cyo and {GawB}Myo31DFNP0001/CyO ( Wang et al . , 2014; Jiang et al . , 2009 ) . P{tubP-GAL4}LL7/TM6 , Tb ( Bloomington #5138 ) was used to drive ubiquitous expression in antennal-eye discs . ChaMI04508-T2A-GAL4 , Gad1MI09277-T2A-GAL4 and vGlutMI04979-T2A-GAL4 driver lines were used to drive expression in cholinergic , GABAergic and glutamatergic neurons , respectively ( Diao et al . , 2015 ) . To induce tissue-specific Dam expression , GAL4 driver lines were crossed to GAL80ts; UAS-LT3-NDam virgin females . Embryos were collected for 4 hr then raised at 18°C . Animals were transferred to 29°C at either 7 days after embryo deposition for 24 hr to obtain third instar larval tissues , or three days after eclosion to obtain adult heads . Fifty brains or thirty antennal-eye discs were dissected in PBS with 100 mM EDTA for each replicate . 71C09-GAL4 > UAS-LT3-NDam ventral nerve cords ( VNCs ) were dissected and central brain and optic lobe regions discarded due to presence of observed 71 C09-GAL4 expression in a small subset of central brain neuroblasts . For midgut experiments , Animals were transferred to 29°C for 24 hr at three days after eclosion , and thirty midguts dissected per genotype . The gut regions dissected were between the crop and malpigian tubules . The amount of tissue dissected was chosen to ensure the presence of an appropriate number of Dam-expressing cells ( ~10000 ( Southall et al . , 2013 ) ) . ( For this study , we chose to dissect all tissues due to non-specificity of some drivers , however , this should not be required if driver expression is specific ) . Genomic DNA extraction and sequencing library preparation was performed as described previously ( Marshall et al . , 2016 ) , with minor modifications - MyTaq ( Bioline ) was used for PCR amplification of adapter ligated DNA . Libraries were sequenced using Illumina HiSeq single-end 50 bp sequencing . Two replicates of at least 10 million reads were acquired for each cell type . A third replicate was acquired for GMR71C09-GAL4 to allow for comparisons in Figure 7 . Sequencing data were mapped back to release 6 . 03 of the Drosophila genome using a previously described pipeline , which was modified to output Dam-only datafiles ( Marshall and Brand , 2015 ) ( available at https://github . com/tonysouthall/damidseq_pipeline_output_Dam-only_data [Southall , 2017a; copy archived at https://github . com/elifesciences-publications/damidseq_pipeline_output_Dam-only_data] ) . This includes mapping reads to the genome using bowtie and assigning to bins delimited by GATC sites . 71C09-GAL4 > UAS-mCD8-GFP third instar larval CNS or adult midgut were dissected in PBS and fixed for 20 min with 4% formaldehyde in PBS , 0 . 5 mM EGTA , 5 mM MgCl2 . Tissues were stained with rat anti-elav ( Developmental Studies Hybridoma Bank ) , chicken anti-GFP ( Thermo scientific ) , mouse anti-Delta ( Developmental Studies Hybridoma Bank ) , and guinea pig anti-deadpan ( kind gift from A . Brand ) . Samples were imaged using a Zeiss LSM510 confocal microscope . Peaks were called and mapped to genes using a custom Perl program ( available at https://github . com/tonysouthall/Peak_Calling_for_CATaDa [Southall , 2017b; copy archived at https://github . com/elifesciences-publications/Peak_Calling_for_CATaDa] ) . In brief , a false discovery rate ( FDR ) was calculated for peaks ( formed of two or more consecutive GATC fragments ) for the individual replicates . Then , each potential peak in the data was assigned a FDR . Any peaks with less than a 1% FDR were classified as significant . Significant peaks present in both replicates were used to form a final peak file . Any gene ( genome release 6 . 11 ) within 5 kb of a peak ( with no other genes in between ) was identified as a potentially regulated gene . Regions of differentially accessible chromatin were identified by calculating GATC fragments for which a difference of >20 rpm was observed between every replicate between at least two cell types . ~97% of these fragments lie within statistically significant ( FDR < 1% ) different ( between respective cell types ) regions . These regions were identified by running the peak calling program on comparison gff files ( generated by subtraction with negative values zeroed ) . The other ~3% are GATC fragments that show >20 rpm difference in isolation , therefore are not identified by the peak calling program . Hierarchical clustering was performed using Morpheus ( Broad Institute , 2017 ) . Sequences for major clusters showing enrichment for a given cell type were then analysed using MEME-ChIP ( Bailey et al . , 2009 ) . Potentially regulated genes , with a peak height of at least 10 rpm were submitted to GOToolBox ( Martin et al . , 2004 ) for GO analysis . GO term enrichments ( frequency in data set divided by expected frequency ) were calculated for each cell type . GATC fragments were identified with at least a 10 rpm difference in all replicates between at least two cell types . Any peak >2 kb from a transcriptional start site that did not overlap with coding sequence was designated an enhancer . Enhancers satisfying these criteria , which were covered by an available Vienna VT ( Kvon et al . , 2014 ) or Janelia FlyLight ( Jenett et al . , 2012 ) GAL4 line were chosen for validation based on the magnitude of the change in accessibility . For comparing Dam data with ATAC and FAIRE , Monte Carlo experiments were performed using a custom Perl script ( available at https://github . com/tonysouthall/Monte_Carlo_simulation [Southall , 2017c; copy archived https://github . com/elifesciences-publications/Monte_Carlo_simulation] ) . Comparisons of areas under curve in Figure 7 were performed using Welch’s ANOVA for heteroscedasticity with Games-Howell post-hoc test in R . For data with equal variances , ANOVA was used with Tukey post-hoc testing ( e . g . Figure 7—figure supplement 1 ) . Results were considered significant at *p<0 . 05 , **p<0 . 01 . Principal component analyses and correlation matrix plots were produced using deepTools ( Ramírez et al . , 2014 ) . Average TSS signal profiles were made using SeqPlots R/Bioconductor package ( Stempor and Ahringer , 2016 ) . All other figures were produced using the ggplot2 package in R . | For an embryo to successfully develop into an adult animal , specific genes must act in different types of cells . Though all the cells have the same genes encoded within their DNA , looking at the way that the DNA is packaged can indicate which parts of the DNA are important for that particular cell type . If regions of DNA are “open” one can infer that those regions are actively involved in gene regulation , whereas “closed” regions are considered less important . It is currently difficult to determine which parts of the DNA are open within an individual cell type in a complex organ , such as the brain . Existing methods require the cells to be physically isolated from the tissue , which is technically challenging . To overcome this issue , Aughey et al . have now developed a method that does not require isolation of the cells . The new technique involves using genetic engineering to introduce an enzyme called Dam into specific cell types in living fruit flies . This enzyme adds a chemical label on regions of open DNA , which can then be detected . Aughey et al . tested this technique on various cells of the developing brain and gut , and were able to see differences in the openness of DNA that corresponded to the action of genes that are important in each cell type . The data also contain trends that help to understand the role of open DNA in development . For example , mature cells were shown to overall have less open DNA than the stem cells that divide to generate them . Aughey et al . hope their new technique will be of use to other researchers working with either fruit flies or mammalian tissues . The knowledge that scientists will gain from identifying how open DNA contributes to gene regulation , in both healthy and diseased tissues , will further our understanding of human development and the biology of diseases such as cancer . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"chromosomes",
"and",
"gene",
"expression",
"developmental",
"biology",
"tools",
"and",
"resources"
] | 2018 | CATaDa reveals global remodelling of chromatin accessibility during stem cell differentiation in vivo |
Hsp104 disaggregase provides thermotolerance in yeast by recovering proteins from aggregates in cooperation with the Hsp70 chaperone . Protein disaggregation involves polypeptide extraction from aggregates and its translocation through the central channel of the Hsp104 hexamer . This process relies on adenosine triphosphate ( ATP ) hydrolysis . Considering that Hsp104 is characterized by low affinity towards ATP and is strongly inhibited by adenosine diphosphate ( ADP ) , we asked how Hsp104 functions at the physiological levels of adenine nucleotides . We demonstrate that physiological levels of ADP highly limit Hsp104 activity . This inhibition , however , is moderated by the Hsp70 chaperone , which allows efficient disaggregation by supporting Hsp104 binding to aggregates but not to non-aggregated , disordered protein substrates . Our results point to an additional level of Hsp104 regulation by Hsp70 , which restricts the potentially toxic protein unfolding activity of Hsp104 to the disaggregation process , providing the yeast protein-recovery system with substrate specificity and efficiency in ATP consumption .
Molecular chaperones maintain protein homeostasis by supporting protein folding and preventing aggregation . However , during severe stress the capacity of this protective system becomes exhausted due to appearance of excess amounts of misfolded proteins and their aggregation . Once physiological conditions are restored , the aggregation process is reversed by specialized chaperones , disaggregases , capable of reactivation of proteins trapped in aggregates ( Aguado et al . , 2015; Doyle et al . , 2013; Kim et al . , 2013; Mogk et al . , 2015 ) . In yeast Saccharomyces cerevisiae this role is played by the Hsp104 chaperone , essential for development of tolerance to stress ( Sanchez and Lindquist , 1990 ) and for propagation of prions ( Chernoff et al . , 1995 ) . Hsp104 is an AAA+ superfamily member ( ATPase Associated with various cellular Activities ) and its homologs , Hsp104/ClpB proteins , are present in fungi , protozoans , plants and bacteria ( Ammelburg and Frickey , 2006; Malinovska et al . , 2015; Mishra and Grover , 2015 ) . Hsp104 consists of four domains: the N-terminal domain , two Nucleotide Binding Domains ( NBD1 and NBD2 ) and a coiled coil region protruding from the NBD1 , called the M-domain ( Lee et al . , 2003 ) . Hsp104 functions as a hexamer in which the Nucleotide Binding Domains form a double ring ( Carroni et al . , 2014; Lee et al . , 2010; Parsell et al . , 1994 ) . Disaggregation is powered by ATP hydrolysis in NBD1 and NBD2 ( Schaupp et al . , 2007; Wendler et al . , 2009 ) . During disaggregation a polypeptide is disentangled from the aggregate , translocated through the central channel , and enabled to fold into the native structure ( Lum et al . , 2004; Tessarz et al . , 2008; Weibezahn et al . , 2004 ) . Hsp104 alone is insufficient for protein recovery from aggregates . A functional disaggregation machinery is composed of Hsp104 cooperating with the Hsp70 chaperone ( Ssa1 in S . cerevisiae ) and its cochaperone Hsp40 ( Ydj1 and Sis1 in S . cerevisiae ) ( Glover and Lindquist , 1998; Krzewska et al . , 2001 ) . Hsp70 interacts with the M-domain of Hsp104 and facilitates Hsp104-mediated protein renaturation at several stages of the process ( Miot et al . , 2011; Rosenzweig et al . , 2013; Schlee et al . , 2004 ) . Firstly , Hsp70 targets Hsp104 to aggregates and stabilizes the Hsp104-aggregate complexes ( Acebrón et al . , 2009; Okuda et al . , 2015; Winkler et al . , 2012 ) . Secondly , Hsp70 activates Hsp104 by affecting the position of the M-domain against the surface of NBD1 ( Haslberger et al . , 2007; Lee et al . , 2013; Lipińska et al . , 2013; Oguchi et al . , 2012; Seyffer et al . , 2012; Sielaff and Tsai , 2010 ) . Finally , Hsp70 might participate in folding events downstream of Hsp104 . The M-domain regulates the activity of the disaggregase . M-domains are assembled around the NBD1 ring in the hexamer . Each M-domain interacts with respective NBD1 via a network of ionic bonds and keeps Hsp104/ClpB proteins in a repressed , disaggregation-compromised state . Hsp70 binding to the M-domain causes its reposition , which transits Hsp104/ClpB into a derepressed state , manifested by highly stimulated ATPase and disaggregation activities ( Oguchi et al . , 2012 ) . The latter state is permanent in the so-called derepressed mutants , which are partially independent from Hsp70 in protein disaggregation and display highly elevated ATPase activity ( Carroni et al . , 2014; Jackrel et al . , 2014; Lee et al . , 2013; Lipińska et al . , 2013; Oguchi et al . , 2012; Seyffer et al . , 2012; Sielaff and Tsai , 2010 ) . The M-domain mediated control of Hsp104 protein is critical for the cell , as the derepressed mutants are toxic ( Lipińska et al . , 2013; Schirmer et al . , 2004 ) . The mechanism coupling the M-domain with the increased Hsp104 activity remains not fully understood . Both Nucleotide Binding Domains of Hsp104 possess typical motifs for an AAA+ protein: Walker A , responsible for ATP binding , Walker B , essential for ATP hydrolysis and arginine fingers , necessary for the intersubunit communication ( Biter et al . , 2012; Schaupp et al . , 2007; Schirmer et al . , 2001; Zeymer et al . , 2014 ) . ATP hydrolysis and translocation of protein substrates through the central channel are in a strict allosteric interdependence: polypeptide processing stimulates the ATPase activity of Hsp104 ( Tessarz et al . , 2008; Woo et al . , 1992; Yamasaki et al . , 2015 ) , while the efficiency of polypeptide binding to Hsp104 is determined by the presence of ATP in NBD1 ( Franzmann et al . , 2011; Schaupp et al . , 2007 ) . Changes of the nucleotide states of NBDs are associated with domain rearrangements and movements of peptide binding loops located in the central channel . These conformational changes couple the energy generated during ATP hydrolysis with the protein-threading force ( Lum et al . , 2004; Wendler et al . , 2009 ) . Remarkably , Hsp104 has low affinity towards ATP and relatively high towards ADP , therefore ADP strongly inhibits ATP hydrolysis ( Franzmann et al . , 2011; Glover and Lindquist , 1998; Grimminger et al . , 2004 ) . For this reason , studies of Hsp104 activity are usually performed under the conditions optimal for its ATPase activity , including saturating ( 5–10 mM ) concentration of ATP and an ATP regeneration system to avoid ADP accumulation . Under such experimental conditions Hsp104 is a potent ATP-hydrolyzing and protein unfolding machine ( Biter et al . , 2012; Desantis et al . , 2014; Franzmann et al . , 2011; Tessarz et al . , 2008 ) . Yet , these nucleotide concentrations are very different from those observed in the cytosol , with ATP at 2 . 1–3 . 4 mM concentration and ADP at 0 . 5–1 . 5 mM , depending on the growth conditions ( Canelas et al . , 2008; Teusink et al . , 2000; Wu et al . , 2006 ) . So far , little has been known about how Hsp104 disaggregase functions under conditions similar to the ones observed in the living cell , where ADP constitutes a significant proportion of adenine nucleotides . Therefore , we addressed the question: how does ADP affect Hsp104 activity ? Here , we report that ADP compromises ATP hydrolysis , substrate binding and translocation by Hsp104 in the absence of the Hsp70 chaperone . We also demonstrate that protein disaggregation at physiological concentrations of ATP and ADP is only effective due to the Hsp70 assistance . Hsp70 promotes Hsp104 binding to the aggregated protein substrate but does not support processing of non-aggregated proteins . Furthermore , we show that the process of polypeptide threading by Hsp104 facilitates binding of ATP , strongly stimulating the ATPase activity of the disaggregase in the presence of ADP . Based on these findings we propose a new level of regulation of the disaggregase by its chaperone partner Hsp70 .
All known Hsp104 activities depend on ATP hydrolysis . In a rough comparison , Hsp104 binds ATP with at least an order of magnitude lower affinity than ADP ( Grimminger et al . , 2004 ) . In the light of the ATP and ADP binding properties we asked how ADP influences Hsp104 functioning . At the saturating ATP concentration ( 10 mM ) , Hsp104 catalyzed the hydrolysis of 90 molecules of ATP per minute ( Figure 1A ) . However , addition of only 1 mM ADP to the reaction mixture reduced the rate of ATP hydrolysis to 30% and under the ATP:ADP ratio 10:4 the ATPase activity dropped to less than 2% ( Figure 1A ) . This illustrates how strong the effect of ADP on Hsp104 activity is , with ATP hydrolysis considerably limited even at the saturating ATP concentration . 10 . 7554/eLife . 15159 . 003Figure 1 . ADP restricts Hsp104 activities . ( A , B ) ADP strongly inhibits the ATPase activity of Hsp104 . The rate of ATP hydrolysis by Hsp104 was assessed ( A ) at 10 mM ATP or ( B ) at 2 . 6 mM ATP and at the indicated concentrations of ADP . Data are the mean of three independent experiments ( ± SD ) . ( C ) ADP inhibits fRCMLa translocation and proteolysis by HAP-ClpP . fRCMLa ( 5 μM ) was incubated at 2 . 6 mM ATP with HAP ( 1 μM ) and ClpP ( 1 . 8 μM ) at the indicated concentrations of ADP and its proteolysis was measured by following changes in fluorescence anisotropy . In a control fRCMLa was incubated with ClpP without HAP ( grey ) . ( D ) The rates of fRCMLa proteolysis by HAP-ClpP were calculated from the slopes of the fluorescence anisotropy curves for each ADP concentration shown in ( C ) and normalized to the HAP activity in the absence of ADP . ( E ) ADP impairs binding of Hsp104 to fRCMLa . Hsp104 E285Q ( 12 μM ) was injected to the reaction mixture containing fRCMLa ( 1 μM ) , at 2 . 6 mM ATP and at the ADP concentrations indicated in the legend . After 200 s , non-labeled RCMLa was added to the final concentration of 40 μM . ( F ) The relative initial rates of fRCMLa binding by Hsp104 E285Q at the indicated ADP concentrations were calculated basing on the fluorescence anisotropy curves . a . u . – arbitrary units . DOI: http://dx . doi . org/10 . 7554/eLife . 15159 . 00310 . 7554/eLife . 15159 . 004Figure 1—figure supplement 1 . ADP inhibition of protein translocation through HAP . ( A ) fRCMLa at the concentrations indicated in the figure legend was incubated with HAP ( 0 . 5 μM ) and ClpP ( 0 . 5 μM ) at 10 mM ATP ( blue ) or at 10 mM ATP and 2 mM ADP ( red ) or in the absence of nucleotides ( grey ) . ( B ) The rates of fRCMLa proteolysis were calculated from the initial slopes of the fluorescence anisotropy curves from ( A ) , normalized to the fRCMLa concentration and shown relatively to the degradation rate at 5 μM fRCMLa and 10 mM ATP . a . u . – arbitrary units . DOI: http://dx . doi . org/10 . 7554/eLife . 15159 . 00410 . 7554/eLife . 15159 . 005Figure 1—figure supplement 2 . ADP effect on ATP hydrolysis by Hsp104 Walker B mutants . ATPase activity of Hsp104 WT , E285Q ( with only NBD2 hydrolytically active ) , and E687Q ( with only NBD1 hydrolytically active ) variants was measured at 2 . 6 mM ATP in the presence ( grey ) or absence ( black ) of 1 mM ADP . Data are the average of three experiments ( ± SD ) . DOI: http://dx . doi . org/10 . 7554/eLife . 15159 . 005 Next , we tested the influence of ADP on the Hsp104 activity at the physiological ATP concentration . Recent measurements in intact yeast cells show that ATP concentration oscillates around 2 . 6 mM ( Ozalp et al . , 2010 ) , which is an intermediate value of the previously reported ATP levels measured in cell extracts ( 2 . 1–3 . 4 mM ) ( Canelas et al . , 2008; Teusink et al . , 2000; Wu et al . , 2006 ) . This ATP concentration will be referred to as physiological thorough this work , although it needs to be kept in mind that the cellular level of ATP changes when yeast metabolism adapts to the growth conditions ( Larsson et al . , 1997; Osorio et al . , 2003; Ozalp et al . , 2010 ) . We measured the ATPase activity of Hsp104 at fixed 2 . 6 mM ATP and increasing ADP concentrations . At the physiological ATP level the rate of ATP hydrolysis was reduced to only 20 molecules of ATP hydrolyzed per minute per Hsp104 monomer ( Figure 1B ) as compared to 90 ATP molecules at the saturating ATP concentration ( Figure 1A ) . ADP in the reaction mixtures caused strong inhibition of the ATPase activity , which dropped to less than 2 ATP molecules per minute at 1 mM ADP . This level of ADP represents the average of the ADP concentrations in yeast cells established in the previous studies ( 0 . 5–1 . 5 mM ) ( Canelas et al . , 2008; Teusink et al . , 2000; Wu et al . , 2006 ) . The degree of inhibition showed similar dependence on ATP:ADP ratio for both ATP levels: 2 . 6 and 10 mM ( Figure 1A , B ) . All the above results show that ADP level is a significant factor determining the ATPase activity of Hsp104 . Considering the strong effect of ADP on Hsp104 ATPase activity we asked how ATP and ADP at the physiological concentrations affect its protein translocation activity . Hsp104 in the absence of the cooperating Hsp70 chaperone is able to bind and translocate proteins which adopt disordered conformation but are not aggregated , such as reduced carboxymethylated lactalbumine ( RCMLa ) ( Bösl et al . , 2005 ) . We labeled RCMLa with fluorescein ( fRCMLa ) and employed it to monitor its translocation by Hsp104 in real time . We used Hsp104 variant termed HAP which was modified to interact with the bacterial protease ClpP so that every polypeptide that has been translocated through the central channel of HAP undergoes proteolysis ( Tessarz et al . , 2008 ) . When fRCMLa was incubated with HAP and ClpP at 2 . 6 mM ATP , we observed a decrease in fluorescence anisotropy , corresponding to fRCMLa degradation and release of short fluorescein-labeled peptides ( Figure 1C ) . We plotted the fRCMLa proteolysis rate against ADP concentrations ( Figure 1D ) . With increasing level of ADP fRCMLa proteolysis by HAP-ClpP slowed down , with a 10-fold reduction of the activity at 1 mM ADP . Thus , under the physiological proportions of adenine nucleotides not only ATP hydrolysis but also translocation of disordered proteins by Hsp104 is impaired . The effect of ADP was similar in an analogous experiments carried out at 10 mM ATP and at different fRCMLa concentrations ( Figure 1—figure supplement 1A , B ) , showing that the degree of inhibition depends predominantly on ATP:ADP ratio . Next , we asked whether the strong inhibition of substrate processing by Hsp104 was associated with a defect in protein binding . We used an Hsp104 E285Q variant in the binding experiments because a complex between this mutant and the substrate is much more stable compared to the WT Hsp104 ( Franzmann et al . , 2011 ) . The E285Q mutation in the Walker B motif affects NBD1 in such way that this domain binds ATP but is not able to hydrolyze it ( Schaupp et al . , 2007 ) . As a result , NBD2 remains the only hydrolytically active domain of Hsp104 . We chose Hsp104 E285Q because its ATPase activity was inhibited by ADP to the same degree as of the WT protein , while Hsp104 with an analogous Walker B substitution in NBD2 ( E687Q ) showed much lower degree of inhibition under the same conditions ( Figure 1—figure supplement 2 ) . We used fluorescence anisotropy to monitor Hsp104 E285Q interaction with fRCMLa . We added Hsp104 E285Q to fRCMLa at constant 2 . 6 mM ATP and changing ADP concentrations and recorded an increase in the anisotropy signal over time , reflecting binding of Hsp104 to fRCMLa ( Figure 1E ) . We plotted the binding rate against ADP concentration ( Figure 1F ) . At 2 . 6 mM ATP we observed fast binding of fRCMLa but with increasing ADP the rate of fRCMLa binding decreased , reaching 3% at 1 mM ADP . Addition of an excess of unlabeled RCMLa to the Hsp104 E285Q-fRCMLa complex caused a drop in fluorescence anisotropy , as the fluorescently labeled substrate was released from Hsp104 ( Figure 1E ) . All the above results show that at the physiological concentrations of adenine nucleotides the biochemical activities of Hsp104 such as ATP hydrolysis , substrate binding and translocation through the central channel are very strongly inhibited . Hsp104 provides thermotolerance in vivo at the physiological concentrations of ATP and ADP , which implies that Hsp104 is active in disaggregation regardless of the disadvantageous nucleotide ratio ( Sanchez and Lindquist , 1990 ) . The recovery of aggregated proteins , however , contrary to the processing of disordered proteins , requires collaboration with the Hsp70-Hsp40 machinery . Therefore , we asked how the physiological concentrations of adenine nucleotides affect protein disaggregation performed by Hsp104 in cooperation with the Hsp70 chaperone . As a substrate we used heat-aggregated GFP , which does not require Hsp70 for folding downstream of Hsp104 . We incubated aggregated GFP with Hsp104 in the presence of Hsp70 ( Ssa1 ) and its co-chaperone Hsp40 ( Sis1 ) at constant 2 . 6 mM ATP and changing ADP concentrations and monitored the recovery of GFP fluorescence ( Figure 2A ) . We plotted the initial GFP folding rate against ADP concentration ( Figure 2B ) . Strikingly , at 1 mM ADP the rate of disaggregation and refolding of GFP was reduced only by 50% . The inhibitory effect of ADP on protein disaggregation by the Hsp104-Hsp70 bi-chaperone system was therefore much milder than the 90% decrease in the protein translocation activity and the 97% reduction in the substrate binding rate observed for Hsp104 alone under the same conditions ( Figure 1D , F ) . This shows that Hsp104 response to ADP is different in the processes dependent on and independent from the Hsp70 chaperone . 10 . 7554/eLife . 15159 . 006Figure 2 . Hsp70 allows Hsp104 to overcome ADP inhibition . ( A ) ADP impact on GFP disaggregation by Hsp104 and the Hsp70 system . Heat-aggregated GFP ( 0 . 04 mg ml-1 ) was incubated with Hsp104 ( 1 μM ) , Ssa1 ( 2 μM ) and Sis1 ( 0 . 4 μM ) at 2 . 6 mM ATP and at the indicated concentrations of ADP . A control experiment was performed without Hsp104 ( grey ) . ( B ) The relative GFP renaturation rates by Hsp104 and Hsp70-Hsp40 at the indicated ADP concentrations were calculated from the initial slope of the fluorescence curves from ( A ) . ( C ) ADP effect on ATP hydrolysis by Hsp104 during the Hsp70-assisted disaggregation . ATPase activity was measured for Hsp104 ( 1 μM ) in the presence of Ssa1 ( 1 μM ) , Sis1 ( 0 . 1 μM ) and aggregated GFP ( 0 . 2 mg ml-1 ) ( green ) or for Hsp104 alone ( blue ) under the conditions as in ( A ) . For comparison , the ATPase activity of Ssa1 , with Sis1 and GFP and without Hsp104 was assessed under the same conditions ( grey ) . Data are the average of three experiments ( ± SD ) . DOI: http://dx . doi . org/10 . 7554/eLife . 15159 . 006 Since protein disaggregation is powered by ATP hydrolysis , we anticipated that the ATPase activity of Hsp104 would also be less affected by ADP during the recovery of aggregated proteins . To verify this , we measured the rate of ATP hydrolysis in the presence of the Hsp70-Hsp40 system ( Ssa1 , Sis1 ) and aggregated GFP ( Figure 2C ) . We halved the concentration of Hsp70 , an ATP consuming protein , to minimize its contribution to ADP production . Under these conditions the ATPase activity of Hsp104 was less sensitive to ADP , similarly as observed in the GFP disaggregation assay ( Figure 2B ) . For example , at the ATP:ADP ratio 2 . 6:1 the rate of ATP hydrolysis was reduced by 50% , whereas for Hsp104 alone the ATPase activity was inhibited by 90% ( Figure 2C ) . Together , these results show that Hsp70 moderates Hsp104 response to ADP . One possible explanation is that Hsp70 releases Hsp104 from ADP inhibition because it interacts with and repositions the M-domain , which abrogates Hsp104 repression and results in the elevated ATPase activity ( Oguchi et al . , 2012 ) . To test this hypothesis we used a derepressed variant of Hsp104 , D484K , in which an ionic interaction between NBD1 and the M-domain is disrupted ( Lipińska et al . , 2013 ) . Hsp104 D484K displayed over five times higher rate of ATP hydrolysis than WT Hsp104 , while the affinity towards ATP was comparable for both protein variants ( the apparent KM values: 4 . 2 mM for WT and 4 . 7 mM for D484K ) . Moreover , the ATPase activity of the derepressed D484K was strongly reduced under the unfavorable ATP:ADP ratio ( Figure 3A ) , similarly as observed for the WT ( Figure 1A ) , regardless of ATP concentration ( Figure 3—figure supplement 1 ) . 10 . 7554/eLife . 15159 . 007Figure 3 . The derepressed D484K variant of Hsp104 is inhibited by ADP . ( A ) The ATPase activity of Hsp104 D484K is strongly affected by ADP . ATPase activity of D484K variant was measured at 10 mM ATP and at the indicated concentrations of ADP . Values are the average of three independent experiments ( ± SD ) . ( B ) ADP inhibits the disaggregation activity of Hsp104 D484K in the absence of Hsp70 . Disaggregation of heat-aggregated GFP ( 0 . 04 mg ml-1 ) by Hsp104 D484K ( 0 . 5 μM ) at 10 mM ATP ( blue ) . After 60 s of the reaction , ADP was added to 2 mM concentration ( red ) . ( C ) ATP regeneration system or ( D ) The Hsp70 system restores the disaggregation activity of Hsp104 D484K . The experiment was initiated as in ( B ) , and after 5 min ( C ) an ATP regeneration system comprising PK ( 0 . 1 mg ml-1 ) and PEP ( 40 mM ) ( purple ) or ( D ) the Hsp70 chaperone system: Ssa1 ( 2 μM ) and Ydj1 ( 0 . 5 μM ) ( green ) was added to the reaction mixture . DOI: http://dx . doi . org/10 . 7554/eLife . 15159 . 00710 . 7554/eLife . 15159 . 008Figure 3—figure supplement 1 . Hsp104 D484K is affected by ADP similarly as the WT Hsp104 . ATPase activity of WT and D484K variant was measured at the physiological ATP concentration ( 2 . 6 mM ) in the absence ( black ) or in the presence of 1 mM ADP ( grey ) . Values are the average of three experiments ( ± SD ) . DOI: http://dx . doi . org/10 . 7554/eLife . 15159 . 00810 . 7554/eLife . 15159 . 009Figure 3—figure supplement 2 . Protein translocation activity of HAP D484K is inhibited by ADP . Proteolysis of fRCMLa ( 5 μM ) by HAP D484K ( 0 . 5 μM ) and ClpP ( 0 . 5 μM ) was carried out at 10 mM ATP ( purple ) or in the absence of nucleotide ( grey ) . Two other reactions were performed at 10 mM ATP and either in the presence of extra 2 mM ADP ( red ) or with an ATP regeneration system ( 0 . 2 mg ml-1 creatine kinase and 120 mM creatine phosphate ) ( blue ) . a . u . – arbitrary units . DOI: http://dx . doi . org/10 . 7554/eLife . 15159 . 00910 . 7554/eLife . 15159 . 010Figure 3—figure supplement 3 . Hsp70 allows efficient disaggregation at low ATP:ADP ratio . Disaggregation of heat-aggregated GFP ( 0 . 04 mg ml-1 ) ( black ) by Hsp104 D484K ( 1 μM ) measured ( A ) in the absence or ( B ) in the presence of the Hsp70 system: Ssa1 ( 2 μM ) and Ydj1 ( 0 . 5 μM ) . The total concentration of adenine nucleotides was 10 mM . At the indicated time points samples were taken from the reaction mixture and subjected to HPLC analysis to assess ADP concentration ( grey ) . DOI: http://dx . doi . org/10 . 7554/eLife . 15159 . 01010 . 7554/eLife . 15159 . 011Figure 3—figure supplement 4 . In the absence of ADP the derepressed Hsp104 D484K is independent of Hsp70 in disaggregation . Disaggregation of heat-aggregated GFP ( 0 . 04 mg ml-1 ) by Hsp104 D484K ( 0 . 5 μM ) was carried out at 10 mM ATP in the presence or absence of Ssa1 ( 2 μM ) and Ydj1 ( 0 . 5 μM ) or an ATP regeneration system ( 0 . 2 mg ml-1 creatine kinase and 120 mM creatine phosphate ) , as indicated . In a control experiment GFP was incubated with Ssa1 ( 2 μM ) , Ydj1 ( 0 . 5 μM ) , without Hsp104 . DOI: http://dx . doi . org/10 . 7554/eLife . 15159 . 011 We also measured ADP impact on the protein translocation activity of Hsp104 D484K . We monitored fRCMLa proteolysis by HAP D484K and ClpP at 10 mM ATP . To compensate for the rapid ADP production by HAP D484K , we included an ATP regeneration system , which increased the rate of fRCMLa degradation three times ( Figure 3—figure supplement 2 ) . At 2 mM ADP a 5-fold inhibition of HAP D484K activity was observed ( Figure 3—figure supplement 2 ) . It is worth noticing , that although the proteolysis rate was much higher for the derepressed D484K variant than for HAP lacking the D484K substitution ( HAP WT ) , a similar degree of inhibition by ADP was observed for both proteins ( Figure 1—figure supplement 1A , B , Figure 3—figure supplement 2 ) . Our results suggest that the reposition of the M-domain , even though stimulating , does not overcome the ADP-dependent inhibition of the ATPase activity and the protein processing activity of Hsp104 . Using Hsp104 D484K variant allowed us to analyze reactivation of aggregated proteins independently of Hsp70 ( Lipińska et al . , 2013 ) . To assess the influence of ADP on this process , we incubated heat-aggregated GFP with Hsp104 D484K at 10 mM ATP . Disaggregation of GFP by Hsp104 D484K proceeded efficiently for only approximately 5 min ( Figure 3B ) . The analysis of ADP concentration in the disaggregation reaction mixture showed that GFP reactivation stopped when ADP exceeded 2 mM concentration ( Figure 3—figure supplement 3A ) . Accordingly , addition of 2 mM ADP after the first minute of the reaction stopped GFP reactivation almost immediately ( Figure 3B ) . These results show that ADP strongly inhibits disaggregation performed by the derepressed Hsp104 variant in the absence of Hsp70 . In agreement with this , when the accumulated ADP was removed from the reaction by an ATP regeneration system , GFP reactivation resumed and proceeded at high rate ( Figure 3C ) . This shows that the derepressed Hsp104 variant efficiently disaggregate proteins from aggregates on its own , as long as it is not inhibited by ADP . Knowing that Hsp70 enables protein reactivation by the WT Hsp104 in the presence of ADP ( Figure 2A ) , we analyzed the influence of Hsp70 on GFP disaggregation by the Hsp104 D484K variant . In the absence of Hsp70 , the recovery of GFP fluorescence stopped almost completely after approximately 5 min of disaggregation . When Hsp70-Hsp40 chaperones were added at this time , GFP reactivation resumed and proceeded efficiently ( Figure 3D ) . In a similar experiment we monitored ADP concentration during GFP reactivation ( Figure 3—figure supplement 3B ) . Hsp70-Hsp40 chaperones allowed Hsp104 to disaggregate GFP in spite of the accumulation of ADP . The reaction proceeded efficiently even at ADP concentration above 2 mM . These results indicate that the Hsp70 chaperone system makes the disaggregation activity of the derepressed Hsp104 D484K variant substantially less sensitive to ADP . Further , we asked if the disaggregation activity of the derepressed Hsp104 D484K variant is additionally stimulated by Hsp70 if the inhibitor , ADP , is removed from the reaction . To address this , we compared renaturation of GFP by Hsp104 D484K performed with and without the Hsp70 system and an ATP regeneration system ( Figure 3—figure supplement 4 ) . When ADP accumulated , Hsp104 D484K was an efficient disaggregase only when its Hsp70 partner was present . However , when ADP was being removed , Hsp104 D484K alone was highly effective in disaggregation and addition of Hsp70 barely stimulated its activity . This indicates that Hsp104 D484K is susceptible to strong inhibition by ADP and only in this context the derepressed disaggregase is Hsp70-dependent . Our results show that the release of the M-domain mediated Hsp104 repression is insufficient for Hsp104 to tolerate ADP . This raises a question , which other function of Hsp70 accounts for overcoming ADP inhibition of protein disaggregation by Hsp104 . Hsp70 not only facilitates disaggregation by the M-domain driven derepression of Hsp104 but also acts in the phase preceding translocation of the substrate by Hsp104 . The upstream role of Hsp70 in disaggregation involves initial aggregate remodeling ( Zietkiewicz et al . , 2006 ) and targeting Hsp104 to aggregates ( Acebrón et al . , 2009; Okuda et al . , 2015; Seyffer et al . , 2012 ) . Therefore , we asked whether the latter function of Hsp70 enables disaggregation in the presence of ADP . To measure the upstream effect of Hsp70 we monitored GFP refolding by the derepressed Hsp104 D484K variant in the presence or absence of the Hsp70 chaperone system ( Ssa1 and Ydj1 ) at the increasing levels of GFP aggregates ( Figure 4A ) . We fitted the Michaelis-Menten curve to the disaggregation rates plotted versus GFP concentrations and calculated the apparent KM values to evaluate Hsp104 affinities for the aggregated substrate . Strikingly , Hsp70 increased Hsp104 affinity towards GFP over 25 times ( Figure 4B ) . When we compared the influence of ADP and Hsp70 on Hsp104 affinity towards aggregates , the effect of Hsp70 was sufficient to compensate for the two-fold increase in the apparent KM caused by ADP ( Figure 4B ) . This result suggests that Hsp70-dependent support in targeting of Hsp104 to aggregates allows overcoming ADP inhibition of protein substrate binding . 10 . 7554/eLife . 15159 . 012Figure 4 . Hsp70 supports Hsp104 in disaggregation , compensating the effect of ADP . ( A ) The opposing effects of ADP and Hsp70 on Hsp104 affinity towards aggregates . Disaggregation of heat-aggregated GFP present at different concentrations by Hsp104 D484K variant ( 0 . 06 μM ) was assessed at 10 mM ATP , with or without ADP ( 1 mM ) , in the presence or absence of Ssa1 ( 2 μM ) and Ydj1 ( 0 . 5 μM ) , as indicated in the legend . The initial rate of recovery of GFP fluorescence was plotted against GFP concentration . Data are the means of three measurements ( ± SD ) . Dashed lines represent Michaelis-Menten curves fitted to each set of experimental data using least squares fitting with the GraphPrism software . a . u . – arbitrary units . ( B ) Apparent KM values calculated for each experiment described in ( A ) . Values are average of three experiments ( ± SD ) . DOI: http://dx . doi . org/10 . 7554/eLife . 15159 . 012 Since Hsp70 recruits Hsp104 to aggregates , we asked whether it also promotes Hsp104 binding to unfolded , but non-aggregated proteins , such as RCMLa . As reported previously , Hsp70 protein interacts with RCMLa ( Bimston et al . , 1998 ) . In accordance , an addition of Hsp70-Hsp40 to fRCMLa resulted in a sharp increase in the fluorescence anisotropy signal , indicating that Hsp70-fRCMLa complex is formed ( Figure 5—figure supplement 1 ) . Next , we examined the Hsp70 and ADP influence on fRCMLa processing by Hsp104 HAP variant accompanied by ClpP . Addition of Hsp70 did not promote fRCMLa translocation and degradation ( Figure 5 ) . Moreover , ADP inhibited the degradation of fRCMLa regardless of the presence of Hsp70 ( Figure 5 ) . Interestingly , addition of Hsp70 did not result in increased fRCMLa proteolysis but inhibition of this process was observed . Since both Hsp104 and Hsp70 are able to bind to the same substrate , the observed inhibition can be explained by competition between Hsp70 and Hsp104 for binding to fRCMLa . 10 . 7554/eLife . 15159 . 013Figure 5 . Hsp70 does not support Hsp104 in processing of disordered , non-aggregated proteins . Proteolysis of fRCMLa ( 5 μM ) by HAP ( 1 μM ) and ClpP ( 1 . 8 μM ) , carried out at 2 . 6 mM ATP with or without 1 mM ADP and in the presence or absence of Ssa1 ( 2 μM ) and Ydj1 ( 0 . 5 μM ) , as indicated . Grey line shows a control experiment , in which HAP was omitted . a . u . – arbitrary units . DOI: http://dx . doi . org/10 . 7554/eLife . 15159 . 01310 . 7554/eLife . 15159 . 014Figure 5—figure supplement 1 . Hsp70 interaction with fRCMLa . Ssa1 ( 2 μM ) and Ydj1 ( 0 . 5 μM ) were injected to the reaction mixture containing fRCMLa ( 5 μM ) at 2 . 6 mM ATP . Complex formation was monitored by following fluorescence anisotropy . a . u . – arbitrary units . DOI: http://dx . doi . org/10 . 7554/eLife . 15159 . 01410 . 7554/eLife . 15159 . 015Figure 5—figure supplement 2 . Hsp70 does not support Hsp104 in processing of α-casein . α-casein ( 20 μM ) was incubated with HAP ( 1 μM ) and ClpP ( 3 . 6 μM ) at 2 . 6 mM ATP in the presence of , optionally , 1 mM ADP , 2 μM Ssa1 and 0 . 5 μM Ydj1 , as indicated , or at 10 mM ATP with an ATP regeneration system comprising 0 . 2 mg ml-1 creatine kinase and 20 mM creatine phosphate ( two last lines ) . In a control experiment α-casein was incubated with ClpP only ( two first lines ) . Proteolysis of α-casein was assessed after 1 hour with SDS-PAGE . DOI: http://dx . doi . org/10 . 7554/eLife . 15159 . 01510 . 7554/eLife . 15159 . 016Figure 5—figure supplement 3 . Derepressed HAP D484K is efficient in translocation of disordered proteins at the physiological ATP and ADP concentrations . fRCMLa ( 5 μM ) proteolysis by HAP ( 1 μM ) or HAP D484K ( 1 μM ) and ClpP ( 1 . 8 μM ) , was carried out at 2 . 6 mM ATP and 1 mM ADP and in the presence of the Hsp70 chaperone system comprising Ssa1 ( 2 μM ) and Ydj1 ( 0 . 5 μM ) . In a control experiment HAP was omitted ( grey ) . DOI: http://dx . doi . org/10 . 7554/eLife . 15159 . 016 To substantiate our results , we assessed Hsp70 influence on HAP-dependent processing of another intrinsically disordered model substrate , α-casein . Hsp70 inhibited its translocation and proteolysis by HAP-ClpP in an analogous way as in case of fRCMLa ( Figure 5—figure supplement 2 ) . The above experiments suggest that while Hsp70 significantly promotes binding of Hsp104 to aggregates , it does not facilitate binding of disordered , non-aggregated protein substrates by the disaggregase . Protein reactivation requires ATP hydrolysis by Hsp104 , an activity that is severely impaired at the physiological concentrations of adenine nucleotides ( Figure 1B ) . However , the impact of ADP on the ATPase activity of Hsp104 becomes weaker during protein reactivation by Hsp104-Hsp70 chaperones ( Figure 2C ) . Considering that: ( i ) Hsp70 acts upstream of Hsp104 ( Figure 4A , B ) , initiating substrate binding and translocation through Hsp104 , and ( ii ) the protein translocation process is associated with elevated ATPase activity of Hsp104 ( Cashikar et al . , 2002; Woo et al . , 1992 ) , we speculated that protein translocation by Hsp104 releases ADP-dependent inhibition of its ATPase activity . To test this hypothesis , we measured the rate of ATP hydrolysis at 2 . 6 mM ATP and 1 mM ADP , with increasing concentrations of the unfolded substrate , RCMLa ( Figure 6A ) . Relatively high concentration of RCMLa was required to compensate for the ADP-dependent inhibition of substrate binding ( Figure 1E , F ) . RCMLa at the saturating concentration stimulated the ATPase activity of the ADP inhibited Hsp104 over 5 times . ( Figure 6A ) . Even higher stimulation ( 19-fold ) was observed for the derepressed Hsp104 D484K variant ( Figure 6B ) . In this case , the ATPase activity reached maximum at lower RCMLa concentration , implying higher affinity of the derepressed Hsp104 for protein substrates ( Figure 6B ) . These results indicate that processing of disordered proteins stimulates the ATPase activity of the disaggregase under the , otherwise limiting , physiological concentrations of adenine nucleotides . 10 . 7554/eLife . 15159 . 017Figure 6 . ADP inhibition is released during protein translocation by Hsp104 . Protein substrate allows efficient ATP hydrolysis at the physiological concentrations of adenine nucleotides . ATPase activity measured for ( A ) Hsp104 or ( B ) Hsp104 D484K variant incubated with different concentrations of RCMLa at 2 . 6 mM ATP and 1 mM ADP . ( C ) Steady-state ATPase activity of Hsp104 plotted against ATP concentration in the presence ( red ) or absence ( blue ) of 50 μM RCMLa . Dashed lines show Michaelis-Menten curves fitted to the experimental data . ( D ) Apparent KM values calculated for the experiments presented in ( C ) . Data represent the mean of three experiments ( ± SD ) . DOI: http://dx . doi . org/10 . 7554/eLife . 15159 . 01710 . 7554/eLife . 15159 . 018Figure 6—figure supplement 1 . Impact of fRCMLa on ATP hydrolysis by Hsp104 is stronger in the presence of ADP . Fold stimulation of the ATPase activity of Hsp104 WT and D484K proteins by the excess of RCMLa ( 100 μM ) . Measurements were carried out at 2 . 6 mM ATP in the presence ( black ) or absence of 1 mM ADP ( grey ) . Data represent the mean of three experiments ( ± SD ) . DOI: http://dx . doi . org/10 . 7554/eLife . 15159 . 018 It is worth noticing that the positive influence of RCMLa on Hsp104 was more pronounced when ADP was present in the reaction mixture ( Figure 6—figure supplement 1 ) . Under the physiological ATP and ADP concentrations , the excess of RCMLa stimulated Hsp104 ATPase activity three times more efficiently as compared to the reaction where no ADP was added . This difference was over two times more pronounced for the derepressed Hsp104 D484K variant . Since ADP competes with ATP for binding to Hsp104 , the observed effects suggest that protein translocation could affect Hsp104 affinity towards nucleotides . To verify it , we measured the ATPase activity of Hsp104 at varying ATP concentrations either at the saturating concentration of RCMLa or in the absence of RCMLa ( Figure 6C ) . This experimental setup allowed us to compare Hsp104 affinities towards ATP in the presence and absence of the translocated substrate ( Figure 6D ) . In the presence of RCMLa , the apparent KM decreased five times , implying that polypeptide translocation highly increases Hsp104 affinity towards ATP . Together , these results show that translocation of the polypeptides through the Hsp104 central channel facilitates ATP binding to Hsp104 and allows to overcome the ADP-dependent inhibition of its ATPase activity . Summarized , our results show that Hsp70 allows Hsp104 to overcome the strong inhibitory effect of ADP on Hsp104 disaggregation activity . While ADP negatively affects binding of the protein substrates , Hsp70 substantially increases Hsp104 affinity towards aggregated proteins . At the same time , Hsp70 does not promote processing of disordered , non-aggregated proteins . As a consequence , at the physiological concentrations of adenine nucleotides and in the presence of the Hsp70 chaperone renaturation of aggregated proteins by Hsp104 is efficient but binding and translocation of non-aggregated proteins is strongly inhibited by ADP .
In this work we investigated in vitro Hsp104 functioning at physiological ATP and ADP concentrations . Under such conditions the activities displayed by Hsp104 alone , including ATP hydrolysis as well as binding and translocation of substrate proteins , are highly limited ( Figure 7 ) . Protein translocation through the central channel is initiated only when Hsp70 specifically recruits Hsp104 for disaggregation by facilitating its interaction with the aggregate . Translocation of the polypeptide substrate leads to the release of ADP inhibition . The release is due to a several-fold increase in Hsp104 affinity towards ATP . Consequently , the process of polypeptide threading highly stimulates the ATPase activity of Hsp104 at non-saturating concentration of ATP and in the presence of the competitive inhibitor , ADP ( Figure 7 ) . The presented results point to a novel , ADP and Hsp70 dependent , level of regulation of the disaggregase . 10 . 7554/eLife . 15159 . 019Figure 7 . The mechanism of Hsp70-mediated Hsp104 activation in the presence of ADP . The cellular level of ADP limits the ATPase activity and restricts Hsp104 interaction with protein substrates . Hsp70 promotes Hsp104 binding to aggregates , but not to intrinsically disordered proteins . When Hsp104 binds a polypeptide and translocation occurs , inhibition by ADP is overcome and the ATPase activity is restored . DOI: http://dx . doi . org/10 . 7554/eLife . 15159 . 019 The strong inhibition by ADP of Hsp104 in the absence of a protein substrate results from the discrepancy between ATP and ADP binding affinities ( Grimminger et al . , 2004 ) . According to the previous studies ( Canelas et al . , 2008; Teusink et al . , 2000; Wu et al . , 2006 ) , the cellular level of ATP ( 2 . 1–3 . 4 mM ) is below saturation for Hsp104 ( ~10 mM ) . Furthermore , depending on growth conditions , residual ADP can constitute a significant fraction of adenine nucleotides in the cytosol ( up to 1 . 5 mM ) ( Canelas et al . , 2008; Larsson et al . , 1997; Osorio et al . , 2003; Ozalp et al . , 2010 ) . The conditions in yeast cytoplasm are therefore far from optimal for the biochemical activity of the disaggregase . In this work we measured the ATPase and the protein-processing activity of Hsp104 under a broad range ADP concentrations , establishing the relation between the degree of inhibition and ATP:ADP ratio . The highest ATP:ADP ratio reported for yeast cells is 7:1 ( Theobald et al . , 1997 ) . Even at slightly higher ATP:ADP ratio 10:1 the activity displayed by Hsp104 alone was highly reduced ( by over 50% ) and under the intermediate physiological ATP:ADP ratio 2 . 6:1 ( Canelas et al . , 2008; Ozalp et al . , 2010 ) it was almost completely abolished ( inhibited by 90% or more ) . In contrast , the disaggregation activity of Hsp104 in the presence of Hsp70 at the ATP:ADP ratio 10:1 was reduced by less than 10% and at 2 . 6:1 - to 50% . 90% degree of inhibition occured below the ATP:ADP ratio 1 . 3:1 , while the lowest proportion of ATP to ADP reported in yeast is 1 . 4:1 ( Hynne et al . , 2001 ) . Therefore , inhibition by ADP and the counteracting effect of Hsp70 regulate Hsp104 activity across a whole range of physiological proportions between ATP and ADP . The mechanism of ADP inhibition of Hsp104 can be explained in the light of the previous studies , which show that ADP predominantly occupies the NBD2 domain . It was proposed that ADP release from NBD2 could be a rate-limiting step in the ATPase cycle ( Franzmann et al . , 2011 ) . Agreeably , we observed that ADP strongly inhibits the ATPase activity in NBD2 in the Hsp104 E285Q variant and exerts much milder effect on ATP hydrolysis by Hsp104 E687Q variant , in which NBD1 is the only active ATPase domain . This shows that ADP inhibits Hsp104 by affecting its NBD2 . Interestingly , WT Hsp104 , in which both NBDs are functional , responds to ADP similarly as the Hsp104 E285Q variant , in which only the NBD2 domain is hydrolytically active . This suggests that in the WT protein the difficulties in ADP release from NBD2 prevent ATP hydrolysis also in NBD1 . It has been established before that when ADP occupies NBD2 , NBD1 is mostly in the nucleotide-free state ( Franzmann et al . , 2011 ) . It has also been reported that the absence of nucleotide in NBD1 affects the ATPase activity of the neighbouring NBDs and interrupts protein binding to Hsp104 ( Schaupp et al . , 2007 ) . This could explain how the inhibitory effect of ADP on NBD2 is allosterically transmitted between all the ATPase domains within the hexamer . Results presented in this work imply an existence of two Hsp104 functional modes , depending on substrate translocation through the central channel: ADP-inhibited and active . In the ADP-inhibited mode both the ATPase activity and polypeptide binding are severely impaired and the protein substrate cannot undergo translocation through Hsp104 . In the active mode the presence of the polypeptide in the central channel facilitates ATP binding , and most likely accelerates the rate-limiting step in the ATPase cycle: ADP release . These findings shed new light on how unfolded substrates stimulate ATP hydrolysis by Hsp104 , however to fully understand the underlying mechanism detailed structural data collected in the presence of a substrate would be required . Our data suggest that transition from ADP-inhibited into the active mode occurs upon a successful interaction with a substrate . Although substrate binding is allowed in the presence of ADP , the probability of this event decreases with the increasing proportion of ADP to ATP . It requires high level of protein substrate to compensate for this effect . This underlines the importance of the upstream role of Hsp70 in disaggregation , which involves initial aggregate remodeling and promoting of interaction between Hsp104 and the aggregate . Due to these activities of Hsp70 , Hsp104 can bind aggregated substrates , which are at low concentration . Curiously , the upstream role of Hsp70 is limited to aggregates . Hsp70 does not support binding of non-aggregated protein substrates , allowing Hsp104 activation only at the face of an aggregate . Thus , by specifically targeting Hsp104 towards aggregates , Hsp70 provides another level of control of the disaggregase . The proposed regulatory mechanism works in addition to the activation of Hsp104 associated with Hsp70-dependent reposition of the M-domain , since the latter effect is not enough to overcome ADP inhibition . This is suggested by the observations that both the wild type Hsp104 and the derepressed D484K mutant of the M-domain are strongly affected by ADP and , in both cases , Hsp70 enables the Hsp104 variants to efficiently disaggregate proteins in the presence of the inhibitor . It needs to be kept in mind , however , that the two roles of Hsp70 are probably interdependent , as stimulation of the ATPase activity via the M-domain might favor formation of the complex between Hsp104 and an aggregate . Interestingly , in the absence of Hsp70 addition of an efficient ATP regeneration system is enough to unleash the disaggregating potential of the derepressed Hsp104 D484K . Thus , when Hsp104 is not subjected to the repression by the M-domain , it depends on the Hsp70 chaperone partner exclusively in the context of the release of ADP inhibition . The restrictive effect of ADP may play an important role in proper and efficient Hsp104 functioning in the cell . Firstly , inhibition of ATP hydrolysis decreases energy dispersing by Hsp104 . Secondly , Hsp104 inhibition by ADP prevents a potentially harmful activity of the disaggregase . Such is manifested upon expression of derepressed Hsp104 variants in yeast , which causes disruption of cytoskeleton and inhibits cell division ( Lipińska et al . , 2013; Schirmer et al . , 2004 ) . This shows that the protein unfolding activity of Hsp104 , although highly advantageous under stress conditions , needs to be under strict control , as instead of aggregates it can turn against other protein targets . Notably , in our in vitro experiments , Hsp70 chaperone supports Hsp104 in processing of aggregates , but not of the model disordered protein substrates . The observation that Hsp70 inhibits both fRCMLa and α-casein proteolysis by the HAP variant of Hsp104 ( and ClpP ) , implies that competition between the chaperones for their binding sites on the unfolded substrates may occur . This points to a plausible shielding role of Hsp70 , protecting non-aggregated , intrinsically disordered proteins against the protein-threading activity of Hsp104 . Translocation of the disordered substrates by HAP WT was at very low level at the physiological concentrations of adenine nucleotides in the presence of Hsp70 ( Figure 5 ) . In comparison , the derepressed HAP D484K variant was still highly active under the same conditions , even upon inhibition by ADP ( Figure 5—figure supplement 3 ) . The high disaggregation activity of the derepressed Hsp104 comes at the price of high and unspecific protein unfolding activity , which might be responsible for the toxic effects of Hsp104 D484K in the cell ( Lipińska et al . , 2013 ) . Meanwhile , the M-domain mediated repression , inhibition by ADP and , supposedly , competition by Hsp70 for binding sites on the partially disordered proteins ensures that the Hsp104 WT is tuned down to the level that is safe for the cell . Due to its remarkable ability to dissolve protein aggregates related to stress , aging , as well as to neurodegenerative diseases ( Erjavec et al . , 2007; Jackrel et al . , 2014; Torrente et al . , 2016 ) , Hsp104 and especially its derepressed variants have been the subject of intense search for their potential therapeutic application ( Castellano et al . , 2015 ) . Our studies show that any plans for application of these protein remodeling machines in the novel environment should involve evaluation of ATP and ADP concentrations , which determine Hsp104 activity . The knowledge of ADP influence on Hsp104 is also important for interpretation of in vitro experiments . As observed during fRCMLa proteolysis by HAP or GFP disaggregation by Hsp104 D484K , accumulating ADP inhibited the reaction before the total pool of the protein substrate has been processed . This affected the renaturation yield , a parameter that should not therefore be used to evaluate Hsp104 activity . Furthermore , the kinetics of the reaction is strongly influenced by the efficiency of different ATP regeneration systems . In the light of Hsp104 dependence on adenine nucleotides , we integrate the established facts and reveal novel aspects of the complex interplay between the processes driving Hsp104-dependent disaggregation: the ATPase cycle , protein translocation trough the central channel , Hsp104 derepression trough the M-domain and the upstream action of Hsp70 . The mechanism which emerges from our work implies that , in combination with the restrictive cellular ATP and ADP concentrations , Hsp70 provides an additional level of Hsp104 control , effectively limiting the toxic unspecific protein unfolding activity of Hsp104 , at the same time supporting efficient recovery of aggregated proteins , the process crucial for cell survival after stress .
Mutations in HSP104 gene were introduced using QuickChange Site Directed Mutagenesis Kit ( Stratagene , USA ) . Published protocols were used for the purification of Hsp104 and its variants , Ydj1 ( Lipińska et al . , 2013 ) , GFP ( Zietkiewicz et al . , 2004 ) , Ssa1 ( Andréasson et al . , 2008 ) and ClpP ( Maurizi et al . , 1994 ) . HAP variant of Hsp104 ( G739I S740G K741F T746A ) was expressed from pET24a-HAP plasmid , a gift from Bernd Bukau ( University of Heidelberg ) . Sis1 was purified from E . coli BL21 ( DE3 ) dnaK carrying plasmid pET11a-SIS1 , a gift from Elizabeth Craig ( University of Wisconsin ) . Purification procedure involved cell lysis with French pressure cell ( Thermo Scientific , USA ) and Q-sepharose FF ( GE-Healthcare , UK ) , SP-sepharose FF ( Sigma-Aldrich , USA ) and hydroxyapatite ( BIO-RAD ) column chromatography . fRCMLa was prepared from α-lactalbumine ( Sigma-Aldrich , USA ) according to ( Bösl et al . , 2005 ) . Non-labelled RCMLa , α-casein , lactic dehydrogenase , pyruvate kinase and creatine kinase were purchased from Sigma-Aldrich , USA . All given protein concentrations refer to monomer . Hsp104 WT , E285Q , E687Q or D484K were incubated at 30°C in the ATPase buffer ( 20 mM HEPES pH 7 . 5 , 25 mM NaCl , 100 mM KCl , 15 mM magnesium acetate , 1 mM DTT , 10% glycerol ) , containing 50 μCi/ml ATPγ33P and the indicated levels of ATP and ADP . Hsp104 concentration in each experiment was adjusted to result in hydrolysis of approximately 20% of ATP in 20 min . Initial rate of release of inorganic phosphate was assessed as described ( Viitanen et al . , 1990 ) . Measurements of Hsp104 ATPase activity during GFP disaggregation were performed in the renaturation buffer ( 40 mM HEPES pH 7 . 5 , 60 mM potassium glutamate , 15 mM magnesium acetate , 5 mM DTT ) . In these experiments , the reaction mixtures containing Ssa1 , Sis1 and aggregated GFP were incubated for 8 min at 30ºC prior to addition of Hsp104 . Steady-state ATPase assay was performed for the assessment of the apparent KM values for Hsp104 and Hsp104 D484K as described previously ( Nørby , 1988 ) . Reaction mixture containing 0 . 25–2 μM Hsp104 , buffer ( 50 mM HEPES pH 7 . 5 , 150 mM KCl , 20 mM magnesium acetate , 10 mM DTT ) , 0 . 265 mM NADH , 100 U/ml lactic dehydrogenase , 70 U/ml pyruvate kinase ( LDH/PK ) and 2 . 8 mM PEP was incubated for 5 min at 30ºC . NADH absorbance was measured at 340 nm in a JASCO V-750 Spectrophotometer . Apparent KM values were calculated by using least squares fitting to the Michaelis–Menten equation , with the GraphPrism software . ATPase activity was calculated for Hsp104 monomer . Samples of the GFP renaturation reaction mixture ( containing 1 μM Hsp104 D484K and optionally 2 μM Ssa1 and 0 . 5 μM Ydj1 ) were analyzed using reversed-phase high performance liquid chromatography ( RP-HPLC ) according to ( Smolenski et al . , 1990 ) on GBC 1150 HPLC Pump , Spectra System AS3000 autosampler , Thermo Finnigan Spectra System UV6000LP . The separation was performed on 50 x 4 , 6 mm HyperClone 3u BDS C18 , 130 A column ( Phenomenex ) . Substrate binding assay was performed as described ( Bösl et al . , 2005 ) . Complex formation between Hsp104 E284Q and fRCMLa or between Ssa1-Ydj1 and fRCMLa was detected by following fluorescence anisotropy signal , measured in a JASCO FP-8000 Fluorescence Spectrometer . fRCMLa ( 1 μM ) was incubated at 30ºC in the renaturation buffer , containing 2 . 6 mM ATP and at the increasing concentrations of ADP . Subsequently , Hsp104 E285Q ( 12 μM ) or Ssa1 ( 2 μM ) with Ydj1 ( 0 . 5 μM ) were added to the reaction mixture . After 200 s , fRCMLa release from Hsp104 was induced by addition of non-labeled RCMLa to the final concentration 40 μM . fRCMLa ( 1 . 25 μM , 5 μM or 20 μM , as indicated ) was incubated for 2 min with ClpP at 30 ºC and subsequently HAP was added . Reaction was carried out in the renaturation buffer , containing 2 . 6 mM or 10 mM ATP and , optionally , ADP or/and Ssa1 and Ydj1 , as indicated . fRCMLa proteolysis by HAP-ClpP was monitored by following fluorescence anisotropy in the JASCO FP-8000 Fluorescence Spectrometer . α-casein ( 20 μM ) was incubated at 30ºC with HAP ( 1 μM ) , ClpP ( 3 . 6 μM ) . Reaction was carried out in the renaturation buffer , containing 2 . 6 mM or 10 mM ATP and , optionally , 1 mM ADP , 2 μM Ssa1 and 0 . 5 μM Ydj1 or an ATP regeneration system comprising 0 . 2 mg ml-1 creatine kinase and 20 mM creatine phosphate . The samples of reaction mixtures before and after incubation were subjected to SDS-PAGE and Coomassie Brilliant Blue staining . GFP renaturation assay was performed as described previously ( Zietkiewicz et al . , 2004 ) , with slight modifications . Briefly , GFP ( 4 mg ml-1 ) was thermally inactivated at 85ºC for 15 min . The reactivation reaction was carried out at 25ºC in the renaturation buffer , with 10 mM or 2 . 6 mM ATP , and , optionally , with the indicated levels of ADP or an ATP regeneration system comprising 0 . 2 mg ml-1 pyruvate kinase and 40 mM phosphoenolpyruvate or 0 . 2 mg ml-1 creatine kinase and 120 mM creatine phosphate . Renaturation was initiated by adding the indicated chaperone proteins: Hsp104 WT ( 1 μM ) , D484K ( 0 . 06–1 μM , as indicated ) , Ssa1 ( 2 μM ) , Ydj1 ( 0 . 5 μM ) or Sis1 ( 0 . 4 μM ) . Sis1 was used instead of Ydj1 for reactions requiring longer incubation times , due to observed low stability of Ydj1 . GFP fluorescence was detected in JASCO FP-8000 Fluorescence Spectrometer or a Beckman Coulter DTX880 microplate reader . | Under stressful conditions , such as high temperatures , many proteins lose their proper structure and clump together to form large irregular aggregates . To combat this effect , living organisms exposed to stress produce specialized proteins called chaperones , which can rescue the damaged proteins from aggregates . Studies into this “disaggregation” process often use budding yeast as a model organism . The protein-recovery machinery in this yeast is composed of a ring-shaped enzyme called Hsp104 , together with a chaperone called Hsp70 and its partner Hsp40 . The Hsp104 enzyme converts molecules of ATP into ADP and uses the energy released from the reaction to move , or “translocate” , damaged proteins through its central channel and release them from the aggregates . Previous studies had reported that ADP negatively affects Hsp104 . Now , Kłosowska et al show that Hsp104 is almost inactive in a test-tube if the concentration of ADP is as high as that found inside a cell . This raises a question: how can Hsp104 efficiently remove proteins from aggregates in cells if the conditions are so unfavorable ? Using purified proteins , Kłosowska et al . go on to show that Hsp104 is able to tolerate the level of ADP found inside cells thanks to the Hsp70 chaperone . The experiments show that ADP weakens Hsp104’s ability to bind proteins while Hsp70 supports this ability and counteracts the negative effect of ADP . Further experiments demonstrate that Hsp104 is less affected by ADP , and binds more readily to ATP , when it is translocating proteins . These findings explain how the yeast disaggregating machinery can work even at relatively high concentrations of ADP , and reveal a new control mechanism in the disaggregation process . Many important proteins have poorly organized fragments that can be recognized by Hsp104 , and if Hsp104 was to bind to and translocate these proteins it could harm the cell . The findings of Kłosowska et al . suggest that Hsp70 helps Hsp104 to specifically bind to and act upon proteins in aggregates , while binding to partly unstructured proteins is limited by the high ADP concentration . Further studies are now needed to understand how the protein-recovery machinery can discriminate between aggregated and non-aggregated proteins . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"biochemistry",
"and",
"chemical",
"biology"
] | 2016 | Adenosine diphosphate restricts the protein remodeling activity of the Hsp104 chaperone to Hsp70 assisted disaggregation |
The control of proliferation and differentiation by tumor suppressor genes suggests that evolution of divergent tumor suppressor repertoires could influence species’ regenerative capacity . To directly test that premise , we humanized the zebrafish p53 pathway by introducing regulatory and coding sequences of the human tumor suppressor ARF into the zebrafish genome . ARF was dormant during development , in uninjured adult fins , and during wound healing , but was highly expressed in the blastema during epimorphic fin regeneration after amputation . Regenerative , but not developmental signals resulted in binding of zebrafish E2f to the human ARF promoter and activated conserved ARF-dependent Tp53 functions . The context-dependent activation of ARF did not affect growth and development but inhibited regeneration , an unexpected distinct tumor suppressor response to regenerative versus developmental environments . The antagonistic pleiotropic characteristics of ARF as both tumor and regeneration suppressor imply that inducing epimorphic regeneration clinically would require modulation of ARF –p53 axis activation .
Urodele amphibians and teleost fish are unique among vertebrates in that they possess the ability to regenerate injured complex structures such as limbs , fins , jaws , and heart by epimorphic regeneration ( Morgan , 1901; Brockes and Kumar , 2008; Poss , 2010 ) . For example , zebrafish fin regeneration proceeds through steps that include wound healing , blastema formation , and regenerative outgrowth to faithfully restore preinjury structures and size of the fin ( Poss et al . , 2003 ) . In such highly regenerative species , the blastema consists of a heterogeneous pool of highly proliferative mesenchymal cells that gives rise to the large amount of new tissue in the regenerate ( Knopf et al . , 2011; Tu and Johnson , 2011 ) . In contrast , absence of a proliferative blastema is a prominent feature of most mammalian solid tissue injury responses ( Muneoka et al . , 2008; Straube and Tanaka , 2006 ) . An open question in biology is how cellular mechanisms controlling proliferation affect the blastema and whether they have evolved to contribute to divergent regenerative capacities among vertebrate species . Tumor suppressor genes control the proliferative and differentiated state of cells , and many are also developmental regulators critical for normal formation of tissues ( Jacks et al . , 1992; Berman et al . , 2008 ) . The complex and precisely controlled proliferation and differentiation that occurs during epimorphic regeneration likely requires similar machinery , and as a group , tumor suppressors are probably necessary for well-orchestrated regeneration to occur ( Pomerantz and Blau , 2013 ) . For example , in eukaryotes the retinoblastoma gene Rb1 regulates the G1/S transition by sequestering E2f transcription factors , and it controls cellular differentiation by associating with chromatin modifiers to regulate activity of tissue-specific transcription factors . Therefore , the role of Rb1 in tumor suppression is likely less important from an evolutionary standpoint than its ancient broad functions in regulating cellular differentiation and tissue formation . In contrast , the mammalian gene Cdkn2a is an essential tumor suppressor in mice and humans , but it is dispensable for mammalian development and tissue formation ( Serrano et al . , 1996 ) . In mammals , Cdkn2a encodes two structurally unrelated proteins translated via alternate reading frames , p16Ink4a and Arf , each of which is a tumor suppressor ( Chin et al . , 1998; Sherr , 2006 ) . While p16Ink4a is a cyclin-dependent kinase inhibitor ( CKI ) that functions upstream of Rb1 , Arf exerts its tumor suppressor function by responding to inappropriate Rb pathway signaling above a presumed threshold ( Lowe and Sherr , 2003 ) . When induced , it stabilizes and activates Tp53 by binding and sequestering Mdm2 , an E3 ubiquitin ligase and negative regulator of Tp53 ( Pomerantz et al . , 1998; Weber et al . , 1999 ) . Depending on the context , stabilized Tp53 either promotes cell cycle arrest or apoptosis . In addition to canonical Tp53-dependent functions , Arf has other important functions including controlling ribosome biogenesis and responding to oxidative stress ( Sherr , 2006; Weber et al . , 2000; Damalas et al . , 2011; Menendez et al . , 2003 ) . The resulting general function of Arf is to maintain the postmitotic state , and we have previously shown that suppression of Arf in the context of compromise of the Rb pathway results in dedifferentiation and proliferation of mammalian muscle cells in culture ( Pajcini et al . , 2010 ) . Unlike in development or in regeneration of epithelial and hematopoietic tissues , reversal of the postmitotic state and dedifferentiation also occur in lower vertebrate epimorphic regeneration scenarios that involve a blastema . Regulation of Tp53 has recently been shown to be important during epimorphic regeneration , where it is downregulated during blastema formation ( Yun et al . , 2013 ) . Although cell cycle reentry of postmitotic cells and dedifferentiation are characteristics of malignant transformation which tumor suppressor mechanisms oppose , why these two processes are permitted to occur in the context of intact tumor suppressor mechanisms during epimorphic regeneration is unknown . How evolution of the central cellular growth and tumor suppressor pathways impacts regenerative capacity is poorly understood . The advent of somatic stem cells in metazoans is thought to have enabled the formation of new types of cancer , thus requiring advanced tumor suppressor mechanisms ( Belyi et al . , 2010; Pearson and Sanchez Alvarado , 2008 ) . Among metazoan species , including vertebrates , selective pressures such as different physiologies and environmental exposures undoubtedly continue to apply pressure to generate species-specific tumor suppressor repertoires . For some tumor suppressor genes such as Tp53 , multiple family members have evolved to carry out certain differentiation functions separately from tumor suppression . For others , such as Arf , a single member has evolved and exists in a limited number of species . Whether such differences in turn relate to distinct regenerative capacities remains unknown . Although tumor suppressors are generally highly conserved in eukaryotes , Arf is unusual in that it is poorly conserved in non-mammalian lineages ( Figure 1A ) . The Cdkn2a/b locus of teleost fish , including zebrafish ( Danio rerio ) and fugu ( Takifugu rubripes ) ( Gilley and Fried , 2001 ) , exists as a single protein-producing unit that only encodes for a CKI . During evolution , Cdkn2a and Cdkn2b developed into two separate but related genes encoding for biochemically related CKIs . Arf is not a CKI and is not closely related to either Cdkn2a or Cdkn2b . Arf is thought to be the product of a genetic duplication caused by either an insertion or transposition into the Cdkn2a/b locus ( Gil and Peters , 2006 ) . Of the highly regenerative species for which genomes have been completely sequenced , none possess an ortholog of Arf ( Figure 1A ) ( Flicek et al . , 2014; Karolchik et al . , 2014 ) . The earliest documented ortholog of Arf exists in the chicken genome ( Kim et al . , 2003 ) . This restricted representation coupled with ARF functions of responding to a high threshold of proliferative signaling and inhibiting dedifferentiation ( Pajcini et al . , 2010; Sherr , 2006 ) is compatible with a hypothesis that the presence of Arf could impact regenerative capacity . 10 . 7554/eLife . 07702 . 003Figure 1 . ARF , not normally present in highly regenerative vertebrates , is specifically activated in blastemas of ARF transgenic zebrafish . ( A ) Comparison of amino acid sequences of proteins produced by the Cdkn2a/b loci of zebrafish ( Danio rerio ) , amphibians including the axolotl ( Ambystoma mexicanum ) and western clawed frog ( Xenopus tropicalis ) , chickens ( Gallus gallus ) , and mammals including the mouse ( Mus musculus ) and human ( Homo sapiens ) . While Cdkn2a and Cdkn2b are conserved and encode Ink4 orthologs , Arf evolved recently and orthologs do not exist in highly regenerative vertebrates including teleost fish and urodele amphibians . ( B ) Schematic of transgene expressing cytoplasmic Green fluorescent protein ( GFP ) under the control of the human ARF promoter ( top ) . The promoter consists of human regulatory sequences 736 bp upstream of the transcriptional start site ( TSS ) of ARF . Immunostaining ( wide-field images ) for GFP at 24 hpf , 48 hpf , and 72 hpf in wild type ( WT ) and ARF:GFP embryos ( bottom ) . Scale bars: 200 μm . GFP expression is visible in the hearts of transgenic fish due to presence of a separate transgene used for selection ( cmlc2:GFP ) . ( C ) Whole-mount in situ hybridization for GFP at 24 hpf , 48 hpf , and 72 hpf in WT and ARF:GFP embryos . Scale bars: 100 μm . Alkaline phosphatase staining is detected in the hearts of transgenic fish ( arrow heads ) because of the selection transgene as in ( B ) . ( D ) Confocal images of coronal vibratome sections immunostained for GFP , Msxb , and Proliferating cell nuclear antigen ( PCNA ) at 2 dpa in WT and ARF:GFP fins . Scale bars: 50 μm . GFP expression is induced in the proliferative blastema of the regenerate , but it is not expressed in the surrounding epithelium . White dashed lines represent amputation planes . ( E ) GFP intensity ( green line ) in the regenerates of ARF:GFP transgenic fish relative to WT fish after amputation . The black line represents the percentage of EdU + cells in the regenerates of WT fish after amputation ( N=3; secondary axis ) . Figure supplement 1 shows in vitro ARF promoter assays . Figure supplement 2 shows additional images for panels B , D , and E . Figure supplement 3 shows wound healing in WT and ARF:GFP fins . Results are shown as mean ± standard deviation . hpa: Hours post amputation . DOI:http://dx . doi . org/10 . 7554/eLife . 07702 . 00310 . 7554/eLife . 07702 . 004Figure 1—figure supplement 1 . In vitro analysis of ARF promoter constructs in zebrafish and human cells . Representative luciferase reporter data of three replicates: relative luminescence generated by ARF promoter-firefly luciferase reporter constructs transfected into zebrafish ( ZKS , ZF4 ) and human ( HeLa ) cells . Two ARF promoter-reporter constructs were tested; one contained sequences up to 3 . 4 kb upstream of the TSS of ARF ( 3 . 4 kb ) , while the other contained sequences up to 736 bp upstream of the TSS of ARF ( 736 bp ) . Relative luminescence was measured by normalizing firefly luciferase values to those detected from a Renilla luciferase construct used as a transfection efficiency control . The relative luminescence values were then normalized to those of the negative control construct , pcDNA . Any values above 2 ( black dashed line ) are significant ( N = 3; pcDNA = 1; p<0 . 05 ) . Results are shown as mean ± standard deviation . ZKS: Zebrafish kidney stromal . DOI:http://dx . doi . org/10 . 7554/eLife . 07702 . 00410 . 7554/eLife . 07702 . 005Figure 1—figure supplement 2 . GFP reporter activity in WT or ARF:GFP zebrafish at multiple developmental and regenerative time points . ( A ) Immunostaining ( sagittal confocal images ) for GFP at 24 hpf , 48 hpf , and 72 hpf in WT and ARF:GFP embryos . Scale bars: 200 μm . GFP expression is restricted to the hearts of transgenic fish due to presence of a separate transgene used for selection ( cmlc2:GFP ) . ( B ) Confocal images from Figure 1 of coronal vibratome sections immunostained for GFP , Msxb , and PCNA at 2 dpa in WT and ARF:GFP fins . Scale bars: 50 μm . Included to the right of the figure are insets showing Msxb + , PCNA + , GFP- blastema cells in WT fins and cytoplasmic GFP expression in Msxb + , PCNA + blastema cells in ARF:GFP fins ( white dashed boxes ) . DAPI is excluded from the inset images to improve clarity of costaining . Scale bars: 10 μm . ( C ) Wide-field epifluorescent images of WT and ARF:GFP fins at multiple time points during fin regeneration . GFP intensity of individual ARF:GFP images was evaluated relative to that of WT images at the same time points , and the resulting values were plotted in Figure 1E . There is a small amount of detectable autofluorescence below the amputation plane in the regenerating wild-type and transgenic fins . Scale bars: 100 μm . Dashed lines represent amputation planes . GFP: Green fluorescent protein; hpa: Hours postamputation; PCNA: Proliferating cell nuclear antigen; WT: Wild type . DOI:http://dx . doi . org/10 . 7554/eLife . 07702 . 00510 . 7554/eLife . 07702 . 006Figure 1—figure supplement 3 . ARF is not activated during wound healing in the absence of a blastema . ( A ) At day 0 , dorsal fin lobes were wounded ( interray laceration , 0 dpi ) , while ventral fin lobes were amputated ( 0 dpa ) . At day 1 , GFP expression was assayed in the healing ( dorsal ) and regenerating ( ventral ) fins . Scale bars: 0 . 5 mm . ( B ) Representative images ( sagittal confocal images ) of cytoplasmic GFP expression in WT and ARF:GFP fins that were either amputated ( 1 dpa ) or wounded ( 1 dpi ) . GFP is only detected in ARF:GFP fins that have been amputated ( N = 5 ) . Scale bars: 50 μm . Arrows point to the interray wound . Dashed lines represent amputation planes . GFP: Green fluorescent protein; WT: Wild type . DOI:http://dx . doi . org/10 . 7554/eLife . 07702 . 006 In this study , we used transgenesis to examine the activity of human ARF in the context of zebrafish fin regeneration . We showed that ARF activated zebrafish Tp53 functions to restrict cellular proliferation and induced apoptosis , which caused a marked suppression of fin regeneration after injury . Remarkably , the human ARF regulatory sequences are dormant throughout zebrafish development but induce ARF expression specifically during regeneration after injury . These findings provide experimental evidence that species-specific tumor suppressors can impact tissue regeneration potential .
Using the Ensembl Genome Browser ( Flicek et al . , 2014 ) , the University of California , Santa Cruz ( UCSC ) Genome Browser ( Karolchik et al . , 2014 ) , and Sal-Site ( Smith et al . , 2005 ) , we analyzed the Ink4b-Arf-Ink4a locus in the genomes of six different vertebrate species including highly regenerative ( teleost fish and urodele amphibians ) and poorly regenerative ( avians and mammals ) vertebrates ( Monaghan and Maden , 2013 ) . Our analysis confirms prior reports ( Kim et al . , 2003 ) that an ARF ancestor exists in chickens . We found that in contrast to Ink4 orthologs , which are pervasive throughout vertebrate genomes , ARF orthologs are not present in the genomes of highly regenerative vertebrates ( Figure 1A ) . The results of this analysis , while not directly demonstrating an association of ARF with regeneration , support the hypothesis and prompted our investigation . To investigate how the ARF gene responds to environmental cues , we generated reporter fish in which green fluorescent protein ( GFP ) is expressed under the control of the human ARF promoter ( Tg ( ARF:GFP ) or ARF:GFP ) ( Figure 1B , top ) . In mammals , ARF expression is regulated by a promoter that contains several putative E2F binding sites , and ARF expression can be regulated by free E2F levels above a threshold ( Gil and Peters , 2006 ) . The ARF promoter has previously been empirically defined ( del Arroyo et al . , 2007 ) and no other regulatory sequences or enhancers have been described to date . We first confirmed that the human ARF promoter can function in zebrafish cells in vitro in transfection experiments using previously described firefly luciferase reporter constructs ( del Arroyo et al . , 2007 ) . Experiments were performed with ZF4 and zebrafish kidney stromal ( ZKS ) ( Stachura et al . , 2009 ) cells with HeLa cells used as a positive control since they express high levels of endogenous ARF ( Figure 1—figure supplement 1 ) . These assays confirmed that the 736 base pair ( bp ) promoter is active in HeLa cells and in zebrafish lines . Therefore , we chose the 736 bp genomic fragment encompassing the human ARF promoter to generate ARF transgenics that mimic regulation of the human ARF gene . Tol2-mediated transgenesis ( Kwan et al . , 2007 ) was used to generate ARF:GFP fish , and transgenic fish were detected using cardiac GFP expression driven by a separate cmlc2:GFP cassette on the transgene . We monitored expression of GFP driven by the human ARF promoter during normal development and in adult fish after injury and during regeneration . To determine if ARF:GFP is active during organogenesis in the zebrafish embryo , we assayed GFP expression at three developmental time points , 24 , 48 , and 72 hr postfertilization ( hpf ) . We were unable to detect ARF:GFP expression in the embryo head , body , or tail by wide field epi or confocal immunofluorescence indicating that the ARF promoter is silent at these developmental stages ( Figure 1B , Figure 1—figure supplement 2A ) . GFP expression in the heart driven by the cmlc2 promoter serves as an internal control . To confirm that our analysis was not significantly compromised by limits of detection , we also performed in situ hybridization for GFP transcripts ( Figure 1C ) . The in situ hybridization results confirmed the immunofluorescence analysis and indicate that ARF is silent or minimally expressed during development , including the developing tail fin region . To investigate ARF activation during regeneration , ARF:GFP regenerates were assessed at the time of amputation and then at time points during which wound healing , blastema formation , and outgrowth of regenerating fins occurs ( Poss et al . , 2003 ) . ARF:GFP transgenic fish regenerated their fins normally . In stark contrast to development , ARF:GFP was induced and highly expressed in the blastema of regenerating adult fins . GFP was specifically detected in ARF:GFP fins after amputation , and GFP colocalized with Msxb and proliferating cell nuclear antigen ( PCNA ) expressing cells ( Figure 1D , Figure 1—figure supplement 2B ) . GFP was not detected in the surrounding epithelium . This observation indicates that the ARF promoter is active in at least a subset of Msxb + blastema cells . GFP signal in the regenerate was first detected at 12 hr postamputation ( hpa ) , peaked at 48 hpa , and then declined to undetectable levels within 6 days ( Figure 1E , Figure 1—figure supplement 2C ) . To correlate GFP expression with proliferation in ARF:GFP fins , we assessed EdU incorporation in regenerates at the above time points . There is a low level of proliferation in uninjured fins ( 0 hpa ) , but proliferation quickly increases to maximal levels within 48 hpa and then decreases ( Figure 1E ) . GFP expression mirrored proliferative changes , suggesting that ARF detects and responds to high proliferative signaling in the regenerate . To further examine the specificity of ARF regulation , we examined the response to the creation of an epithelial laceration wound . In this interray wound model , healing occurs without regeneration or blastema formation ( Gauron et al . , 2013 ) . An epithelial wound was created in the dorsal fin lobe ( Figure 1—figure supplement 3A ) and the ventral fin lobe of the same fish was amputated . GFP expression was evaluated 1 day postinjury ( dpi ) ( Figure 1—figure supplement 3B ) . As expected , GFP was detected in the forming ventral blastema . However , GFP was undetectable in the healing wound . These distinct ARF responses to development and to the two different forms of injury indicate that ARF specifically senses and responds to signals particular to the regeneration environment that differ significantly from those present during wound healing or in the highly proliferative environment of developmental organogenesis . In mammalian cells , ARF detects and responds to aberrant inhibition of the Rb pathway ( Sharpless , 2005; Sherr , 2006 ) . To investigate the specific factors that activate ARF during zebrafish fin regeneration , but not during development , we assessed Rb pathway inhibition by Western blot analysis of E2f1 , Rb1 , and hyperphosophorylated-Rb1 ( p-Rb1 ) in developing embryos ( 72 hpf ) , in adult uninjured fin tissue ( uninj . ) and at 2 days postamputation ( dpa ) . Whereas p-Rb1 levels were relatively low in uninjured adult fins , a modest increase was detected in 72 hpf embryos . However , 2 dpa regenerates contained a dramatic increase in p-Rb1 levels despite stable levels of total Rb1 and total E2f1 ( Figure 2A ) . This reflects a high level of pro-proliferation signaling resulting in inactivation of Rb1 by phosphorylation , as occurs commonly in tumors . To further investigate where in the regenerating fin the changes in p-Rb1 phosphorylation occurred , immunostaining of regenerating and uninjured fins was performed . Similar to Western blot analysis , immunostaining revealed a dramatic increase in p-Rb1 staining during regeneration ( Figure 2B ) . A small amount of p-Rb1 staining was observed in uninjured fins , which is most likely the result of homeostatic proliferation ( Wills et al . , 2008 ) . Co-immunostaining for GFP , and Msxb confirmed that p-Rb1 hyperphosphorylation and GFP were co-expressed in cells specifically localized within the blastema . 10 . 7554/eLife . 07702 . 007Figure 2 . Rb1 hyperphosphorylation and E2f1 binding of the human ARF promoter in the blastema during regeneration . ( A ) Representative Western blot of three experimental replicates of Rb pathway components , E2f1 , Rb1 , and hyperphosphorylated Rb1 ( p-Rb1 ) , before injury ( uninj . ) , at 2 dpa , and during embryogenesis at 72 hpf ( left ) . Quantification of p-Rb1 and Rb1 levels normalized to β-Actin and relative to uninjured tissue . Results are from three independent biological replicate experiments and are shown as mean ratios ± standard deviation . *p<0 . 05; ***p<0 . 001 ( right ) . ( B ) Confocal images of coronal vibratome sections immunostained for Green fluorescent protein ( GFP ) , Msxb , and p-Rb1 in uninjured and regenerating ( 2 dpa ) ARF:GFP fins . Scale bars: 50 μm . Very little p-Rb1 staining is seen in the uninjured fin , but high levels of p-Rb1 staining can be seen in Msxb + cells in the blastema at 2 dpa . The white dashed line represents the amputation plane . ( C ) Representative ChIP qPCR data of three experimental replicates with a pool of 30 fins per experiment . Tissue was collected from ARF:GFP transgenic fish before injury ( uninj . ) , at 2 dpa ( regenerate only ) , and at 72 hpf . Fold enrichment of E2f1 binding was normalized to rabbit IgG . The zebrafish thymidine kinase 1 ( tk1 ) promoter was used as a positive control for E2f1 binding . Sequences 2 kbp upstream of tk1 were used as a negative control ( tk1- ) . Values above twofold ( black dashed line ) are significant ( p<0 . 05 ) . Figure supplement 1 shows promoter sequences for the ARF , tk1 , and tk1- promoters annotated for canonical E2f binding sites . hpa: Hours postamputation . DOI:http://dx . doi . org/10 . 7554/eLife . 07702 . 00710 . 7554/eLife . 07702 . 008Figure 2—figure supplement 1 . Promoter sequences evaluated for E2f1 enrichment using an E2f1 antibody to perform a ChIP assay . Both ARF and tk1 promoters contain E2f binding sites ( bold; del Arroyo et al . , 2007 , Tfsitescan ) , but the tk1- promoter does not . DOI:http://dx . doi . org/10 . 7554/eLife . 07702 . 008 The hyperphosphorylation of Rb1 in the blastema suggested that resulting elevated levels of free E2f could be sensed by ARF resulting in transcriptional activation . Moreover , since E2F can directly activate ARF in mammals ( Gil and Peters , 2006 ) , we evaluated interaction of fish E2f1 with the mammalian ARF promoter by chromatin immunoprecipitation ( ChIP ) experiments using an anti-E2f1 antibody in developing ARF:GFP embryos and uninjured and 2 dpa ARF:GFP fins . We analyzed the precipitated DNA fragments by quantitative polymerase chain reaction ( qPCR ) for three specific genomic regions , the ARF promoter , the tk1 promoter ( a known target gene of E2f1; Wells et al . , 2002 ) , and a region 2 kilobases ( kb ) upstream of the tk1 promoter ( tk1- ) as a negative control ( Figure 2—figure supplement 1 ) . We found that in contrast to the state before amputation , ARF is bound by E2f1 specifically during regeneration as is tk1 but not tk1- ( Figure 2C ) . The ChIP assay showed that binding of the ARF promoter by E2f1 was enriched over sixfold relative to non-amputated controls and the tk1- control . The ARF promoter was even more highly enriched than the tk1+ control . Despite the modestly increased p-Rb1 levels in 72 hpf embryos , which correlated with enrichment of E2f1 at the tk1 promoter , no increase in E2f1 binding of the ARF promoter was observed . This finding suggests that the ARF promoter responds strongly and specifically to suprathreshold free E2f1 levels present during regeneration as opposed to other physiological contexts . This result implies that proliferative signaling during fin regeneration has similarities to that during mammalian tumor formation which elicit the ARF tumor suppressor response . Since ARF is a human protein with no orthologs in zebrafish , we confirmed the expected subcellular localization of ARF and stabilization of Tp53 in zebrafish cells in vitro using the zebrafish cell lines , ZF4 and ZKS ( Stachura et al . , 2009 ) . Cells were transfected with an ARF expression construct ( human ARF cDNA subcloned into pcDNA3 ) to determine the subcellular localization of the protein as well as to confirm its interactions with orthologs of its mammalian partner , Mdm2 ( Sherr , 2006; Sharpless , 2005 ) . Confocal imaging showed that human ARF localized to the nucleolus and co-localized with Mdm2 in zebrafish cells ( Figure 3—figure supplement 1A ) . Tp53 levels were examined in fish cells transfected with ARF expression or control constructs . Elevated Tp53 levels were readily observed in approximately 40% of ARF transfected cells ( Figure 3—figure supplement 1A , B ) . The recapitulation of typical localization and Tp53 upregulation suggested conservation of human ARF functions in zebrafish cells and supported investigation of ARF transgenic fish . To investigate the phenotypic effects of ARF on regeneration in vivo , we first utilized the heat shock protein 70 inducible promoter to drive expression of ARF ( Tg ( hsp70l:ARF ) or hs:ARF ) ( Figure 3 ) . hs:ARF fish were subjected to multiple heat shock regimens to determine ARF expression and stability . Immunostaining showed that when induced with an hour long , 37°C heat shock , ARF is robustly expressed in the fin 3 hr later ( Figure 3 ) . Western blot confirmed induction of ARF protein expression and also showed a rapid decrease almost to baseline at 6 hr ( Figure 3 ) , in accordance with the known 6 hr half-life of the human protein ( Sherr , 2006 ) . 10 . 7554/eLife . 07702 . 009Figure 3 . Expression of the mammalian tumor suppressor ARF in zebrafish driven by heat shock promoter . In vivo analysis of transgenic zebrafish expressing human ARF under the control of an inducible heat shock promoter , Tg ( hsp70l:ARF ) ( hs:ARF ) . Schematic of the hs:ARF transgene ( top left ) . The ARF cassette included in the transgene is a cDNA that consists of human exons 1b , 2 , and 3 of CDKN2A . Representative Western blot of 3 replicates of ARF before ( 0 hr ) and 3 and 6 hr post heat shock induction of ARF expression ( top middle ) . Portion of fin shown for analysis of expression in vivo ( top right; dashed box ) . Scale bar: 1 mm . Immunostaining ( sagittal confocal images ) for ARF in adult hs:ARF zebrafish fins at 0 , 3 , and 8 hr after a single , hour long , 37°C heat shock ( bottom ) . Scale bars: 50 μm . ARF expression is maximal at 3 hr post heat shock , and it is undetectable by 8 hr post heat shock . Figure supplement 1 shows in vitro assays . DOI:http://dx . doi . org/10 . 7554/eLife . 07702 . 00910 . 7554/eLife . 07702 . 010Figure 3—figure supplement 1 . Analysis of ARF expression in zebrafish cells . ( A ) Immunofluorescence for Mdm2 and ARF ( top ) and Tp53 ( bottom ) in zebrafish cells ( ZKS ) transfected with pcDNA-ARF . ARF and Mdm2 co-localize in the nucleolus ( arrow ) when ARF is expressed; in cells without ARF , Mdm2 has a diffuse nuclear staining pattern ( arrow head; top ) . Tp53 upregulation depends on ARF expression ( bottom ) . Scale bars: 10 μm . ( B ) Quantification of Tp53 upregulation in zebrafish cells ( ZKS ) transfected with pcDNA-ARF ( N = 100 , p<0 . 01 ) . Results are shown as mean ± standard deviation . DOI:http://dx . doi . org/10 . 7554/eLife . 07702 . 010 We then examined the effects of inducible , transient ARF expression on fin regeneration . Using a regimen of one heat shock 3 hr prior to amputation and then subsequently every 6 hrs up to 6 dpa ( Figure 4A , top ) , fin regeneration in hs:ARF transgenic fish was compared with non-transgenic wild type ( WT ) clutchmates . hs:ARF and WT fish tolerated heat shock well without overt illness or mortality . ARF expression caused significant inhibition of fin regeneration as evidenced by reduced regenerate length and area; WT regenerates measured 1 . 2 ± 0 . 13 mm in length and 5 . 4 ± 1 . 3 mm2 in area compared with hs:ARF regenerates , which measured 0 . 84 ± 0 . 13 mm in length and 3 . 0 ± 0 . 76 mm2 in area , a reduction of approximately 30% ( p<0 . 001 ) and 45% ( p<0 . 001 ) , respectively ( Figure 4B ) . Inducible ARF expression was confirmed in hs:ARF , but not WT fins exposed to heat shock during regeneration ( 4 dpa ) ( Figure 4—figure supplement 1A ) . After the heat shock regimen ended , fin regeneration resumed to reach full length by 14 dpa ( Figure 4—figure supplement 1B ) . Both hs:ARF transgenic fish maintained at 28–30°C and WT fish exposed to heat shock regenerated their fins normally . When previously heat-shocked hs:ARF fins were reamputated and allowed to regenerate in the absence of heat shock ( Figure 4A , bottom ) , the fins regenerated equally as well as WT fins ( Figure 4C ) . This indicates that ARF inhibits fin regeneration in a reversible manner and that its continued expression is required for regeneration suppression . 10 . 7554/eLife . 07702 . 011Figure 4 . ARF suppresses fin regeneration . ( A ) Schematic of heat shock regimen . An initial hour long , 37°C heat shock is delivered 3 hr prior to amputation ( 0 dpa ) and then every 6 hrs thereafter for 6 days . Regenerates are then assessed ( top ) or fins are reamputated ( 0 dpa ) and allowed to regenerate in the absence of heat shock for 6 days ( bottom ) . ( B ) Quantification of regenerate length and area at 6 dpa in WT and hs:ARF fins exposed to the heat shock regimen ( left; N = 40 fins representing multiple different transgene insertions , p<0 . 001 ) . Representative images of fin regeneration at 6 dpa in WT and hs:ARF fins exposed to the heat shock regimen ( right ) . ( C ) Quantification of regenerate length and area at 6 dpa in reamputated hs:ARF fins not exposed to heat shock ( left; N = 40 fins , p>0 . 05 ) . Representative image of fin regeneration at 6 dpa in a reamputated hs:ARF fin not exposed to heat shock ( right ) . The dashed lines represent amputation planes . Scale bars: 1 mm . Results are shown as mean ± standard deviation . Figure supplement 1 shows ARF and Tp53 immunostaining at 4 dpa , and tp53 and cdkn1a expression changes with ARF expression . It also shows regeneration at 14 dpa after heat shock was discontinued at 6 dpa . hs: Heat shock; WT: Wild type . n . s . : not significant . DOI:http://dx . doi . org/10 . 7554/eLife . 07702 . 01110 . 7554/eLife . 07702 . 012Figure 4—figure supplement 1 . ARF expression during regeneration promotes Tp53 , tp53 , and cdkn1a upregulation and regeneration inhibition is reversible . ( A ) Representative ( sagittal confocal ) images of WT and hs:ARF fins at 4 dpa . Scale bars: 50 μm . Dashed lines represent amputation planes . ARF localizes to the nucleus ( inset ) . Scale bars: 10 μm . ( B ) Quantification of fin regenerate length and area in WT and hs:ARF at 14 dpa after heat shock was discontinued at 6 dpa ( left; N = 10 fins , p>0 . 05 ) . Representative images of WT and hs:ARF fins at 14 dpa after heat shock was discontinued at 6 dpa ( right ) . ( C ) Representative images of ARF and Tp53 in WT and hs:ARF fins at 4 dpa ( left ) . Scale bar: 10 μm . Tp53 expression is only detected in cells with ARF expression . Quantifications of relative tp53 ( middle ) and cdkn1a ( right ) transcript expression in uninjured ( uninj . ) WT and hs:ARF fin and regenerates at 4 dpa ( N = 3 replicates ) . Expression was normalized to β-Actin transcripts and relative to uninjured fins within each condition . Significant increases in tp53 ( N=5 fins , p<0 . 05 ) and cdkn1a ( N =5 fins , p<0 . 01 ) were observed with ARF expression . Results are shown as mean ± standard deviation . hs: Heat shock; WT: Wild type . n . s . : not significant . DOI:http://dx . doi . org/10 . 7554/eLife . 07702 . 012 To assess whether ARF functions through the p53 pathway to inhibit fin regeneration in vivo , we examined Tp53 protein and transcript levels as well as induction of the p53 target gene cdkn1a ( p21 ) in response to ARF expression at 4 dpa ( Figure 4—figure supplement 1C ) . The induction and stabilization of p53 and induction of cdkn1a transcripts by ARF showed that ARF impacts p53 functions in vivo in fish regenerates . To assess the Tp53-dependence of ARF , we first crossed hs:ARF fish with tp53M214K mutant fish to generate hs:ARF fish that are homozygous for the tp53M214K allele ( hs:ARF; tp53M214K/M214K ) . The tp53M214K mutation abrogates Tp53 transactivation functions ( Berghmans et al . , 2005 ) . Using the same amputation and heat shock regimen , we analyzed fin regeneration and found no difference in regenerate length or area despite ARF expression in tp53 mutant fish ( Figure 5A ) . We also tested Tp53-dependence of ARF regeneration suppression by treating zebrafish with either pifithrin-α ( PFTαSigma , St . Louis , MO ) , an inhibitor of Tp53 transactivation ( Komarov et al . , 1999 ) , or nutlin3a , a molecule that disrupts the Mdm2–Tp53 interaction , thereby stabilizing Tp53 levels ( Yun et al . , 2013; Vassilev et al . , 2004 ) . Treatment of hs:ARF and WT fish with 5 μM PFTα and heat shock increased hs:ARF regenerate length from 0 . 44 ± 0 . 04 mm to 0 . 66 ± 0 . 08 mm , an increase of 50% ( p<0 . 01 ) , and area from 1 . 8 ± 0 . 7 mm2 to 3 . 2 ± 0 . 5 mm2 , an increase of approximately 75% ( p<0 . 05 ) , compared with carrier-treated controls . Fin regeneration of WT fish was not affected by PFTα treatments ( Figure 5B ) . Treatment of WT fish with 5 μM nutlin3a reduced fin regenerate length from approximately 0 . 65 ± 0 . 1 mm to 0 . 45 ± 0 . 1 mm , a decrease of 30% ( p<0 . 01 ) , and area from 4 . 7 ± 1 . 2 mm2 to 2 . 8 ± 0 . 7 mm2 , a reduction of 40% ( p<0 . 05 ) ( Figure 5C , left ) , phenocopying the fin regeneration inhibition phenotype of induced hs:ARF fish ( Figure 5C , right ) . Together , these experiments show that ARF functions through Tp53-dependent mechanisms to inhibit fin regeneration , and also demonstrate the importance of active suppression of Tp53 by Mdm2 . 10 . 7554/eLife . 07702 . 013Figure 5 . Human ARF functions through the Tp53 pathway in fish to suppress regeneration . ( A ) Quantification of regenerate length and area at 6 dpa in tp53M214K/M214K , hs:ARF , and hs:ARF; tp53M214K/M214K fins exposed to the heat shock regimen as in Figure 4 ( left; N = 30 fins ) . Representative images of fin regeneration at 6 dpa in tp53M214K/M214K , hs:ARF , and hs:ARF; tp53M214K/M214K fins exposed to heat shock ( right ) . Scale bars: 1 mm . Immunostaining ( sagittal confocal images ) for ARF in tp53M214K/M214K and hs:ARF; tp53M214K/M214K fins 3 hr after a single heat shock ( right inset ) . Scale bars: 10 μm . Fin regeneration proceeds equally well in tp53M214K/M214K and hs:ARF; tp53M214K/M214K fins exposed to heat shock , but fin regeneration inhibition is observed in hs:ARF fins exposed to heat shock . ( B ) Quantification of regenerate length and area at 4 dpa in wild type ( WT ) and hs:ARF fins exposed to heat shock and 5 μM pifithrin-α ( PFTα ) or 0 . 1% Dimethyl sulfoxide ( DMSO ) ( vehicle ) ( left; N = 8 fins , p<0 . 01 ) . Representative images of fin regeneration at 4 dpa in WT and hs:ARF fins exposed to heat shock and 5 μM PFTα or 0 . 1% DMSO ( right ) . Scale bars: 0 . 5 mm . Inhibition of Tp53 activity with PFTα rescues regeneration suppression by ARF . ( C ) Quantification of regenerate length and area at 4 dpa in WT fins exposed to 5 μM nutlin3a or Ethanol ( EtOH ) ( vehicle ) ( left; N = 8 fins , p<0 . 01 ) . Representative images of fin regeneration at 4 dpa in WT fins exposed to 5 μM nutlin3a or EtOH ( right ) . Scale bars: 0 . 5 mm . Inhibition of Mdm2 with nutlin3a phenocopies ARF expression by suppressing fin regeneration . The dashed lines represent amputation planes . Results are shown as mean ± standard deviation . n . s . : not significant . DOI:http://dx . doi . org/10 . 7554/eLife . 07702 . 013 In order to understand the cellular effects of ARF that lead to inhibition of fin regeneration , we examined apoptosis and proliferation in blastema cells during regeneration with and without ARF expression . To estimate cell proliferation differences between WT and hs:ARF fins , EdU pulse-chase experiments were performed at 2 , 4 , and 6 dpa . EdU incorporation was significantly higher in WT fin regenerates compared with hs:ARF regenerates at all time points examined with the greatest difference occurring at 2 dpa ( 171% , p<0 . 001 ) ( Figure 6A ) . Apoptotic cells in hs:ARF and WT regenerates were analyzed using terminal deoxynucleotidyl transferase dUTP nick end labeling ( TUNEL ) staining . When the percent of TUNEL + cells in WT and hs:ARF regenerates was compared , the incidence of apoptosis increased with ARF expression at all time points examined with the greatest difference occurring at 2 dpa ( 210% , p<0 . 01 ) ( Figure 6B ) . To assess whether ARF directly affects proliferating blastema cells , we also measured EdU incorporation with different heat shock regimens starting at 4 dpa . The results showed that either a single heat shock or 24 hr of heat shocks at 4 dpa significantly reduced the number of cycling cells in the regenerate , demonstrating a direct effect of ARF on the regenerating cell population ( Figure 6C ) . 10 . 7554/eLife . 07702 . 014Figure 6 . ARF suppresses fin regeneration by inducing apoptosis and cell-cycle arrest . ( A ) Quantification of EdU staining at 2 , 4 , and 6 dpa in wild type ( WT ) and hs:ARF fins exposed to heat shock ( left ) . At 2 dpa , 6 . 0% ± 1 . 1% of cells in WT regenerates were EdU + compared with approximately 2 . 2% ± 0 . 8% in Heat shock ( hs ) :ARF regenerates . At 4 dpa , approximately 7 . 4% ± 0 . 6% of cells in WT regenerates were EdU + compared with 4 . 2% ± 0 . 6% in hs:ARF regenerates . At 6 dpa , approximately 6 . 4% ± 0 . 9% of cells in WT regenerates were EdU + compared with 2 . 7% ± 0 . 3% in hs:ARF regenerates . Significantly fewer cycling cells are detected with ARF expression ( N = 10 fins , p<0 . 001 ) . Representative ( left – sagittal confocal , right – longitudinal ) images of EdU staining at 2 dpa in WT and hs:ARF fins exposed to heat shock ( right ) . Scale bars: left – 50 μm , right – 25 μm . Dashed lines represent amputation planes . ( B ) Quantification of Terminal deoxynucleotidyl transferase dUTP nick end labeling ( TUNEL ) staining at 2 , 4 , and 6 dpa in WT and hs:ARF fins exposed to heat shock ( left ) . At 2 dpa , 2 . 2% ± 1 . 2% of cells in WT regenerates were TUNEL + , while 6 . 7% ± 3 . 7% of cells in hs:ARF regenerates were TUNEL + . At 4 dpa , only 2 . 7% ± 1 . 2% of cells in WT regenerates were TUNEL + compared with 4 . 8% ± 0 . 8% in hs:ARF regenerates . At 6 dpa , 2 . 4% ± 0 . 6% of cells in WT regenerates were TUNEL + , while 3 . 1% ± 0 . 5% of cell in hs:ARF regenerates were TUNEL + . Significantly more apoptosis is detected with ARF expression ( N = 10 fins , p<0 . 001 ) . Representative images ( left – sagittal , right – longitudinal ) of TUNEL staining at 2 dpa in WT and hs:ARF fins exposed to heat shock ( right ) . Image quantification was performed on regenerates only . Dashed lines represent amputation planes . Scale bars: left – 50 μm , right – 25 μm . ( C ) Quantification of EdU staining in WT and hs:ARF fins 3 hr after a single heat shock or 24 hr of heat shock delivered at 4 dpa . After a single heat shock , 3 . 3% ± 1 . 5% of cells in WT regenerates were EdU + compared with 1 . 9% ± 0 . 6% in hs:ARF regenerates . After 24 hr of heat shock , 3 . 0% ± 0 . 7% of cells in WT regenerates were EdU + compared with 1 . 2% ± 0 . 4% in hs:ARF regenerates . Significantly fewer cycling cells are detected with ARF expression after blastema formation ( N = 10 fins , p<0 . 001 ) . Results are shown as mean ± standard deviation . DOI:http://dx . doi . org/10 . 7554/eLife . 07702 . 014 Since the ARF promoter is activated specifically in the fin during regeneration , we tested how transgenic fish expressing ARF under control of the endogenous ARF promoter would develop and regenerate . To do so , we generated zebrafish lines from independent transgenic insertions that utilize the human ARF promoter to drive ARF expression ( Tg ( ARF:ARF ) or ARF:ARF ) ( Figure 7A , left ) . ARF expression during development would be expected to adversely affect ARF:ARF fish . We observed , however , that ARF:ARF transgenic fish are viable , develop normally , and have no overt size or morphological differences when compared with age- and sex-matched WT counterparts ( Figure 7A , right ) . Furthermore , examination of ARF:ARF embryos confirmed our findings in ARF:GFP transgenics . In agreement with the predictions of ARF:GFP experiments , there was no effect of the ARF:ARF transgene on survival during early embryogenesis compared with WT fish ( Figure 7—figure supplement 1A ) . We also did not detect ARF expression in embryos , as expected given our findings with ARF:GFP fish ( Figure 7—figure supplement 1A ) . To assess whether ARF , if expressed , would interfere with organogenesis or development , we evaluated the effects of induced ARF expression using hs:ARF fish . Upon heat shock , hs:ARF , but not WT clutches , exhibited drastically reduced survival that was associated with high levels of ARF expression throughout the embryo ( Figure 7—figure supplement 1B ) . This finding indicates that ARF expression is very poorly tolerated by developing embryos and clearly implies that in ARF:ARF fish , ARF is not activated significantly during development to affect normal developmental growth and organogenesis . 10 . 7554/eLife . 07702 . 015Figure 7 . ARF senses regenerative signals and suppresses fin regeneration . ( A ) Schematic of transgene expressing human ARF under the control of the human ARF promoter ( left ) . Representative images of age- and sex-matched ARF:ARF and WT zebrafish ( right; 5 months postfertilization , male ) . Scale bar: 1 cm . ARF:ARF fish are viable , grow to adulthood and are of normal size and patterning . ( B ) Immunostaining ( longitudinal confocal images ) for ARF in ARF:ARF transgenic fish before injury ( uninjured ) and at 2 dpa . Scale bars: 50 μm . ARF is specifically expressed upon injury . The dashed line represents the amputation plane . ( C ) Representative images of fin regeneration at 6 dpa in WT and ARF:ARF fins ( top ) . Scale bars: 1 mm . The dashed lines represent amputation planes . Quantification of regenerate length and area at 6 dpa in WT and ARF:ARF fins ( bottom; N= 10 fins , p<0 . 001 ) . The first set of bars in each graph represents the results from one transgenic line ( Line 1 ) , while the second set of bars represents the results from a second , independent transgenic line ( Line 2 ) . ARF causes marked inhibition of fin regeneration . Results are shown as mean ± standard deviation . Figure supplement 1 shows the embryonic viability of ARF transgenic lines . Figure supplement 2 shows the failure of ARF:ARF fins to completely regenerate after 15 days and even 30 days . Figure supplement 3 shows ARF immunostaining at 6 dpa , Tp53 , tp53 , and cdkn1a expression changes with ARF expression in WT and ARF:ARF fins at 4 dpa , fin regeneration rescue in ARF:ARF fins treated with PFTα , and EdU incorporation studies performed in WT and ARF:ARF fins . TSS: Transcriptional state site; uninj . : Uninjured; WT: Wild type . n . s . : not significant . DOI:http://dx . doi . org/10 . 7554/eLife . 07702 . 01510 . 7554/eLife . 07702 . 016Figure 7—figure supplement 1 . The ARF:ARF transgene does not interfere with development , whereas forced ARF expression causes embryonic lethality . ( A ) Quantification of embryonic mortality at 48 hpf and 72 hpf in wild type ( WT ) and ARF:ARF embryos ( left; N = 90 , p>0 . 05 ) . Representative sagittal confocal images of ARF expression at 24 hpf in WT and ARF:ARF ( right ) . ( B ) Quantification of embryonic mortality at 48 hpf and 72 hpf in WT and hs:ARF embryos exposed to heat shock ( left; N = 90 , p<0 . 001 ) . Representative sagittal confocal images of ARF expression at 27 hpf in WT and hs:ARF embryos 3 hr after a single heat shock ( right ) . Scale bars: 200 μm . Results are shown as mean ± standard deviation . n . s . : not significant . DOI:http://dx . doi . org/10 . 7554/eLife . 07702 . 01610 . 7554/eLife . 07702 . 017Figure 7—figure supplement 2 . ARF:ARF fins do not completely regenerate even after 30 days . Representative images of fin regeneration at 15 dpa and 30 dpa in wild type ( WT ) and ARF:ARF fins . Scale bars: 1 mm . DOI:http://dx . doi . org/10 . 7554/eLife . 07702 . 01710 . 7554/eLife . 07702 . 018Figure 7—figure supplement 3 . ARF:ARF expression and p53-dependent functions during regeneration . ( A ) Representative images of wild type ( WT ) ( left ) and ARF:ARF ( right ) fins at 6 dpa . Scale bars: 1 mm . Representative images of ARF expression at 6 dpa in WT ( left ) and ARF:ARF ( right ) fins . Scale bars: 10 μm . Dashed lines represent amputation planes . ( B ) Representative images of ARF and Tp53 in WT and ARF:ARF fins at 4 dpa ( left ) . Scale bar: 10 μm . Tp53 expression is only detected in cells that express ARF . Quantification of relative tp53 ( middle ) and cdkn1a ( right ) transcript expression in uninjured ( uninj . ) WT and ARF:ARF fin and regenerates at 4 dpa ( N = 3 replicates ) . Expression was normalized to β-Actin transcripts and relative to fins within each condition . Significant increases in tp53 ( N = 5 fins , p<0 . 05 ) and cdkn1a ( N = 5 fins , p<0 . 01 ) were observed with ARF expression . ( C ) Quantification of regenerate length and area at 6 dpa in ARF:ARF fins treated with 0 . 1% Dimethyl sulfoxide ( DMSO ) or 5 μM Pifithrin-α ( PFTα ) ( left; N = 8 fins/condition , p<0 . 01 ) . Representative images of fin regeneration at 6 dpa in ARF:ARF fins treated with 0 . 1% DMSO or 5 μM PFTα ( right ) . Scale bars: 1 mm . Dashed lines represent amputation planes . Treatment with PFTα rescues fin regeneration in ARF:ARF transgenic zebrafish . ( D ) Quantification of EdU staining at 2 , 4 , and 6 dpa in WT and ARF:ARF fins ( left ) . At 2 dpa , 5 . 0% ± 0 . 6% of cells in WT regenerates were EdU + compared with approximately 1 . 4% ± 0 . 4% in hs:ARF regenerates . At 4 dpa , approximately 7 . 0% ± 0 . 7% of cells in WT regenerates were EdU + compared with 1 . 3% ± 0 . 3% in hs:ARF regenerates . At 6 dpa , approximately 7 . 0% ± 1 . 1% of cells in WT regenerates were EdU + compared with 1 . 8% ± 0 . 6% in hs:ARF regenerates . Significantly fewer cycling cells are detected with ARF expression ( N = 10 fins , p<0 . 001 ) . Representative ( left – sagittal confocal , right – longitudinal ) images of EdU staining at 2 dpa in WT and ARF:ARF fins ( right ) . Scale bars: left – 50 μm , right – 25 μm . Dashed lines represent amputation planes . Results are shown as mean ± standard deviation . n . s . : not significant . DOI:http://dx . doi . org/10 . 7554/eLife . 07702 . 018 We then performed fin regeneration experiments with ARF:ARF transgenic fish and WT fish . When ARF:ARF fins were injured , ARF was detected in the regenerate ( Figure 7B ) , and the pattern of expression was similar to GFP expression in ARF:GFP regenerates at the same time point ( Figure 1 ) . When fin regeneration was compared between ARF:ARF transgenic fish and WT fish , ARF:ARF regenerates measured 0 . 55 ± 0 . 17 mm in length and 1 . 8 ± 1 . 3 mm2 in area , while WT regenerates measured 1 . 2 ± 0 . 06 mm in length and 6 . 0 ± 1 . 4 mm2 in area . To rule out position effects of the transgene insertion , a second independent transgenic ARF:ARF line was also assessed . Regenerates of this second ARF:ARF line measured 0 . 6 ± 0 . 2 mm in length and 2 . 3 ± 0 . 96 mm2 in area , while WT regenerates measured 1 . 1 ± 0 . 13 mm in length and 5 . 1 ± 0 . 67 mm2 in area . In all , ARF:ARF regenerates were 55% ( p<0 . 001 ) and 44% ( p<0 . 001 ) shorter and 70% ( p<0 . 001 ) and 55% ( p<0 . 001 ) smaller in area than WT regenerates , and anatomical fin defects persisted 1 month after amputation ( Figure 7C , Figure 7—figure supplement 2 ) . ARF expression persisted in ARF:ARF fins but not WT fins at 6 dpa ( Figure 7—figure supplement 3A ) , a time point at which GFP in no longer observed in regenerated ARF:GFP fins , suggesting ongoing regeneration attempts in ARF:ARF fins . We confirmed that fin regeneration inhibition in ARF:ARF fish was p53 dependent as in hs:ARF fish . Tp53 , tp53 , and cdkn1a expression increased with ARF expression in ARF:ARF fins at 4 dpa ( Figure 7—figure supplement 3B ) . As in hs:ARF fish , treatment of ARF:ARF fins with 5 μM PFTα rescued fin regeneration ( Figure 7—figure supplement 3C ) . Finally , we quantified the cell cycle arrest that is a consequence of ARF expression comparing WT and ARF:ARF fins with EdU pulse-chase labeling at 2 , 4 , and 6 dpa . Similar to hs:ARF fins , ARF:ARF fin regenerates had significantly fewer proliferating cells than WT fin regenerates at all time point assessed with the largest difference observed at 4 dpa ( 81% , p<0 . 001 ) ( Figure 7—figure supplement 3D ) . These results confirm that ARF activation is specific to regenerating tissue and show that the magnitude of activation is sufficient to inhibit regeneration . Thus , the presence of a functional human ARF gene in fish results in a diminished regenerative capacity , including absence of epimorphic regeneration , without significantly affecting other major physiological or developmental characteristics .
In this study , we have experimentally tested the hypothesis that tumor suppressor evolution may impact regenerative capacity . We found that the core mammalian tumor suppressor ARF senses regeneration signals and specifically responds to negatively alter the proliferative balance in the zebrafish blastema , greatly perturbing regeneration . Our findings provide the first in vivo experimental evidence that evolution of tumor suppressors can negatively impact solid tissue regeneration potential . Although the core tumor suppressors as a whole support regenerative processes , the properties of ARF identified in this study are at odds with epimorphic regeneration . This new example of antagonistic pleiotropy adds to previously recognized trade-off characteristics of tumor suppressor genes affecting mammalian stem cell function ( Pardal et al . , 2005; Greaves , 2007; Pomerantz and Blau , 2013; Rodier et al . , 2007 ) and shows that ARF antagonistic properties also manifest in the context of the blastema . The evidence that ARF is a critical tumor suppressor in mammals ( Sherr , 2006 , Sharpless , 2005 ) , but opposes regeneration functions ( Sharpless and DePinho , 2007 ) , suggests that the selective pressure that has driven the evolution of ARF has primarily enhanced tumor suppression either at the expense of or in the absence of regeneration pressures . Although our experiments and those of others ( Gemberling et al . , 2013; Poss et al . , 2003 ) show that the regenerative capacity of zebrafish is vulnerable to single gene alterations , whether altering function of a single gene in mammals would induce the emergence of robust epimorphic regenerative capacity is a much more complex issue . Indeed , the multifactorial genetic differences of highly and less regenerative vertebrates make it unlikely that manipulation of a single gene could enable regeneration . It is notable , however , that Cdkn1a ( p21 ) knockout mice do possess a somewhat enhanced ability to regenerate solid structures ( Heber-Katz et al . , 2013; Clark et al . , 1998 ) such as pinnae , which lack the complex tissue structure of a digit , but nonetheless , demonstrate that alteration of cellular growth control mechanisms can impact regeneration . Moreover , the importance of active repression of ARF to maintain stem cell function ( Molofsky et al . , 2005 ) , and of ARF reduction to facilitate dedifferentiation ( Pajcini et al . , 2010 ) have been documented . Among the core tumor suppressor genes that are frequently inactivated in mammalian tumors , ARF is unique in that it does not have orthologs represented in most vertebrates including highly regenerative species . By contrast , Tp53 , Pten , and Ink4a have distant orthologs , present in invertebrates and vertebrates alike . The transgenesis approach we used to study ARF in fin regeneration made it possible for us to study ARF with its human regulatory components but without increasing CDKN2A CKI gene dosage , which could have been a complicating factor in a transgenic harboring the entire CDKN2A ( INK4A/ARF ) locus . This study extends our previous observations ( Pajcini et al . , 2010 ) that ARF prevents dedifferentiation in muscle cells in culture and provides new evidence that ARF functions in vivo to oppose tissue regeneration . Future experiments will determine whether ARF prevents dedifferentiation in vivo , such as the dedifferentiation of osteoblasts in regenerating fins , or whether it acts on proliferating blastema cells after they have dedifferentiated . Combined , our findings suggest that zebrafish cells are more promiscuous in terms of tolerance to high levels of mitogenic activity , thus permitting the cellular processes required for epimorphic regeneration . It follows that regenerating cells in organisms that have an ARF gene would need to prevent ARF activation or would be inherently more restricted in these activities . We found that ARF recapitulates its core mammalian mechanistic functions in zebrafish cells and tissues . As in mammals , when ARF is overexpressed in zebrafish cells , it associates with Mdm2 , stabilizes Tp53 , and induces cell cycle arrest or apoptosis . This functional conservation over an evolutionary distance demonstrates that cross-species genetic variations can be experimentally examined in the study of regeneration . When ARF expression is driven by its endogenous human promoter in zebrafish cells , activation of the p53 axis occurs specifically in the blastema-regeneration scenario . Remarkably , the inhibitory effect on regeneration by ARF:ARF was stronger than with the heat shock promoter , probably reflecting ongoing surveillance of regenerative signals by the ARF promoter in contrast to fluctuating ARF levels obtained with intermittent heat shock induction of a short half-life protein . In the developing or adult uninjured state , E2f1 is sequestered and inhibited by Rb1 , and ARF is inactive . However , during blastema formation and regeneration , Rb1 hyperphosphorylation is associated with sufficient free E2f1 to activate ARF , which inhibits fin regeneration via a Tp53-dependent mechanism ( Figure 8 ) . Our findings and model are in agreement with the recent proposal that in salamanders the absence of ARF permits downregulation of Tp53 during blastema formation ( Yun et al . , 2013 ) . The responsiveness of ARF to the Rb pathway proliferative signaling characteristic of zebrafish fin regeneration implies that similar mitogenic signaling occurring in a mammalian context would be detected as aberrant , activate ARF–MDM2–TP53 tumor suppressor mechanisms , and oppose regeneration . Our findings are compatible with previous mouse studies showing that ARF is a potent tumor suppressor that is dispensable for normal development ( Serrano et al . , 1996; Kamijo et al . , 1997 ) . Moreover , prior observations that ARF is not developmentally expressed in the majority of tissues in the mouse ( Gromley et al . , 2009; Zindy et al . , 2003 ) support the fidelity of the promoter used in this study . Although the majority of tumor suppressors probably function in regeneration as they do in normal development , the findings of the present study indicate that ARF represents an unusual departure from that paradigm in that the properties that cause it to respond specifically to tumorigenesis also cause it to distinguish regeneration contexts from developmental ones . 10 . 7554/eLife . 07702 . 019Figure 8 . Model of ARF function in the context of Rb pathway activity during zebrafish development and fin regeneration . ARF is not active during development during which a moderate level of mitogenic signaling causes modest phosphorylation of Rb1 ( top ) ; however , during regeneration , high mitogenic signaling induces Rb1 hyperphosphorylation and abundant free E2f1 , which activates ARF and leads to inhibition of regeneration ( bottom ) . The dashed lines represent the amputation plane . DOI:http://dx . doi . org/10 . 7554/eLife . 07702 . 019 We show here how examination of zebrafish that are humanized with respect to candidate regeneration modifiers is informative for understanding disparate regenerative capacities . Such an approach should prove useful for examining other candidate genes and pathways of interest . Our findings with respect to ARF strongly suggest that it is a barrier to mammalian epimorphic regeneration because it interprets the regeneration context as similar to tumorigenesis . It follows conceptually that approaches to induce epimorphic regeneration clinically would need to disrupt ARF–MDM2–TP53 axis activation .
Zebrafish maintenance at 28–30°C and all experiments were approved by the Institutional Animal Care and Use Committee of the University of California , San Francisco . Three- to six-month-old WT or transgenic AB zebrafish were used for all experiments . The Tg ( hsp70l:ARF ) , Tg ( ARF:ARF ) , and Tg ( ARF:GFP ) constructs were created by either subcloning the cDNA of human ARF ( exons 1β , 2 , and 3 of CDKN2A ) or a cytoplasmic EGFP cassette downstream of either the promoter sequences of zebrafish hsp70l ( Halloran et al . , 2000 ) or the human CDKN2A promoter ( Robertson and Jones , 1998 ) , respectively . The ARF promoter was subcloned from pKR19 ( Robertson and Jones , 1998 ) using SalI to excise an approximately 1 kb region of the human promoter , which encompassed 736 bp 5′ of the transcriptional state site of ARF . Sequence information can be found in del Arroyo et al . ( 2007 ) . Tol2-mediated transgenesis was used to generate transgenic animals ( Kwan et al . , 2007 ) . Transgenic animals were detected based on their GFP-positive hearts , due to the transgenes containing a cmlc2:GFP cassette . All transgenic strains were analyzed as hemizygotes . For drug treatment experiments , zebrafish were treated with 5 μM αPTFα in dimethyl sulfoxide ( DMSO ) ( 5 mM stock ) or 5 μM ( - ) -Nutlin-3 ( Cayman , Ann Arbor , MI ) in ethanol ( EtOH ) ( 5 mM stock ) . Water was exchanged daily . For EdU pulse-chase experiments , 5 μL of 5 mg/mL of EdU ( Life Technologies , Carlsbad , CA ) in saline was injected intraperitoneally into anesthetized fish 30 min before tissue harvest . Zebrafish fin immunostaining was performed on whole-mounted fins as previously described ( Sousa et al . , 2011 ) . For coronal views , whole-mount stained fins were embedded in 5% agarose , and 200 μm sections were cut with a vibratome . Imaging was performed with a confocal microscope . Zebrafish embryo immunostaining was performed on whole-mounted , 1-phenyl 2-thiourea ( PTU; Sigma ) -treated embryos as previously described ( Macdonald , 1999 ) . Zebrafish cell immunostaining: 4% paraformaldehyde ( PFA ) 10 min , phosphate-buffered saline ( PBS ) 5 min 3× , 0 . 3% PBTx 15 min , PBS 5 min 3× , serum-free protein block ( Dako , Carpinteria , CA ) 1 hr , primary antibodies in antibody diluent ( Dako ) overnight 4°C , PBS 5 min 3× , secondary antibodies in antibody diluent 1 hr , PBS 5 min 3× , mounted with Vectashield mounting medium with DAPI ( Vector Laboratories , Burlingame , CA ) . EdU incorporation was detected using the Click-iT EdU Imaging Kit per the manufacturer’s instructions ( Life Technologies ) . TUNEL detection was performed using the In Situ Cell Death Detection Kit ( Roche , Basel , Switzerland ) per the manufacturer’s instructions . Images were quantified in ImageJ . The percent of EdU + or TUNEL + cells was quantified by first counting the number of positive cells in the regenerate and then dividing that count by the number of nuclei in the field counted . Zebrafish embryo mRNA in situ hybridization was performed on whole-mounted , PTU-treated embryos as previously described ( Chitramuthu and Bennett , 2013 ) . The antisense GFP probe was labeled with digoxigenin-11-UTP ( Roche ) and generated using the following primers: 5′-AAGGGCGAGGAGCTGTTCAC-3′ and 5′-GAACTCCAGCAGGACCATGT-3′ ( MacDonald et al . , 2010 ) . An amount of 50–60 μg of total protein isolated from adult zebrafish fin tissue was loaded per lane , electrophoresed , and transferred to polyvinyl difluoride ( PVDF ) membranes . Protein was visualized using ECL Prime ( GE Healthcare Bio-Sciences , Pittsburgh , PA ) and an ImageQuant LAS 4000 ( GE Healthcare Bio-Sciences ) . Band quantification was performed using ImageQuantTL software . For each condition , Rb1 and p-Rb1 bands were normalized to β-Actin , and the ratio of p-Rb1:Rb1 was calculated and made relative to uninjured tissue ( Table 1 ) . 10 . 7554/eLife . 07702 . 020Table 1 . Primary antibodies . DOI:http://dx . doi . org/10 . 7554/eLife . 07702 . 020Host speciesAntigenCompanyCat . No . DilutionApplicationMouseTp53Abcamab778131:50IHCRabbitMdm2Santa CruzC-181:50IHCRabbitGFPTorrey PinesTP4011:3000IHCChickenGFPAbcamab139701:3000IHCMouseMsxbDSHC4G1-c1:50IHCRabbitPCNAAbcamab24261:500IHCMousep14ARFCell Signaling24071:100/1:500IHC/WBRabbitRb1AnaSpec554321:500WBRabbitp-Rb1 ( S780 ) Abcamab477631:500WBRabbitBeta-actinMilliporeEP1123Y1:1000WBRabbitE2f1Abcamab147691:1000WBIHC: Immunohistochemistry; PCNA: Proliferating cell nuclear antigen; WB: Western blot . Caudal fin amputations were performed with a razor blade on fish anesthetized with 0 . 016% tricaine in aquarium water; consistently only the distal halves of fins were amputated . Heat shocks were delivered by housing fish in a water bath set to 37°C with bidiurnal water exchanges . The water bath achieved 37°C within 15 min , maintained that temperature for 1 hr , and then passively cooled to fish room temperature ( 26–28°C ) . An automatic digital timer ( Intermatic , Spring Grove , IL ) was used to turn on and off the water bath . For heat shock experiments , an initial heat shock was delivered and then fins were amputated 3 hr later . Heat shocks were subsequently delivered every 6 hrs for the duration of the experiment . Quantification of fin regenerate length , area , and GFP intensity was performed in ImageJ . Fin regenerate length was calculated by averaging the length of the longest dorsal and ventral fin ray from the amputation plane . Caudal fin wounding experiments were performed as previously described ( Gauron et al . , 2013 ) . ChIP of zebrafish fin tissue was performed as previously described ( Wehner et al . , 2014 ) with a Bioruptor UCD-200 ( Diagenode , Denville , NJ ) at high power for six 5-min cycles of 30 s ON , 30 s OFF; water was changed after each cycle; 5 μg of rabbit anti-E2f1 antibody or rabbit IgG ( Vector Laboratories ) was used . Promoter annotation was performed by first identifying the sequences amplified by each primer set ( Table 2 ) using the Ensembl Genome Browser and then inputting those sequences into Tfsitescan ( http://www . ifti . org/ ) . E2f binding sites were identified and highlighted in bold ( Figure 2—figure supplement 1 ) . 10 . 7554/eLife . 07702 . 021Table 2 . Chromatin immunoprecipitation primers . DOI:http://dx . doi . org/10 . 7554/eLife . 07702 . 021GeneEnsembl IDTarget siteForward primerReverse primerCDKN2AENSG00000147889TSS5′-GCTGAGGGTGGGAAGATG-3′5′-CCTTAACTGCAGACTGGGA-3′tk1ENSDARG00000086561TSS5′-AGTCACTGTGCCGGTTTATT-3′5′-GTCGTCTGCTTGTTGTCTTTATTT-3′tk1-ENSDARG000000865612 kbp 5′ of TSS5′-CAGGCTTACGGAGACAGCAA-3′5′-AGTGTTTGCTGCTGGATCAC-3′TSS: Transcriptional state site . qPCR assays were performed on 100 ng of cDNA using 1 μL of each primer ( 10 pmol/μL ) and iTaq Universal SYBR Green Supermix ( Bio-Rad , Hercules , CA , United States ) in a 12 μL total reaction volume ( Table 3 ) . The PCR was performed for 40 cycles with annealing temperatures of 58–60°C and elongation times of 1 min . Total RNA was isolated using the RNeasy Mini Kit ( Qiagen , Netherlands ) per the manufacturer’s instructions . cDNA was prepared from total RNA using random hexamer primers and the SuperScript III First Stand Synthesis System for reverse transcription-PCR ( Life Technologies ) per the manufacturer’s instructions . Primers used to quantify tp53 and cdkn1a expression levels have been previously described ( Danilova et al . , 2014 ) . 10 . 7554/eLife . 07702 . 022Table 3 . Quantitative polymerase chain reactionprimers . DOI:http://dx . doi . org/10 . 7554/eLife . 07702 . 022GeneRefSeq IDForward primerReverse primerCDKN2ANM_058195 . 35′-ATGGTGCGCAGGTTCTTGGTGA-3′5′-CACCACCAGCGTGTCCAGGAAG-3′actb2NM_181601 . 45′-CGAGCAGGAGATGGGAACC-3′5′-CAACGGAAACGCTCATTGC-3′tp53NM_131327 . 2 5′-CTGAAGTGGTCCGCAGATG-3′5′-CGTTTGGTCCCAGTGGTGG-3′cdkn1aNM_001128420 . 15′-AGCTGCATTCGTCTCGTAGC-3′5′-TGAGAACTTACTGGCAGCTTCA-3′ Zebrafish cells were cultured at 32°C , 5% CO2 in Dulbecco’s Modified Eagle Medium: Nutrient Mixture F-12 ( DMEM:F-12 ) medium ( ATCC , Manassas , VA ) with 10% fetal bovine serum ( FBS ) , 1% penicillin/streptomycin ( Pen/Strep ) ( ZF4 ) , or 50% L-15 , 35% DMEM , 15% Ham’s F-12 medium with 1 . 8 mM NaHCO3 , 15 mM HEPES , 1% Pen/Strep , 10% FBS , 1% l-glutamine , 0 . 2% gentamicinsulfate ( ZKS ) . HeLa cells were grown at 37°C , 5% CO2 in DMEM with 10% FBS , 1% Pen/Strep . The pcDNA-ARF construct was created by subcloning the cDNA of human ARF ( exons 1β , 2 , and 3 of CDKN2A ) into the multiple cloning site of pcDNA3 . 1 ( + ) . Cells were transfected with either pcDNA-ARF or an empty vector ( pcDNA ) . Transient transfections were performed using the FuGENE 6 transfection reagent ( Promega , Madison , WI ) according to the manufacturer’s instructions . Cells were analyzed 2 days posttransfection . Luciferase assays were performed with pGL3-ARF-736 bp and pGL3-ARF-3 . 4 kb as previously described ( del Arroyo et al . , 2007 ) without activator DNA . Data are presented as mean ± standard deviation . Statistical analyses were performed by using SPSS Statistics Desktop , version 22 . 0 ( IBM , Armonk , NY ) . Statistical differences were analyzed by using a Student’s t-test . A p<0 . 05 was set as the threshold for statistical significance . | Zebrafish are capable of remarkable feats of regeneration in which damaged or lost body parts can be replaced with exact replicas of the original . Humans cannot do this; so one major goal of regenerative medicine is to understand the basis of these differences . This could make it possible to regenerate damaged limbs , heart muscle and other tissues in people . A group of genes called tumor suppressor genes are critical for protecting us from cancer . However , they also limit the ability of cells to grow and divide , which are both important for regeneration processes . Differences in how organisms use tumor suppressor genes could possibly contribute to their differing abilities to regenerate body parts . A tumor suppressor gene called ARF plays an important role in protecting mammals from cancer , but it is not found in zebrafish and other animals that can regenerate body parts . Hesse et al . introduced the human ARF gene into zebrafish to investigate whether it has any effect on tissue regeneration . The experiments show that ARF is silent in growing and uninjured fish , but it is activated when it detects regeneration occurring after an injury . ARF drastically inhibits this process to the point that fish with the ARF gene cannot regenerate damaged tissue . The findings suggest that ARF mistakenly identifies regeneration as the formation of a tumor; therefore therapies that aim to induce regeneration in people may need to control the activity of this gene . Hesse et al . ’s findings show that the specific tumor suppressor genes an organism has can affect its ability to regenerate . Future challenges will be to understand the situations in which ARF is able to interfere with tissue regeneration in mammals , and to learn how to manipulate this process . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"stem",
"cells",
"and",
"regenerative",
"medicine",
"chromosomes",
"and",
"gene",
"expression"
] | 2015 | The human ARF tumor suppressor senses blastema activity and suppresses epimorphic tissue regeneration |
Zika is an emerging virus whose rapid spread is of great public health concern . Knowledge about transmission remains incomplete , especially concerning potential transmission in geographic areas in which it has not yet been introduced . To identify unknown vectors of Zika , we developed a data-driven model linking vector species and the Zika virus via vector-virus trait combinations that confer a propensity toward associations in an ecological network connecting flaviviruses and their mosquito vectors . Our model predicts that thirty-five species may be able to transmit the virus , seven of which are found in the continental United States , including Culex quinquefasciatus and Cx . pipiens . We suggest that empirical studies prioritize these species to confirm predictions of vector competence , enabling the correct identification of populations at risk for transmission within the United States .
In 2014 , Zika virus was introduced into Brazil and Haiti , from where it rapidly spread throughout the Americas . By January 2017 , over 100 , 000 cases had been confirmed in 24 different states in Brazil ( http://ais . paho . org/phip/viz/ed_zika_cases . asp ) , with large numbers of reports from many other counties in South and Central America ( Faria et al . , 2016 ) . Originally isolated in Uganda in 1947 , the virus remained poorly understood until it began to spread within the South Pacific , including an outbreak affecting 75% of the residents on the island of Yap in 2007 ( 49 confirmed cases ) and over 32 , 000 cases in the rest of Oceania in 2013–2014 , the largest outbreak prior to the Americas ( 2016-present ) ( Cao-Lormeau et al . , 2016; Duffy et al . , 2009 ) . Guillian-Barré syndrome , a neurological pathology associated with Zika virus infection , was first recognized at this time ( Cao-Lormeau et al . , 2016 ) . Similarly , an increase in newborn microcephaly was found to be correlated with the increase in Zika cases in Brazil in 2015 and 2016 ( Schuler-Faccini et al . , 2016 ) . For this reason , in February 2016 , the World Health Organization declared the American Zika virus epidemic to be a Public Health Emergency of International Concern . Despite its public health importance , the ecology of Zika virus transmission has been poorly understood until recently . It has been presumed that Aedes aegypti and Ae . albopictus are the primary vectors due to epidemiologic association with Zika virus ( Messina et al . , 2016 ) , viral isolation from and transmission experiments with field populations ( especially in Ae . aegypti [Haddow et al . , 2012; Boorman and Porterfield , 1956; Haddow et al . , 1964] ) , and association with related arboviruses ( e . g . dengue fever virus , yellow fever virus ) . Predictions of the potential geographic range of Zika virus in the United States , and associated estimates for the size of the vulnerable population , are therefore primarily based on the distributions of Ae . aegypti and Ae . albopictus , which jointly extend across the Southwest , Gulf coast , and mid-Atlantic regions of the United States ( Centers for Disease Control and Prevention , 2016 ) . We reasoned , however , that if other , presently unidentified Zika-competent mosquitoes exist in the Americas , then these projections may be too restricted and therefore optimistically biased . Additionally , recent experimental studies show that the ability of Ae . aegypti and Ae . albopictus to transmit the virus varies significantly across mosquito populations and geographic regions ( Chouin-Carneiro et al . , 2016 ) , with some populations exhibiting low dissemination rates even though the initial viral titer after inoculation may be high ( Diagne et al . , 2015 ) . This suggests that in some locations other species may be involved in transmission . The outbreak on Yap , for example , was driven by a different species , Ae . hensilli ( Ledermann et al . , 2014 ) . Closely related viruses of the Flaviviridae family are vectored by over nine mosquito species , on average ( see Supplementary Data ) . Thus , because Zika virus may be associated with multiple mosquito species , we considered it necessary to develop a more comprehensive list of potential Zika vectors . The gold standard for identifying competent disease vectors requires isolating virus from field-collected mosquitoes , followed by experimental inoculation and laboratory investigation of viral dissemination throughout the body and to the salivary glands ( Barnett , 1960; Hardy et al . , 1983 ) , and , when possible , successful transmission back to the vertebrate host ( e . g . Komar et al . , 2003 ) . Unfortunately , these methods are costly , often underestimate the risk of transmission ( Bustamante and Lord , 2010 ) , and the amount of time required for analyses can delay decision making during an outbreak ( Day , 2001 ) . To address the problem of identifying potential vector candidates in an actionable time frame , we therefore pursued a data-driven approach to identifying candidate vectors aided by machine learning algorithms for identifying patterns in high dimensional data . If the propensity of mosquito species to associate with Zika virus is statistically associated with common mosquito traits , it is possible to rank mosquito species by the degree of risk represented by their traits – a comparative approach similar to the analysis of risk factors in epidemiology . For instance , a model could be constructed to estimate the statistical discrepancy between the traits of known vectors ( i . e . , Ae . aegypti , Ae . albopictus , and Ae . hensilli ) and the traits of all possible vectors . Unfortunately , this simplistic approach would inevitably fail due to the small amount of available data ( i . e . , sample size of 3 ) . Thus , we developed an indirect approach that leverages the information contained in the associations among many virus-mosquito pairs to inform us about specific associations . Specifically , our method identifies covariates associated with the propensity for mosquito species to vector any flavivirus . From this , we constructed a model of the mosquito-flavivirus network and then extracted from this model the life history profile and species list of mosquitoes predicted to associate with Zika virus , which we recommend be experimentally tested for Zika virus competence .
Our supplementary and primary models generally concur and their ranking of potential Zika virus vectors are highly correlated ( ρ = 0 . 508 and ρ = 0 . 693 on raw and thresholded predictions , respectively ) . As one might expect , the supplementary model assigned fewer scores of low propensity ( Appendix 1—figure 2 ) , suggesting that incorporating this additional uncertainty in the training dataset eroded the model's ability to distinguish negative links . The supplementary model’s performance on the testing data ( AUC = 0 . 84 ± 0 . 02 ) , however , indicates that the additional uncertainty did not impede model performance . When trained on ‘leave-one-out’ datasets , all three models were able to predict the testing data with high accuracy ( AUC = 0 . 91 , AUC = 0 . 91 , AUC = 0 . 92 for West Nile , dengue , and yellow fever viruses , respectively ) . Performance varied when models were validated against predictions of ‘known outcomes’ . A model trained without West Nile virus predicted highly linked vectors reasonably well ( AUC = 0 . 69 ) , however it assigned low scores to rarer ‘known’ vectors , such as Culiseta inornata , which was only associated with West Nile virus . Similarly , the model trained on the dengue-omitted dataset predicted training data and vectors of dengue itself with high accuracy ( AUC = 0 . 92 ) . While the model trained without yellow fever performed well on the testing data , it performed poorly when predicting vectors of yellow fever virus ( AUC = 0 . 47 ) . Unlike West Nile and dengue viruses , the majority of the known vectors of yellow fever are only associated with yellow fever ( i . e . a single vector-virus link ) , and so were excluded completely from the training data when all yellow fever links were omitted . Additionally , several of the vector species are of the Haemagogus genus , which was completely absent from the training data . Given the importance of phylogeny of the vector species in predicting vector-virus links , it follows that a dataset with a novel subgenus would be difficult for the model to predict on , resulting in low model performance . The low performance of this model illustrates that incorporating common traits and additional vector-virus links improves model prediction . When traits were not available in the training dataset , model performance was much lower , suggesting that there exists a statistical association between a vectors’ traits and its ability to transmit a virus .
Zika virus is unprecedented among emerging arboviruses in its combination of severe public health hazard , rapid spread , and poor scientific understanding . Particularly crucial to public health preparedness is knowledge about the geographic extent of potentially at risk populations and local environmental conditions for transmission , which are determined by the presence of competent vectors . Until now , identifying additional competent vector species has been a low priority because Zika virus has historically been geographically restricted to a narrow region of equatorial Africa and Asia ( Petersen et al . , 2016 ) , and the mild symptoms of infection made its range expansion since the 1950’s relatively unremarkable . However , with its relatively recent and rapid expansion into the Americas and its association with severe neurological disorders , the prediction of potential disease vectors in non-endemic areas has become a matter of critical public health importance . We identify these potential vector species by developing a data-driven model that identifies candidate vector species of Zika virus by leveraging data on traits of mosquito vectors and their flaviviruses . We suggest that empirical work should prioritize these species in their evaluation of vector competence of mosquitoes for Zika virus . Our model predicts that fewer than one third of the potential mosquito vectors of Zika virus have been identified , with over twenty-five additional mosquito species worldwide that may have the capacity to contribute to transmission . The continuing focus in the published literature on two species known to transmit Zika virus ( Ae . aegypti and Ae . albopictus ) ignores the potential role of other vectors , potentially misrepresenting the spatial extent of risk . In particular , four species predicted by our model to be competent vectors – Ae . vexans , Culex quinquefasciatus , Cx . pipiens , and Cx . tarsalis – are found throughout the continental United States . Further , the three Culex species are primary vectors of West Nile virus ( Farajollahi et al . , 2011 ) . Cx . quinquefasciatus and Cx . pipiens were ranked 3rd and 17th by our model , respectively , and together these species were the highest-ranking species endemic to the United States after the known vectors ( Ae . aegypti and Ae . albopictus ) . Cx . quinquefasciatus has previously been implicated as an important vector of encephalitic flaviviruses , specifically West Nile virus and St . Louis encephalitis ( Turell et al . , 2005; Hayes et al . , 2005 ) , and a hybridization of the species with Cx . pipiens readily bites humans ( Fonseca et al . , 2004 ) . The empirical data available on the vector competence of Cx . pipiens and Cx . quinquefasciatus is currently mixed , with some studies finding evidence for virus transmission and others not ( Guo et al . , 2016; Aliota et al . , 2016; Fernandes et al . , 2016; Huang et al . , 2016 ) . These results suggest , in combination with evidence for significant genotype x genotype effects on the vector competence of Ae . aegypti and Ae . albopictus to transmit Zika ( Chouin-Carneiro et al . , 2016 ) , that the vector competence of Cx . pipiens and Cx . quinquefasciatus for Zika virus could be highly dependent upon the genetic background of the mosquito-virus pairing , as well as local environmental conditions . Thus , considering their anthropophilic natures and wide geographic ranges , Cx . quinquefasciatus and Cx . pipiens could potentially play a larger role in the transmission of Zika in the continental United States . Further experimental research into the competence of populations of Cx . pipiens to transmit Zika virus across a wider geographic range is therefore highly recommended , and should be prioritized . The vectors predicted by our model have a combined geographic range much larger than that of the currently suspected vectors of Zika ( Figure 3 ) , suggesting that , were these species to be confirmed as vectors , a larger population may be at risk of Zika infection than depicted by maps focusing solely on Ae . aegypti and Ae . albopictus . The range of Cx . pipiens includes the Pacific Northwest and the upper mid-West , areas that are not within the known range of Ae . aegypti or Ae . albopictus ( Darsie and Ward , 2005 ) . Furthermore , Ae . vexans , another predicted vector of Zika virus , is found throughout the continental US and the range of Cx . tarsalis extends along the entire West coast ( Darsie and Ward , 2005 ) . On a finer scale , these species use a more diverse set of habitats , with Ae . aegypti and Cx . quinquefasciatus mainly breeding in artificial containers , and Ae . vexans and Ae . albopictus being relatively indiscriminate in their breeding sites , including breeding in natural sites such as tree holes and swamps . Therefore , in addition to the wider geographic region supporting potential vectors , these findings suggest that both rural and urban areas could serve as habitat for potential vectors of Zika . We recommend experimental tests of these species for competency to transmit Zika virus , because a confirmation of these vectors would necessitate expanding public health efforts to these areas not currently considered at risk . 10 . 7554/eLife . 22053 . 005Figure 3 . Distribution maps of predicted vectors of Zika virus in the continental US . Maps of Aedes species are based on Centers for disease control and prevention ( 2016 ) . All other species’ distributions are georectified maps from Darsie and Ward ( 2005 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 22053 . 005 While transmission requires a competent vector , vector competence does not necessarily equal transmission risk or inform vectorial capacity . There are many biological factors that , in conjunction with positive vector competence , determine a vector’s role in disease transmission . For example , although Ae . aegypti mosquitoes are efficient vectors of West Nile virus , they prefer to feed on humans , which are dead-head hosts for the disease , and therefore have low potential to serve as a vector ( Turell et al . , 2005 ) . Psorophora ferox , although predicted by our model as a potential vector of Zika virus , would likely play a limited role in transmission because it rarely feeds on humans ( Molaei et al . , 2008 ) . Additionally , vector competence is dynamic , and may be mediated by environmental factors that influence viral development and mosquito immunity ( Muturi and Alto , 2011 ) . Therefore , our list of potential vectors of Zika represents a comprehensive starting point , which should be furthered narrowed by empirical work and consideration of biological details that impact transmission dynamics . Given the severe neurological side-effects of Zika virus infection , beginning with the most conservative method of vector prediction ensures that risk is not underestimated , and allows public health agencies to interpret the possibility of Zika transmission given local conditions . Our model serves as a starting point to streamlining empirical efforts to identify areas and populations at risk for Zika transmission . While our model enables data-driven predictions about the geographic area at potential risk of Zika transmission , subsequent empirical work investigating Zika vector competence and transmission efficiency is required for model validation , and to inform future analyses of transmission dynamics . For example , in spite of its low transmission efficiency in certain geographic regions ( Chouin-Carneiro et al . , 2016 ) , Ae . aegypti is anthropophilic ( Powell and Tabachnick , 2013 ) , and may therefore pose a greater risk of human-to-human Zika virus transmission than mosquitoes that bite a wider variety of animals . On the other hand , mosquito species that prefer certain hosts in rural environments are known to alter their feeding behaviors to bite alternative hosts ( e . g . , humans and rodents ) in urban settings , due to changes in host community composition ( Chaves et al . , 2010 ) . Environmental factors such as precipitation and temperature directly influence mosquito populations , and determine the density of vectors in a given area ( Thomson et al . , 2006 ) , an important factor in transmission risk . Additionally , socio-economic factors such as housing type and lifestyle can decrease a populations’ contact with mosquito vectors , and lower the risk of transmission to humans ( Moreno-Madriñán and Turell , 2017 ) . Effective risk modeling and forecasting the range expansion of Zika virus in the United States will depend on validating the vector status of these species , as well as resolving behavioral and biological details that impact transmission dynamics . Although we developed this model with Zika virus in mind , our findings have implications for other emerging flaviviruses and contribute to the recently developed methodology applying machine learning methods to the prediction of unknown agents of infectious diseases . This technique has been used to predict rodent reservoirs of disease ( Han et al . , 2015 ) and bat carriers of filoviruses ( Han et al . , 2016 ) by training models with host-specific data . Our model , however , incorporates additional data by constructing a vector-virus network that is used to inform predictions of vector-virus associations . The combination of common virus traits with vector-specific traits enabled us to predict potential mosquito vectors of specific flaviviruses , and to train the model on additional information distributed throughout the flavivirus-mosquito network . Uncertainty in our model arises through uncertainty inherent in our datasets . Vector status is not static ( e . g . mutation in the chikungunya virus to increase transmission by Ae . albopictus [Weaver and Forrester , 2015] ) and can vary across vector populations ( Bennett et al . , 2002 ) . When incorporating uncertainty in vector status through our supplementary model , our predictions generally agreed with that of our original model . However , the increased uncertainty did reduce the models’ ability to distinguish negative links , resulting in higher uncertainty in propensity scores ( as measured by standard deviation ) and a larger number of predicted vectors . Additionally , the model performs poorly when predicting on vector-virus links with trait levels not included in the training data set , as was the case when omitting yellow fever virus . Another source of uncertainty is regarding vector and virus traits . In addition to intraspecific variation in biological traits , many vectors are understudied , and common traits such as biting activity are unknown to the level of species . Additional study into the behavior and biology of less common vector species would increase the accuracy of prediction techniques such as this , and allow for a better of understanding of species’ potential role as vectors . Interestingly , our constructed flavivirus-mosquito network generally concurs with the proposed dichotomy of Aedes species vectoring hemorrhagic or febrile arboviruses and Culex species vectoring neurological or encephalitic viruses ( Grard et al . , 2010 ) ( Figure 1 ) . However , there are several exceptions to this trend , notably West Nile virus , which is vectored by several Aedes species . Additionally , our model predicts several Culex species to be possible vectors of Zika virus . While this may initially seem contrary to the common phylogenetic pairing of vectors and viruses noted above , Zika’s symptoms , like West Nile virus , are both febrile and neurological . Thus , its symptoms do not follow the conventional hemorrhagic/encephalitic division . The ability of Zika virus to be vectored by a diversity of mosquito vectors could have important public health consequences , as it may expand both the geographic range and seasonal transmission risk of Zika virus , and warrants further empirical investigation . Considering our predictions of potential vector species and their combined ranges , species on the candidate vector list need to be validated to inform the response to Zika virus . Vector control efforts that target Aedes species exclusively may ultimately be unsuccessful in controlling transmission of Zika because they do not control other , unknown vectors . For example , the release of genetically modified Ae . aegypti to control vector density through sterile insect technique is species-specific and would not control alternative vectors ( Alphey et al . , 2010 ) . Additionally , species’ habitat preferences differ , and control efforts based singularly on reducing Aedes larval habitat will not be as successful at controlling Cx . quinquefasciatus populations ( Rey et al . , 2006 ) . Predicted vectors of Zika virus must be empirically tested and , if confirmed , vector control efforts would need to respond by widening their focus to control the abundance of all predicted vectors of Zika virus . Similarly , if control efforts are to include all areas at potential risk of disease transmission , public health efforts would need to expand to address regions such as the northern Midwest that fall within the range of the additional vector species predicted by our model . An understanding of the capacity of mosquito species to vector Zika virus is necessary to prepare for the potential establishment of Zika virus in the United States , and we recommend that experimental work start with this list of candidate vector species .
Our dataset comprised a matrix of vector-virus pairs relating all known flaviviruses and their mosquito vectors . To construct this matrix , we first compiled a list of mosquito-borne flaviviruses to include in our study ( Van Regenmortel et al . , 2000; Kuno et al . , 1998; Cook and Holmes , 2006 ) . Viruses that only infect mosquitoes and are not known to infect humans were not included . Using this list , we constructed a mosquito-virus pair matrix based on the Global Infectious Diseases and Epidemiology Network database ( GIDEON , 2016 ) , the International Catalog of Arboviruses Including Certain Other Viruses of Vertebrates ( ArboCat ) ( Karabatsos , 1985 ) , The Encyclopedia of Medical and Veterinary Entomology ( Russell et al . , 2013 ) and Mackenzie et al . ( 2012 ) . We defined a known vector-virus pair as one for which the full transmission cycle ( i . e , infection of mosquito via an infected host ( mammal or avian ) or bloodmeal that is able to be transmitted via saliva ) has been observed . Basing vector competence on isolation or intrathoracic injection bypasses several important barriers to transmission ( Hardy et al . , 1983 ) , and may not be true evidence of a mosquito’s ability to transmit an arbovirus . We found our definition to be more conservative than that which is commonly used in disease databases ( e . g . Global Infectious Diseases and Epidemiology Network database ) , which often assumes isolation from wild-caught mosquitoes to be evidence of a mosquito’s role as a vector . Therefore , a supplementary analysis investigates the robustness of our findings with regards to uncertainty in vector status by comparing the analysis reported in the main text to a second analysis in which any kind of evidence for association , including merely isolating the virus in wild-caught mosquitoes , is taken as a basis for connection in the virus-vector network ( see Appendix 1 for analysis and results ) . Fifteen mosquito traits ( Appendix 2—table 1 ) and twelve virus traits ( Appendix 2—table 2 ) were collected from the literature . For the mosquito species , the geographic range was defined as the number of countries in which the species has been collected , based on Walter Reed Biosystematics Unit , ( 2016 ) . While there are uncertainties in species’ ranges due to false absences , this represents the most comprehensive , standardized dataset available that includes both rare and common mosquito species . A species’ continental extent was recorded as a binary value of its presence by continent . A species’ host range was defined as the number of taxonomic classes the species is known to feed on , with the Mammalia class further split into non-human primates and other mammals , because of the important role primates play in zoonotic spillovers of vector-borne disease ( e . g . dengue , chikungunya , yellow fever , and Zika viruses ) ( Weaver , 2005; Diallo et al . , 2005; Weaver et al . , 2016 ) . The total number of unique flaviviruses observed per mosquito species was calculated from our mosquito-flavivirus matrix . All other traits were based on consensus in the literature ( see Appendix III for sources by species ) . For three traits – urban preference , endophily ( a proclivity to bite indoors ) , and salinity tolerance – if evidence of that trait for a mosquito was not found in the literature , it was assumed to be negative . 10 . 7554/eLife . 22053 . 006Table 1 . Predicted vectors of Zika virus , as reported by our model . Mosquito species endemic to the continental United States are bolded . A species is defined as a known vector of Zika virus if a full transmission cycle ( see main text ) has been observed . DOI: http://dx . doi . org/10 . 7554/eLife . 22053 . 006SpeciesGBM prediction ±SDKnown vector ? Aedes aegypti0 . 81 ± 0 . 12YesAe . albopictus0 . 54 ± 0 . 14YesCulex quinquefasciatus0 . 38 ± 0 . 14NoAe . polynesiensis0 . 36 ± 0 . 13NoAe . scutellaris0 . 33 ± 0 . 13NoAe . africanus0 . 32 ± 0 . 11NoAe . furcifer0 . 31 ± 0 . 16YesAe . vittatus0 . 30 ± 0 . 20YesAe . taylori0 . 30 ± 0 . 16YesAe . luteocephalus0 . 25 ± 0 . 12YesAe . tarsalis0 . 18 ± 0 . 11YesAe . metallicus0 . 16 ± 0 . 08NoAe . minutus0 . 16 ±0 . 09NoAe . opok0 . 14 ± 0 . 06NoAe . bromeliae0 . 11 ± 0 . 06NoAe . scapularis0 . 10 ± 0 . 04NoCx . pipiens0 . 10 ± 0 . 04NoAe . hensilli0 . 10 ± 0 . 06YesAe . vigilax0 . 10 ± 0 . 05NoCx . annulirostrix0 . 08 ± 0 . 03NoPsorophora ferox0 . 08 ± 0 . 05NoCx . rubinotus0 . 08 ± 0 . 07NoCx . tarsalis0 . 08 ± 0 . 03NoAe . occidentalis0 . 08 ± 0 . 05NoAe . flavicolis0 . 07 ± 0 . 04NoAe . serratus0 . 07 ± 0 . 04NoCx . p . molestus0 . 07 ± 0 . 04NoAe . vexans0 . 06 ± 0 . 04NoCx . neavei0 . 06 ± 0 . 02NoRunchomyia frontosa0 . 06 ± 0 . 04NoAe . neoafricanus0 . 06 ± 0 . 03NoAe . chemulpoensis0 . 06 ± 0 . 03NoCx . vishnui0 . 05 ± 0 . 01NoCx . tritaeniorhynchus0 . 05 ± 0 . 01NoAe . fowleri0 . 04 ± 0 . 03Yes We collected data on the following virus traits: host range ( Mahy , 2009; Mackenzie et al . , 2012; Chambers and Monath , 2003; Cook and Zumla , 2009b ) , disease severity ( Mackenzie et al . , 2012 ) , human illness ( Chambers and Monath , 2003; Cook and Zumla , 2009 ) , the presence of a mutated envelope protein , which controls viral entry into cells ( Grard et al . , 2010 ) , year of isolation ( Karabatsos , 1985 ) , and host range ( Karabatsos , 1985 ) . Disease severity was based on Mackenzie et al . ( 2012 ) , ranging from no known symptoms ( e . g . Kunjin virus ) to severe symptoms and significant human mortality ( e . g . yellow fever virus ) . For each virus , vector range was calculated as the number of mosquito species for which the full transmission cycle has been observed . Genome length was calculated as the mean of all complete genome sequences listed for each flavivirus in the Virus Pathogen Database and Analysis Resource ( http://www . viprbrc . org/ ) . For more recently discovered flaviviruses not yet cataloged in the above databases ( i . e . , New Mapoon Virus , Iquape virus ) , viral traits were gathered from the primary literature ( sources listed in Appendix 3 ) . Following Han et al . ( 2015 ) , boosted regression trees ( BRT ) ( Friedman , 2001 ) were used to fit a logistic-like predictive model relating the status of all possible virus-vector pairs ( 0: not associated , 1: associated ) to a predictor matrix comprising the traits of the mosquito and virus traits in each pair . Boosted regression trees circumvent many issues associated with traditional regression analysis ( Elith et al . , 2008 ) , allowing for complex variable interactions , collinearity , non-linear relationships between covariates and response variables , and missing data . Additionally , this technique performs well in comparison with other logistic regression approaches ( Friedman , 2001 ) . Trained boosted regression tree models are dependent on the split between training and testing data , such that each model might predict slightly different propensity values . To address this , we trained an ensemble of 25 internally cross-validated BRT models on independent partitions of training and testing data . The resulting model demonstrated low variance in relative variable importance and overall model accuracy , suggesting models all converged to a similar result . Prior to the analysis of each model , we randomly split the data into training ( 70% ) and test ( 30% ) sets while preserving the proportion of positive labels ( known associations ) in each of the training and test sets . Models were trained using the gbm package in R ( Ridgeway , 2015 ) , with the maximum number of trees set to 25 , 000 , a learning rate of 0 . 001 , and an interaction depth of 5 . To correct for optimistic bias ( Smith et al . , 2014 ) , we performed 10-fold cross validation and chose a bag fraction of 50% of the training data for each iteration of the model . We estimated the performance of each individual model with three metrics: Area Under the Receiver Operator Curve , specificity , and sensitivity . For specificity and sensitivity , which require a preset threshold , we thresholded predictions on the testing data based on the value which maximized the sum of the sensitivity and specificity , a threshold robust to the ratio of presence to background points in presence-only datasets ( Liu et al . , 2016 ) . Variable importance was quantified by permutation ( Breiman , 2001 ) to assess the relative contribution of virus and vector traits to the propensity for a virus and vector to form a pair . Because we transformed many categorical variables into binary variables ( e . g . , continental range as binary presence or absence by continent ) , the sum of the relative importance for each binary feature was summed to obtain a single value for the entire variable . Each of our twenty-five trained models was then used to predict novel mosquito vectors of Zika by applying the trained model to a data set consisting of the virus traits of Zika paired with the traits of all mosquitoes for which flaviviruses have been isolated from wild caught individuals , and , depending on the species , may or may not have been tested in full transmission cycle experiments ( a total of 180 mosquito species ) . This expanded dataset allowed us to predict over a large number of mosquito species , while reasonably limiting our dataset to those species suspected of transmitting flaviviruses . The output of this model was a propensity score ranging from 0 to 1 . In our case , the final propensity score for each vector was the mean propensity score assigned by the twenty-five models . To label unobserved edges , we thresholded propensity scores at the value of lowest ranked known vector ( Liu et al . , 2013 ) . In addition to conventional performance metrics , we conducted additional analyses to further validate both this method of prediction , and our model specifically . To account for uncertainty in the vector-virus links in our initial matrix , we repeated our analysis for a vector-virus matrix with a less conservative definition of a positive link ( field isolation and above ) , referred to as our supplementary model . Vector competence is a dynamic trait , and there exists significant intraspecific variation in the ability of a vector to transmit a virus for certain species of mosquitoes ( Diallo et al . , 2005; Gubler et al . , 1979 ) . Our supplementary model is based on a less conservative definition of vector competence and includes species implicated as vectors , but not yet verified through laboratory competence studies , and therefore accounts for additional uncertainty such as intraspecific variation . While this approach is well-tested in epidemiological applications ( Parascandola , 2004 ) , it has only recently been applied to predict ecological associations , and , as such , has limitations unique to this application . To further evaluate this prediction method , we performed a modified ‘leave-one-out’ analysis , whereby we trained a model to a dataset from which a well-studied virus had been omitted , and then predicted vectors for this virus and compared them against a list of known vectors . We repeated this analysis for West Nile , dengue , and yellow fever viruses , following the same method of training as for our original model . While this analysis differs from our original method , it provides a more stringent evaluation of this method of prediction because the model is trained on an incomplete dataset and predicts on unfamiliar data , a more difficult task than that posed to our original model . | Mosquitoes carry several diseases that pose an emerging threat to society . Outbreaks of these diseases are often sudden and can spread to previously unaffected areas . For example , the Zika virus was discovered in 1947 , but only received international attention when it spread to the Americas in 2014 , where it caused over 100 , 000 cases in Brazil alone . While we now recognize the threat Zika can pose for public health , our knowledge about the ecology of the disease remains poor . Nine species of mosquitoes are known to be able to carry the Zika virus , but it cannot be ruled out that other mosquitoes may also be able to spread the disease . There are hundreds of species of mosquitoes , and testing all of them is difficult and costly . So far , only a small number of species have been tested to see if they transmit Zika . However , computational tools called decision trees could help by predicting which mosquitoes can transmit a virus based on common traits , such as a mosquito's geographic range , or the symptoms of a virus . Evans et al . used decision trees to create a model that predicts which species of mosquitoes are potential carriers of Zika virus and should therefore be prioritized for testing . The model took into account all known viruses that belong to the same family as Zika virus and the mosquitoes that carry them . Evans et al . predict that 35 species may be able to carry the Zika virus , seven of which are found in the United States . Two of these mosquito species are known to transmit West Nile Virus and are therefore prime examples of species that should be prioritized for testing . Together , the ranges of the seven American species encompass the whole United States , suggesting Zika virus could affect a much larger area than previously anticipated . The next step following on from this work will be to carry out experiments to test if the 35 mosquitoes identified by the model are actually able to transmit the Zika virus . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"ecology",
"computational",
"and",
"systems",
"biology"
] | 2017 | Data-driven identification of potential Zika virus vectors |
The number of precaudal vertebrae in all extant crocodylians is remarkably conservative , with nine cervicals , 15 dorsals and two sacrals , a pattern present also in their closest extinct relatives . The consistent vertebral count indicates a tight control of axial patterning by Hox genes during development . Here we report on a deviation from this pattern based on an associated skeleton of the giant caimanine Purussaurus , a member of crown Crocodylia , and several other specimens from the Neogene of the northern neotropics . P . mirandai is the first crown-crocodylian to have three sacrals , two true sacral vertebrae and one non-pathological and functional dorsosacral , to articulate with the ilium ( pelvis ) . The giant body size of this caiman relates to locomotory and postural changes . The iliosacral configuration , a more vertically oriented pectoral girdle , and low torsion of the femoral head relative to the condyles are hypothesized specializations for more upright limb orientation or weight support .
The Neogene crocodylian fauna of northern South America is remarkable in terms of species richness , levels of species sympatry , and ecomorphological specialization ( e . g . , Riff et al . , 2010; Sánchez‐Villagra and Aguilera , 2006; Scheyer et al . , 2013; Scheyer and Delfino , 2016 ) . Among a plethora of taxa , Purussaurus was an exceptionally large caiman ( Alligatoroidea ) that lived in the northern neotropics of South America in the middle and late Miocene ( ca . 13–5 Ma ) . Its impressively large skull is the basis of its taxonomy , which encompasses three species distributed in localities in Brazil , Peru , Colombia and Venezuela ( Aguilera et al . , 2006; Langston , 1965; Mook , 1921a; Mook , 1942; Riff et al . , 2010; Salas-Gismondi et al . , 2015 ) . The postcranial anatomy of Purussaurus , as that of most extinct crocodylians , is still poorly known . We report here on the discovery of an exceptional skeleton of Purussaurus mirandai and several other remains from the late Miocene in Venezuela . Archosauria are represented today by 10 , 000+ species of birds ( crown Aves ) , but less than 30 species of crocodylians ( crown Crocodylia ) . As sister taxa , both groups have undergone profound changes in their body plans throughout ca . 250 million years of evolutionary history . In comparing the morphology of extant Archosauria , birds reveal much larger variation in body shape , musculoskeletal form and function , ecology , and lifestyle; whereas crocodylian species resemble each other much more closely due to their shared amphibious and overall carnivorous lifestyle . When examining the underlying developmental patterns in archosaurs , the difference in variation is also reflected in the axial patterning of the vertebral column in Aves and Crocodylia , with the former having variable precaudal vertebral numbers and the latter showing a conserved pattern ( Mansfield and Abzhanov , 2010; Müller et al . , 2010 ) . Finally , extant crocodylians show distinctly low genome-wide evolutionary rates compared to those of birds , which could be linked to prolonged generation times in the former clade ( Green et al . , 2014 ) . These low evolutionary rates could potentially underlie the generally lower morphofunctional disparity seen in the post-Cretaceous crocodylian body plans ( Brusatte et al . , 2010; Stubbs et al . , 2013 ) . Vertebrate axial patterning by means of Homeobox ( Hox ) gene expression has been extensively studied in model organisms , including the chick , since the 1980 s ( see Favier and Dollé , 1997; McGinnis and Krumlauf , 1992; Wellik , 2007 for overviews ) . The study of the developmental patterning and associated gene expressions in crocodylians ( among other extant reptiles ) , on the other hand , has only recently received attention ( Böhmer , 2013; Böhmer et al . , 2015; Mansfield and Abzhanov , 2010 ) , with a focus on the presacral patterning of the body . These developmental studies as well as comparative anatomy ( Hoffstetter and Gasc , 1969; Mook , 1921b ) corroborate the general precaudal count of all crown Crocodylia to consist of nine cervicals , 15 dorsals ( thoracic and lumbar ) and two sacrals . Various pathological conditions have been reported using classical dissection ( e . g . , Baur , 1886; Baur , 1889; Reinhardt , 1873; Reinhardt , 1874 ) . Here , we present all relevant axial and appendicular material of the extinct giant caimanine Purussaurus mirandai , including a revised character scoring and phylogenetic analysis for the species , as the first non-pathological case within crown-Crocodylia that deviates from the highly conserved precaudal count of the group . Comparison with pathological ( e . g . , the presence of congenital malformation lumbosacral transitional vertebrae ) and non-pathological extant crocodylians served as the basis for elucidating developmental patterns for the sacralisation of the last dorsal ( i . e . , lumbar ) vertebra in the extinct species .
Crocodylia Gmelin , 1789 Alligatoridae Gray , 1844 Caimaninae Brochu , 2003 ( following Norell , 1988 ) PURUSSAURUS Barbosa-Rodrigues , 1892 P . mirandai Aguilera , Riff and Bocquentin-Villanueva , 2006 Holotype material: UNEFM-CIAAP-1369 , comprising the skull and associated lower jaw material , as well as a femur and ischium , which , according to Aguilera et al . ( 2006 ) was collected at El Hatillo locality ( see Scheyer and Delfino , 2016 for locality information ) . New referred material: AMU-CURS-541 , an associated but disarticulated skeleton , preserving cranial material and much of the postcranium , which was embedded normally with its abdomen in the sediment . The specimen was preliminarily assigned to Purussaurus cf . P . mirandai ( Scheyer and Delfino , 2016 ) , but given the overall shape and proportions of the lower jaw ( see below ) , the lightly wavy lateral outline of the dentary , the low premaxilla and low and slender jugal ( indicating a rather flat skull profile ) it is herein referred to as Purussaurus mirandai . The shape of the mandible , the teeth , and of the tooth row in general are otherwise also congruent with that of the holotype specimen of P . mirandai . There is so far no indication for the presence of a second species of Purussaurus in the Urumaco Formation . The phylogenetic context of the species among Caimaninae is shown in Figure 1 . Locality and age: The specimen comes from sediments of the Upper Member of the Urumaco Formation ( late Miocene ) , from ‘Norte El Picache’ locality ( 11° 15’ 09 . 00’’ N; 70° 13’ 40 . 00’’ W ) , Urumaco , Falcón state , Venezuela . Its most novel feature found in the sacral region ( juncture of vertebral column and pelvis ) expands the morphological diversity of the axial skeleton in crown crocodylians and suggests developmental changes correlated with biomechanical demands for support and locomotion , likely correlated with giant body size . Furthermore , we infer that the gigantic size of up to 10–12 meters of total body length and related mass of these animals ( see also Aureliano et al . , 2015 ) influenced the peculiar pectoral and pelvic morphology . In the following , we thus provide an abbreviated description of the girdle bones and the sacral region of P . mirandai pertinent to the discussion ( Figure 2 , Figure 2—figure supplement 1 , Figure 3; Figure 3—figure supplement 1 ) . An exhaustive description of all studied specimens , including the postcranial bones of AMU-CURS-541 , will be published elsewhere . Of the pectoral girdle elements ( Figure 2 ) , isolated scapulae are known from two specimens ( AMU-CURS-541; UNEFM-CIAAP-1367 ) and a well-preserved and isolated coracoid is known from UNEFM-CIAAP-1367 . Similar to extant alligatorids such as Alligator mississippiensis and Caiman crocodilus , Purussaurus mirandai has narrow scapular blades . In contrast to the extant taxa , the extinct species has the scapulae oriented more dorsally and somewhat posteriorly , as well as ventromedially and slightly posteriorly oriented coracoids . The long , flat and narrow blade , which is set off by a constriction from the proximal articulation and the glenoid fossa , extends only slightly towards its distal end . The deltoid crest ( acromion of Cong , 1998 ) of the scapula of Purussaurus from the Urumaco Formation ( well preserved in UNEFM-CIAAP-1367 ) is a very robust and distinct process , and of similar proximodistal expansion as the width of the scapular articulation with the coracoid . The coracoid has a prominent and robust shaft , a narrow articular facet for the scapula , an ovoid coracoid foramen close to the articular facet , and an expanded , flaring distal part . In general , these elements ( Figure 3 ) resemble those of other crocodylians in overall shape , but show also some specializations . As in A . mississippiensis ( e . g . , Brochu , 1999 ) , the ilium of P . mirandai ( AMU-CURS-541 , UNEFM-CIAAP-1369 ) has a weak indentation at the dorsal margin of the postacetabular process . Most strikingly , however , the medial surface of the Purussaurus ilium shows three rugose concavities separated by ridges forming a ‘π' ( Figure 3C–E ) , instead of a ‘τ' as in extant crocodylians with two sacral vertebrae . These concavities are the attachment sites for the sacral ribs and lie close together , their size responding to the anteroposterior width of each of the three sacral ribs as is visible in UNEFM-CIAPP 1369 . All of the four recovered ilia assignable to Purussaurus from the Urumaco Formation show the three articulation sites separated by low bony ridges . To date , there is only limited comparative information on the pelvic girdle morphology of other giant crocodylian forms . The well-preserved right ilium of UCMP 38012 ( holotype ) of Mourasuchus atopus from La Venta , Colombia , however , with 137 mm length ( Langston , 1965 ) ; much smaller than Purussaurus; also has ‘π'-shaped ridges on its medial side . Given the lack of other specimens corroborating the potential for three sacrals , we conservatively treat the condition in Mourasuchus as being inconclusive . In some mekosuchine crocodylians ( i . e . , morphotype described as ‘pelvic forms three and four’; see also Figure 3—figure supplement 2 ) , the ‘τ'-shape separation is not obvious , because the attachment sites of the two sacral ribs are shifting apart , thus creating a space in between ( Stein et al . , 2017 ) . This space , however , consists of a smooth and flat bone surface , which should not be confused with an additional rugose and excavated attachment site . The pubis is hatchet-shaped in P . mirandai , with a slender shaft and distal flaring blade . The proximal bone surface of the pubis differs from that of other crocodylians ( Claessens and Vickaryous , 2012; Mook , 1921b; Stein et al . , 2017 ) in that the lateral part of the surface is angled . The ischium is known in the holotype UNEFM-CIAAP-1369 and non-holotype material of P . mirandai , as well as in AMU-CURS-528 and AMU-CURS-541 . The ischium has a slender shaft , and proximally two articulation facets for the ilium , which form the acetabulum , and a distally flaring broad shield . Medially the ischia have a straight contacting margin . Isolated sacrals are known in specimens AMU-CURS-541 ( Figure 3—figure supplement 1 ) and MCNC-URU-111–72V , but it is the complete articulated sacrum in UNEFM-CIAAP-1369 ( non-holotype material ) , consisting of three moderately preserved but articulated sacral vertebrae that is the most informative . All three sacral vertebrae have the centrum and sacral ribs preserved in articulation , showing the detailed morphology of the pelvic girdle and the connection of the ilium with the three sacral vertebrae in Purussaurus . The sacral ribs of the first sacral ( =dorsosacral ) vertebra are tilted posteriorly . The sacral ribs of the second sacral extend laterally and the sacral ribs of the third sacral are tilted anteriorly . Of the three sacrals , the ribs of the mid-sacral show the largest anteroposterior expansion , followed by the third sacral and then those of the first sacral , which are the least expanded . With the exception of a single prezygapophysis , the other zygapophyses and the neural spines of the sacrals are missing or strongly distorted . Of the isolated sacrals , the 3rd sacral ( =primordial sacral 2 ) centrum of AMU-CURS-541 shows the best-preserved articular surfaces , indicating that the anterior condyle is round in shape and flattened ( rather than strongly convex ) , whereas the posterior cotyle is slightly oval-shaped and concave . The cotyle also shows a thick marginal rim , but a flange seems to be absent ( Figure 3—figure supplement 1D , E and G ) . All above-mentioned specimens are considered large adults , and they vary only modestly in shape . The posterior articular surface of the last vertebral centra in the sacral series ( i . e . , primordial sacral 2 ) of the non-holotype material accessioned under UNEFM-CIAAP-1369 is 12 cm wide and 6 cm tall ( distance between sacral rib ends is 30 cm ) , whereas that of primordial sacral 2 of AMU-CURS-541 is about 12 . 8 cm wide and 8 . 8 cm tall ( distance between sacral rib ends is 28 cm , but the ends are not complete ) . The strongly weathered posterior articular surface of the sacral rib of MCNC-URU-111–72V is 9 cm wide and 5 . 5 cm tall ( sacral rib ends are 36 cm apart ) . Given the strongly concave posterior margins and posterior tilting of its sacral ribs , and in comparison with the complete sacral series of UNEFM-CIAAP-1369 described above , this specimen is identified as an isolated dorsosacral , that is the first sacral in the series , as well ( Figure 3—figure supplement 1J–L ) . The osteological description and analysis provided 43 characters that could be scored based on AMU-CURS-541 , with 19 ( 15 postcranial , four mandibular ) new in comparison to the previously available scoring of P . mirandai ( Salas-Gismondi et al . , 2015 ) . The TNT analysis recovered 20 most parsimonious trees ( tree length = 687 steps; CI = 0 . 383 , RI = 0 . 806 ) , for which a strict consensus tree was computed . The analysis ( Figure 1 ) shows the best resolution of those performed herein . It yielded a similar topology to that presented by Salas-Gismondi et al . ( 2015 ) : fig . 4 ) with the following exceptions: Within Gavialoidea , the ‘Pebas gavialoid’ Gryposuchus pachakamue ( Salas-Gismondi et al . , 2016 ) , fell into a polytomy with Siquisiquesuchus , Ikanogavialis , Piscogavialis , Gavialis , and a Gryposuchus clade ( consisting of G . croizati and G . colombianus ) . The Purussaurus clade – sister to the genus Mourasuchus - is fully resolved with P . neivensis being the sister to the clade P . mirandai – P . brasiliensis . The Jacarea clade is better resolved compared to the original analysis of Salas-Gismondi et al . ( 2015 ) , but Bremer support for the clades within Caimaninae was also generally low ( Figure 1; Figure 1—figure supplement 1 ) . This analysis recovered 60 most parsimonious trees ( tree length = 687 steps; consistency index CI = 0 . 383 , retention index RI = 0 . 806 ) , for which a strict consensus tree was computed . The analysis shows a similar topology as presented in the main text ( see also Salas-Gismondi et al . , 2015 , fig . 4 ) . The sister grouping of P . mirandai and P . brasiliensis could not be recovered and instead the Purussaurus clade collapsed into a polytomy . This analysis yielded 160 most parsimonious trees ( tree length = 688 steps; CI = 0 . 382 , RI = 0 . 805 ) . The strict consensus of those trees showed the same topology as in a previous analysis of Salas-Gismondi et al . ( 2015 ) , with the exception of the polytomy with Siquisiquesuchus , Ikanogavialis , Piscogavialis , Gavialis , a Gryposuchus clade within Gavialoidea , and a poorly resolved Jacarea clade .
The scapula and coracoid remained separate in Purussaurus , even in large specimens ( Figures 2 and 4A ) , whereas in extant caimans , scapulocoracoid synchondrosis closure begins relatively early in ontogeny . This early closure was suggested as a peramorphic feature of Caimaninae ( Brochu , 1995 ) , as well as an ambiguous character support of the group ( Brochu , 1999 ) . The scapula blade is oriented only slightly posteriorly , and the coracoid is ventromedially and slightly posteriorly oriented , giving the whole pectoral girdle a rather upright or straight ( subvertical ) appearance in lateral view ( Figure 4A ) . The scapulocoracoid suture in the Purussaurus species is narrow in comparison to that in extant species because both pectoral girdle bones lack a proximal anterior expansion ( Figure 4A ) . The very wide and robust deltoid crest of the scapula of Purussaurus indicates a well-developed origination site of the deltoideus clavicularis muscle ( Meers , 2003; = M . deltoideus scapularis inferior sensu Brochu , 1999 ) . The deltoideus clavicularis likely strengthened the anchoring of the humerus to the shoulder girdle . Analogous expansions of the insertion of this muscle onto the proximal portion of the humerus in the baurusuchid crocodyliform Stratiotosuchus were cited as evidence for improved parasagittal limb function ( Riff and Kellner , 2011 ) . The deltoid crest may also have hosted parts of the insertions of the levator scapularis and trapezius/cucullaris muscles ( Cong , 1998 ) , which should have had functions related to stabilization/mobilization of the pectoral girdle or the neck . The pelvic girdle bones of Purussaurus articulate in similar fashion to those of extant crocodylians ( Figure 4B , C; see Claessens and Vickaryous , 2012 ) , with a double articulation between ilium and ischium and a sole articulation of the pubis with the anterior margin of the ischium . The angle of the proximal articular facet in the pubis in P . mirandai is peculiar , because the standard articular surface in crocodylian taxa has a sub-circular outline and a single , weakly concavoconvex articular surface with the ischium ( e . g . , Claessens and Vickaryous , 2012 ) . The lateral , angled part of the proximal articular facet in the Purussaurus pubis is thus impossible to articulate with the ischium when the medial part is in articulation ( Figure 4C ) . The medial attachment areas between the left and right distal pubes were likely a point contact ( Figure 4C ) ; compared to extant crocodylians , the medial contact between the ischia was reduced in length . In addition to isolated sacrals , the complete sacral region with the three sacrals of UNEFM-CIAAP-1369 is known for P . mirandai . Together with the evidence from the medial articulation sites on the ilia , there is strong support for the ‘three sacral’ condition in this extinct caimanine as a unique trait within crown-Crocodylia ( Böhmer et al . , 2015; Mook , 1921b; Romer , 1956 ) . Based on comparison with extant crocodylian skeletons ( e . g . , Crocodylus niloticus , Caiman yacare; Figure 4D ) and the shapes of the sacral ribs and associated attachment sites on the ilia , we show that the last dorsal ( i . e . , lumbar ) vertebra has been sacralised into a dorsosacral in P . mirandai . This leads to a vertebral count of nine cervicals , 14 dorsals , and three sacrals . The first caudal in P . mirandai has , as in extant crocodylians , a biconvex centrum shape , with the transverse processes being posteriorly oriented . The femoral head of the best-preserved P . mirandai femur appears to have a more medially directed orientation than in most other crown-Crocodylia , at only ~12° relative to the mediolateral axis of the femoral condyles ( Figure 4E ) . While bone distortion and wear of the condyles might have altered this orientation , the low torsion of the femoral head compared to condyles contrasts even with the relatively medial 36° orientation in Stratiotosuchus or the larger angles found in extant crocodylians ( Riff and Kellner , 2011 reported an angle of 52° in a specimen of Caiman yacare ) . In humans , the pathological condition of a congenital malformation lumbosacral transitional vertebra ( LSTV ) is widespread and frequently studied in the medical and veterinary medical literature . This condition is proposed to be linked to lower back pain symptoms referred to as Bertolotti’s syndrome in humans ( see Holm et al . , 2017; Jancuska et al . , 2015 ) for overviews ) or predispose for cauda equina syndrome in dogs ( e . g . , Flückiger et al . , 2006 ) . The developmental mechanisms underlying these pathologies occur early during embryogenesis , when Homeobox gene expression , that is Hox8 , Hox10 , and Hox11 ( the latter two also playing an important role in limb patterning ) , induces axial patterning of the posterior dorsal ( =lumbar ) , sacral , and anterior caudal regions in the embryo ( Casaca et al . , 2014; Swinehart et al . , 2013; Wellik , 2007; Wellik and Capecchi , 2003 ) . Hoxc8 is expressed throughout the dorsal ( =thoracolumbar ) region of the crocodylian skeleton , but not in the sacral region ( Böhmer et al . , 2015; Böhmer , 2013; Mansfield and Abzhanov , 2010 ) . Slight heterochronic shifts in the interplay of Hox gene expressions are known to cause drastic shifts in development ( Gérard et al . , 1997 ) , as in Hoxa10 and Hoxa11 expression in the sacral region ( Carapuço et al . , 2005 ) . The latter study showed that changes in the timing of expression of Hoxa10 and Hoxa11 within the presomitic mesoderm versus expression in the somites influenced rib development , sacralisation , and caudal formation in the sacral region . As such , Hoxa10 expression in the presomitic mesoderm led to formation of vertebrae without ribs , whereas the same expression in the somites led to vertebrae with ribs; similarly , Hoxa11 expression in the presomitic mesoderm caused sacralisation , whereas in the somites it induced caudal formation . Among extant crocodylians , numerous pathological conditions affecting the axial skeleton have been described ( Kälin , 1933; Rothschild et al . , 2012 ) . These reportedly include changes in the number and identity of vertebrae , such as vertebral addition in the dorsal series , shifting of the sacral region , sacralisation of last dorsals and first caudals ( Baur , 1886; Baur , 1889; Reinhardt , 1873; Reinhardt , 1874 ) , but images or drawings were usually not provided , and only occasionally the studied specimens were directly referred to ( e . g . , Baur , 1889 ) , p . 240 ) . A unilateral articulation of a left transverse process of the last , hemisacralised dorsal vertebra with the left ilium has also been recently noted and figured in a pathological specimen of Alligator mississippiensis , while discussing the repeated independent appearance of non-pathological dorsosacrals among Triassic Archosauriformes ( Griffin et al . , 2017 ) . The latter study discussed the potential role of Hox genes , especially Hox11 paralogs , and changes in the timing of expression thereof , for compartmentalizing the dorsal and sacral series and shifting the boundary between them in the phytosaur Smilosuchus from the Upper Triassic Chinle Formation of northeastern Arizona , USA . In our sample of comparative materials , we discovered a hemisacralised last dorsalsacral in a juvenile specimen of the dwarf caiman Palaeosuchus palpebrosus ( Figure 5; specimen RVC-JRH-PP4 ) . This specimen exhibits the girdle bones and stylopodial elements in life position/articulation and thus serves well for comparison to the Purussaurus fossils . It has a wide scapulocoracoid synchondrosis , and a more strongly torsioned femoral head ( Figure 5C; with 53 . 5° similar to angle reported for Caiman in Riff and Kellner , 2011 ) compared to the distal condylar plane ( Figure 5C , D ) . As reported for the A . mississippiensis specimen ( Griffin et al . , 2017 ) mentioned above , the primordial sacral 1 of P . palpebrosus ( Figure 5D , E ) shows the development of a shallow flange onto the dorsosacral and a slight posterior shift of its sacral rib base on the side where the hemisacralization occurred . In addition , another pathological specimen of P . palpebrosus ( MACV-6139; image courtesy of Mariano Padilla Cano ) shows a pathological ‘three-sacral’ condition in which the size and shape of the sacral ribs and the articulation sites on the medial surfaces of the ilia varies between the left and right side ( Figure 5—figure supplement 1 ) . To our knowledge , Purussaurus mirandai is the first member of crown-group Crocodylia with three non-pathological sacrals ( see Müller et al . , 2010 ) . A similar condition was reported in the terrestrial notosuchian crocodylomorph Notosuchus from the Late Cretaceous of Argentina , which has the second and third sacral fused , in some atoposaurid neosuchians with three unfused sacrals ( Fiorelli and Calvo , 2008; Nobre and Carvalho , 2013; Pol , 2005; Tennant and Mannion , 2014 ) , and in some teleosauroid Machimosaurini , a group of large-sized marine crocodylomorphs from the Jurassic , in which the first caudal was sacralised ( e . g . , Johnson et al . , 2018; Jouve et al . , 2016 ) . All of these groups , however , differ strongly from the conserved body plan of extant crocodylians . Molnar et al . ( 2014 ) identified the lumbosacral joint ( between dorsal 15 and sacral 1 ) of the vertebral column of Crocodylus niloticus as having the highest intervertebral joint stiffness , which had been hypothesized as beneficial in supporting a large tail and countering hind limb forces ( Willey et al . , 2004 ) . Based on a study of five small specimens ( between 40 and 50 cm snout-vent length ) of Alligator mississippiensis , Willey et al . ( 2004 ) found the centre of mass to be situated about 70% along the trunk length , and the tail mass to be up to 28% of the total body mass . In addition , Aureliano et al . ( 2015 ) estimated the largest specimens of Purussaurus brasiliensis to have reached more than 12 m in body length and over eight metric tons in mass . Given the 145 cm length of the preserved lower jaw and the size of the vertebrae with up to 11 . 8 cm centrum length , AMU-CURS-541 is estimated to have ranged between eight to ten meters in total length . Unfortunately , the associated femur of AMU-CURS-541 was only partially preserved , making size estimations based on this skeletal element difficult ( Farlow et al . , 2005 ) . Comparisons to other specimens of Purussaurus from the Urumaco Formation indicate , however , that AMU-CURS-541 is not the largest individual recovered so far . In AMU-CURS-20 , the largest preserved vertebra ( 13 . 6 cm centrum length ) is about 15% larger compared to the vertebrae of AMU-CURS-541 . In addition , the dimensions of the associated and fairly complete right femur of AMU-CURS-20 ( 54 cm length; compared to 51 cm of the femur of the holotype UNEFM-CIAAP-1369 of P . mirandai ) are among the largest crocodylian femora known ( Salas-Gismondi et al . , 2007 reported on a 54 . 5 cm long femur of Purussaurus from Peru ) . With 210 mm , the femur of AMU-CURS-20 has a similar minimum shaft circumference as the femoral shaft fragment MNN G102–2 of Sarcosuchus imperator , which was used by Farlow et al . ( 2005 ) to estimate the total length of that specimen to range between 7 . 2 and 9 . 1 m . We used the dataset and R software code from O’Brien et al . ( 2019 ) with P . mirandai input into the phylogeny as per Figure 1 and with a branch length of 0 . 01 ( assuming 25% corrected mass as per their methods ) to estimate the total body mass and total length of specimen AMU-CURS-541 ( head width ~0 . 88 m ) . Based on this more conservative approach , our analysis estimated 1686–2637 kg and 7 . 11–8 . 01 m for total body mass and length , respectively ( from lower to upper interquartile ranges of Bayesian analysis ) . This compares favorably with other large Crocodyliformes such as Sarcosuchus imperator ( as per O’Brien et al . , 2019 ‘longirostrine’ calculations ) , estimated at 1976–2981 kg and 8 . 50–9 . 47 m ( their Table 5 ) . Compared to the more massive and robust appearance of P . brasiliensis , the more ‘gracile’ P . mirandai specimen AMU-CURS-541 could have had the tail alone weighing up to 472–738 kg ( based on Willey et al . , 2004; values that exceed most average masses of extant crocodylians ( Grigg and Kirschner , 2015 ) ; the exceptionally large ‘Lolong’ , an old Crocodylus porosus 6 . 17 m long , was reported to be 1075 kg: Britton et al . , 2012 ) . Regardless of the mass estimation used , we speculate that the ‘three sacrals’ condition , together with the robustness of the hind limb bones encountered in P . mirandai , was related to these animals’ giant size and body mass ( Aureliano et al . , 2015; Salas-Gismondi et al . , 2007 ) . A longer sacrum with three instead of only two sacrals articulating with the ilium expands the connection between the axial skeleton and the posterior appendicular skeleton , thus potentially increasing stability of this region in the vertebral column and allowing for better transmission of forces across the pelvic and caudal regions . Comparative data from other giant crocodylians , however , are either scarce or remain largely undescribed ( e . g . , Deinosuchus: Colbert and Bird , 1954; Holland , 1909; Rivera-Sylva et al . , 2011; Schwimmer , 2002; Euthecodon: Storrs , 2003; Gryposuchus: Langston , 1965; Riff and Aguilera , 2008; Laganosuchus: Sereno and Larsson , 2009; Mourasuchus: Langston , 1965; Sarcosuchus: Dridi , 2018; Sereno et al . , 2001 ) . Thus , the influence of body size and mass on the sacral region of giant crocodylians remains largely unexplored . Of those taxa , at least one well-preserved ilium from Mourasuchus atopus shows a ‘π’ pattern of ridges on its medial surface ( UCMP 38012 , Figure 3—figure supplement 2 ) . This could potentially indicate the presence of three sacrals in this additional giant caimanine if confirmed by further specimens . The ‘three-sacral’ condition described herein is purely of functional origin , as both Mourasuchus and Purussaurus are giants ( although likely of distinctly different body masses ) , or there may also be a phylogenetic signal involved , as both taxa are sister genera within Caimaninae . Other studies of Crocodylomorpha/Archosauria have inferred that expanded ilia and sacralisation correlate with more upright limb posture and/or improved support against gravity ( e . g . , Riff and Kellner , 2011 ) . The above evidence from the iliosacral region corroborates other specializations for more upright limb orientation or simply weight support in P . mirandai , including the more vertically oriented pectoral girdle and low torsion of the femoral head relative to the condyles . The pectoral and pelvic regions of Purussaurus mirandai show a peculiar morphology in comparison to extant taxa , including a narrow scapulocoracoid contact , wide and prominent deltoid crest on the scapula serving as a muscle origin , more vertically oriented pectoral girdle , and autapomorphic ‘three sacral’ condition in the pelvis , interpreted herein to be linked to the giant body size . We infer that the underlying developmental mechanisms , especially an earlier timing of expression of Hox11 and partial suppression of Hox10 ( Wellik and Capecchi , 2003 ) , resulting in shifted domains of Hoxc8 and Hox11 , led to the formation of the dorsosacral in P . mirandai . Discovery and examination of additional specimens of Purussaurus with exceptional preservation to document genus variation , and of other giant crocodylians would lead to a better understanding of the anatomical peculiarities of these taxa . Such studies should inspect the medial side of the ilium to provide indirect information on the sacral count .
The numerous postcranial elements of the new specimen AMU-CURS-541 were used to re-score some hitherto unknown characters for the species Purussaurus mirandai . All characters are taken from Salas-Gismondi et al . ( 2015 ) , a matrix largely based on and modified from Brochu ( 2011 ) and Jouve et al . ( 2008 ) . Three maximum parsimony analyses were performed using TNT v . 1 . 5 ( Goloboff and Catalano , 2016 ) . In the first analysis , AMU-CURS-541 was scored as a separate terminal taxon besides the three species of Purussaurus: P . mirandai , P . neivensis , and P . brasiliensis . For the second analysis , the new scorings based on AMU-CURS-541 were added to the previous scorings of P . mirandai . Consistency indices and Bremer support values were calculated in TNT v . 1 . 5 using the ‘stats . run’ and ‘Bremer . run’ scripts ( downloaded on 02 . 11 . 2018 from http://phylo . wikidot . com/tntwiki; see also http://gensoft . pasteur . fr/docs/TNT/1 . 5/ ) and the latter was checked manually by collapsing tree topologies through inclusion of successive suboptimal trees . In both phylogenetic analyses , the extinct taxon Melanosuchus fisheri was removed due to recently revealed inconsistencies between the holotype and referred material ( Bona et al . , 2017; Foth et al . , 2018 ) . In the third analysis , the updated scoring of P . mirandai was used , but Me . fisheri was left in , to see how the addition or the removal of that taxon influenced the results . In all analyses , the heuristic search ( traditional search; space 60000 trees in memory , random seed = 1 , Tree Bisection Reconnection ( TBR ) mode activated ) was run with Bernissartia fagesii as outgroup , 1000 random additional sequence replicates and 100 trees saved per replication . Characters were equally weighted , unordered , and set as non-additive . Scorings of 201 characters used for P . mirandai updated by the scores of AMU-CURS-541 in the matrix of Salas-Gismondi et al . ( 2015 ) : ? ? ? ? 1 110 ? ? 00001 0 ? 0 ? 0 11110 01 ? ? ? ? ? 111 0 ? 11 ? ? 01 ? ? ? 1100 00 ? 2 ? ? 1 ? 11 ? 1201 100 ? 1 1001 ? ? ? ? 0 ? 112 ? 0 0021 ? 10000 11100 ? ? ? 10 00 ? ? ? ? 0000 00011 11 ? 1 ? 11021 0 ? 111 110 ? 1 10200 10112 000 ? ? 10104 ? ? ? ? ? ? ? ? ? ? 0011 ? 2 ? 00 ? 10 ? ? 0 01000 00 ? ? 1 00 ? 10 0 Scorings of 201 characters used for AMU-CURS-541 as terminal taxon in the matrix of Salas-Gismondi et al . ( 2015 ) : ? ? ? ? 1 110 ? ? 00001 0 ? 0 ? 0 11110 0 ? ? ? ? ? ? 111 ? ? 11 ? ? 01 ? ? ? 1100 00 ? 2 ? ? ? ? ? 1 ? ? 2 ? ? ? 00 ? 1 1001 ? ? ? ? 0 ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? 1 ? ? Colección Paleontológica de la Alcaldía Bolivariana de Urumaco , Estado Falcón , Venezuela; MACUT , Collections of the Museo di Anatomia Comparata dell’Università di Torino hosted by the Museo Regionale di Scienze Naturali Torino , Italy; MACV , Museum of Comparative Anatomy of Vertebrates , Complutense University of Madrid , Spain . MCNC , Museo de Ciencias Naturales de Caracas , Venezuela; MNN , Musée National du Niger , Niamey , Niger; PIMUZ , Palaeontological Institute and Museum , University of Zurich , Switzerland; QM , Queensland Museum , Brisbane , Australia; RVC , Royal Veterinary College , London , UK; UNEFM-CIAAP , Universidad Nacional Experimental Francisco de Miranda/Centro de Investigaciones Antropológicas , Arqueológicas y Paleontológicas , Coro , Venezuela . | All living crocodiles , alligators , caimans and gharials – collectively called crocodylians – have a similar body plan that suits their lifestyles as aquatic ambush predators . This similarity extends throughout their bodies , including the skeleton . Their backbones , for example , always have nine vertebrae in the neck , 15 in the trunk , and two in the pelvis . Closely related extinct crocodylians also organize their spines in the same way . Scheyer et al . , however , now report that one extinct caiman called Purussaurus mirandai had a spine that was structured unlike any other known crocodylian . Weighing almost three tons ( ~2 , 600 kg ) , the 10-meter-long Purussaurus was more than twice as heavy as the largest living crocodylian , the saltwater crocodile . When Scheyer et al . examined fossilized remains from Venezuela that are estimated to be between 7–5 million years old , they found an extra vertebra in the creature's pelvic area and one less in its trunk . Scheyer et al . speculate that this unusual arrangement may have helped the extinct creature to support its massive weight and compensate for the strain imposed on its skeleton . Within the animal kingdom , so-called homeobox genes dictate how different body structures , including the spine , develop in embryos . Shifts in where these genes are active in the embryo can result in an extra pelvic vertebra in humans and other animals . Scheyer et al . conclude that changes in the boundaries of the activity of homeobox genes may also explain the extra pelvic vertebra in this ancient caiman . It is not yet clear if other extinct crocodylians had extra pelvic vertebrae as well . But these new findings are likely to lead to more research on related giant crocodylian fossils to find out . Such research could help scientists to better understand the biomechanics of crocodylians and may lead to new insights on caimans , which have thrived in the tropics of northern South America for the past seven million years . Further research in this area may also help explain how these reptiles have adapted to their environment and the role they play in their ecosystems , which is currently threatened by human activity . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"evolutionary",
"biology"
] | 2019 | Giant extinct caiman breaks constraint on the axial skeleton of extant crocodylians |
The seemingly limitless diversity of proteins in nature arose from only a few thousand domain prototypes , but the origin of these themselves has remained unclear . We are pursuing the hypothesis that they arose by fusion and accretion from an ancestral set of peptides active as co-factors in RNA-dependent replication and catalysis . Should this be true , contemporary domains may still contain vestiges of such peptides , which could be reconstructed by a comparative approach in the same way in which ancient vocabularies have been reconstructed by the comparative study of modern languages . To test this , we compared domains representative of known folds and identified 40 fragments whose similarity is indicative of common descent , yet which occur in domains currently not thought to be homologous . These fragments are widespread in the most ancient folds and enriched for iron-sulfur- and nucleic acid-binding . We propose that they represent the observable remnants of a primordial RNA-peptide world .
The origin of most present-day proteins can be attributed to combinatorial shuffling and differentiation events involving a basic set of domain prototypes , which act as the unit of protein evolution ( Anantharaman et al . , 2001; Apic et al . , 2001; Ponting and Russell , 2002; Orengo and Thornton , 2005 ) . Many of these domains can be traced back to the time of the Last Universal Common Ancestor ( LUCA ) ( Kyrpides et al . , 1999; Koonin , 2003; Ranea et al . , 2006 ) , a hypothetical primordial organism from which all life on earth is thought to have descended approximately 3 . 5 billion years ago ( Glansdorff et al . , 2008 ) . The origin of domains themselves , however , is poorly understood . An origin de novo , by random concatenation of amino acids , appears impossible due to the high sequence complexity and low folding yield of polypeptides , as well as to the absence of abiotic processes that could produce peptide chains of more than 5–10 residues . The abiotic scenario would also leave open the fundamental question as to how the information contained in successful polypeptides could have been passed on . Many lines of evidence , including the identification of local sequence and structure similarity within domains of different fold ( Brennan and Matthews , 1989; Doherty et al . , 1996; Copley et al . , 2001; Grishin , 2001b; Friedberg and Godzik , 2005; Alva et al . , 2007; Andreeva et al . , 2007 ) , or the frequent construction of domains by repetition of subdomain-sized fragments ( McLachlan , 1987; Andrade et al . , 2001; Hocker et al . , 2002; Chaudhuri et al . , 2008; Remmert et al . , 2010 ) , show that domains might not constitute the only evolutionary unit of protein structure . These observations led to the proposal that the first folded domains arose by repetition , fusion , recombination , and accretion from an ancestral set of peptides ( Fetrow and Godzik , 1998; Lupas et al . , 2001; Soding and Lupas , 2003 ) that emerged in the RNA world ( Gilbert , 1986 ) , in which RNA served both as carrier of genetic information and catalyst of metabolic reactions ( Jeffares et al . , 1998 ) . According to this model , the local similarities found in modern proteins represent the observable remnants of such peptides . In the RNA world , which is widely thought to have been an important intermediate stage in the origin of cellular life , simple peptides may have been recruited by RNA to expand its functional repertoire . The catalytic range of RNA molecules is restricted ( Joyce , 2002 ) and peptides are good chelators of metals and small molecules . Peptides are also beneficial for RNA thermostability and folding specificity , and for the formation of oligomeric complexes . For these reasons , peptides of initially abiotic origin may have been co-opted as cofactors . In time , selective pressures on availability , interaction specificity , and functional effectiveness would have driven the emergence of longer , RNA-encoded and -produced peptides , optimized for the formation of secondary structure by exclusion of water with the RNA scaffold . It is known that many peptides formed of the 20 proteinogenic amino acids have an intrinsic affinity for RNA and form secondary structures upon binding ( Patel , 1999; Das and Frankel , 2003 ) . By repetition , fusion , recombination , and accretion , these preoptimized peptides would have reached a level of complexity enabling them to exclude water by hydrophobic contacts , making them independent of the RNA scaffold . In this theory , protein folding would have been an emergent property of peptide-RNA coevolution .
Given the striking conservation of many proteins in sequence and structure across evolutionary time , we conjectured that if this hypothesis is true , we might be able to see remnants of these primordial peptides in modern proteins . To this end , we decided to take a comparative approach based on the systematic analysis of present-day domains , similar to the approach taken by linguists to reconstruct ancient vocabularies by comparing modern languages . The comparative studies of languages and of proteins , in fact , exhibit many parallels because of their shared linear nature of information storage , the high conservation of evolutionary modules , and the similarity of evolutionary constraints acting upon them ( Gray and Atkinson , 2003; Pagel et al . , 2007; Searls , 2013; List et al . , 2014 ) ( Figure 1A ) . Today , for instance , it is known that hundreds of words in European languages contain the conserved Semitic root qnw ( *qanaw- ) meaning ‘reed’ ( Huehnergard , 2011 ) , despite it having diversified into a wide range of functional forms by the same processes as already familiar from biological evolution ( e . g . , orthology , paralogy , horizontal transfer ) . 10 . 7554/eLife . 09410 . 003Figure 1 . The evolution of words and proteins shows many parallels . ( A ) The Semitic root qnw ( *qanaw- ) , meaning reed , is the ancestor of hundreds of words in many different languages , following the same mechanisms as already known from biological evolution . Here we track the descendants of this root in the English language , arisen through the intermediary of Latin and Greek . In addition to the orthologous Greek word kanna ( reed ) , paralogous cognates arose in antiquity based on certain attributes of reed , e . g . , the levelling rule kanon ( taking the straight and rigid attribute of reed ) , the wicker basket kanastron ( flexible ) , and the Latin water duct canalis ( round and hollow ) . A few examples of analogous words , which appear to be related to the descendants of qnw but have different evolutionary origins , are shown in green . ( B ) The primordial β-hammerhead motif ( shown in red ) is seen in four different folds , which cover a wide array of functions . Following our hypothesis of an origin in the RNA world , we propose that RNA binding is the orthologous function of this peptide , seen today in ribosomal protein L27 and exosome subunit RRP4 . Paralogous functions arose around the time of the Last Universal Common Ancestor from its ability to form a biotin-binding domain by duplication , yielding the biotin-dependant enzymes of the barrel-sandwich hybrid fold , and to serve as a structural element in domains formed by accretion , yielding a domain of RNA-polymerase β’ subunit , as well as a range of enzymes with an α/β-hammerhead fold . By our analysis , enzymes classified in the α/β-hammerhead fold superfamily d . 41 . 5 , such as MoaE , are analogous to the other superfamilies in this fold , due to a lack of detectable sequence similarity , but nevertheless contain a supersecondary structure resembling the β-hammerhead . DOI: http://dx . doi . org/10 . 7554/eLife . 09410 . 003 To reconstruct the ‘vocabulary’ of ancient peptides , we aimed at finding local similarities in sequence and structure within globally different folds , which are presently thought to have arisen independently , by convergent evolution . Since the events that led to the emergence of domains took place before the time of LUCA , modern domains might only retain weak signals thereof in their sequences . As protein structures diverge much more slowly than their sequences , structural similarity is often used for identifying such distant events . However , similar structures may have arisen convergently , owing to the limited number of conformations available to a folded polypeptide chain , particularly at the supersecondary structure level ( Salem et al . , 1999; Kim et al . , 2009; Fernandez-Fuentes et al . , 2010 ) . Consequently , structure similarity alone does not provide conclusive evidence of common ancestry . In contrast , the combinatorial sequence space is enormous and many sequences are compatible with a particular local structure , so that sequence convergence is rare . Thus sequence similarity is a more reliable marker for common ancestry ( Soding , 2005; Kim et al . , 2009 ) . We have therefore employed sequence similarity , as evaluated by the comparison of profile hidden Markov models ( HMMs ) ( Soding et al . , 2005 ) , as a first criterion for inferring common ancestry of domains in this study . Due to the large evolutionary divergence of sequences , we used structural similarity as a second criterion to confirm the potential homology relationships . To implement this comparative approach between domains of different fold , we needed a reference database for the assignment of domains to fold types . For this we used the SCOPe ( Murzin et al . , 1995; Fox et al . , 2014 ) database ( release 2 . 03 ) , which is a point of reference in the classification of protein folds . In this database , the first two classification levels ( family and superfamily ) capture homologous relationships , while the grouping of structurally similar superfamilies into the same fold reflects convergent evolution , i . e . analogy . In order to reduce the very large background of obvious homologous matches , we filtered SCOPe to a maximum of 30% sequence identity ( SCOPe30 ) . At this level , many relationships considered homologous by SCOPe are removed , whereas representatives for all families and superfamilies are still retained . As detailed in the ‘Materials and methods’ , we then compared the resulting domain set in sequence space using HHsearch with stringent settings , and subsequently in structure space using TM-align ( Zhang and Skolnick , 2005 ) . We plotted the obtained scores separately for comparisons within families ( presumed homologous relationships; Figure 2A ) and between folds ( presumed analogous relationships; Figure 2B ) . The expected score distributions would have been at the top , right ( high HHsearch probabilities , high TM-scores ) for homologous relationships and bottom , left ( low HHsearch probabilities , low TM-scores ) for analogous ones . In fact , the distributions we obtained were bimodal , with a higher incidence of scores at both the top , right and bottom , left of the plots . The presence of some low-scoring relationships in the homologous set ( Figure 2A ) did not appear too surprising , given the considerable evolutionary distance of some relationships captured in SCOPe , but the presence of some high-scoring relationships in the analogous set ( Figure 2B ) did , as it suggested the presence of hitherto unrecognized homologous relationships . We decided to explore these further , using cut-offs at HHsearch probabilities of 70% ( corresponding to P-values < 5e-05 ) and TM-scores of 0 . 5 . At these threshold values , about a fourth of presumed homologous relationships in SCOPe30 and >99 . 95% of matches between folds , which are presumed to be analogous , are omitted . This provided a substantial margin of safety in evaluating the high-scoring relationships between domains presumed to be analogous ( Figure 2B ) . 10 . 7554/eLife . 09410 . 004Figure 2 . Estimation of cut-offs for HHsearch probability and TM-score . We compared all domains in the SCOPe30 set in sequence space using HHsearch and subsequently in structure space using TM-align ( see ‘Materials and methods’ ) , and plotted the obtained scores . Separate plots for comparisons of domains within families ( A ) and between folds ( B ) were generated . Scores would have been expected at high HHsearch probabilities and TM-scores for intrafamily comparisons ( presumed homologs , Panel A ) and low HHsearch probabilities and TM-scores for interfold comparisons ( presumed analogs , Panel B ) , but the score distributions were in fact bimodal , as also illustrated by the histograms top and right in each panel , which are plotted as probability density functions . In the comparison of domains of different fold , matches with an HHsearch probability of < 10% are not plotted . DOI: http://dx . doi . org/10 . 7554/eLife . 09410 . 004 Since the comparisons made in the analogous set were between domains of different fold , the high-scoring matches always involved subdomain-sized fragments . Those fragments that satisfied the cut-offs and a number of additional criteria , as detailed in the ‘Materials and methods’ , were clustered together automatically if they overlapped by at least 80% of their length . Following upon this step , clusters were inspected individually and merged further where appropriate , as described in the ‘Materials and methods’ . We obtained 65 clusters , of which 20 relied on a global similarity of the folds and were omitted from further consideration . Such cases of globally similar folds that are nevertheless classified as different are mainly due to homologous fold change events , such as strand invasion , hairpin swapping , or circular permutation ( Grishin , 2001a; Andreeva and Murzin , 2006; Alva et al . , 2008 ) . Thus , for instance , SCOPe superfamilies d . 12 . 1 and d . 58 . 7 are clearly related globally by circular permutation , despite being classified into different folds . A further 5 groups were due to artifacts in domain boundaries assigned by SCOPe and were also omitted . For example , the d . 79 . 4 . 1 family makes connections to the a . 5 . 10 . 1 family , as its sequence in SCOPe encompasses the latter . This yielded a final set of 40 clusters , within which all fragments were inspected individually , superimposed , and trimmed to a consensus length where appropriate ( Figure 3 , Figure 3—source data 1; Table 1 ) . As examples , we show the different embodiments of the β-hammerhead motif , found in 4 folds comprising 8 superfamilies ( Figure 1B , Figure 3: fragment 12 ) , and the nucleic acid-binding helix-hairpin-helix motif ( Doherty et al . , 1996; Shao and Grishin , 2000 ) , found in 8 folds comprising 15 superfamilies ( Figure 3: fragment 2 , Figure 4 ) . The median length of the 40 consensus fragments is 24 residues , with the shortest fragment comprising 9 and the longest 38 residues . This is in accord with the expectation that ancient peptides were simple and subdomain-sized . Of these fragments , only about half correspond to typical supersecondary structure elements , such as α-hairpins , β-hairpins , β-meanders , and βαβ-elements ( Salem et al . , 1999 ) , whereas the others are unusual fragments with odd shapes , which frequently do not form compact structures . This supports the notion that they predate the emergence of hydrophobic contacts as a driving force for protein folding and that their open structures reflect the association with an RNA scaffold , as still seen in ribosomal proteins today ( Soding and Lupas , 2003 ) . 10 . 7554/eLife . 09410 . 005Figure 3 . Vocabulary of primordial peptides that gave rise to folded proteins . The 40 peptides we detected are shown as ensembles in backbone representation; α-helices are coloured in yellow , β-strands in green , and loops in gray . Detailed information on each fragment is provided in Table 1 and Figure 3—source data 1 . The fragments are numbered sequentially and their occurrence in different folds and superfamilies of SCOPe is given . Fragments reported individually before are indicated by a dot . Nucleic-acid binding , nucleotide-binding , and metal-binding motifs are highlighted in yellow , blue , and red , respectively . Fragments found in ribosomal proteins are indicated by red font colour . Fragments that form folds by repetition are boxed . DOI: http://dx . doi . org/10 . 7554/eLife . 09410 . 00510 . 7554/eLife . 09410 . 006Figure 3—source data 1 . Multiple sequence alignments and accession details for the 40 primordial fragments shown in Figure 3 and the 5 B-set fragments shown in Figure 3—figure supplement 1 . Multiple homologous copies of a fragment found within the same domain are indicated by a * . Fragments reported individually before are indicated by a + . DOI: http://dx . doi . org/10 . 7554/eLife . 09410 . 00610 . 7554/eLife . 09410 . 007Figure 3—figure supplement 1 . Vocabulary of primordial peptides ( B-set ) . In addition to the 40 fragments described in the main text , we detected five further fragments after relaxing the sequence similarity requirement to an HHsearch probability of 60% . One of these fragments , the Asp box ( B1 ) , has been previously described and it forms folds by repetition ( emphasized by dotted boxes ) . Detailed information is provided in Figure 3—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 09410 . 00710 . 7554/eLife . 09410 . 008Figure 4 . The nucleic-acid binding helix-hairpin-helix motif is found in 8 different folds comprising 15 superfamilies . ( A ) Representative domains from the eight SCOPe folds . The motif is coloured in red and the remainder of the structure is shown in gray . The SCOPe family a . 60 . 2 . 1 contains two copies of this motif , whereas the remaining folds contain one copy each . ( B ) Structural superimposition of the helix-hairpin-helix motifs displayed in panel A . ( C ) Sequence alignment of the motifs shown in panel A . Residues conserved in at least half of the aligned sequences are highlighted in black and similar residues are highlighted in gray . DOI: http://dx . doi . org/10 . 7554/eLife . 09410 . 00810 . 7554/eLife . 09410 . 009Table 1 . Data on the 40 primordial fragments . DOI: http://dx . doi . org/10 . 7554/eLife . 09410 . 009FragmentNumber offolds , superfamiliesRepetition in SCOPe foldor superfamilyRibosomal ( SCOPe id; protein ) Ligands9 most ancient and basal folds ( SCOPe id ) Occurrence in the 10 folds with the largest number of superfamilies in SCOPe1+14 , 20a . 4 . 5; S19eDNAa . 42+8 , 15a . 60 . 2a . 156 . 1; S13DNA , RNAa . 603+3 , 3a . 174 . 1DNAc . 37 . 1 . 2042 , 2a . 2 . 2; L29RNAa . 252 , 2d . 14 . 1; S5RNA6+2 , 2d . 52 . 3; S3DNA , RNA7+6 , 8all folds containing itDNA , CTP , SAM , FMN8+10 , 10FAD , NAD , NEA , NAJ , NAP , NAI , NDP , AMP , NMN , APR , LNC , A3D , ODP , UMA , SAH , SAM , COA , c . 2 , c . 6692 , 2DNA103 , 4d . 66 . 1; S4RNA112 , 2b . 53 . 1; TL5RNA124 , 8b . 84 . 1b . 84 . 4; L27RNA13+5 , 30all folds containing itDNA142 , 2ZN , DNA155 , 7d . 37 . 1FMN , AMP , FAD16+3 , 3ATP , GTP , ADP , CTD , DGP , DCP , ANP , TMP , GDPc . 3717+3 , 3a . 138 . 1HEM , HECa . 2418+2 , 2a . 1 . 2 , d . 58 . 1SF4 ( Fe4S4 ) d . 58 . 1d . 5819+3 , 3ZN202 , 2FES212 , 2a . 39 . 1 , a . 139 . 1CA222 , 2a . 25 . 1FEC , FE , FE2 , HEM232 , 2COA24+2 , 2d . 45 . 1; L7/1225+2 , 7f . 4263 , 3g . 327+2 , 228+7 , 12a . 118 . 8a . 24 , a . 118292 , 2b . 34 . 4b . 34302 , 2a . 2312 , 2322 , 2c . 55 . 1332 , 234+3 , 4352 , 2d . 58 . 49d . 58362 , 2c . 55 . 5372 , 2382 , 2a . 60392 , 2a . 118402 , 2a . 20 . 1+previously reported fragments To evaluate whether the sequence similarity exhibited by these fragments could be the result of biophysical constraints , rather than common descent , we searched SCOPe30 with each fragment for structurally similar matches: 36 of the 40 had at least one match to another superfamily with a TM-score ≥ 0 . 5 , but undetectable sequence similarity ( HHsearch probability <10% ) . Indeed , of these 36 fragments , 34 had more than half of their structure matches to fragments with undetectable sequence similarity . This shows that essentially every one of the structures we detected can be formed by fundamentally different sequences . To verify this anecdotal observation in a systematic way , we analyzed the relationship between sequence and structure similarity in our fragments by comparison with a reference set of 40 of the most frequent supersecondary structure elements from the Smotifs library ( Fernandez-Fuentes et al . , 2010 ) ( assembled as described in the ‘Materials and methods’ ) . We reasoned that a correlation should be detectable in homologs , as these would have started ancestrally with identical sequences and structures , before gradually diverging towards a baseline of similarity . There should be no correlation however in analogs , unless structural constraints had limited the number of residues allowed at specific positions , causing a convergence of the sequences ( Remmert et al . , 2010; Kopec and Lupas , 2013 ) . We therefore computed sequence similarity scores using structure-based sequence alignments for each fragment in the two sets against all domains in SCOPe30 . For fragments in our set , we considered matches to other superfamilies in which we had detected the respective fragment ( listed in the Figure 3—source data 1 ) as homologous and to all other folds as analogous ( Figure 5A ) . For the Smotifs fragments , we considered matches to the same superfamily as homologous and to all other folds as analogous ( Figure 5B ) . For both fragment sets , the presumed homologous matches show a strong correlation between sequence and structure similarity . The correlation for the Smotifs reference set is expected , as the homologs there follow the generally accepted criteria laid out in the SCOPe classification . Observing a nearly equivalent correlation for our fragment set therefore underscores our inference of homology . Note that , although the correlation is slightly weaker in our fragments than in the Smotifs , this is likely due to the fact that comparisons in our set are made between superfamilies , whereas in the Smotifs they are made within superfamilies , i . e . across smaller evolutionary distances . In contrast , the correlation between sequence and structure similarity in the presumed analogous matches is very weak for both our fragments and the Smotifs . This shows that structurally induced sequence convergence is very low and the fragments can be formed by a broad range of different sequences . Biophysical constraints can therefore not explain the sequence similarity exhibited by our fragments . 10 . 7554/eLife . 09410 . 010Figure 5 . Sequence similarity of our fragments cannot be explained by structural constraints . ( A ) For each occurrence of any of our 40 fragments , we searched for structural matches in SCOPe30 and plotted the TM-align score versus the profile-similarity score for the fixed alignment given by TM-align . The putatively homologous matches to occurrences of the same fragment in another superfamily are shown in red . Matches to fragments outside the list of folds in which the query fragment was found to occur ( i . e . non-homologous matches ) are blue . ( B ) Same as A , but using the Smotifs reference fragments as queries instead of our set . Matches within superfamilies ( homologs ) are shown in red , matches between fragments from different folds ( analogs ) are shown in blue . For both sets , sequence and structure similarity scores are significantly correlated for presumably homologous matches ( our set: r=0 . 38; Smotifs: r=0 . 56 , see linear regression lines ) but not for analogous matches ( our set: r=0 . 14; Smotifs: r=0 . 12 ) . ( C , D ) Distribution of profile similarity scores for matches with a TM-score ≥ 0 . 5 , for the homologous and analogous distribution in the plots ( A ) and ( B ) , respectively . The means of the Gaussian fits are exactly the same in C and D . DOI: http://dx . doi . org/10 . 7554/eLife . 09410 . 010 We are of course aware that molecular homology cannot be proven rigorously by scientific standards and that the boundaries of what constitutes evidence of common descent evolve continuously as extrapolation from increasingly distant connections are found to yield useful structural and functional predictions . Due to the conservative nature of our sequence comparisons and the range of additional criteria we applied in order to eliminate potentially spurious matches , we expect that we did not extend these boundaries substantially beyond what is already established in the field . Given the depth of analysis in many structural studies , we therefore surmised that at least some of our fragments should have been noted previously . In fact , 40% ( 16 ) have been described individually before and connected to deep evolutionary events ( Figure 3 , indicated by a dot; Table 1 ) , showing that we are moving within the boundaries of established sequence-structure analyses . To our knowledge , our fragments comprise all but one previously reported case . We failed to detect the ASP-box ( Figure 3—figure supplement 1: B1 ) , found in six different folds ( Copley et al . , 2001 ) , owing to the stringency of our search criteria ( if the HHsearch probability cut-off is lowered to 60% , this fragment is included , along with four others ) . Examples of previously described fragments include the dinucleotide-binding β-α-β motif ( Buehner et al . , 1973; Rossmann et al . , 1974; Wierenga et al . , 1986; Dym and Eisenberg , 2001 ) ( Figure 3: 8 ) , the KH motif of type I and type II KH domains ( Grishin , 2001b ) ( Figure 3: 6 ) , the helix-hairpin-helix motif of nucleic acid-binding domains ( Doherty et al . , 1996; Shao and Grishin , 2000 ) ( Figure 3: 2 , Figure 4 ) , the EF-Tu-binding α-hairpin of elongation factor EF-Ts and ribosomal protein L7/12 ( Wieden et al . , 2001 ) ( Figure 3: 24 ) , and the P-loop of PEP carboxykinases and P-loop NTPases ( Walker et al . , 1982; Matte et al . , 1996 ) ( Figure 3: 16 ) . For many of these fragments , our systematic approach , the growth of structure databases , and improved sequence comparison methods allowed us to identify new instances of occurrence . For example , another occurrence of the P-loop is in the catalytic domain of MurD-like peptide ligases ( SCOP c . 72 . 2 ) , where it is involved in binding the phosphate group of a mononucleotide as well . If our starting assumption of ancestral peptides in the context of an RNA world is correct , some of the most basic functions would be nucleic-acid binding and catalysis . We would therefore expect these properties to be enriched in our set . Indeed , we find that about a third of our fragments make contact with nucleic acids ( 13 ) , based on evidence from known structures ( fragments with a yellow background in Figure 3; Table 1 ) . For comparison , the Smotifs reference set yielded only 2 nucleic-acid binders , even though it covers about three times as many superfamilies and folds as our fragment set ( see ‘Materials and methods’ ) . Our nucleic acid-binding fragments include four of the six most highly represented fragments in our dataset , particularly the helix-turn-helix motif ( Sauer et al . , 1982; Pabo and Sauer , 1984; Brennan and Matthews , 1989; Suzuki and Brenner , 1995; Aravind et al . , 2005 ) ( Figure 3: 1- found in 14 folds and 20 superfamilies , abbreviated in the following as 14 , 20 ) and the helix-hairpin-helix motif ( Figure 3: 2- 8 , 15; Figure 4 ) . A special case of nucleic-acid interaction is provided by ribosomal proteins , which contain 8 of our 13 nucleic acid-binding fragments ( indicated by red font color in Figure 3; Table 1 ) . Ribosomal proteins are likely to represent the oldest proteins observable today and , for the most part , still require the RNA scaffold to become structured , providing a window into the time when protein domains were being established ( Hsiao et al . , 2009 ) . For comparison , of the nine folds proposed to be the most ancient ( Caetano-Anolles et al . , 2007 ) , based on the comparative analysis of proteomes from diverse branches of life , six encompass at least one of our 40 fragments ( Table 1 ) . The second basic function peptides would plausibly have had in the RNA world would have been catalytic , as coordinators of metals , iron-sulfur clusters , nucleotides , and nucleotide-derived cofactors ( coenzyme A , NAD ( P ) , FAD ) , all of which are thought to already have played an essential role in prebiotic metabolism ( White , 1976; Wachtershauser , 1992 ) . Here , however , an enrichment relative to the Smotifs reference set is less clearly apparent , being primarily seen for iron-sulfur clusters which are absent in the reference set . Seven of our fragments coordinate metal ions and iron-sulfur clusters ( red background in Figure 3; Table 1 ) , e . g . , the cytochrome-heme-attachment motif ( Figure 3: 17 ) ( Mathews et al . , 1985 ) and the 4Fe-4S coordinating peptide ( Figure 3: 18 ) ( Lupas et al . , 2001 ) , and five bind nucleotides and nucleotide-derived cofactors ( blue background in Figure 3; Table 1 ) . Particularly two of these fragments , the P-loop motif ( Figure 3: 16 ) and the dinucleotide-binding β-α-β motif ( Figure 3: 8 ) , are at the core of some of the largest enzyme superfamilies in nature and play a central role throughout metabolic processes . We note that in all these fragments , the contribution of the peptides is the coordination of catalytic cofactors and not the provision of catalytic residues per se , in accordance with a primordial role as cofactors of RNA-driven catalysis . Thus , for example , the role of the P-loop in nucleotidases is to coordinate the nucleotide , the mechanism for its hydrolysis having evolved independently in different lineages of this superfamily . When we initiated this study , we believed that assembly from non-identical fragments may have been one of the primary forces in the evolution of domains ( Lupas et al . , 2001; Soding and Lupas , 2003 ) , and we expected to find many examples demonstrating it . However , we did not find even one domain that contained two or more different fragments from our set . Our fragments either form their folds by repetition or in single copy , decorated by heterologous structural elements . At present , we find the reasons for the lack of fragment combinations unclear , but we note that many of our fragments might represent dominant cores that guide the folding of the remainder of the polypeptide chain ( Religa et al . , 2007 ) and would , as such , not be generally compatible with each other . In the few cases where they would be sufficiently compatible to produce an initial fold capable of entering biological selection , they would be under considerable pressure to adapt to the new structural environment . This , in most cases , might lead to the retention of only one dominant fragment , the other ( s ) adjusting by rapid divergence , making them undetectable by our methods . For an initial exploration of this possibility , we asked whether any fold in SCOPe30 could be found to contain two or more of our fragments if we relaxed the sequence similarity requirement for all but one of them . We analyzed in detail the largest family from each superfamily containing one of our fragments ( 188 in all ) for such combinations of 'significant' and 'non-significant' fragments . As long as we retained any sequence similarity cutoff for the 'non-significant' fragments , even as low as HHsearch probabilities of 10% , no combinations could be found . If we however removed the sequence similarity requirement entirely for the 'non-significant' fragments , asking only for structural similarity ( TM-score ≥ 0 . 5 ) , over 50% of the families showed fragment combinations . In Figure 6 , we have collected examples , in which the fold is substantially formed by the combination of a 'significant' with a 'non-significant' fragment , showing that the incompatibility we observe for our fragments is not of a geometrical nature . 10 . 7554/eLife . 09410 . 011Figure 6 . Folds showing two nonidentical fragments , one of which is which is not significant by our criteria . No SCOP fold combines two of our fragments at the cutoffs used in this study ( TM-score ≥ 0 . 5 and HHsearch probability ≥ 70% ) . If we however omit the sequence cutoff entirely for the second fragment , combinations become apparent . In these three examples , the 'significant' fragments are colored in red , the 'nonsignificant' ones in blue , and the remainder in gray . The structures are: ( A ) a . 60 . 6 . 1 , N-terminal domain of polymerase β ( 4KLI , A: 10-91 ) , ( B ) a . 2 . 2 . 1 , 50S ribosomal protein L29 ( 1VQ8 , V: 1-65 ) , and ( C ) d . 51 . 1 . 1 , KH domain-like hypothetical protein APE0754 ( 1TUA , chain A: 1-84 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 09410 . 011 While we were unable to detect fragment combinations , repetition is wide-spread , seen for 14 of our 40 fragments ( 35% ) ( indicated by a dotted box in Figure 3; Table 1 ) . Nine of these are also capable of forming folds in single copy , with additional decoration . For example , the TPR element ( Figure 3: 28 ) is found in multiple copies in domains belonging to the TPR-like superfamily ( a . 118 . 8 ) ( D'Andrea and Regan , 2003 ) , but only in single copy in other folds that contain it ( Figure 7A ) . Also , the β-hammerhead motif ( Figure 1B , Figure 3: 12 ) , which resembles a hammerhead , is duplicated in the barrel-sandwich hybrid fold ( b . 84 ) , but occurs in single copy in the α/β-hammerhead fold ( d . 41 ) and in two other folds ( e . 29 and f . 46 ) . Two further fragments also occur in variable numbers per fold , but never in single copy . These are the outer membrane β-hairpin ( Figure 3: 25 ) ( Remmert et al . , 2010 ) , which forms β-barrels of between four and twelve hairpins ( Figure 7B ) , and the four-stranded β-meander ( Figure 3: 13 ) ( Chaudhuri et al . , 2008; Kopec and Lupas , 2013 ) , which forms propeller-like toroids containing between four and twelve copies . The remaining two fragments always form folds by duplication ( or homo-oligomerization ) . For example , the cradle-loop barrels , whose evolution we have studied in detail ( Coles et al . , 1999; 2005; 2006; 2007; 2008 ) , encompass a broad range of topological variants , which are however all built from two copies of a β-α-β fragment ( Figure 3: 7 , Figure 7C ) . 10 . 7554/eLife . 09410 . 012Figure 7 . Amplification and accretion are key forces in the emergence of domains . Of the 40 fragments in our set , 14 form folds by repetition . The fragments are coloured in red in the shown structures . ( A ) The TPR element ( Figure 3: 28 ) occurs repetitively in the TPR-like superfamily ( a . 118 . 8; 1ELW , shown on the left side ) and singly in six other folds ( e . g . , a . 7 . 16 , 2CRB , right ) . ( B ) Outer membrane β-barrels comprise 4–12 homologous copies of a β-hairpin element ( Figure 3: 25 ) ; examples include the eight-stranded OmpA ( 1QJP , left ) and the twelve-stranded NalP ( 1UYN , right ) . The entire barrels are formed by repetition , but the strands of the hairpin split by the N- and C- termini are left gray . ( C ) The transcription factor AbrB ( 1YFB , left ) is a homodimer and contains one copy of the β-α-β motif ( Figure 3: 7 ) per subunit . MraZ has internal sequence symmetry and contains two homologous copies of the β-α-β motif ( 1N0E , right ) . DOI: http://dx . doi . org/10 . 7554/eLife . 09410 . 012 Our findings thus indicate that repetition and accretion must have been the key forces in the emergence of domains . The importance of repetition has been pointed out by many earlier studies . Starting in the 1970's Andrew McLachlan charted out in more than 20 publications the origin of many proteins by repetition ( McLachlan , 1972; 1987 ) , in some cases clearly of subdomain-sized fragments ( Blundell et al . , 1979; McLachlan , 1980; 1980 ) . Aided by computational tools to detect repeats in sequence ( Heger and Holm , 2000; Szklarczyk and Heringa , 2004; Biegert and Soding , 2008 ) and structure ( Kim et al . , 2010; Myers-Turnbull et al . , 2014 ) , analyses of fold space have shown the high incidence of repetitive folds , amounting to as many as one fifth of all folds ( Andrade et al . , 2001; Balaji , 2015; Forrest , 2015 ) . These include some of the most frequent folds , such as ferredoxins ( Eck and Dayhoff , 1966 ) , immunoglobulins ( Huang and Xiao , 2007 ) , β-propellers ( Murzin , 1992 ) and TIM-barrels ( Soding et al . , 2006 ) ; indeed , of the 10 most populated folds in SCOPe , 6 ( including the top 5 ) , have repetitive structures and all of them have members in which the repetition is also detectable at the sequence level . In most cases , the repetitive structure of these folds has been interpreted as evidence for their origin by amplification of a subdomain-sized fragment and in some cases , such as for β-propellers ( Yadid et al . , 2010; Voet et al . , 2014 ) , β-trefoils ( Lee and Blaber , 2011; Broom et al . , 2012 ) , and TIM-barrels ( Bharat et al . , 2008; Richter et al . , 2010 ) , this process has also been explored by protein engineering . Only a subset of these fragments that form folds by repetition are present in our set , since we require them to occur in at least two folds to consider them antecedent to folded proteins , but many are currently only detectable in one fold . Our fragments are spread across a total of 130 folds and 188 superfamilies in SCOPe . Given that the current version of SCOPe comprises 1194 folds and 1961 superfamilies , they could be considered to cover only a small of fraction . However , of the 25 most populated folds in SCOPe , which comprise about 25% of all superfamilies , 14 contain one of the fragments and 7 of the top 10 do . We conclude that , far from being anecdotal , our fragments are indeed widespread in today’s domains . The 40 fragments we describe clearly represent a lower bound , given the stringent significance cut-offs in our study , the as yet incomplete fold assignment for known structures , and the fact that some ancestral fragments may have survived only in a single fold and are thus invisible to our method . Indeed , our goal in this study was not completeness , given the extensive overlap in statistical scores that can be expected between the least likely false negatives and the most likely true positives . Rather , it was to assemble a comprehensive set of confident positives in order to show that a signal of homology that predates folding as a widespread property is still detectable in proteins today . How many fragments may we be missing ? With respect to the significance cut-offs we find , for example , that relaxing the sequence similarity requirement to a probability of 60% leads to the inclusion of 5 further fragments , some with substantial arguments in their favor , such as the aforementioned ASP-box ( Copley et al . , 2001 ) ( Figure 3—figure supplement 1: B1 , Figure 3—source data 1 ) . Also , the availability of more sequences and structures , particularly for proteins from poorly sampled branches of life , should allow to bridge more widely separated folds by increasing the significance of their matches . For the fragments that occur in a single fold , we note that because of the importance of repetition , it might be possible to reconstruct further fragments by specifically analyzing repetitive folds . Certainly some of the chemical activities provided by these folds appear sufficiently ancient to justify the expectation that they were covered early in the genesis of folded proteins ( Farias-Rico et al . , 2014; Lupas et al . , 2015 ) . For these reasons we think that the number of ancestral fragments that could be reconstructed at a satisfactory level of confidence may well approach 100 in the next decades . Nevertheless , the set of 40 fragments we have described here should cover both the most frequent and best conserved ancestral peptides , and support a viable theory for the emergence of folded proteins .
To assemble domains representative of all known fold types , we chose the SCOPe database ( release 2 . 03 ) ( Fox et al . , 2014 ) and filtered it to a maximum of 30% sequence identity , obtaining 9452 domains . Multiple alignments were built for each of these domains using the buildali . pl script ( with default parameters ) from the HHsearch package ( Soding , 2005 ) . This script uses PSI-BLAST ( Altschul et al . , 1997 ) and contains heuristics to reduce the inclusion of nonhomologous sequence segments at the ends of PSI-BLAST sequence matches , the leading cause of high-scoring false positive matches . We used PSI-BLAST , rather than the more sensitive HHblits ( Remmert et al . , 2012 ) , because in our experience the sensitivity of HHblits leads it to occasionally assign elevated probabilities to analogous matches . Profile HMMs were calculated from the alignments using hhmake , also from the HHsearch package , and subjected to pairwise comparisons with HHsearch . Comparisons were thus always made with the full domains , never with fragments thereof . We used default settings , but switched off secondary structure scoring ( option ssm 0 ) in order to reduce the likelihood that matches were scored highly because of a chance similarity of their ( predicted ) secondary structures . The HHsearch probabilities we obtained are therefore conservative with respect to the ones obtained for example from the HHpred server in default settings ( Soding et al . , 2005 ) , which is the most frequent source of HHsearch-based deep homology analyses in publications today . We only considered reciprocal matches ( of the form domain A matches domain B and domain B matches domain A ) , and assigned these the average of the two obtained HHsearch probabilities . The structures of the aligned segments of SCOPe30 domains were subsequently compared using TM-align ( Zhang and Skolnick , 2005 ) . We filtered out all matches in which the aligned segment involved only a single secondary structure element . To establish cut-offs for the comparison of domains of different fold , we plotted comparisons of domains within families ( presumed homologs ) ( Figure 2A ) . The plot shows a bimodal distribution , with the highest representations at the top and bottom . At HHsearch probabilities of ≥ 70% and TM-scores of ≥ 0 . 5 , about a fourth of all homologous relationships in SCOPe30 are filtered out . We used these cut-offs for the comparison of domains of different fold , as it gives us a substantial margin of safety in the interpretation of relationships presumed analogous by SCOPe . We gathered all matches between domains of different fold with an HHsearch probability of ≥ 70% and a TM-score of ≥ 0 . 5 . Next , we eliminated all matches between larger superfamilies that nevertheless only made a single connection . This was done in order to minimize the number of potential false positives . We then applied single-linkage clustering to this resulting set of matches . Fragments from matches between folds were pooled together if they overlapped by at least 80% of their length , i . e . domains A , B , and C were combined together if domain A matched domain B and domain B matched domain C such that the boundaries in B overlapped by at least 80% . The resulting clusters were analyzed interactively and merged together if their similarity was based on the presence of a shared sequence- and structure-similar fragment . This yielded a final set of 64 potentially interesting clusters . We generated sequence and structure alignments for the fragments that formed the basis of these clusters and assigned boundaries by manual inspection . To obtain a reference set of fragments for our study , we assembled the 40 most frequent supersecondary structure motifs seen in proteins , according to the Smotifs database ( Fernandez-Fuentes et al . , 2010 ) . This is an exhaustive library of geometrically defined local supersecondary structure motifs , composed of two consecutive secondary structures connected by a loop . In this library , motifs of varying lengths , but with similar geometry and secondary structures , are grouped together into clusters , 2296 in total . The Smotifs library is available as a MySQL database; however , the frequency of occurrence of the motif clusters cannot be directly calculated from this database . We were therefore kindly provided with the frequency table by Andras Fiser ( Albert Einstein College of Medicine , New York ) . Using this table , we randomly selected one representative each for the 50 most frequent motif clusters; we required the representative to comprise 24 residues , the median length of our fragments , with at least 6 residues in each secondary structure element and a connecting loop of at most 6 residues . Next , we inspected these representatives and picked the 40 most frequent fragments after eliminating non-compact ones . Finally , for each of these 40 fragments , we picked a prototype in the SCOPe30 database by performing structure searches using TM-align and by selecting the match with the best TM-score . These 40 fragments formed our reference set . To evaluate whether the sequence similarity shown by our fragments could be the result of structural convergence , rather than origin from a common ancestor , we searched the SCOPe30 database with each fragment from our set and the Smotifs reference set using TM-align at a cut-off of 0 . 5 . Since our 40 fragments are structurally very similar in all their respective embodiments , we randomly picked one representative of each type . To calculate a sequence-versus-structure plot for our set and the Smotifs set , we followed the methodology described by us in previous studies ( Remmert et al . , 2010; Kopec and Lupas , 2013 ) . We used TM-align to compare the Smotifs set with domains from within the respective superfamily ( presumed homologous set ) and with the background set that comprised all other folds ( presumed analogous set ) to search for structurally similar fragments . For our fragments , we compared each representative fragment with domains of all other superfamilies in which we had detected it ( homologous ) and with all other folds ( analogous ) . In both cases , the TM-score was normalized based on the length of the query motifs . For each structure match , we calculated the profile-profile sequence similarity score with HHalign from the HHsearch package ( Soding , 2005 ) , however based on the fixed structural alignment obtained from TM-align . The HHalign score was normalized based on the number of aligned residues . Next , each pair of structure and sequence scores was plotted in a scatter plot . To calculate the correlation between TM- and HHalign-scores , we assumed a linear dependency between them and performed linear regression using SciPy ( Jones et al . , 2001 ) . We performed a t-test to determine whether the slope of the regression line differs significantly from zero; we chose a significance level of 1e-10 . First , we searched the SCOPe30 database for occurrences of each of the 40 reference Smotifs using TM-align at a TM-score cut-off of 0 . 5 and length coverage of 100% . This yielded occurrences in 323 folds and 447 superfamilies . For comparison , our 40 fragments occur in 130 folds and 188 superfamilies . Following this step , for all occurrences of our fragments and the Smotifs fragments in SCOPe30 , we searched for potential interactions with nucleic acids , metals , iron-sulfur clusters , nucleotides , and nucleotide-derived cofactors . A fragment was deemed to interact with these molecules if it made at least three inter-atomic contacts to them within a distance cut-off of 3Å . In addition , we included one of our fragments , the β-α-β motif seen in cradle-loop barrels ( Alva et al . , 2008 ) ( Figure 3: 7 ) into the nucleic acid binding set , even though only a high-resolution model of its interaction with DNA is currently available , based on NMR data ( Zorzini et al . , 2015 ) . Even discounting this addition , the 12 other nucleic acid-binding fragments of our set substantially exceed the two nucleic acid binders found in the Smotifs set , even though these cover about three times as many superfamilies and folds . | Life as we know it today is largely the result of the chemical activity of proteins . Much research suggests that the ancestors for most modern proteins were already present in the ‘Last Universal Common Ancestor’ , a theoretical ancient organism from which all life on earth descended and which lived around 3 . 5 billion years ago . Today , related versions of these ancestral proteins are found in organisms as different as bacteria , humans and plants . While they seem highly diverse , these proteins were all assembled from only a few thousand modular units , termed domains . However , it is not clear how the first domains emerged . Previously , in 2001 and 2003 , researchers hypothesized that the first protein domains arose by joining and swapping short lengths of proteins called peptides that had emerged before there were living cells on earth – a time that is often called the “RNA world” . Now , Alva et al . – including the researchers involved in the 2003 work – have attempted to detect remnants of these ancient peptides in modern proteins . Alva et al . first compared modern proteins in a way that is similar to how linguists have compared modern languages to reconstruct ancient vocabularies . This revealed 40 fragments that occur in seemingly unrelated proteins , but are very similar in their sequence and structure . These fragments are commonly found in what are likely the oldest observable proteins , and are involved in the activities that are most fundamental to life ( for example , binding to DNA and RNA ) . This led Alva et al . to propose that these fragments represent the observable remnants of a primordial “RNA-peptide world” . The hypothesis that proteins evolved from peptides provides a number of predictions that can be tested in experiments . These fragments open avenues to explore in the laboratory the origin of modern proteins and to build new proteins not seen in nature . | [
"Abstract",
"Introduction",
"Results",
"and",
"discussion",
"Materials",
"and",
"methods"
] | [
"biochemistry",
"and",
"chemical",
"biology",
"structural",
"biology",
"and",
"molecular",
"biophysics"
] | 2015 | A vocabulary of ancient peptides at the origin of folded proteins |
The Melanocortin Receptor Accessory Protein 2 ( MRAP2 ) is an important regulator of energy homeostasis and its loss causes severe obesity in rodents . MRAP2 mediates its action in part through the potentiation of the MC4R , however , it is clear that MRAP2 is expressed in tissues that do not express MC4R , and that the deletion of MRAP2 does not recapitulate the phenotype of Mc4r KO mice . Consequently , we hypothesized that other GPCRs involved in the control of energy homeostasis are likely to be regulated by MRAP2 . In this study we identified PKR1 as the first non-melanocortin GPCR to be regulated by MRAP2 . We show that MRAP2 significantly and specifically inhibits PKR1 signaling . We also demonstrate that PKR1 and MRAP2 co-localize in neurons and that Mrap2 KO mice are hypersensitive to PKR1 stimulation . This study not only identifies new partners of MRAP2 but also a new pathway through which MRAP2 regulates energy homeostasis .
G-Protein Coupled Receptors ( GPCRs ) are responsible for a wide variety of physiological functions , and their activity is precisely regulated , not only by their ligands ( agonists and in some cases antagonists ) but also by other factors including G-proteins , β-arrestins , kinases , lipids , ions and accessory proteins ( van der Westhuizen et al . , 2015 ) . GPCR accessory proteins are single-pass transmembrane proteins that regulate trafficking and/or signaling of the receptors to which they bind ( Cooray et al . , 2009 ) . A few examples include: the receptor activity modifying proteins ( RAMPs ) , which promote the trafficking and ligand specificity of the calcitonin receptor-like receptor/adrenomedullin receptor and several other GPCRs ( Parameswaran and Spielman , 2006; Sexton et al . , 2006 ) ; the receptor transporting proteins ( RTPs ) , which promote the trafficking of odorant and taste receptors ( Saito et al . , 2004 ) ; and the melanocortin receptor accessory proteins ( MRAPs ) , whose known roles prior to this study were limited to the regulation of trafficking and signaling of melanocortin receptors ( Metherell et al . , 2005; Sebag and Hinkle , 2007; 2009a; Sebag et al . , 2013; Cerdá-Reverter et al . , 2013; Asai et al . , 2013 ) . Two MRAPs exist in mammals . MRAP1 is expressed in few tissues like adipocytes ( Xu et al . , 2002 ) and the adrenal glands , where it is essential for proper trafficking and signaling of the melanocortin-2 receptor ( Metherell et al . , 2005; Sebag and Hinkle , 2007; Sebag and Hinkle , 2009a; 2009b ) . MRAP2 is likewise expressed in the adrenal glands , but also in the brain and other tissues . Unlike MRAP1 , the physiological functions and binding partners of MRAP2 are largely unknown . MRAPs display a remarkable topology: they are inserted in the plasma membrane in both the Nout/Cin and Nin/Cout orientation , with MRAPs of opposite orientation forming anti-parallel homodimers ( Sebag and Hinkle , 2007; 2009b ) . This structure is important for MRAP function ( Sebag and Hinkle , 2009b ) , and so far appears to be unique to these proteins in the eukaryotic proteome . MRAP2 is an important regulator of energy homeostasis and Mrap2 KO mice develop severe obesity ( Asai et al . , 2013 ) . The mechanisms through which MRAP2 regulates energy balance have not yet been fully identified , however , they include the potentiation of the melanocortin-4 receptor ( MC4R ) ( Sebag et al . , 2013; Asai et al . , 2013 ) , a protein central to the regulation of food intake and energy expenditure . Notably , like their Mrap2 KO counterparts , Mc4r KO mice are severely obese ( Butler and Cone , 2003 ) . There are however key differences between the obesity phenotypes of the two strains . In particular , the Mc4r KO mice are hyperphagic , have decreased energy expenditure and are insulin resistant ( Butler and Cone , 2002; 2003 ) , characteristics that are absent in the Mrap2 KO mice ( Asai et al . , 2013 ) . These phenotypic differences suggest that MC4R is not the only effector through which MRAP2 regulates the energy state , a conclusion consistent with the fact that MRAP2 is expressed in tissues that do not express MC4R ( Asai et al . , 2013 ) . Food intake is regulated by the activity of several GPCRs including the prokineticin receptor 1 ( PKR1 ) . Activation of PKR1 in vivo , through central or peripheral injection of its ligand prokineticin 2 ( PK2 ) , was shown to significantly decrease food intake ( Gardiner et al . , 2010; Beale et al . , 2013 ) . In addition to food intake , PKR1 plays important roles in the regulation of a variety of physiological functions including energy expenditure ( Zhou et al . , 2012 ) , insulin sensitivity ( Dormishian et al . , 2013 ) , gastrointestinal contraction ( Li et al . , 2001 ) , nociception ( Negri and Lattanzi , 2011 ) , cardiovascular function and angiogenesis ( Boulberdaa et al . , 2011; Urayama et al . , 2007 ) . Meanwhile , its orthologue PKR2 regulates placentation ( Hoffmann et al . , 2007 ) , inflammation ( Denison et al . , 2008 ) and nociception ( Negri and Lattanzi , 2011 ) . PKR1 and 2 couple to both the Gαs and Gαq proteins ( Ngan and Tam , 2008 ) , and consequently signal through the cAMP as well as the IP3/calcium pathways . Even though PKR1 and PKR2 appear to have some redundant physiological functions , it was shown that only PKR1 regulates food intake since injection of PK2 retains its full anorexigenic effect in PKR2 KO mice but does not decrease food intake in PKR1 KO mice ( Beale et al . , 2013 ) . In this study we identify PKR1 as the first non-melanocortin receptor to be regulated by MRAP2 and discover a novel mechanism of regulation of energy homeostasis by MRAP2 through the modulation of PKR1 signaling .
For PKR1 signaling to be regulated by MRAP2 in-vivo , the latter needs to be expressed along with the receptor . To determine what organs express both proteins , we performed RT-PCR on mRNA extracted from several mouse tissues . MRAP2 was readily detectable in the brain ( hypothalamus and pituitary gland ) , the adrenal glands , the lungs , the spleen and the kidneys , but also , at lower level , in the heart and the pancreas ( Figure 1A ) . Both PKR1 and PKR2 seem to be expressed in a large number of tissues including brain , heart , lungs , stomach , colon , kidneys , adrenals , fat and testis ( Figure 1A ) , thus confirming that MRAP2 and PKRs expression overlap in several organs . Because of the known involvement of both MRAP2 and PKR1 in the regulation of energy homeostasis , and the fact that both PKR1 and MRAP2 mRNA were detected in the hypothalamus , we tested if both proteins co-localized in hypothalamic neurons . In order to detect MRAP2 in brain slices we validated a commercial antibody by western blot ( Figure 1C ) and by immunofluorescence ( Figure 1D–E ) . The MRAP2 antibody was validated by western blot using lysates from CHO cells transfected with mouse MRAP2-V5 or empty vector as a control . Both the MRAP2-antibody and the V5-antibody detected the same bands and no signal was detectable in the lysate of mock transfected cells ( Figure 1C ) . We also validated the MRAP2 antibody for immunofluorescence using a GT1-1 hypothalamic neuronal cell line stably expressing GFP ( GT1-1-GFP ) as a control , or MRAP2 ( GT1-1-MRAP2 ) . We show that the MRAP2 antibody specifically labeled GT1-1-MRAP2 cells ( Figure 1G , H and E ) but not GT1-1-GFP cells ( Figure 1D , E and F ) , further validating the specificity of the antibody . Due to the fact that the cell lines used are not clonal , not all the cells show the same intensity of staining . Unfortunately , the MRAP2 antibody could not be validated on the Mrap2 KO mouse because of the way this mouse model was generated . Indeed , to produce this mouse , the cassette was inserted in the intron between exon 3 and 4 before the region coding for the transmembrane domain of MRAP2 , thus allowing the N-terminal region to be produced ( details can be found on the EUCOMM website ) . This region by itself is soluble and non-functional but can be recognized by the polyclonal MRAP2 antibody . Due to the lack of specific antibody to detect PKR1 in-vivo , we generated a mouse expressing a tagged PKR1 . In order for PKR1 expression level , localization and activity to remain unchanged in this mouse model , we used the CRISPR / Cas9 technology to insert 2XHA tag immediately after the start codon of the endogenous PKR1 gene in chromosome 6 ( Figure 1—figure supplement 1 and Figure 1M ) . The insertion of the HA tags was verified by sequencing ( Figure 1—figure supplement 1 ) . Using immunofluorescence method we show that MRAP2 is readily detectable in cells of the arcuate nucleus in both WT and 2HA-Prokr1 mice ( Figure 1J and N ) . In contrast , PKR1 was only detected , using the HA antibody , in the 2HA-Prokr1 mouse , therefore validating the specificity of PKR1 staining ( Figure 1K and O ) . Finally we show that PKR1 and MRAP2 colocalize in a large number of cells in the arcuate nucleus , making it physiologically relevant to study the regulation of PKR1 by MRAP2 ( Figure 1P and Q ) . 10 . 7554/eLife . 12397 . 003Figure 1 . Tissue distribution of PKRs and MRAP2 . ( A ) Expression of MRAP2 , PKR1 and PKR2 mRNA , measured by RT-PCR in tissues harvested from male mice . ( B ) DAPI-stained mouse brain section containing the hypothalamus . Square depicts the position of the arcuate nucleus . ( C ) Validation of anti-MRAP2 antibody by western blot on lysates from CHO cells transfected with empty vector ( EV ) or mouse MRAP2 ( M2 ) . ( D-I ) Validation of the MRAP2 antibody by immunofluorescence using GT1-1 cells stably expressing GFP ( D , E and F ) or MRAP2 ( G , H and E ) . ( J–L ) Confocal images of immunofluorescence detecting MRAP2 ( in green ) or HA ( in red ) in the arcuate nucleus of a wild type mouse . ( M ) Schematic representation of the insertion of 2XHA tags after the start codon of the PKR1 gene in mice using CRISPR/Cas9 technology . A more detailed description is depicted in Figure 1—figure supplement 1 . ( N-Q ) Confocal images of immunofluorescence detecting MRAP2 ( in green ) or HA-PKR1 ( in red ) in the arcuate nucleus of the 2HA-Prokr1 mouse . DOI: http://dx . doi . org/10 . 7554/eLife . 12397 . 00310 . 7554/eLife . 12397 . 004Figure 1—figure supplement 1 . Generation of the 2HA-Prokr1 mouse model . ( A ) Schematic representation of the CRISPR/Cas9 strategy targeting the Prokr1 gene for inserting the HA tags . ( B ) Genotyping of the 2HA- Prokr1 mice . The higher band represents the region around the start codon with the 2HA insertion . ( C ) Sequencing result of the Prokr1 gene with the 2XHA insert in the successfully targeted mouse . DOI: http://dx . doi . org/10 . 7554/eLife . 12397 . 004 To determine if MRAP2 is a regulatory protein of PKR1 , we first assessed the ability of those two proteins to form a complex . To this end , we transfected CHO cells with 2HA-PKR1 and MRAP2-3Flag and pulled down either PKR1 with a monoclonal anti-HA antibody or MRAP2 with a monoclonal anti-Flag antibody from the cell lysates . The precipitated proteins were then identified by western blot using an anti-HA antibody to detect the receptor and an anti-flag antibody to detect MRAP2 . We observed that a significant fraction of PKR1 co-immunoprecipitates with MRAP2 ( Figure 2A ) and that a large fraction of MRAP2 co-immunoprecipitates with PKR1 ( Figure 2B ) , thus demonstrating that PKR1 and MRAP2 interact . Like MRAP1 , MRAP2 is detected as two bands , one band representing the non-glycosylated form of MRAP2 and a higher band corresponding to the glycosylated form as confirmed by treatment with the deglycosylation enzyme , PNGase F ( Figure 2E ) . In some cases a higher molecular weight smear is detectable but the molecular identity of this band remains unclear . We verified that the bands corresponding to PKR1 in cells transfected with both PKR1 and MRAP2 matched the bands in cells transfected with PKR1 and empty vector ( Figure 2C ) , and that the bands corresponding to MRAP2 in cells transfected with both PKR1 and MRAP2 matched the bands in cells transfected with MRAP2 and empty vector ( Figure 2D ) in order to further confirm that the bands observed in the co-immunoprecipitation were in fact PKR1 and MRAP2 . We then determined whether this interaction happens in live cells and identified the sub-cellular localization of the complex . To this end we used a bi-molecular fluorescence complementation assay , or BiFC . PKR1 C-terminally fused to a fragment of YFP ( Y2 ) was co-expressed with MRAP2 C-terminally fused to the complementary fragment of YFP ( Y1 ) . Using this method fluorescence can only be achieved if PKR1 and MRAP2 come in close proximity and allow the YFP to complement ( Figure 2F ) . We found that , as predicted , expressing only one of the two fusion proteins did not yield any fluorescence ( Figure 2G and H ) , whereas YFP fluorescence was readily detectable in intracellular compartments as well as on the plasma membrane of cells expressing both fusion proteins ( Figure 2I ) . This result confirms that PKR1 and MRAP2 can interact in live cells . 10 . 7554/eLife . 12397 . 005Figure 2 . PKR1 and MRAP2 coimmunoprecipitation and localization in live cells . ( A ) and ( B ) Co-immunoprecipitation of 2HA-PKR1 and MRAP2-3Flag from transfected CHO cells . PKR1 was detected using rabbit anti-HA antibody ( A ) , and MRAP2 was detected using anti-Flag antibody ( B ) . X= beads only , no antibody was used for the IP . ( C ) and ( D ) Immunoprecipitation of PKR1 ( C ) or MRAP2 ( D ) from cells expressing PKR1 alone , MRAP2 alone or both . PKR1 was immunoprecipitated and detected with mouse anti-HA and MRAP2 was immunoprecipitated and detected with mouse anti-Flag . ( E ) Western blot detecting MRAP2 treated or not with PNGase F . ( F ) Schematic representation of bimolecular fluorescence complementation ( BiFC ) between YFP fragments fused to PKR1 and MRAP2 . ( G , H and E ) CHO cells were transfected with MRAP2-Y1 ( G ) , PKR1-Y2 ( H ) or both ( E ) Nuclei stained with Hoechst 33 , 342 are shown in blue and YFP fluorescence in green . DOI: http://dx . doi . org/10 . 7554/eLife . 12397 . 005 We then assessed the effect of MRAP2 on PKR1 signaling . Prokineticin receptors are coupled to both the Gαs and the Gαq signaling pathways . In the former they activate adenylyl cyclase , leading to the production of cAMP . In the latter they activate phospholipases , leading to the production of IP3 and to the release of calcium from the ER . In order to measure the effects of MRAP2 on PKRs signaling through the Gαs pathway , CHO cells were transfected with the CRE-luciferase reporter and either PKR1 or PKR2 in the presence or absence of MRAP2 . We find that MRAP2 dose-dependently inhibits PKR1 efficacy and decreases its potency four to five fold with a maximum effect at a receptor to MRAP2 DNA ratio between 1:7 and 1:10 ( Figure 3A ) . For this reason all subsequent experiments were performed at a 1:10 ratio . To further validate the inhibitory effect of MRAP2 on PKR1 , we tested the effect of both human and mouse MRAP2 on cAMP production stimulated by both agonists ( PK1 and PK2 ) . Both MRAP2 isoforms inhibited PKR1 responses stimulated by either PK1 ( Figure 3B ) or PK2 ( Figure 3C ) and in both cases mouse MRAP2 was a more effective inhibitor of PKR1 than human MRAP2 . In the rest of the study we only use the human MRAP2 since the receptors used are the human isoforms . To rule out the possibility that MRAP2 interfered with the luciferase reporter rather than inhibiting PKR1-mediated cAMP production , we stimulated CHO cells expressing PKR1 with empty vector at a 1:10 ratio or with MRAP2 at the same ratio with increasing concentration of PK2 in the presence of the phosphodiesterase inhibitor IBMX . Accumulated cAMP was measured using the Perkin Elmer LANCE cAMP assay . This assay directly measures cAMP concentration using a TR-FRET technology and does not involve luminescence . MRAP2 inhibited PK2 responses similarly in the two assays ( Figure 3D ) , confirming its inhibitory role in PKR1-mediated cAMP signaling . 10 . 7554/eLife . 12397 . 006Figure 3 . MRAP2 inhibits PKR1 signaling . CHO cells in a 10 cm dish were transfected with 2 . 5 μg CRE-luciferase , 225 ng PKR1 and a total of 2 . 25 μg of plasmids encoding MRAP2 and/or empty vector in the ratios shown . After 24 hr cells were subcultured for use as follows: ( A ) Cells expressing PKR1 and different amounts of hMRAP2 and/or empty vector were stimulated with PK2 and cAMP responses measured with the CRE-luciferase cAMP assay . ( B and C ) Cells expressing PKR1 with empty vector , mouse MRAP2 ( mMRAP2 ) or human MRAP2 ( hMRAP2 ) at a 1:10 ratio were stimulated with ( B ) PK1 or ( C ) PK2 and cAMP responses measured with CRE-luciferase . ( D ) . Cells expressing PKR1 with empty vector or hMRAP2 were stimulated with PK2 in the presence of 0 . 1 mM isobutylmethylxanthine and cAMP concentrations measured with the LANCE cAMP assay . ( E and F ) Cells expressing PKR1 with empty vector or mMRAP2 were stimulated with ( E ) PK1 or ( F ) PK2 in the presence of LiCl and IP3 production measured with the IP-One assay . ( G ) Surface or ( H ) total expression of PKR1 in cells transfected with empty vector , 2HA-PKR1 and empty vector , or 2HA-PKR1 and hMRAP2 using cell ELISA assays . One-way ANOVA with Tukey post test *p<0 . 05 , **p<0 . 01 , ***p<0 . 001DOI: http://dx . doi . org/10 . 7554/eLife . 12397 . 006 Because PKR1 can also couple to Gαq , we tested the effect of MRAP2 on PKR1-mediated production of IP3 . For those experiments , we used CHO-M1 cells , which stably express the Gαq-coupled muscarinic M1 receptor; this allowed us to use the response to the M1R agonist carbachol to normalize the results and correct for possible differences in growth rates in the different transfection conditions . The assay used measures the accumulation of IP1 , a degradation product of IP3 , in the presence of lithium , an inhibitor of the inositol monophosphatase . We found that , as with the cAMP pathway , MRAP2 significantly inhibits PKR1-mediated IP3 production in response to both PK1 ( Figure 3E ) and PK2 ( Figure 3F ) . This result suggests that MRAP2 can inhibit both known signaling pathways downstream of PKR1 . Since several GPCR accessory proteins have been shown to modulate GPCR trafficking ( Cooray et al . , 2009; Sexton et al . , 2006; Metherell et al . , 2005; Sebag and Hinkle , 2007; Matsunami et al . , 2009 ) , we tested the impact of MRAP2 on PKR1 trafficking to the plasma membrane . To this end we measured both the surface density ( in non-permeabilized cells ) and total expression ( in permeabilized cells ) of 2HA-PKR1 when expressed with empty vector or with MRAP2 by fixed-cell ELISA . We found that MRAP2 decreased PKR1 expression at the cell surface by approximately 35% ( Figure 3G ) . Analysis of total expression revealed that the level of PKR1 is reduced in the presence of MRAP2 by 27% ( Figure 3H ) . The fact that the decreases in surface and total expression of PKR1 in the presence of MRAP2 are similar suggests that MRAP2 does not significantly impact PKR1 trafficking . We also investigated the ability of MRAP2 to regulate PKR2 , a GPCR closely related to PKR1 . We transfected CHO cells with 2HA- PKR2 and MRAP2-3Flag and performed a co-immunoprecipitation as described earlier . A significant fraction of PKR2 was pulled down with MRAP2 ( Figure 4A ) and conversely , a significant fraction of MRAP2 co-immunoprecipitated with PKR2 ( Figure 4B ) . As for PKR1 , we confirmed that the bands observed for PKR2 and MRAP2 in the co-immunoprecipitation match the bands obtained with cells expressing either PKR2 ( Figure 4C ) or MRAP2 ( Figure 4D ) alone . The interaction of PKR2 with MRAP2 was also confirmed by BiFC in cells expressing PKR2 fused to a fragment of YFP and MRAP2 fused to the complementary fragment of YFP . As expected , no YFP fluorescence was detected when only one of the two fusion proteins was expressed ( Figure 2G and Figure 4F ) , however , YFP fluorescence was readily detectable when both fusion proteins were expressed ( Figure 4F ) , especially in intracellular compartments , thus confirming that PKR2 and MRAP2 can form complexes in live cells . We then measured the effect of MRAP2 on PKR2 signaling in response to PK2 through both the cAMP and the IP3 pathway . MRAP2 significantly inhibited PKR2-mediated cAMP and IP3 signaling ( Figure 4G and H ) . It is , however , noticeable that MRAP2 inhibited the cAMP and IP3 pathways to the same extent ( about 50% ) with PKR1 ( Figure 3B , C , E and F ) , whereas MRAP2 inhibited the cAMP pathway more effectively than the IP3 pathway downstream of PKR2 ( about 70% for cAMP vs . about 20% for IP3 ) ( Figure 4G and H ) . Unlike for PKR1 , MRAP2 decreased PKR2 efficacy but did not significantly affect its potency . We also measured the effect of MRAP2 on PKR2 trafficking and showed that , unlike with PKR1 , MRAP2 decreased the surface expression of PKR2 by over 70% ( Figure 4I ) with only 20% to 30% change in total receptor expression ( Figure 4J ) , suggesting a possible effect of MRAP2 on PKR2 trafficking . In fact , in the case of PKR2 , the extent of the trafficking inhibition is comparable to the inhibition in cAMP signaling . 10 . 7554/eLife . 12397 . 007Figure 4 . Specificity of MRAP2 regulation . ( A and B ) Co-immunoprecipitation of 2HA-PKR2 and MRAP2-3Flag from transfected CHO cells . PKR2 was detected using rabbit anti-HA antibody ( A ) , and MRAP2 was detected using mouse anti-Flag antibody ( B ) . X= beads only , no antibody was used for the IP . ( C and D ) Immunoprecipitation of PKR2 ( C ) or MRAP2 ( D ) from cells expressing PKR2 alone , MRAP2 alone or both . PKR2 was immunoprecipitated and detected using mouse anti-HA antibody . MRAP2 was immunoprecipitated and detected with mouse anti-Flag ) . ( E and F ) CHO cells were transfected with PKR2-Y2 ( E ) or with PKR2-Y2 and MRAP2-Y1 ( F ) . Nuclei stained with Hoechst 33 , 342 are shown in blue and YFP fluorescence in green . ( G and H ) cAMP production stimulated by ( G ) PK1 or ( H ) PK2 in cells expressing PKR2 with empty vector or hMRAP2 . ( I ) Surface and ( J ) total expression of PKR2 in cells transfected with empty vector , 2HA-PKR2 alone or 2HA-PKR2 and MRAP2 using cell ELISA assays . ( K ) PK2-stimulatedcAMP in cells expressing PKR1 with empty vector or hMRAP1α . ( L ) Isoproterenol-stimulatedcAMP in cells expressing the β2-ADR with empty vector or MRAP2 . cAMP responses were measured with the CRE-luciferase assay . One-way ANOVA with Tukey post test *p<0 . 05 , **p<0 . 01 , ***p<0 . 001DOI: http://dx . doi . org/10 . 7554/eLife . 12397 . 007 It is clear that MRAP2 can regulate multiple GPCRs , including PKRs , MC4R ( Sebag et al . , 2013; Asai et al . , 2013 ) and MC5R ( Sebag and Hinkle , 2009a; Chan et al . , 2009 ) . Nonetheless , MRAP proteins are highly selective for their targets since MRAP1 did not modulate PKR1 signaling ( Figure 4K ) and MRAP2 did not affect β2-adrenergic receptor signaling ( Figure 4L ) . We previously reported that MRAP2 does not modify signaling by several other GPCRs involved in the regulation of energy homeostasis including the Y2 receptor , Y1 receptor , glucagon-like peptide 1 receptor and the MC3R ( Sebag et al . , 2013 ) . To further validate our findings , we assessed the regulatory action of MRAP2 on PKR1 and PKR2 signaling ( Figure 5A and B ) and trafficking ( Figure 5C and D ) in a more relevant cell line , the hypothalamic neuronal GT1-1 cell line , and found very similar results to those obtained in CHO cells . 10 . 7554/eLife . 12397 . 008Figure 5 . Regulation of PKR1 and PKR2 signaling and trafficking in GT1-1 cells . ( A ) GT1-1cells expressing CRE-Luciferase , 2HA-PKR1 and either empty vector or hMRAP2 were stimulated with PK2 and cAMP responses measured with the CRE-luciferase cAMP assay . ( B ) Cells expressing CRE-Luciferase , 2HA-PKR2 and either empty vector or were stimulated with PK2 and cAMP responses measured with CRE-luciferase . ( C and D ) ELISA measuring the surface expression of PKR1 ( C ) or PKR2 ( D ) in GT1-1 cells in the presence or absence of MRAP2 . One-way ANOVA with Tukey post test *p<0 . 05 , **p<0 . 01 , ***p<0 . 001DOI: http://dx . doi . org/10 . 7554/eLife . 12397 . 008 We then investigated if MRAP2 regulates PKR1 activity in-vivo in mice . Male and female WT and Mrap2 KO sibling mice were cannulated in the lateral ventricle of the brain . After recovering from the surgery , mice were fasted overnight and injected ICV with vehicle or the indicated dose of PK2 10 min prior to being given access to food . Food intake was then measured at 30 min , 1 , 2 , 4 and 6 hr . As expected the highest dose of PK2 significantly inhibited food intake in both WT and Mrap2 KO mice , however , the anorexigenic effect was more pronounced in Mrap2 KO mice ( Figure 6 ) . This is especially evident at the lowest PK2 dose injected since this dose has no significant effect in both male and female WT mice but inhibits up to 70% of food intake in Mrap2 KO mice ( Figure 6 ) . The effect of PK2 seemed to be more potent and longer lasting in female compared to male Mrap2 KO mice . These results confirm that MRAP2 significantly inhibits the anorexigenic signal mediated by PKR1 in vivo . Both in published studies and in our hands , PK2 injections decreased food intake without causing malaise or sickness like behavior in the animals ( Gardiner et al . , 2010; Beale et al . , 2013 ) . 10 . 7554/eLife . 12397 . 009Figure 6 . MRAP2 mediated regulation of PKR1 anorexigenic effect . Cumulative food intake in male ( A ) and female ( B ) WT and Mrap2 KO mice after overnight fast and ICV injection of saline or the indicated dose of PK2 . One-way ANOVA with Tukey post test *p<0 . 05 , **p<0 . 01 , ***p<0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 12397 . 009 Because MRAP2 has been shown to regulate the MC4R ( Sebag et al . , 2013 ) and that it had been suggested that the anorexigenic effect of PKR1 may be working upstream of the MC4R ( Gardiner et al . , 2010 ) , we tested if MC4R signaling was required for PKR1 stimulation to decrease food intake . To this end we injected overnight fasted WT and Mc4r KO mice ICV with vehicle or 0 . 65 μg PK2 before giving them access to food . We show that , like for WT mice , PK2 caused a very significant decrease in food intake over 6 hr in both male and female Mc4r KO mice ( Figure 7 ) , demonstrating that PKR1 can inhibit food intake independently of the MC4R ( Figure 6 ) . Like observed in the previous experiment PK2 seems to be more efficacious in female mice . The difference in food intake between Mc4r KO and WT mice injected with saline may be attributed to the documented enhanced stress-induced anorexia in Mc4r KO mice ( Liu et al . , 2007; Vergoni et al . , 1999 ) . Additionally , a more potent anorexigenic effect of PK2 was measured in the Mc4r KO colony compared to the Mrap2 KO colony , possibly due to slightly different genetic backgrounds . 10 . 7554/eLife . 12397 . 010Figure 7 . The Anorexigenic effect of PKR1 does not require MC4R . Cumulative food intake in male ( A ) and female ( B ) WT and Mc4r KO mice after overnight fast and ICV injection of saline or the indicated dose of PK2 . ( C and D ) Same results as those depicted in A and B but normalized to the food intake of the 'Vehicle' injected mice of the same genotype at the 6 hr time point . T-test *p<0 . 05 , **p<0 . 01 , ***p<0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 12397 . 010
MRAP proteins are known to interact with all five melanocortin receptors and to regulate the trafficking and/or signaling of MC2R , MC4R and MC5R ( Metherell et al . , 2005; Sebag and Hinkle , 2007; 2009a; Sebag et al . , 2013; Cerdá-Reverter et al . , 2013; Asai et al . , 2013 ) . In this study we identify a non-melanocortin GPCR regulated by a member of the MRAP protein family , i . e . the prokineticin receptors . This finding not only demonstrates the importance of considering the expression of MRAP2 when studying PKRs , but also establishes that MRAPs , especially MRAP2 due to its broader tissue distribution , can regulate various GPCRs and are involved in numerous physiological functions . We also show that , unlike all the other GPCR accessory proteins ( RAMPs , REEP and RTPs ) , which to date have been shown only to promote GPCR trafficking or signaling , MRAPs can either potentiate or inhibit the GPCRs they regulate . For example , whereas MRAP1 is required for MC2R activity , it prevents MC5R trafficking and homodimerization ( Sebag and Hinkle , 2009a ) . In zebrafish , the two isoforms of MRAP2 , zMRAP2a and zMRAP2b , have opposite effects on MC4R signaling ( Sebag et al . , 2013 ) and here we demonstrate that mammalian MRAP2 , which had previously been shown to potentiate MC4R signaling ( Asai et al . , 2013 ) , strongly inhibits PKR1 and PKR2 signaling . GPCR accessory proteins have so far been described as potentiators of GPCR trafficking and/or signaling ( Ritter and Hall , 2009 ) with the exception of MRAPs . Indeed , to our knowledge , the ability to selectively potentiate a subset of GPCRs while inhibiting another is unique to MRAP proteins . As such , it will be exciting to pursue the study of the mechanisms underlying MRAP2 targeting of specific GPCRs and the directionality of the resulting regulation . The finding that MRAP2 inhibits rather than activates PKRs was unexpected since both PKRs and MRAP2 are thought to promote leanness . In fact , knocking out PKR1 ( Szatkowski et al . , 2013 ) or MRAP2 ( Asai et al . , 2013 ) in mice causes obesity , and central administration of the PKR1 agonist PK2 leads to a significant decrease in food intake ( Gardiner et al . , 2010; Beale et al . , 2013 ) and Figures 6 and 7 . It is , however , important to note that Asai et al . showed that MC4R / MRAP2 double KO mice have an intermediate obesity phenotype between the Mrap2 KO and the Mc4r KO mice ( Asai et al . , 2013 ) . In other words , losing the function of both MRAP2 and MC4R causes a milder obesity than losing the function of MC4R alone ( Asai et al . , 2013 ) , suggesting that MRAP2 can promote weight gain through a mechanism that is distinct and independent from the regulation of the MC4R . Further arguing this hypothesis , we showed that the anorexigenic effect of PKR1 is independent of MC4R since central injection of PK2 in fasted Mc4r KO mice significantly decreased food intake . Consequently , we conclude that MRAP2-mediated inhibition of PKR1 is an important mechanism through which MRAP2 promotes hunger , energy intake and weight gain and that MRAP2 is a major component of the machinery that controls energy homeostasis . The cause of the more potent anorexigenic effect of PK2 in female compared to male Mrap2 KO mice remains unclear and will require further studies . It is also important to note that PKR1 and PKR2 are expressed in many tissues and play roles in numerous physiological processes for which negative regulation by MRAP2 could be critical . Increasingly , GPCRs are found to be regulated by accessory proteins , thus strongly suggesting that the importance of this mode of receptor regulation is underestimated . Moreover , our study indicates that accessory proteins not only potentiate , but can also inhibit receptor signaling . For these reasons , identifying the accessory proteins that are associated with specific GPCRs is an important step in accurately determining their pharmacology . Additionally , a better understanding of GPCR regulation by accessory proteins will allow the development of more accurate and more relevant model systems for the identification of small molecules targeting receptors and will increase the likelihood of such compounds to retain their activity in vivo . Finally , this study illustrates the important role of the interaction of PKR1 and MRAP2 in the regulation of energy homeostasis and suggests that this complex may be a valuable new target for the treatment of obesity .
Mrap2 KO mice are on a C57BL/6 background and were obtained from the EUCOMM program . LoxTB Mc4r KO mice were a generous gift from Dr . Lutter , University of Iowa . Both male and female mice were used and all experiments using mice were approved by the animal care and use committee at the University of Iowa . The 2HA-Prokr1 mouse was generated by the genome editing core facility of the University of Iowa using pronuclear injection of a guide RNA targeting the first exon of the Prokr1 gene , Cas9 and a repair DNA containing the sequence of 2XHA tags immediately after the start codon . Mice were screened for the insertion by PCR and validated by sequencing . The Prokr1 and Prokr2 plasmids , encoding PKR1 and PKR2 respectively , were obtained from the Missouri S&T cDNA Resource Center . The N-terminal 2HA tag was added to Prokr1 and Prokr2 by PCR using the Prokr1 and Prokr2 plasmids as templates and the tagged receptor gene inserts were cloned into pcDNA5 . hMrap2 plasmid was a generous gift from Dr . Patricia Hinkle , University of Rochester . Plasmids encoding the split-YFP were a kind gift from Dr . Catherine Berlot , Weis Center for Research , Geisinger Clinic , Danville , PA . Mrap2 , Prokr1 and Prokr2 were amplified and inserted , in frame , 5’ of the YFP fragments . The plasmid encoding CRE-luc was a kind gift from Dr . George Holz , New York University School of Medicine , New York , NY . The validity of all constructs was verified by sequencing . The antibodies used in this study were mouse monoclonal anti-HA ( HA11 ) ( Biolegend , San Diego , CA ) , rabbit anti-HA ( cell signaling , Danvers , MA ) , mouse monoclonal anti-V5 ( AbDSerotec , Raleigh , NC ) , M2 anti-Flag ( Sigma-Aldrich , St . Louis , MO ) , horseradish peroxidase ( HRP ) -conjugated antibodies against mouse and rabbit immunoglobulin ( Biorad , Hercules , CA ) and anti-MRAP2 rabbit polyclonal antibody ( Novus Biological , Littleton , CO ) . HEK293T and CHO-K1 were purchased from ATCC , CHO-M1 cells ( stably expressing the M1 muscarinic receptor ) were obtained from Molecular Devices , GT1-1 cells were kindly provided by Dr . Richard Weiner , University of California San Francisco School of Medicine . All cell lines are mycoplasma free . HEK293T , CHO-K1 , CHO-M1 and GT1-1 cells were cultured in Dulbecco’s Modified Eagle’s Medium ( DMEM ) /F-12 supplemented with 5% fetal bovine serum and 1% penicillin-streptomycin , in a humidified atmosphere consisting of 5% CO2 at 37°C . For transfection , cells were grown to 70% confluency . Transfections were performed using LipoD293 in vitro transfection reagent ( Signagen , Rockville , MD ) for all cells except GT1-1 cell that were transfected with Lipofectamine 3000 . Total plasmid concentration was kept identical for all transfections by the addition of empty vector . HEK293T cells transfected with the indicated plasmids were lysed in 0 . 1% n-dodecyl-β-maltoside in PBS with protease inhibitors . Lysates were centrifuged and supernatants were incubated with the indicated antibody ( mouse anti-HA or mouse anti-Flag M2 ) at 1/5000 dilution overnight at 4°C . Immune complexes were collected with protein-G Dynabeads ( Life Technologies , Carlsbad , CA ) at 4°C for 1 hr . Beads were washed three times and resuspended in LDS loading buffer with 5% β-mercaptoethanol and boiled 5 min . Proteins were resolved by SDS/PAGE and detected by western blot using rabbit anti-HA for PKR1 and PKR2 coimmunoprecipitation experiments , mouse anti-HA or anti-Flag for all other experiments . For MRAP2 deglycosylation , samples were incubated with PNGaseF for 1h at 37°C after the immunoprecipitation step . Tissues were harvested from mice and mRNA was extracted using Trizol . Reverse transcription was carried out using 1 μg of mRNA per sample using the iScript kit ( Biorad , Hercules , CA ) . Prokr1 , Prokr2 , Mrap2 and Gapdh were then amplified using the prepared cDNA as template and the following primers: mProkr1 Forward: 5’ CAC CAA CTT GCT TAT CGC CAA CC 3’; mProkr1 Reverse: 5’ GGC CAG ATC TGA CCA CAG AAG AT 3’; mProkr2 Forward: 5’ TGG CCA TCT CTG ACT TCC TGG T 3’; mProkr2 reverse:: 5’ TAG GAT TTG TAG TAG AGC TGC TGG T 3’; mMrap2 Forward: 5’TGT AAA GCC TGC GGT AAC CC 3’; mMrap2 Reverse: 5’ AGG ACT CCG CGT TGT CTT G 3’; Gapdh Forward: 5’ GGA GAG TGT TTG CTC GTC CC 3’ and Gapdh Reverse: 5’ ACT GTG CCG TTG AAT TTG CC 3’ . CHO-K1 cells were plated in 10 cm dishes . The next day , they were transiently transfected with a plasmid encoding the CRE-luciferase reporter ( firefly luciferase driven by 4 repeats of a non-palindromic cAMP responsive element ) , along with PKR1 or PKR2 and either empty vector or MRAP2 ( 1:10 ratio receptor to MRAP2 ) . 24 hr following transfection , the cells were transferred to a white 96-well plate and left to adhere overnight . Cells were then incubated with vehicle or varying concentrations of peptide agonist ( PK1 or PK2 ) or 20 μM forskolin , in DMEM/F-12 supplemented with 0 . 1% bovine serum albumin ( BSA ) for 4 hr at 37°C . The medium was removed and 100 μl of luciferin in lysis buffer , 200 mM Tris-HCl , 10 mM MgCl2 , 300 uM ATP , 1% Igepal , protease inhibitor cocktail ( Roche , Basel , Switzerland ) , 12 . 2 mM Acetyl Coenzyme A; 30 μg/ml luciferin ( Goldbio , Olivette , MO ) , was added . Samples were incubated at room temperature for 5 min and luminescence was then measured using a Spectramax I3 plate reader ( Molecular Devices , Sunnyvale , CA ) . Because CRE drives the expression of luciferase , the level of expression of the enzyme is proportional to the cAMP produced in the cell and luminescence serves as a reporter of cAMP production . In addition to the CRE-luciferase reporter , cells were transfected with a constant amount of PKR1 and increasing concentrations of MRAP2 plus enough empty vector to maintain a constant total DNA concentration . Cells were then treated with vehicle or with the indicated concentration of PK2 prior to measuring the cAMP production . In order to correct for possible variations in transfection efficiencies , the results were normalized to the signal obtained in response to 20 μM forskolin in each condition . The results were then normalized to the highest signal obtained in cells expressing PKR1 and CRE-luc in the absence of MRAP2 . CHO cells in 10 cm dishes were transfected with indicated plasmids . The next day , cells were plated in a white 96 well plate and allowed to adhere overnight . Cells were then incubated with agonist at the indicated concentration in the presence of 0 . 1 mM 3-isobutyl-1-methylxanthine for 30 min at 37°C . Cells were then processed using the LANCE cAMP assay kit ( PerkinElmer , Waltham , MA ) following manufacturer’s manual . Plate was read with a Spectramax i3 plate reader equipped with a HTRF ( homogeneous time resolved fluorescence ) module . Cells were grown in a 24-well plate and transiently transfected with empty vector ( mock ) , with 2HA-PKR1 or 2HA-PKR2 and empty vector , or with 2HA-PKR1 or 2HA-PKR2 and MRAP2 . 24 hr after transfection , the cells were washed with PBS , fixed for 10 min with 4% paraformaldehyde in PBS , washed , blocked in either 5% milk in PBS ( surface-expression assay ) or 5% milk in RIPA buffer ( 150 mM NaCl , 50 mM Tris , 1 mM EDTA , 1% Triton X-100 , 0 . 1% SDS , 0 . 5% sodium deoxycholate , pH 8 . 0 ) ( total-expression assay ) . Cells were then incubated with 1/5000 anti-HA antibody in blocking buffer for 2 hr at RT , washed three times for 5 min in PBS , incubated with 1/5000 anti-mouse-HRP antibody for 1 hr RT , and washed three times with PBS for 5 min . ELISA substrate ( 3 , 3′ , 5 , 5′-Tetramethylbenzidine , Sigma-Aldrich ) was then added until blue color was visible , and the reaction was stopped with 10% sulfuric acid . Absorbance was measured at 450 nm using a Spectramax I3 plate reader . CHO-K1 cells were seeded on sterilized coverslips and transfected with MRAP2-Y1 , PKR1-Y2 or PKR2-Y2 plus empty vector , or PKR1-Y2 or PKR2-Y2 plus MRAP2-Y1 . The next day , cells were incubated at room temperature for 1 hr and nuclei were stained with 1 μg/mL Hoechst 33 , 342 ( Life Technologies ) for 15 min . Coverslips were then mounted in imaging chambers , and imaging was performed using a 60X objective with an Olympus IX83 inverted fluorescence microscope . Inositol phosphate assay was performed using the IP-One-Tb HTRF kit ( Cisbio Bioassays , France ) following the manufacturer’s instructions . CHO-M1 cells were transfected with PKR1 or PKR2 plus either empty vector or MRAP2 at a ratio of 1:10 ( receptor to MRAP2 ) . The day before the experiment , the transfected cells were plated in a white 384-well plate at a density of 20 , 000 cells/well . The next day the cells were treated with different concentrations of PK1 , PK2 or 1 μM carbachol in stimulation buffer containing lithium for 1 hr at 37°C , and then incubated with IP1-d2 conjugate and Anti-IP1 cryptate Tb in lysis buffer for 1 hr at room temperature . HTRF signal was read at 615 nm and 665 nm using a Spectramax I3 plate reader equipped with the HTRF Cisbio cassette . For each condition , signal obtained with 1 μM carbachol , an agonist of the M1 muscarinic receptor , was used for normalization . Results were then normalized to the highest signal from the control ( receptor + empty vector ) . 2HA-Prokr1 mice and wild type littermates were deeply anesthetized with isoflurane and perfused transcardially with PBS and subsequently with 4% PFA in PBS . The whole brains were dissected out and post-fixed in 4% PFA overnight at 4°C . The brains were put in 30% sucrose in PBS until sunk . 40 μm frozen brain sections were cut by a cryostat and treated with 0 . 4% Triton X-100 in PBS for 1 hr , subsequently incubated in 3% hydrogen peroxide in PBS for 30 min and rinsed two times with PBS before incubation in 10 mM sodium citrate at 80°C for 30 min . The sections were blocked in 5% normal goat serum in 0 . 4% Triton X-100 in PBS and then in Mouse on mouse ( MOM ) Ig block reagent at room temperature for 1 hr according to the manufacturing protocol ( Vector Laboratories , Burlingame , CA ) , rinsed with PBS twice , further incubated in MOM diluent for 5 min and subsequently incubated in 1:300 mouse anti-HA antibody in MOM diluent for 30 min . The free floating sections were washed twice with PBS and incubated in 1:250 biotinylated anti mouse IgG antibody for 10 min , washed with PBS three times and incubated in 1:1000 MRAP2 rabbit antibody in 0 . 4% Triton X-100 in PBS supplemented with 5% normal goat serum overnight at 4°C . The sections were washed with 0 . 4% Triton X-100 in PBS three times and incubated in 1:250 Pierce high sensitivity streptavidin-HRP ( Cat . No . 21130 , Thermo Scientific ) mixed with 1:250 Alex Fluor 546 goat anti rabbit antibody for 1 hr in 5% normal goat serum , 0 . 4% triton X-100 in PBS . The sections were washed four times and incubated in 1:100 Alexa Fluor 647 tyramide in amplification buffer for 10 min according to the manufacturer protocol ( Cat . No . T20926 , molecular probes life technologies ) and washed with 0 . 4% Triton X-100 in PBS four times . The sections were mounted on a glass slide and air-dried . Prolong diamond antifade mountant with DAPI was applied on the sections before images were acquired with a Leica SP8 STED confocal microscope . For immunofluorescence in cells , GT1-1-GFP and GT1-1-MRAP2 cells were plated on sterile coverslips , fixed in 4% PFA , blocked in 5% goat serum in PBS supplemented with 0 . 3% Triton X-100 for 1 hr , then incubated with anti-MRAP2 antibody at 1/1000 dilution in blocking buffer for 2 hr at room temperature , washed with PBS , incubated with anti-rabbit-alexa 546 sary antibody for 1 hr , washed with PBS . The coverslips were then mounted on slides with prolong diamond with dapi and imaged . Individually housed male and female MRAP2 knockout mice and sibling WT or loxTB Mc4r KO and sibling WT at age 6–7 week-old were anesthetized with 17 . 5 mg/ml ketamine/ 2 . 5 mg/ml xylazine mix at dose 0 . 1 ml per 20 g body weight . A stainless steel guide cannula ( Plastics One , USA ) was stereotaxically placed into the lateral ventricle ( 1 mm lateral , 0 . 3 mm posterior and 2 . 5 mm ventral from bregma ) . Perioperatively , mice were administered subcutaneously 0 . 1 mg/kg buprenorphine . Animals were kept in a temperature and humidity controlled room under a 12 hr light/dark cycle and allowed to recover for seven days after surgery and were handled for five days before the start of the experiment . Animals were fasted overnight and injected ICV with vehicle , 0 . 25 μg or 0 . 65 μg PK2 in 3 µl saline 10 min before food was returned to the cages ( only vehicle or 0 . 65 μg PK2 were used for the Mc4r KO experiments ) . Food in each cage was weighed at 30 min , 1 , 2 , 4 and 6 hr following re-feeding . A minimum of three days drug-free period was maintained between infusions . During this period , the animals were handled but not tested . The mice that had received PK2 at the first injection received saline at the second and the mice that received saline the first time received PK2 . The cannula placement was verified using histological methods . Animals were euthanized with carbon dioxide and received an ICV injection with 2 µl of ink . Brains were removed and the lateral ventricles opened to check for ink staining . The minimal number of number of animals to be used was determined by power analysis . All experiments were repeated separately a minimum of three times . Statistics were calculated as follow . For ELISA histograms , one-way ANOVA with Tukey post-test was used and significance was measured between the control ( no MRAP2 ) and the test ( with MRAP2 ) . For food intake studies in WT and Mrap2 KO mice , statistics were measured using a one way ANOVA within the result for the different doses of PK2 at a single time point and for one genotype and one gender . Lines connect the results compared for which the statistical significance is noted . For food intake studies in WT and Mc4r KO mice , statistics were measured a T-test between the result for vehicle or PK2 injected within the same gender and same time point , asterisk were color coded for clarity . *p<0 . 05 , **p<0 . 01 and ***p<0 . 001 . Results are shown as mean ± SEM . | The brain plays a major role in controlling how much food animals eat . The nerve cells ( neurons ) involved in this process contain “receptors” that respond to cues from various parts of the body . For example , a receptor called PKR1 acts to limit food intake . The activities of PKR1 and other receptors are tightly regulated in cells , but it is not clear how this works . A protein called MRAP2 is known to regulate the activity of a receptor that regulates food intake and energy use in the brain . However , MRAP2 may also interact with other receptors to control food intake . Here , Chaly , Srisai et al . investigated whether MRAP2 can regulate the activity of PKR1 in animal cells and rodents . The experiments show that MRAP2 can interact with and inhibit the activity of PKR1 . Furthermore , both MRAP2 and PKR1 can be found in the same neurons . Mutant mice that lack the gene that encodes MRAP2 have higher levels of PKR1 activity and eat less than normal mice when PKR1 is stimulated . Together the experiments suggest that MRAP2 can increase food intake by preventing PKR1 from being activated in the brain . The next steps are to find out if this protein regulates other receptors involved in the control of food intake , and to test whether PKR1 and MRAP2 also play a role in regulating energy usage . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"cell",
"biology",
"neuroscience"
] | 2016 | The Melanocortin Receptor Accessory Protein 2 promotes food intake through inhibition of the Prokineticin Receptor-1 |
Context is information linked to a situation that can guide behavior . In the brain , context is encoded by sensory processing and can later be retrieved from memory . How context is communicated within the cortical network in sensory and mnemonic forms is unknown due to the lack of methods for high-resolution , brain-wide neuronal recording and analysis . Here , we report the comprehensive architecture of a cortical network for context processing . Using hemisphere-wide , high-density electrocorticography , we measured large-scale neuronal activity from monkeys observing videos of agents interacting in situations with different contexts . We extracted five context-related network structures including a bottom-up network during encoding and , seconds later , cue-dependent retrieval of the same network with the opposite top-down connectivity . These findings show that context is represented in the cortical network as distributed communication structures with dynamic information flows . This study provides a general methodology for recording and analyzing cortical network neuronal communication during cognition .
Context is the contingent sensory or cognitive background for a given situation . Different contexts can dramatically alter perception , cognition , or emotional reactions and decision-making , and in the brain network context can be represented during sensory encoding or mnemonic retrieval . The study of context is important for understanding the link between perception and cognition , in terms of both behavioral and neural processing , and the neural mechanisms underlying contextual information processing have been studied in a variety of domains including visual perception ( Bar , 2004; Schwartz et al . , 2007 ) , emotion ( De Gelder , 2006; Barrett et al . , 2007; Barrett and Kensinger , 2010 ) , language ( Hagoort , 2005; Aravena et al . , 2010 ) , and social cognition ( Ibañez and Manes , 2012 ) . In the brain , context is proposed to require an interplay between bottom-up and top-down information processing in distributed neural networks ( Tononi and Edelman , 1997; Friston , 2005 ) . However , a comprehensive functional view of the brain circuits that mediate contextual processing remains unknown because bottom-up and top-down processes are often concurrent and interdependent , making the temporal and spatial resolution of their neural network organization difficult to separate . To understand contextual information processing , we developed a fundamentally new approach to study high-resolution brain network architecture . The approach combines broadband neural recording of brain activity at high spatial and temporal resolution with big data analytical techniques to enable the computational extraction of latent structure in functional network dynamics . We employed this novel methodological pipeline to identify functional network structures underlying fast , internal , concurrent , and interdependent cognitive processes during context processing in monkeys watching video clips with sequentially staged contextual scenarios . Each scenario contained a conspecific showing emotional responses preceded by different situational contexts . With specific combinations of context and response stimuli , this paradigm allowed an examination of context-dependent brain activity and behavior by isolating context processing as a single variable in the task . To measure large-scale brain network dynamics with sufficient resolution , we used a 128-channel hemisphere-wide high-density electrocorticography ( HD-ECoG ) array to quantify neuronal interactions with high spatial , spectral , and temporal resolution . This ECoG system has wider spatial coverage than conventional ECoG and LFP ( Buschman and Miller , 2007; Pesaran et al . , 2008; Haegens et al . , 2011 ) and single–unit activity ( Gregoriou et al . , 2009 ) , higher spatial resolution than MEG ( Gross et al . , 2004; Siegel et al . , 2008 ) , broader bandwidth than EEG ( Hipp et al . , 2011 ) , and superior temporal resolution to fMRI ( Rees and House , 2005; Freeman et al . , 2011 ) . After recording , we interrogated large-scale functional network dynamics using a multivariate effective connectivity analysis to quantify information content and directional flow within the brain network ( Blinowska , 2011; Chao and Fujii , 2013 ) followed by big data analytical approaches to search the database of broadband neuronal connectivity for a latent organization of network communication structures .
Monkeys watched video clips of another monkey ( video monkey , or vM ) engaging with a second agent ( Figure 1 ) while cortical activity was recorded with a 128-channel ECoG array covering nearly an entire cerebral hemisphere . Three monkeys participated , one with a right hemisphere array ( Subject 1 ) , and two in the right ( Subjects 2 and 3 ) ( Figure 1—figure supplement 1 ) . The data are fully accessible online and can be downloaded from the website Neurotycho . org . 10 . 7554/eLife . 06121 . 003Figure 1 . Subjects observe situational contexts with high-density electrocorticography ( HD-ECoG ) recording . We recorded 128-channel HD-ECoG signals from monkeys viewing video clips of a conspecific under three different situational contexts and two responses . The subject ( lower-left , green circles represent ECoG electrodes ) was seated in front of a TV monitor showing video clips consisting of a Waiting period of 2 . 5 s followed by a Context period of 1 . 5 s with one of three interactions between a video monkey ( vM ) on the left and a second agent on the right: vM threatened by a human ( Ch ) , threatened by another monkey ( Cm ) , or an empty wall ( Cw ) . Next , a curtain closed to conceal the second agent followed by a Response period of 3 s with the vM showing either a frightened ( Rf ) , or neutral expression ( Rn ) . Pairwise combination of the contexts and responses produced six different video clips . DOI: http://dx . doi . org/10 . 7554/eLife . 06121 . 00310 . 7554/eLife . 06121 . 004Figure 1—figure supplement 1 . Electrode locations in 3 subjects . In Subject 1 , electrodes ( green dots ) were placed to cover most of the lateral surface of the right hemisphere , also the medial parts of the frontal and occipital lobes . In Subject 2 , a similar layout was used , but in the left hemisphere . In Subjects 3 , all electrodes were placed on the lateral surface of the left hemisphere , and no medial parts were covered . For brain map registration , the electrode locations and the brain outlines from Subjects 1 and 3 were manually registered to those from Subject 2 based on 13 markers ( red circles ) in the lateral hemisphere and five markers in the medial hemisphere . DOI: http://dx . doi . org/10 . 7554/eLife . 06121 . 004 The video clips started with a context between the two agents ( Context period ) followed by a response to the context ( Response period ) . Six different video clips were created from three contexts , vM threatened by a human ( Ch ) , threatened by another monkey ( Cm ) , or facing an empty wall ( Cw ) , combined with two responses , vM showing a frightened expression ( Rf ) , or neutral expression ( Rn ) , which were termed ChRf , CmRf , CwRf , ChRn , CmRn , and CwRn ( see Videos 1–6 ) . Each video contained audio associated with the event , for example , sounds of a threatening human ( Ch ) and a frightened monkey ( Rf ) . Each video represents a unique social context-response scenario , For example , ChRf shows a human threatening a monkey ( vM ) followed by the monkey's frightened response . These staged presentations were designed to examine whether different contexts ( Ch , Cm , or Cw ) would give rise to context-dependent brain activity even with the same responses ( Rf or Rn ) . During the task , subjects freely moved their eyes to observe the video interactions . We monitored eye movements to examine these spontaneous behavioral reactions and the associated zones in the video . We divided the trials into two conditions based on whether the context stimulus was visually perceived: C+ where the subject was looking at the screen during the Context period , and C− where the subject was either closing its eyes or looking outside of the screen . Example eye movements are shown in Figure 2—figure supplement 1 . We first investigated which side of the video monitor the monkey attended . When the context was perceived ( C+ ) and the response stimulus was Rf , subjects focused more on the right section during the Response period , indicating interest in the curtain , or the threat behind the curtain , than the frightened vM ( Figure 2A ) . This preference was absent when the response stimulus was Rn or the context was not visually perceived ( C− ) . This is behavioral evidence that gaze direction preference required not only the vM response , but also perception of the preceding context , which demonstrated a cognitive association between the perception of the context and response stimuli . 10 . 7554/eLife . 06121 . 011Figure 2 . Context- and response-dependent eye movements . ( A ) Measurements of gaze shifting revealed a behavioral association between the context and response phases of the task . The gaze positions averaged from three subjects are shown for each trial type ( ChRf , CmRf , CwRf , ChRn , CmRn , and CwRn ) and condition ( C+ and C− ) . Gaze shifting was quantified by gaze positions significantly different from baseline values ( αBonf = 0 . 05 , baseline: gray bar ) , and was found only in Rf trials under the C+ condition ( upper-left panel , the timing of gaze shifts are indicated on top , where the color represents the trial type indicated on the right ) . Black vertical lines represent the following events ( see labels on the x-axis ) : ( a ) onset of the Context period , ( b ) the curtain starts closing , ( c ) the curtain is fully closed , ( d ) onset of the Response period , and ( e ) end of the Response period ( onset of the next trial ) . ( B ) Context and response dependence in gazing behavior . Gaze positions between different trial types were compared , separately in C+ and C− . For each comparison ( y-axis ) , the timing of significant differences are shown as circles ( αBonf = 0 . 05 ) , where blue , green , and red circles represent context dependence in Rf , context dependence in Rn , and response dependence , respectively . Gazing behavior showed both response dependence and context dependence , but only in C+ . DOI: http://dx . doi . org/10 . 7554/eLife . 06121 . 01110 . 7554/eLife . 06121 . 012Figure 2—figure supplement 1 . Examples of eye movement . Examples of eye position during CmRf scenario under C+ condition from Subject 2 . Sampled eye positions ( circles ) in the three periods ( Waiting , Context , and Response ) are shown , where the corresponding timings are indicated by the colorbar . Snapshots of the videos during the corresponding periods are also shown . DOI: http://dx . doi . org/10 . 7554/eLife . 06121 . 012 We then compared gazing behaviors from different trial types to identify behaviors selective to different scenarios ( ChRf , CmRf , CwRf , ChRn , CmRn , or CwRn ) and conditions ( C+ or C− ) . We performed nine pairwise comparisons on gaze positions from different scenarios , separating C+ and C− conditions , to examine their context and response dependence . For context dependence , we compared behaviors from trials with different context stimuli but the same response stimulus ( 6 comparisons: CmRf vs CwRf , CwRf vs ChRf , and ChRf vs CmRf for context dependence in Rf; CmRn vs CwRn , CwRn vs ChRn , and ChRn vs CmRn for context dependence in Rn ) . For response dependence , we compared behaviors from trials with the same context stimulus but with different response stimuli ( 3 comparisons: CmRf vs CmRn , CwRf vs CwRn , and ChRf vs ChRn ) . A context and response dependence was found in gazing behavior ( Figure 2B ) . In C+ , significant differences in gaze position were found during the Response period between CmRf and CwRf , and CwRf and ChRf , but not between ChRf and CmRf ( blue circles in left panel ) . This indicated that gaze shifting in CmRf and ChRf was comparable and stronger than in CwRf . This context dependence was absent when the response stimuli were Rn ( green circles in left panel ) . Furthermore , a significant response dependence was found during the Response period for all contexts ( CmRf vs CmRn , CwRf vs CwRn , and ChRf vs ChRn ) ( red circles in left panel ) consistent with the results described in Figure 2A . In C− , the context and response dependence found in C+ was absent ( right panel ) . These results indicated that the subjects' gaze shift during the Response period showed both response dependence ( Rf > Rn ) and context dependence ( Cm ≈ Ch > Cw ) , but only when context was perceived ( C+ > C− ) . To analyze the large-scale ECoG dataset , we identified cortical areas over the 128 electrodes in the array by independent component analysis ( ICA ) . Each independent component ( IC ) represented a cortical area with statistically independent source signals ( Figure 3—figure supplement 1 , and experimental parameters in Table 1 ) . 10 . 7554/eLife . 06121 . 013Table 1 . Experimental parametersDOI: http://dx . doi . org/10 . 7554/eLife . 06121 . 013Subject 1Subject 2Subject 3ExperimentHemisphere implantedRightLeftLeft# of electrodes128128128# of trials per class150150150# of trials preserved per class ( mean ± std ) ( see trial screening in ‘Materials and methods’ ) 117 . 7 ± 3 . 5122 . 2 ± 3 . 1109 . 5 ± 3 . 6# of C+ trials per class ( mean ± std ) 64 . 8 ± 5 . 260 . 3 ± 6 . 757 . 3 ± 5 . 6# of C− trials per class ( mean ± std ) 52 . 8 ± 1 . 961 . 8 ± 8 . 152 . 2 ± 3 . 5ICA ( see Figure 3—figure supplement 1 ) # of ICs for 90% variance explained583839# of ICs preserved ( see IC screening in ‘Materials and methods’ ) 49 ( removed ICs 1 , 2 , 3 , 4 , 5 , 11 , 44 , 46 , and 47 ) 33 ( removed ICs 1 , 2 , 7 , 8 , and 29 ) 36 ( removed ICs 2 , 10 , and 27 ) We then measured the causality of a connection from one cortical area ( source area ) to another ( sink area ) with a multivariate effective connectivity measure based on Granger causality: direct directed transfer function ( dDTF ) ( Korzeniewska et al . , 2003 ) , which can represent phase differences between the two source signals to provide a time-frequency representation of their asymmetric causal dependence . We acquired dDTFs from all connections for each trial type ( 12 types: six scenarios and two conditions ) , and measured event-related causality ( ERC ) , by normalizing the dDTF of each time point and each frequency bin to the median of the corresponding baseline control values . Thus , ERCs represent the spectro-temporal dynamics of network interactions evoked by different scenarios and conditions . Examples of ERCs are shown in Figure 3A . 10 . 7554/eLife . 06121 . 014Figure 3 . Identification of latent structures in context- and response-dependent cortical network interactions . ( A ) Event-related causalities ( ERCs ) between cortical areas . Example ERCs for a connection ( IC 8 to IC 14 , the corresponding cortical areas shown on the top ) in two scenarios ( CmRf and CwRf in C+ ) from Subject 1 are shown . Each ERC represents the spectro-temporal dynamics of causality evoked by a scenario , calculated as the logarithmic ratio between the direct directed transfer function ( dDTF ) and corresponding baseline values ( baseline: gray bar ) , and measured in decibel ( dB ) . Black vertical lines represent task events explained in Figure 2 . ( B ) ∆ERCs , or the significant differences in ERCs between the two trial types ( CmRf − CwRf ) ( αFDR = 0 . 05 , false discovery rate correction ) are shown . The results were either 0 ( no significant difference ) , +1 ( significantly greater ) , or −1 ( significantly weaker ) . ( C ) 3D tensor of ∆ERCs . The data for the entire study were organized in three dimensions: dynamics ( top ) , function ( middle ) , and anatomy ( bottom ) . Top: ∆ERCs shown in B describe the dynamics of difference in causality of a connection between two trial types , presented as a vector in 3D space ( illustrated as a bar , where each segment represents a ∆ERC value ) . Middle: For the same connection , ∆ERCs from other comparisons were pooled to describe the functional dynamics of the connection ( illustrated as a plate ) . Bottom: Functional dynamics from all connections were pooled to summarize the functional network dynamics in a subject ( illustrated as a block ) . The data from all subjects were further combined to assess common functional network dynamics across subjects . ( D ) Parallel factor analysis ( PARAFAC ) extracted five dominant structures from the 3D tensor with consistency ( >80% , also see Figure 3—figure supplement 2 ) . Each structure represented a unique pattern of network function , dynamics , and anatomy ( e . g . , Func . 1 , Dyn . 1 , and Anat . 1 for Structure 1 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 06121 . 01410 . 7554/eLife . 06121 . 015Figure 3—figure supplement 1 . Independent component analysis ( ICA ) results from 3 subjects . The spatial distribution of each IC and its time course are shown for each subject . For each IC , the size of each circle represents the relative contribution of the activity from the electrode to the IC , where red and blue colors represent the positive and negative contributions , respectively . For clarity , the time courses shown were obtained by averaging source signals of the IC over one trial type ( CmRf trials under C+ condition ) , and the y-axis is not shown . For each time course , three red vertical lines represent the events ( a ) , ( d ) and ( e ) described in Figure 2 . The ICs that were removed from analysis are labeled ( see Table 1 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 06121 . 01510 . 7554/eLife . 06121 . 016Figure 3—figure supplement 2 . PARAFAC revealed five dominant structures in the 3D tensor . A core consistency diagnostic was used to evaluate how well the tensor can be represented by different numbers of structures . During deconvolving the tensor into different numbers of structures , core consistencies were measured by two methods: DTLD/GRAM and random values ( see the ‘Materials and methods’ ) . For the random values method , 100 core consistency values were measured and their means and standard deviations are shown . In both methods , a sharp decrease in consistency was found when the number of structures increased from 5 to 6 , indicating that five structures yielded the optimal fit . DOI: http://dx . doi . org/10 . 7554/eLife . 06121 . 01610 . 7554/eLife . 06121 . 017Figure 3—figure supplement 3 . Five latent network structures were robust against ICA model order selection . ( A ) PARAFAC results ( with DTLD/GRAM method ) from data obtained from ICA with 90% ( ICA90% , as in Figure 2D ) , 80% ( ICA80% ) , and 75% ( ICA75% ) of total variance preserved . In all cases , five latent network structures yelled the optimal fits . ( B ) Similarities among structures obtained from different ICA results . Top row: We first compared Structure i to Structure j obtained from ICA90% ( 90% vs 90% ) . Correlations were evaluated in four different domains: Comparison ( the first tensor dimension ) , Time and Frequency ( the second tensor dimension ) , and Causal outflow ( the third tensor dimension ) . The significant correlations ( α = 0 . 05 ) are indicated as asterisks , and the correlations with high correlation coefficients ( Pearson , ρ > 0 . 8 ) are indicated as circles . Second row: Correlations between Structure i obtained from ICA80% and Structure j obtained from ICA90% ( 80% vs 90% ) . The structures obtained from ICA80% were reordered so that the correlations in the diagonals were maximized . Bottom row: Correlations between Structure i obtained from ICA75% and Structure j obtained from ICA90% ( 75% vs 90% ) . High correlations were found in the diagonals in 80% vs 90% and 75% vs 90% , and high similarities were found among 90% vs 90% , 80% vs 90% , and 75% vs 90% . These results indicate that the five latent network structures were similar under different selections of ICA model order . DOI: http://dx . doi . org/10 . 7554/eLife . 06121 . 017 We compared ERCs from different trial types to identify networks selectively activated in different scenarios ( ChRf , CmRf , CwRf , ChRn , CmRn , or CwRn ) and conditions ( C+ or C− ) . We performed nine pairwise comparisons on ERCs from different scenarios , separating C+ and C− trials , to examine their context and response dependence . To examine context dependence , we compared ERCs from trials with different contexts but the same response ( 6 comparisons ) . In contrast , to examine response dependence , we compared ERCs from trials with the same context but with different responses ( 3 comparisons ) . This approach is similar to the eye movement analysis ( Figure 2B ) . The comparisons were performed with a subtractive approach to derive a significant difference in ERCs ( ∆ERCs , Figure 3B ) . Hence , ∆ERCs revealed network connections , with corresponding time and frequency , where ERCs were significantly stronger or weaker in one scenario compared to another . We pooled ∆ERCs from all comparisons , conditions , connections , and subjects , to create a comprehensive broadband library of network dynamics for the entire study . To organize and visualize the dataset , we created a tensor with three dimensions: Comparison-Condition , Time-Frequency , and Connection-Subject , for the functional , dynamic , and anatomical aspects of the data , respectively ( Figure 2C ) . The dimensionality of the tensor was 18 ( nine comparisons under two conditions ) by 3040 ( 160 time windows and 19 frequency bins ) by 4668 ( 49 × 48 connections for Subject 1 , 33 × 32 for Subject 2 , and 36 × 35 for Subject 3 ) . To extract structured information from this high-volume dataset , we deconvolved the 3D tensor into multiple components by performing parallel factor analysis ( PARAFAC ) , a generalization of principal component analysis ( PCA ) to higher order arrays ( Harshman and Lundy , 1994 ) and measured the consistency of deconvolution under different iterations of PARAFAC ( Bro and Kiers , 2003 ) . Remarkably , we observed five dominant structures from the pooled ∆ERCs that represented functional network dynamics , where each structure contained a comprehensive fingerprint of network function , dynamics , and anatomy ( Figure 3D , and Figure 3—figure supplement 2 ) . These five structures were robust against model order selection for ICA ( Figure 3—figure supplement 3 ) . The five structures are shown in Figure 4 ( Structures 1 and 2 ) and Figure 5 ( Structures 3 , 4 , and 5 ) . Each structure represented a unique functional network dynamics , described by its compositions in the three tensor dimensions . The first tensor dimension ( panel A ) represented the differences across comparisons for each structure . We identified the significant differences and reconstructed the activation levels to show how each structure was activated under different scenarios and conditions ( see the ‘Materials and methods’ ) . The second tensor dimension ( panel B ) represented spectro-temporal dynamics for each structure . The third tensor dimension ( panel C ) represented the anatomical connectivity pattern for each structure . We measured three connectivity statistics: ( 1 ) causal density is the sum of all outgoing and incoming causality for each area , showing areas with busy interactions; ( 2 ) causal outflow is the net outgoing causality of each area , indicating the sources and sinks of interactions; and ( 3 ) maximum flow between areas is the maximal causality of all connections between cortical areas ( 7 areas found with busy interactions were chosen ) ( see results for individual subjects in Figure 4—figure supplements 1–3 ) . The extracted statistics were robust across all subjects with different electrode placements suggesting that the structures were bilaterally symmetric across hemispheres ( Figure 4—figure supplement 4 ) . 10 . 7554/eLife . 06121 . 018Figure 4 . Network structures for perception of context and response . Each structure was defined by three dimensions: function , dynamics , and anatomy . ( A ) Function: The function dimension showed each structure's context and response dependence . Top: For each structure , the first tensor dimension contained 18 differences for nine pairwise comparisons in the C+ or C− condition . Significant differences are highlighted ( * , α = 0 . 05 , see the ‘Materials and methods’ ) . Bottom: The comparisons with significant differences were used to reconstruct how each structure was selectively activated . Each oval and its vertical position represent the trial type and its activation level , respectively . Blue , green , or red arrows indicate significant context dependence under Rf , significant context dependence under Rn , and significant response dependence , respectively ( each corresponds to a significance highlighted in the top panel ) . ( B ) Dynamics: The dynamics dimension indexed each structure's activation in different times and frequencies . Black vertical lines represent events , as explained in Figure 2 . ( C ) Anatomy: The anatomy dimension showed each structure's activation in different connections . Three connectivity statistics , averaged across subjects after brain map registration , are shown on the lateral and medial cortices . Top: Cortical areas with greater causal density represent areas with busier interactions . Middle: Cortical areas with positive ( red ) and negative ( blue ) causal outflows represent the sources and sinks of interactions , respectively . Bottom: The direction and strength of each maximum flow between areas are indicated by the direction and size ( and color ) of an arrow , respectively . Seven cortical areas were determined for visualization: the visual ( V ) , parietal ( P ) , prefrontal ( PF ) , medial prefrontal ( mPF ) , motor ( M ) , anterior temporal ( aT ) , and posterior temporal ( pT ) cortices . DOI: http://dx . doi . org/10 . 7554/eLife . 06121 . 01810 . 7554/eLife . 06121 . 019Figure 4—figure supplement 1 . Causal density in individual subjects . Cortical areas with high causal density ( busy traffic ) are indicated as red . Results from individual subjects and the summarized results after brain map registration are shown for each latent network structure . DOI: http://dx . doi . org/10 . 7554/eLife . 06121 . 01910 . 7554/eLife . 06121 . 020Figure 4—figure supplement 2 . Causal outflow in individual subjects . Cortical areas with overall positive causal outflow ( source areas ) are indicated shown in red , and those with overall negative causal outflow ( sink areas ) are shown in blue . Results from individual subjects and the summarized results after brain map registration are shown for each latent network structure . DOI: http://dx . doi . org/10 . 7554/eLife . 06121 . 02010 . 7554/eLife . 06121 . 021Figure 4—figure supplement 3 . Maximum flow between areas in individual subjects . ( A ) The maximum information flows between cortical areas in all subjects . For each subject ( column ) and each structure ( row ) , the maximal loadings of connections from each of the seven cortical areas ( source , y-axis ) to the others ( sink , x-axis ) are shown . Seven cortical areas were manually identified: the visual cortex ( V ) , the parietal cortex ( P ) , the posterior temporal cortex ( pT ) , the anterior temporal cortex ( aT ) , the motor cortex ( M ) , the prefrontal cortex ( PFC ) , and the medial PFC ( mPF ) . The seven areas identified in all subjects are shown on the registered brain map on the right . ( B ) The average maximum information flow between areas across subjects presented in two different formats: matrices ( left ) and arrows ( right ) . DOI: http://dx . doi . org/10 . 7554/eLife . 06121 . 02110 . 7554/eLife . 06121 . 022Figure 4—figure supplement 4 . Robust connectivity across subjects . For each connectivity statistics ( causal density , causal outflow , or maximum flow between areas ) from each structure , we compared its values ( after brain map registration ) between each subject pair . As results , 15 correlation coefficients were acquired for each connectivity statistics to show how similar the connectivity statistics across subjects in all structures . The average correlation coefficients for causal density , causal outflow , and maximum flow between areas were 0 . 66 ± 0 . 08 , 0 . 60 ± 0 . 05 , and 0 . 69 ± 0 . 26 , respectively ( mean ± std , n = 15: 3 pairs , 5 structures ) . High correlations in connectivity statistics among subjects indicate that connectivity in each structure is robust across subjects . DOI: http://dx . doi . org/10 . 7554/eLife . 06121 . 02210 . 7554/eLife . 06121 . 023Figure 5 . Network structures for context representation and modulation . The function ( A ) , dynamics ( B ) , and anatomy ( C ) dimensions of Structures 3 , 4 , and 5 . Structures 3 and 4 represent the initial formation/encoding and later reactivation/retrieval of abstract context information , respectively , and Structure 5 represents context-dependent top-down feedback that modulates eye gaze or visual attention . Same presentation details as in Figure 4 . DOI: http://dx . doi . org/10 . 7554/eLife . 06121 . 02310 . 7554/eLife . 06121 . 024Figure 5—figure supplement 1 . Spatial and spectral characteristics of network structures . ( A ) The correlations between the causal outflows between structures . For each structure , causal outflows were measured for all ICs in all subjects , which resulted in a 1 by 118 vector ( = 49 + 33 + 36 , see the numbers of ICs in Table 1 ) . The first two principal components ( PC 1 and PC 2 ) of the causal outflows of the five structures are shown in the inset . The correlations between the causal outflows were also measured . The significant correlations ( α = 0 . 05 ) are indicated as asterisks , and the correlations with high correlation coefficients ( Pearson , ρ > 0 . 8 ) are indicated as circles . ( B ) The correlations between the frequency profiles of different structures . The frequency profile of each structure , shown in the inset , was quantified by averaging the corresponding loadings in the second tensor dimension ( Time-Frequency ) across time points , which resulted in a 1 by 19 vector . The correlations between the frequency profiles were then measured and shown . DOI: http://dx . doi . org/10 . 7554/eLife . 06121 . 024 Structure 1 was activated first , with context dependence only in the Context period ( Cm > Ch > Cw ) suggesting sensory processing that can discriminate between contextual stimuli , that is , context perception . The context dependence was weaker but remained in C− ( C+ > C− ) , suggesting that auditory information in context stimuli was processed . The spectral dynamics of Structure 1 emerged primarily in the high-γ band ( >70 Hz ) , and contained mostly bottom-up connections from the posterior to anterior parts of temporal cortex . Structure 2 was the earliest activated in the Response period , with only response dependence ( Rf > Rn ) , and was independent of whether context stimuli were visually perceived ( C+ ≈ C− ) . Thus , Structure 2 corresponds to sensory processing that can discriminate between response stimuli , that is , response perception . Spectrally , Structure 2 emerged in both high-γ and β bands , and contained connections similar to those in Structure 1 , with an additional communication channel from anterior temporal cortex to the prefrontal cortex ( PFC ) . Therefore , Structures 1 and 2 represent the multisensory processing of audiovisual stimuli , and Structure 2 could underlie the additional evaluation of emotional valence associated with the stimuli . Structure 3 was activated the second earliest in the Context period showing a generalized context dependence ( Cm ≈ Ch > Cw ) representing the abstract categorization of the context ( ‘an indeterminate agent is threatening vM’ ) . Similar to Structure 1 , the dependence in Structure 3 was weaker in C− ( C+ > C− ) , which suggested that the creation of abstract contextual information depended on the initial perception of context stimuli . Structure 3 appeared mainly in the β band ( 10–30 Hz ) , and contained primarily bottom-up connections from the posterior temporal cortex ( mainly the area TEO ) to the anterior temporal cortex ( mainly the temporal pole ) and the lateral and medial PFC . Structure 4 showed the same generalized context dependence as Structure 3 , but during the Response period when context stimuli were absent and only in C+Rf ( not in C+Rn and C− ) . The absence of context dependence in C+Rn and C− suggested that Structure 4 required both vM responses with high emotional valence and its context . Moreover , Structure 4 exhibited spatial and spectral characteristics similar to Structure 3 ( Figure 5—figure supplement 1 ) . We conclude that Structures 3 and 4 represent the same or very similar neural substrate , differing only in when and how they were activated . Structure 3 corresponds to the initial formation/encoding of the contextual information , while Structure 4 represents the Rf -triggered reactivation/retrieval of the contextual information . Therefore , Structures 3 and 4 represent the generalized , abstract perceptual and cognitive content of the context . Structure 5 showed context dependence ( Cm ≈ Ch > Cw ) in C+Rf ( not in C+Rn and C− ) , and response dependence ( Rf > Rn ) in C+ ( not in C− ) during the Response period , and appeared mainly in α and low-β bands ( 5–20 Hz ) . Anatomically , the structure showed primarily top-down connections between posterior temporal cortex , the anterior temporal cortex , and the lateral and medial PFC . Remarkably , Structure 5 is the only one demonstrating clear top-down connections , with the same context and response dependence as the gaze behavior ( see Figure 2B ) . These results suggest that Structure 5 corresponds to a network for the context-dependent feedback modulation of eye gaze or visual attention during the task , and the other four structures index internal processes that lead to this behavioral modulation . We investigated how the structures coordinated with each other during the task by examining how they correlated with each other in the functional , dynamical , and anatomical domains . To study function , we evaluated how each structure's context and response dependence correlated with others' , by measuring correlation coefficients of structures' differences in comparisons across contexts in Rf , across contexts in Rn , and across responses ( Figure 6A ) ( detailed in the ‘Materials and methods’ ) . Significant correlations between two structures indicated that one structure's activation affected another's , and vice versa , demonstrating a causal interdependence or a common external driver . Across contexts in Rf ( Figure 6A , left ) , Structure 1 significantly correlated to Structures 3 and 4 , which were themselves significantly correlated to Structure 5 . However , across contexts in Rn ( Figure 6A , middle ) , a significant correlation was found only between Structures 1 and 3 . These results confirmed that sensory perception of the context stimuli could be significantly correlated to the formation of an abstract context , and , in turn this abstract context could be significantly correlated to its reactivation and top-down modulation when a response had high emotional valence . Across responses ( Figure 6A , right ) , Structure 2 significantly correlated to Structure 4 , which was itself significantly correlated to Structure 5 . This indicated that that top-down modulation is the integration of response information and abstract context information . 10 . 7554/eLife . 06121 . 025Figure 6 . Coordination and co-activation of network structures . ( A ) Functional coordination: The coordination between structures was evaluated by the correlation coefficients between structures' context and response dependence ( the differences shown in Figures 4A , 5A ) . Each panel illustrates how Structure i ( y-axis ) correlated with Structure j ( x-axis ) in context dependence in Rf ( left ) , context dependence in Rn ( middle ) , and response dependence ( right ) . Significant correlations are indicated as asterisks ( α = 0 . 05 ) ( see ‘Materials and methods’ ) . ( B ) Dynamic co-activation: The dynamics correlation was shown by correlation coefficients between structures' temporal and spectral activation . Each panel shows how Structure i correlated with Structure j in temporal dynamics ( left ) and frequency profile ( right ) . Significant correlations are indicated as asterisks ( α = 0 . 05 ) . ( C ) Anatomical overlap: The anatomical similarity was indexed by the ratio of shared anatomical connections between structures . Each panel illustrates the ratio of the number of shared connections between Structures i and j and the total number of connections in Structure i . Results obtained from three subjects are shown separately . ( D ) Undirected pathways of connections shared by all structures for each subject ( top ) , and those appearing in at least one structure for each subject ( bottom ) . The lateral cortical surface is shown on the left for Subject 1 , and on the right for Subjects 2 and 3 . Shared pathways ( lines ) between two cortical areas ( circles ) of the top 1 , 5 , 10 , and 25% connections are shown . Pathways with greater strengths are overlaid on those with weaker strengths . DOI: http://dx . doi . org/10 . 7554/eLife . 06121 . 025 To examine dynamics , we tested whether network structures had mutually exclusive or overlapping spectro-temporal dynamics . We measured the temporal dynamics of each structure by summing up the activation in the second tensor dimension across frequencies . Significant correlations in temporal dynamics were found between structures activated in the Context period ( Structures 1 and 3 ) and the Response period ( Structures 2 , 4 , and 5 ) ( Figure 6B , left ) . We then measured the spectral profile of each structure by summing up the activation in the second tensor dimension over time . Significant correlations in spectral profiles were found among structures with β band activation ( Structures 2 , 3 , 4 , and 5 ) ( Figure 6B , right ) . To investigate anatomy , we identified directed connections with the top 10% strengths in the third tensor dimension , and examined the shared top 10% connections between structures for each subject ( Figure 6C ) . The numbers of shared connections between all structures were particularly high ( >70% shared ) between Structures 3 and 4 ( abstract contextual information ) , and Structures 1 and 2 ( perception ) . We examined the undirected pathways that exclude the directionality of connections and found pathways shared by all structures and subjects in and from the temporal cortex to PFC ( Figure 6D , top ) . Pathways appearing in at least one structure were widespread across cortex ( Figure 6D , bottom ) . These results demonstrate the functional coordination and spatio-spectro-temporal co-activation of the five identified network structures , and reveal the multiplexing property of large-scale neuronal interactions in brain: simultaneous information transfer in similar frequency bands along similar anatomical pathways could be functionally reconstituted into distinct cognitive operations depending on other networks' ongoing status . This type of information would be difficult to extract from traditional EEG/MEG/fMRI analyses .
These findings show that context can be encoded in large-scale bottom-up interactions from the posterior temporal cortex to the anterior temporal cortex and the lateral and medial PFC . The PFC is an important node in the ‘context’ network ( Miller and Cohen , 2001; Bar , 2004 ) , where the lateral PFC is believed to be critical for establishing contingencies between contextually related events ( Fuster et al . , 2000; Koechlin et al . , 2003 ) , and the medial PFC is involved in context-dependent cognition ( Shidara and Richmond , 2002 ) and conditioning ( Fuster et al . , 2000; Koechlin et al . , 2003; Frankland et al . , 2004; Quinn et al . , 2008; Maren et al . , 2013 ) . Our results indicate that abstract contextual information can be encoded not only within the PFC , but in PFC interactions with lower-level perceptual areas in the temporal cortex . These dynamic interactions between unimodal sensory and multimodal association areas could explain the neuronal basis of why context networks can affect a wide range of cognitive processes , from lower-level perception to higher-level executive functions . Apart from the bottom-up network structure that encodes abstract context , we discovered other network structures that process either lower-level sensory inputs for context encoding or integrate contextual information for behavioral modulation . Evidently , brain contextual processing , from initial perception to subsequent retrieval , is represented not by sequential activation but rather sequential modular communication among participating brain areas . Thus , we believe that the network structures we observed represent a module of modules , or ‘meta-module’ for brain communication connectivity . Further investigation of this meta-structure organization for brain network communication could help determine how deficits in context processing in psychiatric disorders such as schizophrenia ( Barch et al . , 2003 ) and post-traumatic stress disorder ( Milad et al . , 2009 ) could contribute to their etiology . These results suggest a basic structural organization of large-scale communication within brain networks that coordinate context processing , and provide insight into how apparently seamless cognition is constructed from these network communication modules . In contrast to previous studies where brain modularity is defined as a ‘community’ of spatial connections ( Bullmore and Sporns , 2009; Sporns , 2011 ) , or coherent oscillations among neuronal populations in overlapping frequency bands ( Siegel et al . , 2012 ) , our findings provide an even more general yet finer grained definition of modularity based on not only anatomical and spectral properties , but also temporal , functional , and directional connectivity data . The relationships among network properties in the functional , temporal , spectral , and anatomical domains revealed network structures whose activity coordinated with each other in a deterministic manner ( Figure 7A ) , despite being highly overlapping in time , frequency , and space ( Figure 7B ) . Such multiplexed , yet large-scale , neuronal network structures could represent a novel meta-structure organization for brain network communication . Further studies will be needed to show whether these structures are components of cognition . 10 . 7554/eLife . 06121 . 026Figure 7 . Context as a sequence of interactions between network structures . ( A ) Coordination between network structures ( S1 to S5 , circles ) , under Rn ( top ) or Rf ( bottom ) responses . In both response contingencies , context perception ( S1 ) encoded contextual information ( S3 ) . However , when the response stimulus contained high emotional valence ( Rf , bottom ) , response perception ( S2 ) reactivates the contextual information ( S4 ) , resulting in top-down modulation feedback ( S5 ) that shares the same context and response dependence as the gazing behavior ( black arrow and rounded rectangles ) . Green , blue , and red arrows represent correlations in context dependence in Rn , context dependence in Rf , and in response dependence , respectively ( see Figure 6A ) . ( B ) Temporal , spectral , and spatial profiles and overlap in defined network structures . Network structures can be characterized by frequency range ( labeled on the left ) and connectivity pattern ( shown on the right ) . Their temporal activations are plotted over trial time , with a ‘sound-like’ presentation , where a higher volume represents stronger activation . Black vertical lines represent the events as indicated in Figure 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 06121 . 026 We developed an analytical approach using an unbiased deconvolution of comprehensive network activity under well-controlled and staged behavioral task conditions . This workflow enabled us to identify novel network structures and their dynamic evolution during ongoing behavior . In principle , this approach can be generally useful to investigate how network structures link neural activity and behavior . However , we caution that the latent network structures we identified were extracted computationally , and therefore will require further confirmatory experiments to verify their biological significance , particularly the causality of the connectivity patterns within each structure and the functional links bridging different structures . The biological meaning of the identified network structures could be achieved by selective manipulation of neuronal pathways by electrical or optogenetic stimulation linked to the ECoG array by neurofeedback , or neuropharmacological manipulations . The general class of network structures we identified are not necessarily unique to context , By recording with a hemisphere-wide ECoG array and applying our analytical methodologies to other cognitive behaviors and tasks in non-human primates , we fully expect to observe similar network structures . Our approach of pooling large-scale data across subjects may be useful to extract network structures that are generalizable , because neural processes specific to individual subjects or trials will cancel . Indeed , the stable and consistent trial responses across subjects in our chronic ECoG recordings suggest that the network structures we isolated may be candidate innate , elemental units of brain organization . Conversely , future identification of unique differences in network structures between subjects could offer insight into structures related to individual trait and state variability , and the network-level etiology of brain diseases ( Belmonte et al . , 2004; Uhlhaas and Singer , 2006 ) .
Customized 128-channel ECoG electrode arrays ( Unique Medical , Japan ) containing 2 . 1 mm diameter platinum electrodes ( 1 mm diameter exposed from a silicone sheet ) with an inter-electrode distances of 5 mm were chronically implanted in the subdural space in three Japanese macaques ( Subjects 1 , 2 , and 3 ) . The details of surgical methods can be found on Neurotycho . org . In Subject 1 , electrodes were placed to cover most of the lateral surface of the right hemisphere , also the medial parts of the frontal and occipital lobes . In Subject 2 , a similar layout was used , but in the left hemisphere . In Subjects 3 , all electrodes were placed on the lateral surface of the left hemisphere , and no medial parts were covered . The reference electrode was also placed in the subdural space , and the ground electrode was placed in the epidural space . Electrical cables leading from the ECoG electrodes were connected to Omnetics connectors ( Unique Medical ) affixed to the skull with an adaptor and titanium screws . The locations of the electrodes were identified by overlaying magnetic resonance imaging scans and x-ray images . For brain map registration , the electrode locations and the brain outlines from Subjects 1 and 3 were manually registered to those from Subject 2 based on 13 markers in the lateral hemisphere and 5 markers in the medial hemisphere ( see Figure 1—figure supplement 1 ) . All experimental and surgical procedures were performed in accordance with the experimental protocols ( No . H24-2-203 ( 4 ) ) approved by the RIKEN ethics committee and the recommendations of the Weatherall report , ‘The use of non-human primates in research’ . Implantation surgery was performed under sodium pentobarbital anesthesia , and all efforts were made to minimize suffering . No animal was sacrificed in this study . Overall care was managed by the Division of Research Resource Center at RIKEN Brain Science Institute . The animal was housed in a large individual enclosure with other animals visible in the room , and maintained on a 12:12-hr light:dark cycle . The animal was given food ( PS-A; Oriental Yeast Co . , Ltd . , Tokyo , Japan ) and water ad libitum , and also daily fruit/dry treats as a means of enrichment and novelty . The animal was occasionally provided toys in the cage . The in-house veterinary doctor checked the animal and updated daily feedings in order to maintain weight . We have attempted to offer as humane treatment of our subject as possible . During the task , each monkey was seated in a primate-chair with its arms and head gently restrained , while a series of video clips was presented on a monitor ( Videos 1–6 ) . In one recording session , each of six video clips was presented 50 times , and all 300 stimuli were presented in a pseudorandom order in which the same stimulus would not be successively presented . In order to keep the monkey's attention to the videos , food items were given after every 100 stimuli . Each monkey participated three recording sessions within a week . Each stimulus consisted of three periods: Waiting , Context , and Response periods . During the Waiting period , a still picture created by pixel-based averaging and randomizing the all frames of stimuli was presented without sound for 2 . During the first 0 . 5 s of the Context period , a still image of an actor ( a monkey ) and an opponent ( a monkey , a human , or wall ) was presented with the sound associated with the opponents . The actor was always positioned on the left side of the image . Then a curtain in the video started to close from the right side toward the center to cover the opponent . The curtain closing animation took 0 . 5 s , and the curtain stayed closed for another 0 . 5 s . During the Response period , one of two emotional expressions of the actor ( frightening or neutral ) was presented with sound for 3 s , followed by the Waiting period of the next trial . An iMac personal computer ( Apple , USA ) was used to present the stimuli on a 24-in LCD monitor ( IOData , Japan ) located 60 cm away from the subject . The sound was presented through one MA-8BK monitor speaker ( Roland , Japan ) attached to the PC . The experiments were run by a program developed in MATLAB ( MathWorks , USA ) with Psychtoolbox-3 extensions ( Brainard , 1997 ) . The same PC was used to control the experiments and the devices for recording monkey's gaze and neural signals via USB-1208LS data acquisition device ( Measurement Computing Co . , USA ) . A custom-made eye-track system was used for monitoring and recording the monkey's left ( Subject 1 ) or right ( Subjects 2 and 3 ) eye at 30 Hz sampling ( Nagasaka et al . , 2011 ) . Cerebus data acquisition systems ( Blackrock Microsystems , USA ) were used to record ECoG signals with a sampling rate of 1 kHz . Trials during which the subject's eye position was within the screen area more than 80% of the time during the first 0 . 5 s of the Context period were classified as C+ trials . The rest of the trials were identified as C− trials , where the subject either closed its eyes or the eye position was outside the screen or outside the recording range ( ±30° ) . | If we see someone looking frightened , the way we respond to the situation is influenced by other information , referred to as the ‘context’ . For example , if the person is frightened because another individual is shouting at them , we might try to intervene . However , if the person is watching a horror video we may decide that they don't need our help and leave them to it . Nevertheless , it is not clear how the brain processes the context of a situation to inform our response . Here , Chao et al . developed a new method to study electrical activity across the whole of the brain and used it to study how monkeys process context in response to several different social situations . In the experiments , monkeys were shown video clips in which one monkey—known as the ‘video monkey’—was threatened by a human or another monkey , or in which the video monkey is facing an empty wall ( i . e . , in three different contexts ) . Afterwards , the video monkey either displays a frightened expression or a neutral one . Chao et al . found that if the video monkey looked frightened by the context , the monkeys watching the video clip shifted their gaze to observe the apparent threat . How these monkeys shifted their gaze depended on the context , but this behavior was absent when the video monkey gave a neutral expression . The experiments used an array of electrodes that covered a wide area of the monkeys' brains to record electrical activity of nerve cells as the monkeys watched the videos . Chao et al . investigated how brain regions communicated with each other in response to different contexts , and found that the information of contexts was presented in the interactions between distant brain regions . The monkeys' brains sent information from a region called the temporal cortex ( which is involved in processing sensory and social information ) , to another region called the prefrontal cortex ( which is involved in functions such as reasoning , attention , and memory ) . Seconds later , the flow of information was reversed as the monkeys utilized information about the context to guide their behavior . Chao et al . 's findings reveal how information about the context of a situation is transmitted around the brain to inform a response . The next challenge is to experimentally manipulate the identified brain circuits to investigate if problems in context processing could lead to the inappropriate responses that contribute to schizophrenia , post-traumatic stress disorder and other psychiatric disorders . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"neuroscience"
] | 2015 | Cortical network architecture for context processing in primate brain |
We determined differential gene expression in response to high glucose in lymphoblastoid cell lines derived from matched individuals with type 1 diabetes with and without retinopathy . Those genes exhibiting the largest difference in glucose response were assessed for association with diabetic retinopathy in a genome-wide association study meta-analysis . Expression quantitative trait loci ( eQTLs ) of the glucose response genes were tested for association with diabetic retinopathy . We detected an enrichment of the eQTLs from the glucose response genes among small association p-values and identified folliculin ( FLCN ) as a susceptibility gene for diabetic retinopathy . Expression of FLCN in response to glucose was greater in individuals with diabetic retinopathy . Independent cohorts of individuals with diabetes revealed an association of FLCN eQTLs with diabetic retinopathy . Mendelian randomization confirmed a direct positive effect of increased FLCN expression on retinopathy . Integrating genetic association with gene expression implicated FLCN as a disease gene for diabetic retinopathy .
Almost all individuals with diabetes will develop some form of diabetic retinopathy over time ( National Diabetes Fact Sheet , 2011 ) . In the United States diabetic retinopathy is the most frequent cause of blindness among working age individuals ( Centers for Disease Control and Prevention , 2018 ) . Interindividual variation contributes significantly to susceptibility of the severe manifestations of diabetic retinopathy , which results in vision impairment . Epidemiological studies suggest that phenotypic variation is influenced by two primary risk factors: the duration of diabetes and an individual’s level of glycemia ( HbA1c ) ( DCCT/EDIC Research Group et al . , 2017 ) . However , these two factors do not completely explain an individual’s risk for developing diabetic retinopathy . For instance , a common anecdotal clinical experience is the comparison of patients with similar durations of diabetes and similar levels of glycemic control who have tremendously disparate clinical outcomes for diabetic retinopathy . Moreover , some individuals with diabetes develop very minimal retinopathy ( Sun et al . , 2011 ) , whereas others clearly seem to have a predisposition for severe retinopathy ( Gao et al . , 2014 ) . Together , these observations in conjunction with the high concordance of diabetic retinopathy between family members support an underlying genetic mechanism . Familial aggregation and twin studies estimate that genetic factors account for 25–50% of an individual’s risk of developing severe diabetic retinopathy ( Arar et al . , 2008; Hietala et al . , 2008 ) . Unfortunately , little is known about the genetic architecture that contributes to susceptibility for diabetic retinopathy . Genetic studies suggest that it is a highly polygenic trait influenced by multiple genetic variants of small effect . Our group and others have performed genome-wide association studies to better delineate the molecular factors that predispose to diabetic retinopathy ( Grassi et al . , 2011; Grassi et al . , 2012; Pollack et al . , 2019 ) . However , these studies have had limited success , likely due to insufficient study sample sizes and the phenotypic heterogeneity of diabetic retinopathy . Notably , like other complex disease traits including age-related macular degeneration ( Fritsche et al . , 2014; Fritsche et al . , 2016 ) , a majority of genetic variants nominally associated with diabetic retinopathy are located in intronic or inter-genic regions ( Risch and Merikangas , 1996 ) . Most of these variants appear to play critically important functional roles in regulating gene expression . In fact , several of the top associated SNPs identified in our meta-GWAS of diabetic retinopathy ( Grassi et al . , 2011 ) are present in DNase hypersensitivity sites and affect gene expression levels by altering the allelic chromatin state or the binding sites of transcription factors ( Maurano et al . , 2012 ) . The observation that disease-associated genetic loci often influence gene expression levels ( Gamazon et al . , 2018 ) led us to postulate that integrating gene expression with genetic association would be a powerful approach to identify susceptibility genes for diabetic retinopathy . We hypothesized that cell lines derived from individuals with diabetes with and without retinopathy could be used to uncover genetic variation that explain individual differences in the response to diabetes . Culturing two sets of cell lines under controlled , identical conditions from individuals with diabetes who did and those who did not develop retinopathy could unmask molecular differences in how these groups respond to glucose ( Grassi et al . , 2016; Grassi et al . , 2014 ) . We presumed that a portion of those differences would have a genetic basis . In this article , we identify genes whose expression responds differently to glucose in cells derived from T1D individuals with and without diabetic retinopathy . We show that one of these genes , folliculin ( FLCN ) , is causally implicated in diabetic retinopathy based on results from genetic association testing and Mendelian randomization .
Matched DCCT/EDIC participants ( for age , sex , treatment group , cohort , and diabetes duration ) from whom the gene expression profiling was obtained are detailed in Supplementary file 1a . All individuals had T1D and were Caucasian , and 60% were female . As anticipated , notable differences were observed between individuals with and without retinopathy ( PDR vs . nDR ) for mean duration of diabetes ( 53 ± 43 . 4 months vs . 27 ± 13 . 4 months ) . as it was also not possible to completely match participant pairs for this covariate or for level of glycemia ( HbA1c ) . mean HbA1c ( 9 . 71 ± 2 . 37 vs . 7 . 62 ± 1 . 07 ) given their significant impact on retinopathy . We quantified gene expression levels from lymphoblastoid cell lines ( LCLs ) of all study individuals ( nDM , PDR , and nDR ) in both standard glucose ( SG ) and high glucose ( HG ) conditions and determined the genome-wide transcriptional response to glucose for each individual ( RGall ) . We observed that 22% of 11 , 548 examined genes demonstrated a differential response in expression between the two conditions ( true positive rate; π1 = 0 . 22 ) ( Storey and Tibshirani , 2003; Figure 1—figure supplement 1 ) , with 299 of those at an FDR < 0 . 05 ( Figure 1a ) , supporting a significant impact of glucose on the LCL transcriptome . We confirmed that interindividual transcriptome response to HG is greater than the intraindividual response ( p=2×10−16 ) ( Figure 1—figure supplement 2 , Figure 1—figure supplement 3 , Figure 1—figure supplement 4 ) . Interestingly , TXNIP , the most highly glucose-inducible gene in multiple cell types ( Devi et al . , 2017; Chen et al . , 2016 ) , exhibited the largest ( log2 ( FC ) difference = 0 . 2 ) and most significant ( p=3 . 2×10−12 , FDR = 5 . 1×10−8 ) transcriptional response to glucose . Pathway analysis using gene set enrichment analysis ( GSEA ) revealed dramatic upregulation of genes involved in structural changes to DNA ( DNA packaging , FDR < 0 . 0001 ) and in genes such as transcription factors that modulate the cellular response to environmental stimuli ( protein DNA complex , FDR < 0 . 0001 ) ( Figure 1b ) . Conversely , genes that modulate the cellular response to infection were considerably downregulated ( type 1 interferon , FDR < 0 . 0001; gamma interferon , FDR < 0 . 0001; leukocyte chemotaxis genes , FDR < 0 . 0001 ) potentially supporting earlier work that chronic glucose exposure depresses cellular immune responsiveness ( Delamaire et al . , 1997; Al-Mashat et al . , 2006 ) . We observed differences in the transcriptional response to glucose between matched individuals with and without diabetic retinopathy ( RGpdr–ndr ) . Principal component analysis ( PCA ) demonstrated that the observed interindividual variance is dominated by randomized DCCT treatment ( intensive vs . conventional ) group effects based on retinopathy status ( p=3×10−6 ) ( Figure 2—figure supplement 1 ) and is not confounded by LCL growth rate ( p>0 . 05 ) or EBV- ( Epstein Barr virus ) copy number ( p>0 . 05 ) . Using a gene-wise analysis we identified 103 genes exhibiting a differential glucose response between individuals with and without retinopathy ( p<0 . 01 ) ( Figure 2; Supplementary file 1b ) . Some of these genes and pathways have previously been shown to play a role in diabetic retinopathy . One of the top differential response genes was IL1B ( p=0 . 008 , log2 ( FC ) response difference = 0 . 289 ) . Expression of IL1B has been previously reported to be induced by HG ( Shanmugam et al . , 2003 ) . Additionally , the expression of IL1B is upregulated in the diabetic retina and has been implicated in the pathogenesis of diabetic retinopathy ( Liu et al . , 2012 ) . Likewise , the top GSEA pathway has also previously been implicated in the pathogenesis of diabetic retinopathy . We identified PDGF signaling as the most significant differential response pathway ( FDR = 0 . 012 ) ( Figure 2—figure supplement 2 ) . Elevated levels of PDGF are present in the vitreous of individuals with proliferative diabetic retinopathy ( PDR ) compared to individuals without diabetes ( Freyberger et al . , 2000 ) . As PDGF is required for normal blood vessel maintenance , it is thought to contribute to the pericyte loss , microaneurysms , and acellular capillaries that are key features of the diabetic retina ( Hammes et al . , 2002 ) . Interestingly , despite our model utilizing lymphoblastoid cells , it was able to reveal the upregulation of PDGF which is primarily a vascular factor that also plays a key role in neuronal tissue . We sought to assess whether the most significant differential response genes ( RGpdr–ndr ) could yield novel insights into diabetic retinopathy . An overview of our approach is presented in Figure 3a . First , we selected the top 103 genes ( p<0 . 01 ) that showed the largest difference in gene expression response to glucose between individuals with diabetes with and without retinopathy . We next identified all of the significant expression quantitative trait loci ( eQTLs ) for these genes in GTEx ( version 7 ) ( GTEx Consortium , 2015 ) . In total , we found 7253 unique eQTL SNPs ( hereafter referred to as eSNPs ) in at least one of the 48 tissues investigated by GTEx . Differential response genes are more likely to harbor eSNPs , and hence be eGenes , compared to the genome-wide average ( p=2 . 0×10−16 ) ( Figure 3—figure supplement 1 ) . This suggests that differential response genes are more likely to be genetically regulated and may contribute to interindividual differences in the development of diabetic retinopathy . To test if the eSNPs for the 103 differential response genes were more associated with diabetic retinopathy than expected , we evaluated the association between the 7253 differential response gene eSNPs and diabetic retinopathy using our published GWAS of diabetic retinopathy ( Grassi et al . , 2011 ) . The 7253 eSNPs from the differential response genes are enriched for association with diabetic retinopathy ( FDR < 0 . 05 ) ( Figure 3b ) . To further assess the significance of this enrichment , we performed permutation testing of eSNPs from random sets of 103 genes which demonstrated that less than 1% contained the same proportion of similarly skewed GWAS p-values ( Figure 3—figure supplement 2 ) . The eSNPs for differential response genes were enriched among diabetic retinopathy meta-GWAS p-values relative to all eSNPs ( p=0 . 0012 ) and all SNPs ( p=0 . 0023 ) ( Figure 3c ) . Thus , some of the genes exhibiting a differential response to glucose ( RGpdr–ndr ) are associated with the development of severe diabetic retinopathy . The most significant retinopathy-associated eSNP among the set of 7253 eSNPs tested is rs11867934 ( Figure 4a ) ; FDR < 0 . 05; meta-GWAS p=6 . 7×10−6<Bonferroni adjusted p-value of 6 . 9 × 10−6; OR = 0 . 86 , 95% confidence interval ( CI ) = 0 . 71 , 1 . 00; minor allele frequency = 0 . 22 . rs11867934 is an intergenic eSNP for FLCN in multiple biologically relevant tissues including artery and nerve . We confirmed FLCN expression in the retina of human donor eyes ( Figure 4—figure supplement 1 ) . In the LCLs derived from individuals with diabetes , FLCN was upregulated in response to glucose to a greater extent in individuals with diabetic retinopathy than in individuals with diabetes without retinopathy ( log2FC difference = 0 . 27 , p=2 . 5×10−3 ) ( Figure 2 , Supplementary file 1b , and Figure 4—figure supplement 2 ) . eQTLs in retina have recently been mapped ( Ratnapriya et al . , 2019 ) . We determined that at least 43% of retina eQTLs are also eQTLs in GTEx LCLs . Examining the genome-wide association signal for a disease from eQTLs in aggregate can be a more powerful strategy to discern a heterogenous genetic signal than testing each of these SNPs individually . We collated all the eSNPs for FLCN in the retina . We assessed the aggregated association of FLCN eSNPs ( n = 272 eSNPs significant in the retina and 20 or more GTEx tissues ) with diabetic retinopathy in the meta-GWAS and observed an enrichment for association with diabetic retinopathy ( π1 = 0 . 9; Figure 4b , Figure 4—figure supplement 3 ) . We then validated the FLCN association with diabetic retinopathy in a third cohort , the UK Biobank ( UKBB ) ( Supplementary file 1c ) , and found that the FLCN eSNPs were enriched for association with diabetic retinopathy in the UKBB ( π1 = 0 . 73 ) ( Figure 4—figure supplement 4 ) . We applied Mendelian randomization to assess whether the level of FLCN expression affects the development of diabetic retinopathy . We first imputed retinal FLCN expression in the UKBB , and then estimated the effects of the estimated FLCN expression on diabetic retinopathy using summary data-based Mendelian randomization analysis ( Zhu et al . , 2016 ) ( SMR ) . Mendelian randomization treats the genotype as an instrumental variable . A one standard deviation ( SD ) increase in the predicted retinal expression of FLCN increases the risk of diabetic retinopathy by 0 . 15 SD ( 95% CI: 0 . 02–0 . 29 , standard error 0 . 07 , p=0 . 024 ) . Individuals with diabetes with high predicted retinal FLCN expression have increased odds of developing retinopathy ( 1 . 3 OR increase per SD increase in FLCN expression ) ( Chinn , 2000 ) . We did not observe any evidence of horizontal pleiotropy ( in which FLCN eSNPs are independently associated with both FLCN expression and diabetic retinopathy ) confounding the analysis [HEIDI p>0 . 05 ( p=0 . 2 ) ] ( Zhu et al . , 2016 ) . We detected an aggregated effect of 14 independent FLCN eQTLs ( r2 < 0 . 2 ) on the development of diabetic retinopathy through FLCN expression using multi-SNP Mendelian randomization ( p=0 . 04 ) ( Wu et al . , 2018a ) . Together , these findings support the presence of genetic variation at the FLCN locus affecting both FLCN expression and the development of diabetic retinopathy through the expression of FLCN .
The cellular response to elevated glucose is an increasingly important pathway to understand in light of the emerging epidemic levels of diabetes worldwide ( National Diabetes Fact Sheet , 2011 ) . Variations in the cellular response to glucose at a molecular level have not been well characterized between cell types , and to an even lesser degree between individuals . In prior work , we characterized robust , repeatable interindividual differences in transcriptional response to glucose in LCLs of individuals with diabetic retinopathy ( Grassi et al . , 2016 ) . As an LCL generated from each individual is genetically unique , it follows that the gene expression response to glucose between individuals should be phenotypically heterogeneous and that a portion of the interindividual variability will be genetically determined . We hypothesized that interindividual variation in the cellular response to glucose may reveal clues to the genetic basis of diabetic retinopathy , thereby providing insights into its predisposition . We demonstrated that different individual-derived cell lines treated under identical culture conditions reveal an individual-specific transcriptional response to glucose and this signal far exceeds accompanying experimental noise . Transformation and multiple freeze/thaw passages do not homogenize the individualized response to HG-induced gene expression in LCLs . Analyzing the individual glucose-stimulated transcriptional response revealed several insights into the pathophysiology of the diabetic state and how it relates to the development of retinopathy . For instance , TXNIP was identified as the top differential response gene to glucose in all individuals ( RGall ) . TXNIP is a key marker of oxidative stress . It is upregulated in the diabetic retina where it induces Muller cell activation ( Devi et al . , 2017 ) . HG treatment has been shown to increase TXNIP expression ( Chen et al . , 2016 ) . TXNIP is a glucose sensor whose expression has been strongly associated with both hyperglycemia and diabetic complications . Specifically , the TXNIP locus was differentially methylated in the primary leukocytes of EDIC cases and controls ( Chen et al . , 2016 ) . A key mechanism by which cells respond to stress is through changes in genome configuration . Conformational alterations in DNA packaging influence the accessibility of DNA for transcription . Structural changes in DNA conformation facilitate cellular adaptation and response to stimuli which can enable transcriptional changes . The GSEA showed that the cellular response to chronic glucose stress involves alterations in DNA accessibility which facilitates the gene expression response to this environmental stimulus ( Smith and Workman , 2012 ) . The transcriptional response to glucose in part manifests as diminished immune responsiveness , a well-characterized feature of diabetes ( Shanmugam et al . , 2003; Mowat and Baum , 1971 ) . Further , we considered that the genetic component of an individual’s response to glucose may influence their susceptibility to diabetic complications like retinopathy . Cell lines from individuals with diabetes with and without retinopathy reveal differences in the response to glucose at a molecular level . In addition , not only were some of these differential response genes biologically relevant to diabetic retinopathy as exemplified by IL1B and PDGF , but also many had a genetic basis for their differential response . By integrating the gene expression findings with GWAS data , we implicated FLCN as a putative disease gene in diabetic retinopathy . Mendelian randomization provided evidence that genetic variation affects diabetic retinopathy through alterations in FLCN expression thereby suggesting that FLCN expression is a mediator of diabetic retinopathy . FLCN is a biologically plausible diabetic retinopathy disease gene since its expression is present in both neuronal and vascular cells of the retina . Current evidence suggests that FLCN is a negative regulator of AMPK which helps to modulate the energy sensing ability of AMPK and plays a role in responding to cellular stress ( Hasumi et al . , 2012 ) . AMPK plays an important role in providing resistance to cellular stresses by regulating autophagy and cellular bioenergetics to avoid apoptosis . Loss of FLCN results in constitutive activation of AMPK . Higher levels of FLCN would suggest less cellular capacity to deal with stress ( Possik et al . , 2014 ) . Interestingly , the protective effect of agents such as metformin and fenofibrate on diabetic retinopathy might be mediated through AMPK ( Kim et al . , 2007; Joe et al . , 2015 ) . Our study design had several advantages over prior approaches aimed at revealing the genetic basis of diabetic retinopathy . First , we utilized white blood cells which are readily accessible from the peripheral circulation of human patients ( Epidemiology of Diabetes Interventions and Complications ( EDIC ) Research Group , 1999 ) and can reveal differential molecular characteristics depending on the stage of diabetic retinopathy ( Tang and Kern , 2011; Gubitosi-Klug et al . , 2008; Kern , 2007 ) . LCLs are derived from white blood cells making them a relevant cellular population to study for diabetic retinopathy . LCLs have been shown to be a powerful model system for functional genetic studies in humans ( Tang and Kern , 2011; Kern , 2007 ) . Second , an LCL was generated for every individual enrolled in the landmark DCCT/EDIC study . DCCT/EDIC is the best-characterized prospective interventional cohort ever created to follow systemic complications of long-standing diabetes . DCCT/EDIC allows for detailed stratification of individuals , each of whom has had extensive prospective clinical phenotyping . Third , glucose was employed to elicit a provocative response in LCLs . By focusing on a secondary sequela of diabetes like retinopathy , the cellular response to glucose stimulation through transcription became a meaningful and directly relevant reflection of the stress each cell in the body encounters from diabetes . Insights into glucose-stimulated gene expression in LCLs have broad applicability to multiple tissues of interest for diabetic complications ( even in the retina as we have shown ) due to significant evidence supporting a shared framework for gene regulation among tissues ( GTEx Consortium , 2015 ) . Finally , disease-associated eQTL provide functional insights into the pathogenesis of a condition . We show that altering the levels of FLCN expression impacts risk of diabetic retinopathy . Aggregating independent eQTLs for the same gene ( that are not in high linkage disequilibrium ) revealed an enriched association that may otherwise have been missed by a conventional GWAS approach ( Wu et al . , 2018b ) . Treating the associated eQTL as an instrumental variable , Mendelian randomization supported the potential causality of FLCN in the pathogenesis of the disease . Inherently , this approach yielded all three M’s of target modulation: mechanism , magnitude , and markers ( Plenge , 2019 ) . The present work had inherent limitations . First , LCLs are not primary cells but rather a transformed cell line . The Multiple Tissue Human Expression Resource ( MuTHER ) LCL study revealed a large impact of common environmental exposure , stemming from shared sample handling , on gene expression in twin LCLs ( Wright et al . , 2014 ) . The significant correlation of these extrinsic factors on LCL gene expression emphasizes the importance of randomization and biological replicates which we implemented in this study . Moreover , as a cell line , heterogeneous genomic alterations have been identified in lymphoblastoid cells that increase with passaging , thereby raising the concern that this can lead to variability in their transcriptome ( Ben-David et al . , 2018 ) . Importantly , the EDIC cell lines employed in this study were only passaged once previously . Additionally , genomic changes have only a minor effect on genotypic frequencies with a 99 . 63% genotype concordance between lymphoblastoid cells and their parent leukocytes . Mendelian error rates in levels of heterozygosity are not significantly different between LCLs and their primary B-lymphocyte cells of origin ( McCarthy et al . , 2016 ) . Second , it is not possible to delineate cause from effect in gene expression studies . Gene expression changes may be causal , epiphenomena , or due to reverse causality ( the disease causing the gene expression changes rather than the other way around ) . In this study , by integrating genetic analyses with gene expression and recognizing that variation in the underlying genome precedes disease onset and can therefore be considered an instrumental variable , we identified through Mendelian randomization potentially causal gene expression changes in FLCN that act as a mediator for retinopathy thereby avoiding the trap of reverse causality . Finally , eQTL found in LCLs may not be relevant to diabetic retinopathy . As noted previously we found 43% of retina eQTL are shared with LCLs . We demonstrated that independent FLCN eQTLs found both in the retina and GTEx tissues showed an enriched association with diabetic retinopathy , a finding that was replicated in a large independent cohort from the UKBB . For complex trait associations in general and for those specifically in the retina , eQTL that are shared between tissues explain a greater proportion of associations than tissue-specific eQTL ( Gamazon et al . , 2018 ) . For instance , shared tissue eQTL are enriched among genetic associations with age-related macular degeneration , another common retinal disease , despite the high tissue specificity of the disease ( Ratnapriya et al . , 2019; Unlu et al . , 2019 ) . In summary , integration of gene expression from a relevant cellular model with genetic association data provided insights into the functional relevance of genetic risk for a complex disease . Using disease-associated differential gene and eQTL-based genome-wide association testing , we identified possible causal genetic pathways for diabetic retinopathy . Specifically , our studies implicated FLCN as a putative diabetic retinopathy susceptibility gene . Future work that incorporates more extensive molecular profiling of the cellular response to glucose in conjunction with a greater number of cell lines may yield further insights into the underlying genetic basis of diabetic retinopathy .
In this study we profiled the transcriptomes of cell lines derived from 22 individuals ( seven individuals with no diabetes [nDM] , eight with T1D with PDR , and seven with T1D with no retinopathy [nDR] ) utilizing gene expression microarrays to characterize the transcriptional response to glucose . Specifically , we cultured LCLs derived from each individual in SG and HG medium and measured gene expression for each gene in each sample , as well as the difference ( Δ = response to glucose [RG] ) in each gene’s expression for each individual ( Figure 5a ) . We compared the differential response in gene expression to glucose for all individuals with and without proliferative retinopathy . ‘Differential response’ in gene expression refers to the difference in gene expression response to glucose between groups . Specifically , we identified genes with a significant differential response in expression between individuals with diabetes with and without PDR ( RGpdr–ndr ) . We followed up genes showing differential response using the results of both a prior genome-wide association study ( GWAS ) meta-analysis of diabetic retinopathy ( in the GoKinD and EDIC cohorts ) ( Grassi et al . , 2011 ) and the results of a multi-tissue eQTL analysis from GTEx ( GTEx Consortium , 2015 ) to identify potential diabetic retinopathy susceptibility genes ( Figure 5b ) . All cell lines were de-identified prior to their arrival at the University of Illinois at Chicago; therefore , this proposal qualified as non-human subjects research according to the guidelines set forth by the institutional review board at the University of Illinois at Chicago . As the data were analyzed anonymously , no participant consent was required . DCCT participants previously provided consent for their samples to be used for research . Matching of participants was performed at George Washington University Biostatistics Center and did not involve protected health information as the phenotypic data were de-identified . The George Washington University institutional review board has approved all analyses of EDIC data of this nature . All protocols used for this portion of the study are in accordance with federal regulations and the principles expressed in the Declaration of Helsinki . Specific approval of the study design and plan was obtained from the EDIC Research Review Committee . Twenty-two LCLs were used in the study as described previously ( Grassi et al . , 2016 ) . Briefly , we included 15 of the 1441 LCLs generated from individuals with type 1 diabetes from the DCCT/EDIC cohort ( The Writing Team for the Diabetes Control and Complications Trial/Epidemiology of Diabetes Interventions and Complications Research Group , 2002; Epidemiology of Diabetes Interventions and Complications ( EDIC ) Research Group , 1999 ) , consisting of eight matched cases with PDR and seven without retinopathy ( nDR ) as the controls ( Grassi et al . , 2016; Supplementary file 1d ) . Whole blood samples were ascertained from DCCT study participants between 1991 and 1993 . White blood cells from the samples were transformed into LCLs in the early 2000s . The 15 LCLs from individuals with diabetes consisted of matched cases and controls . Cases were defined by the development of PDR by EDIC Year 10 ( 2004 ) , whereas controls had no retinopathy through EDIC Year 10 ( 2004 ) . Retinopathy grade was determined by seven-field stereoscopic photos . Control participants had an ETDRS ( Early Treatment Diabetes Retinopathy Score ) of 10 and case participants had an ETDRS score of ≥61 . All eight pairs were matched by age , sex , treatment group ( intensive vs . conventional ) , cohort ( primary vs . secondary ) , and diabetes duration ( The DCCT Research Group , 1986; The diabetes control and complications trial , 1995 ) , except one pair that was matched by age , sex , and treatment group only . Diabetes duration was defined as the number of months since the diagnosis of diabetes at DCCT baseline which was the time at participant enrollment ( 1983–1989 ) . For the seven pairs matched on duration , four pairs were matched by duration quartiles ( baseline duration 0–4 years , 4–8 years , 8–12 years , or 12–15 years ) and three pairs were matched by duration halves ( <8 years vs . ≥ 8 years ) . Matching by age was done similar to duration: four pairs by quartile ( <21 years , 21–25 years , 26–31 years , and ≥31 years ) and the remaining four by halves ( <26 years vs . ≥26 years ) . The remaining seven LCLs were purchased from the Coriell Institute for Medical Research NIGMS Human Genetic Cell Repository ( http://ccr . coriell . org/ ) ( GM14581 , GM14569 , GM14381 , GM07012 , GM14520 , GM11985 , and GM07344 ) . None of these individuals had a history of diabetes ( nDM ) . The covariates available for these seven individuals were age and sex; male and female individuals were included . All of these individuals without diabetes were unrelated and of European ancestry ( Grassi et al . , 2016; Grassi et al . , 2014; Supplementary file 1d ) . All 22 cell lines underwent Hoechst staining to ensure they were free from mycoplasma contamination . All LCLs were maintained in conventional lymphocyte cell culture conditions of RPMI 1640 with 10% FBS in a 25 cm2 cell culture flask . The cells were incubated at 37°C in 5% CO2 and the media was changed twice each week . Prior to the experiments ( below ) , lymphoblastoid cells following serum starvation were passaged for a minimum of 1 week in either SG RPMI 1640 ( 11 mM glucose ) or HG RPMI media ( 30 mM glucose ) ( Caramori et al . , 2012 ) . Quality control from RNA extraction was performed using the Agilent bio-analyzer , processed using the Illumina TotalPrep−96 RNA Amplification Kit ( ThermoFisher 4393543 ) , hybridized to Illumina HT12v4 microarrays ( Catalog number: 4393543 ) , and scanned on an Illumina HiScan scanner ( Du et al . , 2008; Lin et al . , 2008 ) . For each of the 22 individuals , three biological replicates were profiled , with each sample assessed at both SG conditions ( 11 mM of glucose ) and HG conditions ( 30 mM of glucose ) . Biological replicates were split from the same mother flask; cells were grown in separate flasks and run on different microarray plates on different days . Each biological replicate was generated from a separate frozen aliquot of that cell line . The gene expression profiling was performed in a masked fashion for both the case/control ( PDR , nDR , and nDM ) status of the individual and the glucose treatment ( SG and HG ) of the sample in order to reduce any bias . Standard TaqMan qPCR was performed using EBV and NRF1 probes and primers ( Choy et al . , 2008 ) . To calculate real-time PCR efficiencies a standard curve of 10 points of twofold dilution of 156 . 7 ng of gDNA was used from the Raji cell line ( ATCC CCL-86 ) . Probes were designed for the target , EBV , and a reference gene , NRF1 . Final concentrations of the probes and primers were 657 nM and 250 nM , respectively . EBV probe: 5′6FAM-CCACCTCCACGTGGATCACGA-MGBNFGQ3′; EBV forward primer: 5′ GAGCGATCTTGGCAATCTCT; EBV reverse primer: 5′ AGTAGCCAGGCACCTACTGG; NRF1 probe: 5′VIC-CACTGCATGTGCTTCTATGGTAGCCA-MGBNFQ3′; NRF1 forward primer: 5′ ATGGAGGAACACGGAGTGAC; NRF1 reverse primer: 5′ CATCAGCTGCTGTGGAGTTG . Cycle number of crossing point versus DNA concentration were plotted to calculate the slope . The real-time PCR efficiency ( E ) was calculated according to the equation: E = 10 ( −1/slope ) . Triplicates were done for each data point . Genomic DNA ( 78 . 3 ng ) from each LCL was used in a standard TaqMan qPCR reaction with EBV as target gene and NRF1 as reference gene . The sequences and concentrations of the probes and primers were as shown above . LCLs were thawed and cultured in RPMI and 10% FBS until they reached over 85% cell viability . Cells were seeded in a T25 flask . Two replicates were performed per cell line . Cells were counted every day or every other day for 5–10 days and recorded . The gene expression data comprised a total of 144 samples from 22 individuals ( three replicates per individual and treatment , except for three individuals with five replicates ) . Gene expression was assessed in two conditions , SG and HG , and generated in four batch runs that were carefully designed to minimize potential batch effects . BeadChip data were extracted using GenomeStudio ( version GSGX 1 . 9 . 0 ) and the raw expression and control probe data from the four different batches were preprocessed using a lumiExpresso function in the lumi R package version 2 . 38 . 0 ( Grassi et al . , 2011; Grassi et al . , 2012 ) in three steps: ( i ) background correction ( lumiB function with the bgAdjust method ) ; ( ii ) variance stabilizing transformation ( lumiT function with the log2 option ) ; and ( iii ) normalization ( lumiN function with the robust spline normalization [rsn] algorithm that is a mixture of quantile and loess normalization ) . To remove unexpressed probes , we applied a detection filter to retain probes with strong true signal by applying Illumina BeadArrays detection p-values <0 . 01 followed by removing probes that did not have annotated genes , resulting in a total of 15 , 591 probes . The study design is portrayed in Figure 5a . For a given individual Si ( i = 1 , … , 22 ) and gene Gk ( k = 1 , . . . , 15591 ) , we calculated ∆i , k = HGi , k− SGi , k , where ∆ is the individual’s response to glucose ( RG ) , HG is gene expression in high glucose culture , and SG is gene expression in standard glucose culture . All replicate data were fit using a mixed model that accounted for the correlation between repeated measures within individuals . The design matrix was constructed and analysis performed using the R version 3 . 5 . 1 package limma ( Ritchie et al . , 2015 ) . We built a design matrix using the model . matrix function , and accounted for correlation between biological triplicates using limma’s duplicate correlation function . A mixed linear model was then fit that incorporates this correlation and ∆i , k using the lmFit function . PCA of gene expression was run with the prcomp function in R ( Becker , 1988 ) . For each gene , we calculated moderated t- and f-statistics and log-odds of expression by empirical Bayes moderation of the standard errors toward a common value . Differential response reflects fold change ( FC ) differences between matched individuals in the two groups in their paired response to glucose . The power to detect a 2 FC difference in gene expression between the two retinopathy groups ( retinopathy vs . no retinopathy ) ( RGpdr–ndr ) with a paired analysis given our sample size and using a type I error rate of 0 . 05 is 95% ( as supported by our prior work Grassi et al . , 2016 ) . GSEA was performed using pre-ranked gene lists ( Subramanian et al . , 2005 ) . We ranked all analyzed genes based on sign ( fold change ) × ( –log10 ( p-value ) ) ( Grassi et al . , 2012 ) . Duplicated genes were removed . The gene ranking resulted in the inclusion of 11 , 579 genes . Enrichment statistics were calculated using rank weighting and the significance of enrichment was determined using permutations performed by gene set . The gene sets included c2 . all . v6 . 0 and c5 . all . v6 . 0 . The minimum gene set size was 15 and the maximum gene set size was 500 . GSEA was used to identify significant gene sets for the response to glucose in all study participants ( RGall: nDM + PDR + nDR ) . To determine if the genes showing a differential response in gene expression ( RGpdr–ndr ) is driven by germline genetic variation , we tested if the eQTLs for these genes are enriched for small diabetic retinopathy GWAS p-values ( Grassi et al . , 2011 ) . We use the term ‘differential response gene’ for those genes identified by the RGpdr–ndr analysis . All statistically significant eSNPs ( false discovery rate [FDR] threshold of ≤0 . 05 ) ( single nucleotide polymorphisms , SNPs , corresponding to cis-eQTLs from the GTEx and EyeGEx datasets ) were collated for the glucose response genes in any of the 48 GTEx ( version 7 ) tissues and the retina ( GTEx Consortium , 2015; Ratnapriya et al . , 2019 ) . We use the term eGene for any gene with at least one significant eSNP in any tissue . Meta-analysis p-values were ascertained from our prior GWAS for diabetic retinopathy ( Grassi et al . , 2011 ) . The study assessed the genetic risk of sight threatening complications of diabetic retinopathy as defined by the presence of diabetic macular edema or PDR ( cases ) compared to those without ( controls ) in two large type 1 diabetes cohorts of 2829 total individuals ( 973 cases and 1856 controls ) taken from the Genetics of Kidney in Diabetes ( GoKinD ) and the Epidemiology of Diabetes Interventions and Complications study ( EDIC ) cohorts . We sought to determine whether there is enrichment of small p-values for diabetic retinopathy meta-GWAS among the significant eQTLs for the glucose response genes that show a significant differential glucose response between individuals with and without retinopathy ( RGpdr–ndr ) . We used Benjamini–Hochberg adjusted p-values ( FDR ) to account for multiple testing given the high level of linkage disequilibrium between many eSNPs within an eQTL . SNPs from the three studies ( expression , eQTL , and GWAS ) were matched by mapping all SNPs to dbSNP v . 147 ( Grassi et al . , 2013 ) . We determined the corresponding FDR for each glucose response gene’s eSNPs in the diabetic retinopathy meta-GWAS . The Bonferroni correction was used to establish the threshold for significance . To assess enrichment , we first determined the observed proportion of meta-GWAS FDR values <0 . 05 among the statistically significant eQTLs of the glucose response genes ( RGpdr–ndr ) . Next , we took 10 , 000 random samples of 103 GTEx eGenes ( genes with an eQTL in any GTEx tissue ) and identified corresponding eSNPs across all GTEx tissues . We calculated the GWAS FDR for these eSNPs and recorded the proportion of FDR values <0 . 05 . Validation for the association of glucose response gene eSNPs with diabetic retinopathy was performed in the UKBB GWAS ( Supplementary file 1c; Sudlow et al . , 2015 ) . Only individuals of northern European ancestry were analyzed . Quality control excluded individuals who were outliers based on relatedness , exhibited an excess of missing genotype calls , had more heterozygosity than expected , or had sex chromosome aneuploidy . A total of 337 , 147 individuals were available for analysis . Case participants were defined as those who answered ‘yes’ to questionnaire data eyesight field 6148 ‘Diabetes related eye disease’ ( n = 2332 ) . Our prior work validated the utility of self-report for the presence of severe diabetic retinopathy ( Grassi et al . , 2013; Grassi et al . , 2009 ) . Control participants were defined as those who answered ‘yes’ to data field 2443 ‘Diabetes diagnosed by doctor’ ( n = 14 , 680 ) , excluding case participants . SNPs were excluded according to the following: minor allele frequency <0 . 004; missing rate >0 . 015; HWE p<1×10−10; INFO score <0 . 8 . We performed logistic regression as implemented in Plink2 ( Chang et al . , 2015 ) on this set of cases and controls . The logistic regression , including the following covariates: first 10 genotype-based principal components , chromosomal sex ( as defined by XX , XY status ) , age , type of diabetes , HbA1c , and genotyping array type . To explore a possible causal effect of increased FLCN expression on diabetic retinopathy , we employed Mendelian randomization ( Davies et al . , 2019 ) . Effects were estimated with summary data-based Mendelian randomization analysis ( Zhu et al . , 2016 ) ( SMR ) . We estimated the effect of increasing levels of FLCN expression on diabetic retinopathy in the UKBB GWAS for diabetic retinopathy ( described above ) utilizing 272 SNPs that were significant cis-eSNPs ( FDR ≤ 0 . 05 ) for FLCN in retina and also in at least 20 GTEx tissues . A total of 246 SNPs remained after removing those SNPs or their proxies ( r2 > 0 . 8 ) not genotyped in the UKBB . For each individual , the exposure was based on the genetically predicted gene expression of FLCN in retina and the outcome was the likelihood of having diabetic retinopathy . Heterogeneity in dependent instruments ( HEIDI ) ( Zhu et al . , 2016 ) was used to investigate the possibility of confounding bias from horizontal pleiotropy with 14 independent ( r2 < 0 . 2 ) FLCN eQTLs . As multiple independent ( r2 < 0 . 2 , n = 14 ) FLCN eQTLs exist , we also employed multi-SNP Mendelian randomization to assess for an aggregated effect ( Wu et al . , 2018a ) of the eQTLs on diabetic retinopathy mediated through FLCN expression . A whole eye from a 69-year-old Caucasian female post-mortem donor without diabetes was obtained from National Disease Research Interchange ( NDRI ) . Findings were replicated in an additional five post-mortem donors without diabetes from the NDRI . The eye was cut in half in a horizontal plane , and each half was placed in an individual cassette . Samples were processed on ASP300 S automated tissue processor ( Leica Biosystems ) using a standard overnight processing protocol and embedded into paraffin blocks . Tissue was sectioned at 5 µm , and sections were de-paraffinized and stained on BOND RX autostainer ( Leica Biosystems ) following a preset protocol . In brief , sections were subjected to EDTA-based ( BOND ER2 solution , pH9 ) antigen retrieval for 40 min at 100°C , washed , and incubated with protein block ( Background Sniper , Biocare Medical , BS966 ) for 30 min at room temperature . For immunofluorescence ( IF ) , sequential staining with rabbit polyclonal anti-FLCN antibody ( 1:50 , Abcam #ab93196 ) and mouse monoclonal anti-CD31 antibody ( 1:50 , DAKO , M0823 ) was conducted using goat-anti-rabbit Alexa-488 and goat-anti-mouse Alexa-555 secondary antibodies ( Molecular Probes ) for detection . DAPI ( Invitrogen , #D3571 ) was used to stain nuclei . The slides were mounted with ProLong Diamond Antifade mounting media ( ThermoFisher , #P36961 ) . Images were taken at 20× magnification on Vectra three multispectral imaging system ( Akoya Biosciences ) . A spectral library acquired from mono stains for each fluorophore ( Alexa-488 , Alexa-594 ) , DAPI , and human retina background fluorescence slide was used to spectrally unmix images in InForm software ( Akoya Biosciences ) for visualization of each color . | One of the side effects of diabetes is loss of vision from diabetic retinopathy , which is caused by injury to the light sensing tissue in the eye , the retina . Almost all individuals with diabetes develop diabetic retinopathy to some extent , and it is the leading cause of irreversible vision loss in working-age adults in the United States . How long a person has been living with diabetes , the extent of increased blood sugars and genetics all contribute to the risk and severity of diabetic retinopathy . Unfortunately , virtually no genes associated with diabetic retinopathy have yet been identified . When a gene is activated , it produces messenger molecules known as mRNA that are used by cells as instructions to produce proteins . The analysis of mRNA molecules , as well as genes themselves , can reveal the role of certain genes in disease . The studies of all genes and their associated mRNAs are respectively called genomics and transcriptomics . Genomics reveals what genes are present , while transcriptomics shows how active genes are in different cells . Skol et al . developed methods to study genomics and transcriptomics together to help discover genes that cause diabetic retinopathy . Genes involved in how cells respond to high blood sugar were first identified using cells grown in the lab . By comparing the activity of these genes in people with and without retinopathy the study identified genes associated with an increased risk of retinopathy in diabetes . In people with retinopathy , the activity of the folliculin gene ( FLCN ) increased more in response to high blood sugar . This was further verified with independent groups of people and using computer models to estimate the effect of different versions of the folliculin gene . The methods used here could be applied to understand complex genetics in other diseases . The results provide new understanding of the effects of diabetes . They may also help in the development of new treatments for diabetic retinopathy , which are likely to improve on the current approach of using laser surgery or injections into the eye . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"genetics",
"and",
"genomics"
] | 2020 | Integration of genomics and transcriptomics predicts diabetic retinopathy susceptibility genes |
Brains organize behavior and physiology to optimize the response to threats or opportunities . We dissect how 21% O2 , an indicator of surface exposure , reprograms C . elegans' global state , inducing sustained locomotory arousal and altering expression of neuropeptides , metabolic enzymes , and other non-neural genes . The URX O2-sensing neurons drive arousal at 21% O2 by tonically activating the RMG interneurons . Stimulating RMG is sufficient to switch behavioral state . Ablating the ASH , ADL , or ASK sensory neurons connected to RMG by gap junctions does not disrupt arousal . However , disrupting cation currents in these neurons curtails RMG neurosecretion and arousal . RMG signals high O2 by peptidergic secretion . Neuropeptide reporters reveal neural circuit state , as neurosecretion stimulates neuropeptide expression . Neural imaging in unrestrained animals shows that URX and RMG encode O2 concentration rather than behavior , while the activity of downstream interneurons such as AVB and AIY reflect both O2 levels and the behavior being executed .
Mammals adopt different global states in response to threats or opportunities by coordinated changes in physiology and neurochemistry that optimize and focus their response to the situation at hand ( LeDoux , 2012 ) . Hallmarks of such global states include arousal , a reconfiguring of the relative importance given to different sensory cues , and altered physiology due to endocrine feedback ( LeDoux , 2012 ) . Examples of global states include those evoked by a potential mate ( Pfaff et al . , 2008 ) , by predators ( Martinez et al . , 2008; Motta et al . , 2009 ) , or by nutritional state ( Atasoy et al . , 2012; Sternson et al . , 2013 ) . Understanding how global states are encoded in neural circuits is of interest because they may provide insights into subjective behaviors: for example , fear , aggression , and hunger ( LeDoux , 2012; Damasio and Carvalho , 2013 ) . The exact nature of global organismic states is poorly understood . A switch in global state is thought to involve recruitment of many brain circuits whose individual activities are dynamically assembled to address the circumstances faced by the animal . A powerful entry point to study circuits orchestrating such states has involved identifying small populations of neurons whose activation or inhibition can evoke features of the states . For example , in mouse , optogenetic activation of neurons in the ventromedial hypothalamus can induce aggressive behavior ( Lin et al . , 2011 ) , while optogenetic control of basolateral terminals in the amygdala's central nucleus can regulate anxiety-like states ( Tye et al . , 2011 ) . How these small populations of neurons modulate different areas of the brain , how their functional effects depend on the state of other circuits , and how their activity is itself controlled , are open questions . Invertebrates also adopt different behavioral states , for example , in response to potential mates ( Villella and Hall , 2008 ) , predators ( Henschel , 1990; Uma and Weiss , 2012 ) , conspecific rivals ( Dow and von Schilcher , 1975 ) , and nutritional state ( Gaudry and Kristan , 2012 ) . Some of these responses share features of global organismic states in mammals , and may provide insights into how the latter are encoded . A complication in understanding how circuits encode global animal states is that the cues evoking them are often complex . For example , a male Drosophila courting a female fly responds to visual , olfactory , and gustatory cues ( Villella and Hall , 2008 ) ; in Caenorhabditis elegans , starvation resets many sensory responses , including gustation ( Saeki et al . , 2001 ) , olfaction ( Tsunozaki et al . , 2008 ) , and thermotaxis ( Hedgecock and Russell , 1975 ) , but how starvation is sensed and communicated to sensory circuits is poorly understood ( Milward et al . , 2011 ) . A way around this problem is to identify animal state changes that are robustly linked to single identified sensory inputs , providing a defined entry point to dissect how neural networks are reconfigured to encode global states ( e . g . , Chamero et al . , 2007; Kubli and Bopp , 2012; Dewan et al . , 2013 ) . Surface exposure is hazardous for some terrestrial invertebrates , for example , due to desiccation and predation , and triggers a switch in behavioral state as animals seek to escape it . Oxygen concentration [O2] is 21% at the surface but lower in buried spaces due to biomass respiration . The nematode C . elegans can recognize surface exposure by measuring [O2] ( Gray et al . , 2004; Persson et al . , 2009 ) . An increase in O2 from 19% to 21% elicits avoidance behaviors: animals reverse and change their direction of travel ( McGrath et al . , 2009; Persson et al . , 2009 ) . If after these maneuvers C . elegans fails to find an environment with lower [O2] , they become highly active , suggesting a simple form of arousal associated with escape behavior ( Busch et al . , 2012 ) . This activated state is sustained for at least 2 hr , and potentially until animals locate an environment with lower O2 levels ( Busch et al . , 2012 ) . Avoidance and escape from 21% O2 are driven principally by O2-sensing neurons called URX , AQR , and PQR ( Gray et al . , 2004; Cheung et al . , 2005; Zimmer et al . , 2009; Busch et al . , 2012; Couto et al . , 2013 ) . The pair of URX neurons is the most important , and is necessary and sufficient to mediate these responses . In these neurons , rising O2 stimulates an atypical O2-binding soluble guanylate cyclase composed of GCY-35 and GCY-36 subunits ( gcy , guanylate cyclase ) , leading to cGMP channel opening , and activation of L-type Ca2+ channels ( Cheung et al . , 2004; Gray et al . , 2004; Zimmer et al . , 2009; Busch et al . , 2012; Couto et al . , 2013 ) . The response of URX neurons ( as well as AQR and PQR ) to 21% O2 is tonic , that is , non-adapting , and alters other behaviors besides promoting rapid movement . At 19% O2 , C . elegans strongly avoids carbon dioxide ( CO2 ) , but at 21% O2 tonic URX signaling suppresses this avoidance ( Carrillo et al . , 2013; Kodama-Namba et al . , 2013 ) . Tonic signaling from the O2 sensors induces C . elegans to leave depleting food patches at much higher rates in 21% O2 ( Lima and Dill , 1990; Milward et al . , 2011 ) . This behavior is consistent with ecological studies of many species showing that animals tend to leave foraging sites when threat levels rise ( Lima and Dill , 1990 ) . The O2-sensing neurons also promote aggregation and accumulation where bacterial food is thickest ( Coates and de Bono , 2002; Gray et al . , 2004; Rogers et al . , 2006 ) . Besides reconfiguring behavior , O2-sensing neurons alter physiology: they regulate lifespan ( Liu and Cai , 2013 ) and body size ( Mok et al . , 2011 ) . In summary , C . elegans perceives 21% O2 as a threat , and responds to it via a discrete set of O2-sensing neurons whose tonic activity coordinates an altered global organismic state . The striking behavioral switch observed in natural C . elegans isolates at 21% O2 cannot be studied in the laboratory reference strain , N2 ( Bristol ) , due to a gain-of-function mutation in a FMRFamide-like peptide ( FLP ) receptor , called NPR-1 ( neuropeptide receptor family ) that arose during domestication ( de Bono and Bargmann , 1998; Rockman and Kruglyak , 2009; Weber et al . , 2010 ) . Knocking out npr-1 restores to N2 ( Bristol ) animals strong responses to 21% O2 . NPR-1 is expressed in about 20 neural types , including the O2-sensing neurons AQR , PQR , and URX , and their post-synaptic partners RMG and AUA . The NPR-1 215V receptor alters the function of several of these neurons , including URX and RMG , although the mechanisms involved are unclear ( Coates and de Bono , 2002; Macosko et al . , 2009; Carrillo et al . , 2013; Kodama-Namba et al . , 2013 ) . Here , we investigate how O2-sensing neurons evoke a change in global C . elegans state in response to 21% O2 . We trace information flow from a defined entry point—the URX O2 sensors—to downstream circuits that implement the change in state .
We hypothesized that tonic signaling from the O2-sensing neurons URX , AQR , and PQR could reprogram gene expression according to ambient O2 levels . To investigate this possibility , we analyzed the transcriptomes of npr-1 ( null ) mutants ( referred to as npr-1 ) and gcy-35; npr-1 young adult animals grown at 21% and 7% O2 using RNA sequencing ( see ‘Materials and methods’ ) . URX , AQR , and PQR neurons do not respond to O2 changes in gcy-35; npr-1 mutants , and exhibit activity levels found in npr-1 animals kept at 7% O2 ( Persson et al . , 2009; Zimmer et al . , 2009; Busch et al . , 2012 ) . To prevent npr-1 animals from aggregating , which induces gene expression changes that would confound our analysis ( Andersen et al . , 2014 ) , we grew animals at low density . We carried out three sets of comparisons . To identify genes whose expression is O2-modulated we compared the transcriptomes of npr-1 animals grown at 21% and 7% O2 ( Supplementary file 1 ) . To identify genes regulated by GCY-35-dependent signaling , we compared the transcriptomes of npr-1 and gcy-35; npr-1 animals grown at 21% O2 ( Supplementary file 2 ) . And to identify and exclude genes whose expression is altered by growth at different O2 levels independently of gcy-35 , we compared the transcriptomes of gcy-35; npr-1 animals grown at 21% and 7% O2 ( Supplementary file 3 ) . We then intersected the three comparisons , using as a cut-off q value of 0 . 05 ( see ‘Materials and methods’ ) . If our hypothesis was correct , many of the genes identified as differentially expressed in Supplementary file 1 ( npr-1 21% O2 vs npr-1 7% O2 ) should also be identified in Supplementary file 2 ( npr-1 21% O2 vs gcy-35; npr-1 21% O2 ) but not in Supplementary file 3 ( gcy-35; npr-1 21% O2 vs gcy-35; npr-1 7% O2 ) . Consistent with this , we found that 152/210 genes differentially expressed between npr-1 7% O2 versus npr-1 21% O2 were also found in the gcy-35; npr-1 21% O2 versus npr-1 21% O2 comparison . Expression of 72/152 of these genes was not significantly altered by O2 experience in gcy-35; npr-1 animals ( Supplementary file 4 ) . For almost all of these genes , 71/72 , disrupting gcy-35 had a similar effect on gene expression as growing npr-1 animals at 7% O2 ( Supplementary file 4 ) . Genes whose expression was regulated by O2 experience in a GCY-35-dependent way encoded proteins involved in metabolism ( e . g . , elo-6 , a polyunsaturated fatty acid elongase; folt-2 , a folate transporter; mai-2 , a mitochondrial intrinsic ATPase inhibitor protein ) , oxidation–reduction reactions ( e . g . , the cytochrome p450s , cyp-35C1 and cyp-34A2 ) , and proteolysis ( e . g . , the cathepsins cpr-4 and cpr-6 ) , suggesting broad changes not limited to the nervous system ( Supplementary file 4 ) . Efficient escape from aversive environments requires coordinated movement that avoids conflicting behaviors . When exposed to 21% O2 in the presence of food , npr-1 animals not only increased their speed of movement ( Busch et al . , 2012 ) ( Figure 1A ) , but also increased the persistence of their forward travel by inhibiting spontaneous short reversals ( Figure 1B , C ) . Mutations in gcy-35 , which abolish the O2 responsiveness of URX , AQR , and PQR neurons , disrupted O2 modulation of both speed ( Busch et al . , 2012 ) and spontaneous reversal frequency ( Figure 1B ) . Expressing gcy-35 selectively in URX neurons restored rapid movement and sustained inhibition of reversals at 21% O2 . URX neurons thus provide a defined entry point to the circuit controlling the behavioral state switch evoked by 21% O2 in npr-1 animals . 10 . 7554/eLife . 04241 . 003Figure 1 . RMG activation induces rapid and persistent forward movement . ( A and B ) URX O2 sensors provide an entry point to the circuit controlling response to 21% O2 . Selective expression of GCY-35 in URX neurons restores rapid ( A ) and persistent ( B ) forward movement at 21% O2 to gcy-35; npr-1 animals on food . Statistics compare rescued ( blue ) and mutant ( red ) animals at time points indicated by the black bars . ( C ) 21% O2 causes npr-1 animals on food to suppress the short , frequent reversals observed at 7% O2 . Reversal probability is calculated per 1 s . ( D ) Ablating URX abolishes the Ca2+ responses evoked in RMG by 21% O2; unc-64 syntaxin loss-of-function mutants show a partial reduction in this response . Here , and in subsequent panels , black bars indicate time intervals for statistical comparison of responses at 21% and 7% O2 using the Mann–Whitney U test . ***p<0 . 001; **p<0 . 01; *p<0 . 05; NS , not significant . ( E and F ) Stimulating RMG using channelrhodopsin evokes rapid movement ( E ) and inhibits backward movement ( F ) in npr-1 animals kept at either 7% or 21% O2 with food . Here and in subsequent panels , red bars indicate time intervals used for statistical comparisons of responses when light is on to when it is off . ***p<0 . 001 , **p<0 . 01; *p<0 . 05; NS , not significant . ( G–I ) Channelrhodopsin stimulation of RMG can induce rapid movement when AQR , PQR , and URX neurons are ablated . Data plotted in ( I ) are replotted from G and H . ( J ) URX Ca2+ responses evoked by 21% O2 are strongly attenuated when RMG is ablated . ( K ) URX neurons retain O2-evoked Ca2+ responses in unc-64 syntaxin mutants , although baseline Ca2+ is reduced . Each line in this panel ( and , unless specified , in subsequent panels ) represents the mean response of all animals of one genotype or condition . Error bars ( lighter shading ) in all panels show standard error of the mean . Gray areas indicate periods of higher O2 concentration; blue areas indicate periods with blue light ( 0 . 26 mW/mm2 ) on; orange areas indicate periods with green light ( 0 . 64 mW/mm2 ) on . DOI: http://dx . doi . org/10 . 7554/eLife . 04241 . 00310 . 7554/eLife . 04241 . 004Figure 1—figure supplement 1 . Simplified circuitry associated with the URX , AQR , and PQR neurons dissected in this paper . DOI: http://dx . doi . org/10 . 7554/eLife . 04241 . 004 Among the three O2-sensing neurons , URX neurons uniquely make gap junctions and reciprocal synaptic connections with the RMG interneurons ( White et al . , 1986; wormwiring . org; Figure 1—figure supplement 1 ) . The RMG interneurons respond to a 7–21% rise in O2 with a sustained increase in Ca2+ , and ablating RMG disrupts behavioral responses evoked by this O2 switch ( Busch et al . , 2012 ) . To show that O2-evoked Ca2+ responses in RMG reflect input from URX , we ablated URX and imaged RMG using the YC2 . 60 Ca2+ sensor ( Horikawa et al . , 2010 ) . When URX was ablated , O2 stimuli failed to evoke Ca2+ responses in RMG ( Figure 1D ) , and RMG Ca2+ resembled that found in tax-4 mutants , which lack the cGMP-gated channel required for URX , AQR , and PQR neurons to transduce O2 stimuli ( Figure 1D ) . These data suggest that O2-evoked Ca2+ responses in RMG are driven by URX input . Is depolarizing RMG sufficient to switch C . elegans behavioral state ? To address this we used channelrhodopsin-2 ( ChR2; Nagel et al . , 2005 ) to selectively stimulate RMG in npr-1 animals kept at 7% O2 . Stimulating RMG inhibited spontaneous reversals and induced rapid and persistent forward movement for as long as blue light was on ( Figure 1E , F ) . Stimulating RMG neurons is thus sufficient to confer a highly active locomotory state on npr-1 animals kept at 7% O2 . In npr-1 animals , URX neurons respond in a graded manner to changes in O2 between 7% and 21% , and evoke graded increases in locomotory rate according to final O2 concentration ( Cheung et al . , 2005; Busch et al . , 2012 ) . To investigate the dynamic range of URX–RMG signaling , and to examine if graded RMG activity can evoke graded changes in the animal's speed , we stimulated RMG in animals kept at either 7% O2 or 21% O2 . The effects of activating RMG using ChR2 summated with input from the O2 sensory circuit: animals kept at 21% reached higher speeds and inhibited reversals more strongly than animals kept at 7% ( Figure 1E , F ) . These data suggest that the dynamic range of the circuit permits higher levels of RMG activation to evoke qualitatively similar but quantitatively stronger behavioral responses . RMG and URX are connected by gap junctions and by reciprocal chemical synapses . Stimulating RMG using ChR2 could therefore alter behavior by activating URX . To test this , we selectively activated RMG using ChR2 in animals ablated for URX ( and AQR and PQR ) . Ablated animals robustly increased their speed and suppressed reversals upon light-activation of RMG , both at 7% and at 21% O2 ( Figure 1G , H ) . As expected if O2 input was lost , the speed responses of the ablated animals were not influenced by the O2 levels ( Figure 1I ) . The ability of RMG stimulation to induce rapid movement and inhibit reversals in the absence of URX , AQR , and PQR neurons suggests that feed-forward signaling from RMG to other neurons evokes these behaviors . Our results did not exclude that RMG influences URX activity . To investigate this possibility , we ablated RMG neurons in L1 larvae and imaged URX Ca2+ responses in young adults 2 days later . npr-1 animals lacking RMG showed a much smaller URX Ca2+ response to a 7–21% O2 stimulus compared to mock ablated controls ( Figure 1J ) . Thus , not only does URX activate RMG in response to rising O2 , but RMG input somehow regulates URX excitability . Although the gap junctions and synaptic connections between URX and RMG provide a direct route for such communication , we cannot exclude more indirect mechanisms . However , consistent with a role for synaptic input in sustaining URX excitability , partial loss-of-function mutants of unc-64 syntaxin , which have deficits in synaptic release , showed reduced Ca2+ levels in URX and RMG both at 7% and 21% O2 ( Figure 1D , K ) . URX neurons could still evoke Ca2+ responses in RMG in unc-64 mutants ( Figure 1D ) . However , since the unc-64 allele we used is a partial loss-of-function ( null mutants are dead ) , we cannot draw firm conclusions about the relative roles of gap junctions and synaptic transmission in mediating URX–RMG communication . Excessive signaling from a hyperactive neuropeptide receptor , NPR-1 215V , prevents N2 ( Bristol ) animals from moving rapidly on food at 21% O2 ( de Bono and Bargmann , 1998; Cheung et al . , 2005 ) . The RMG neurons are one site of action for this receptor: expressing NPR-1 215V selectively in RMG reduces the locomotory activity of npr-1 animals ( Macosko et al . , 2009 ) . Does NPR-1 215V expression in RMG inhibit responses evoked by a switch from 7% to 21% O2 ? npr-1 animals that expressed NPR-1 215V in RMG did not exhibit long-lasting changes in locomotion in response to changes in O2 ( Figure 2 , Figure 2—figure supplement 1A , B ) . However , they retained the transient bout of reversals and reorientation triggered by a sharp rise in O2 ( Figure 2—figure supplement 1C ) . Thus , NPR-1 215V receptor signaling in RMG selectively suppresses the sustained switch in locomotory behavior evoked by high and low O2 . A different circuit may act downstream of O2 sensors to evoke transient avoidance responses , consistent with the observation that activating RMG using ChR2 failed to induce reversals . 10 . 7554/eLife . 04241 . 005Figure 2 . FLP-21/NPR-1 ligand/receptor signaling limits RMG circuit output downstream of RMG cell body Ca2+ . ( A–C ) Disrupting the flp-21 FMRF-like neuropeptide enables N2 animals to respond to 21% O2 with a persistent increase in locomotory activity ( A ) , and increases the amplitude of such responses in animals expressing the natural npr-1 215F receptor allele ( B ) but not in npr-1 ( ky13null ) mutants ( C ) . ( D ) The npr-1 215V allele only slightly reduces the Ca2+ responses evoked by 21% O2 in RMG . ( E ) RMG stimulation using ChR2 only weakly stimulates movement in N2 animals , both at 7% and 21% O2 , contrasting with its effects in npr-1 animals . ( F–H ) Knocking out eat-16 , a member of the RGS7 family that inhibits Gq signaling , or its interacting partner rsbp-1 , enables N2 animals to switch to rapid movement at 21% O2 . DOI: http://dx . doi . org/10 . 7554/eLife . 04241 . 00510 . 7554/eLife . 04241 . 006Figure 2—figure supplement 1 . npr-1 GPCR signaling inhibits long-lasting but not transient responses to 21% O2 in feeding animals . npr-1 215V signallingsignaling in RMG prevents 21% O2 from evoking persistent rapid movement ( A ) , and sustained inhibition of reversals ( B ) , but does not affect the initial bout of reversals evoked by a rise in O2 ( C ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04241 . 00610 . 7554/eLife . 04241 . 007Figure 2—figure supplement 2 . Differences in YC2 . 60 sensor expression may partially account for differences in YFP/CFP ratios in RMG between N2 and npr-1 animals . ( A ) Average YFP/CFP ratios in RMG at 21% O2 in N2 animals and npr-1 mutants plotted against average YFP expression levels . ( B ) Scatter plot of YFP/CFP ratios in RMG at 21% O2 against YFP expression levels for individual animals shows a positive correlation . *** , p < 0 . 001 . YC2 . 60 expression is restricted to RMG using a Cre–Lox combination in which sensor expression is driven from the flp-21 promoter ( see ‘Materials and methods’ ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04241 . 007 The FMRFamide neuropeptide encoded by the flp-21 gene is an in vivo ligand for NPR-1 ( Rogers et al . , 2003 ) . flp-21 is expressed in several neurons , including RMG itself ( Kim and Li , 2004; Macosko et al . , 2009 ) . FLP-21 peptide release could therefore serve to limit the O2-evoked behavioral state switch by activating NPR-1 signaling in RMG . Consistent with this , deleting flp-21 caused animals expressing the natural npr-1 215F or derived npr-1 215V allele , but not npr-1 ( null ) mutants , to increase the amplitude of their O2-evoked behavioral switch ( Figure 2A–C ) . How does the NPR-1 neuropeptide receptor inhibit RMG function ? Using the YC2 . 60 sensor , we compared the Ca2+ responses evoked in RMG by a 7–21–7% O2 stimulus train in N2 and npr-1 mutants . N2 animals exhibited reduced Ca2+ responses in RMG compared to npr-1 mutants ( Figure 2D ) , which would be consistent with NPR-1 215V inhibiting RMG Ca2+ signaling . However , the effects , although statistically significant , were surprisingly small compared to the strong inhibitory effects of NPR-1 215V signaling on O2-evoked behaviors . More importantly , the promoter combination targeting YC2 . 60 expression to RMG ( pncs-1::cre and pflp-21::LoxP-STOP-LoxP::YC2 . 60 ) drove lower expression in N2 compared to npr-1 animals ( Figure 2—figure supplement 2A ) . Lower sensor expression correlated with lower YFP/CFP ratio at 21% O2 ( Figure 2—figure supplement 2B ) , suggesting that sensor expression differences could account for much of the YFP/CFP ratio differences between N2 and npr-1 animals ( see below for an explanation of why pflp-21 drove lower RMG expression in N2 compared to npr-1 animals ) . These observations led us to speculate that the main inhibitory role of NPR-1 was downstream of Ca2+ influx . To test this , we compared the effects of stimulating RMG neurons with ChR2 in N2 and npr-1 animals . If NPR-1's inhibitory role is predominantly downstream of Ca2+ entry , ChR2-activation of RMG should have little effect in N2 animals . Consistent with this , light activation of RMG triggered only a small increase in N2 locomotory activity , regardless of whether animals were kept at 21% or 7% O2 , and in marked contrast to its effects in npr-1 mutants ( Figure 2E ) . Most compellingly , although npr-1 mutants kept at 7% O2 had much lower Ca2+ in RMG compared to N2 animals kept at 21% O2 ( Figure 2D ) , they responded to RMG ChR2 activation significantly more strongly ( Figure 2E ) . These results suggest that the inhibitory effects of npr-1 215V occur predominantly downstream of the Ca2+ responses in RMG , possibly at the presynaptic level . A network of G protein pathways modulates neurotransmission presynaptically: Gq and Gs signaling stimulates neurotransmitter release whereas Go/i signaling inhibits it , possibly by negatively regulating Gs and Gq signaling ( Miller et al . , 1999; Nurrish et al . , 1999 ) . In vitro work suggests that the NPR-1 215V and NPR-1 215F receptors can each couple to Go/i signaling following stimulation by FLP-21 ( Rogers et al . , 2003 ) . If NPR-1 215V inhibits neurotransmission by activating Go , then mutations that promote Gq/Gs signaling over Go signaling should restore O2 control of locomotory state and mimic npr-1 mutants . The RGS ( regulator of G protein signaling ) protein EAT-16 terminates Gq signaling by activating intrinsic Gq GTPase activity ( Hajdu-Cronin et al . , 1999 ) ; RSBP-1 ( R7 binding protein 1 homolog ) interacts with and is required for EAT-16 activity ( Porter and Koelle , 2010 ) . Disrupting either eat-16 or rsbp-1 enabled N2 animals , which express npr-1 215V , to behave like npr-1 mutants in response to a change in O2 ( Figure 2F–H ) . Together , these results suggest that a change in the balance of Go/Gq signaling in a subset of neurons explains how the NPR-1 215V neuropeptide receptor controls behavior . Animals integrate information across multiple sensory modalities to respond appropriately to changing environments . The ASK sensory neurons respond to pheromones and food ( Macosko et al . , 2009; Wakabayashi et al . , 2009 ) , and , like the URX O2 sensors , make gap junctions with RMG ( White et al . , 1986; wormwiring . org; Figure 1—figure supplement 1 ) . Previous work suggested that cGMP signaling in ASK is required for npr-1 animals to move rapidly and to aggregate ( Tremain , 2004; Macosko et al . , 2009 ) , leading to the hypothesis that ASK neurons are a major output of the RMG hub-and-spoke circuit ( Macosko et al . , 2009 ) . The ASH and ADL nociceptive neurons also make gap junctions with RMG ( Figure 1—figure supplement 1 ) . Previous work suggested that TRPV signaling in ASH and ADL promotes aggregation behavior and avoidance of high O2 ( de Bono et al . , 2002; Chang et al . , 2006; Rogers et al . , 2006 ) . Together , these results suggested that gap-junctional communication across the RMG circuit integrates multiple sensory cues and is necessary for npr-1 animals to move rapidly at 21% O2 and to aggregate ( Macosko et al . , 2009 ) . To investigate this model in the context of O2-evoked responses , we used the YC3 . 60 reporter to examine if changing [O2] altered [Ca2+] in ASK or ASH neurons in npr-1 animals . A switch from 7% to 21% O2 elicited a small increase in YFP/CFP FRET in ASK and ASH , indicating a rise in [Ca2+] ( Figure 3A , B ) . The responses were sustained while animals were at 21% O2 . Deleting gcy-35 abolished the O2-evoked Ca2+ responses in ASK and ASH ( Figure 3A , B ) , consistent with these responses being driven by URX , and therefore potentially via RMG . These results indicate that O2 input can indeed modify ASK and ASH Ca2+ levels , but the effect is small—at least at the cell body , where we made our measurements . 10 . 7554/eLife . 04241 . 008Figure 3 . ASK , ASH , and ADL neurons are not necessary for the RMG circuit to stimulate rapid movement at 21% O2 . ( A and B ) In npr-1 animals a 7% O2 to 21% O2 stimulus evokes a small rise in Ca2+ in ASK ( A ) and ASH ( B ) neurons . These responses are abolished in gcy-35; npr-1 animals . ( C and D ) Ablating ASK ( C ) , or acutely inhibiting its activity using halorhodopsin ( D ) did not alter the locomotory behavior of npr-1 animals on food at 7% or 21% O2 . ( E ) Stimulating RMG using ChR2 can stimulate locomotion in npr-1 animals kept at 7% O2 in the absence of ASH and ADL neurons . ( F ) Inhibiting synaptic release from ASH and ADL using tetanus toxin disrupts avoidance of Cu2+ ( inset ) but does not inhibit the behavioral state switch evoked by changing O2 . DOI: http://dx . doi . org/10 . 7554/eLife . 04241 . 00810 . 7554/eLife . 04241 . 009Figure 3—figure supplement 1 . Ablating ASK neurons does not reduce aggregation behavior of npr-1 animals . DOI: http://dx . doi . org/10 . 7554/eLife . 04241 . 009 To ask if ASK neurons contributed to O2-evoked behavioral states , we ablated ASK in npr-1 animals by targeted expression of the egl-1 cell death gene ( Wakabayashi et al . , 2004 ) . ASK ablation was confirmed by dye filling and using a psra-9::YC3 . 60 fiduciary marker . npr-1 animals lacking ASK neurons responded to a 7–21–7% O2 regime indistinguishably from control npr-1 animals ( Figure 3C ) . Optogenetic inhibition of ASK , in an npr-1 strain expressing halorhodopsin ( NpHR ) specifically in ASK , had no effect on speed ( Figure 3D ) . Ablating ASK neurons in npr-1 animals also did not impair aggregation behavior ( Figure 3—figure supplement 1 ) . Aggregation is highly sensitive to O2 circuit function and is disrupted in gcy-35 mutants ( Cheung et al . , 2004; Gray et al . , 2004; Rogers et al . , 2006 ) . These data suggest that ASK neurons are not necessary for O2-evoked behavioral responses , or for aggregation . To investigate if ASH and ADL are required for the O2-controlled behavioral state switch , we ablated these neurons in npr-1 animals expressing channelrhodopsin in RMG , using targeted expression of a miniSOG gene and light-induced ablation ( Qi et al . , 2012 ) . The ASH and ADL ablated animals responded to high and low O2 and to current injection into RMG like npr-1 controls ( Figure 3E ) . Tetanus toxin in the ASH and ADL neurons also did not disrupt O2 responses ( Figure 3F ) . To monitor toxin expression , we used a polycistronic construct that also expressed RFP , and we confirmed that tetanus toxin disrupted synaptic release from ASH and ADL by monitoring reversals in response to a 3 mM Cu2+ drop ( Figure 3F , inset ) . Thus , ablating ASK alone , or ASH and ADL together , does not disrupt relay of O2-modulated RMG activity to downstream circuits that promote rapid movement . The ASH and ADL neurons express the TRPV1 homolog ocr-2 , and disrupting ocr-2 , or its partner subunit osm-9 , attenuates animals' ability to navigate spatial O2 gradients and to aggregate ( de Bono et al . , 2002; Chang et al . , 2006; Rogers et al . , 2006 ) . ocr-2; npr-1 animals only weakly modulated their locomotory activity when switched between 7% and 21% O2 ( Figure 4A ) . Expressing ocr-2 cDNA in ASH or ADL or the serotonergic ADF neurons increased the speed of ocr-2; npr-1 animals at 21% O2 and , as expected , rescued the bordering and aggregation phenotype ( Figure 4A , Figure 4—figure supplement 1A ) ( de Bono et al . , 2002; Chang et al . , 2006 ) . The ability of ocr-2 expression in ASH or ADL neurons to rescue the speed response and aggregation behavior of ocr-2; npr-1 animals was retained when neurotransmission from these neurons was blocked using tetanus toxin ( Figure 4B , C , Figure 4—figure supplement 1B ) . These data suggest that ocr-2 TRPV activity in ASH and ADL is unlikely to promote O2 responses by facilitating these neurons' neurosecretory activity , but may instead influence their gap junctional communication , for example , with RMG . 10 . 7554/eLife . 04241 . 010Figure 4 . Disrupting a TRPV channel in ASH , ADL , and ADF neurons attenuates locomotory responses to 21% O2 . ( A ) The switch to rapid movement evoked in npr-1 animals by 21% O2 is attenuated in the absence of the OCR-2 TRPV channel . This defect of ocr-2; npr-1 animals can be rescued to varying extents by expressing ocr-2 cell-specifically in ASH , ADL , or ADF neurons . ( B and C ) The ability of pADL::ocr-2 ( B ) or pASH::ocr-2 ( C ) transgenes to rescue the ocr-2; npr-1 O2 phenotype does not depend on synaptic release . Synaptic release was inhibited by expressing tetanus toxin in ASH and ADL neurons , using the gpa-11 promoter . ( D ) Channelrhodopsin stimulation of ASH neurons restores modulation of locomotory activity by 21% O2 to ocr-2; npr-1 animals . ( E and F ) 2 mM Cu2+ elicited strong Ca2+ responses in ASH neurons of N2 and npr-1 animals ( E ) but did not alter Ca2+ in the RMG neurons ( F ) . ( G and H ) Disrupting ocr-2 did not significantly reduce O2-evoked Ca2+ responses in the cell bodies of URX ( G ) or RMG ( H ) neurons . ( I–K ) Stimulating RMG in ocr-2; npr-1 animals kept on food using channelrhodopsin can partly restore O2 modulation of locomotory activity . DOI: http://dx . doi . org/10 . 7554/eLife . 04241 . 01010 . 7554/eLife . 04241 . 011Figure 4—figure supplement 1 . OCR-2 expression in ASH or ADL can restore aggregation behavior to ocr-2; npr-1 animals ( A ) even when synaptic transmission in these neurons is inhibited by tetanus toxin expression ( B ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04241 . 011 The OCR-2/OSM-9 TRPV channel is required for the ASH and ADL neurons to respond to sensory inputs ( Hilliard et al . , 2004 ) . If disrupting ocr-2 reduces ASH and ADL tonic or chronic activity , it could decrease RMG activity by reducing current input into RMG , or by shunting current away from RMG through the ASH–RMG and ADL–RMG gap junctions ( Figure 1—figure supplement 1 ) . If this model is correct , injecting current into ASH or ADL using ChR2 should restore to ocr-2; npr-1 mutants rapid movement at 21% O2 . To test this , we expressed ChR2 specifically in ASH by using the FLP/FRT system and by monitoring expression using YFP-tagged ChR2 ( see ‘Materials and methods’ ) . We exposed transgenic animals to constant blue light of different intensities , ranging from 0 . 005 to 0 . 05 mW/mm2 . As expected , control animals not exposed to blue light did not strongly modulate their speed when switched between 7% and 21% O2 ( Figure 4D ) . By contrast , animals exposed to 0 . 005 , 0 . 015 or 0 . 05 mW/mm2 of continuous blue light modulated their locomotory state according to O2 levels ( Figure 4D ) . The amplitude of the O2-evoked change in locomotory activity was similar to that obtained when we selectively restored ocr-2 expression to ASH in ocr-2; npr-1 animals ( compare Figure 4C , D ) . Thus , injecting a constant Ca2+/Na+ current into ASH neurons is sufficient to restore O2 modulation to the circuit in ocr-2; npr-1 animals . This suggests that the effects of the ocr-2 mutation are not developmental , as they can be rescued in adult animals . Our results support a model in which tonic or chronic ASH ( and ADL ) activity facilitates O2-evoked behavioral switches either by injecting current into RMG or by reducing shunting from the RMG circuit . A rise in O2 evoked a small increase in ASH Ca2+ ( Figure 3B ) , suggesting that depolarizing current flows from RMG to ASH in npr-1 animals at 21% O2 . To test explicitly if current can flow in the reverse direction , from ASH to RMG , we asked if strongly activating ASH using a noxious stimulus evoked a rise in RMG Ca2+ . As expected , a 10 mM Cu2+ stimulus resulted in a large , sharp rise in ASH Ca2+ that was easily detectable using YC3 . 60 ( Hilliard et al . , 2004 ) ( Figure 4E ) . By contrast , this stimulus failed to evoke any Ca2+ responses in the RMG cell body that could be measured with YC2 . 60 ( Figure 4F ) . These results suggest that the anatomically defined gap junctions may not allow significant Ca2+ current to flow from the ASH to the RMG cell body under our imaging conditions . How , then , does OCR-2 promote RMG activity ? An alternative model is that by tonically/chronically depolarizing ASH and ADL neurons , OCR-2 reduces the current flowing from RMG to ASH and ADL via gap junctions . In ocr-2; npr-1 mutants , a more negative membrane potential in ASH and ADL leads to more current being shunted from RMG to the nociceptive neurons , reducing RMG activity . To test this hypothesis , we compared O2-evoked Ca2+ responses in URX and RMG in npr-1 and ocr-2; npr-1 animals ( Figure 4G , H ) . Surprisingly , the steady state Ca2+ levels in the URX or RMG cell bodies were not significantly affected by the ocr-2 mutation . Since the tetanus toxin experiments suggested that rescue of the ocr-2; npr-1 phenotype by a pASH::ocr-2 transgene did not require synaptic transmission , we speculate that the Ca2+ effects of disrupting ocr-2 are local to the gap junctions and not visible at the RMG cell body . Alternatively , ASH and ADL have the potential to leach away some other excitatory factor from RMG via gap junctions , and OCR-2 activity in ASH and ADL can attenuate this . If loss of the OCR-2 TRPV channel somehow reduced the functionality of RMG , injecting current directly into RMG should rescue the phenotype of ocr-2; npr-1 animals . Consistent with this , ChR2-activation of RMG in ocr-2; npr-1 animals stimulated rapid movement , and inhibited reversals , both at 21% and 7% O2 ( Figure 4I , J and data not shown ) . This contrasts with the failure of ChR2-driven RMG activation to drive tonic changes in behavioral state in N2 animals ( which express the NPR-1 215V receptor ) ( Figure 2F ) . These results suggest that the OCR-2-expressing neurons can facilitate RMG signaling in a way that is upstream of NPR-1 signaling . Genetically encoded Ca2+ sensors are blind to signaling mechanisms that modulate neurosecretion without altering Ca2+ ( Miller et al . , 1999; Nurrish et al . , 1999; Rhee et al . , 2002; Oda et al . , 2011 ) . This limitation prompted us to seek a way to monitor enduring changes in neurosecretion as readouts of different global states . In vertebrates , several studies have reported that increased neural activity is associated with increased neuropeptide gene transcription ( Uhl and Nishimori , 1990 ) . In pancreatic β cells , feedback mechanisms couple insulin production and release ( Borgonovo et al . , 2006 ) . If transcription of most neuropeptide genes is coupled to peptide release , promoter::GFP reporters for such genes should provide a readout of peptidergic circuit activity . To test this , we first studied a pflp-11::GFP reporter expressed in the URX , AUA , and SAB neurons ( Kim and Li , 2004 ) . URX and AUA both show tonically elevated Ca2+ at 21% O2 compared to 7% O2 ( Busch et al . , 2012 ) . Cultivating animals in 7% O2 , or deleting gcy-35 , reduced pflp-11::gfp expression in both URX and AUA ( Figure 5A ) . By contrast , knocking out ocr-2 did not , suggesting that TRPV signaling did not alter URX or AUA peptide expression ( Figure 5A ) . The genotype at the npr-1 locus did not affect pflp-11::gfp expression in URX , but expression in the AUA neurons was reduced in animals encoding the npr-1 215V allele compared to npr-1 null mutants ( Figure 5A ) . pflp-11::gfp expression in SAB , a neuron not known to be modulated by O2 , was not affected either by O2 experience or the genotypes tested , providing an internal control . These data are consistent with pflp-11 expression reporting neural activity . 10 . 7554/eLife . 04241 . 012Figure 5 . Neuropeptide gene expression levels report neurosecretory activity . ( A ) Expression of a pflp-11 reporter in URX , AUA , and SAB neurons in different genotypes and at different O2 tensions . In this and subsequent panels , AU = arbitrary units . ( B ) Expression of pflp-11::GFP in URX is inversely related to expression of a gain-of-function K+ channel , EGL-2 ( GF ) , in the same neuron . Expression of the K+ channel can be tracked due to co-expression of RFP in an operon . ( C ) Blocking exocytosis from URX by targeted expression of tetanus toxin strongly reduces expression of pflp-11::GFP in URX . Expression of tetanus toxin can be tracked due to co-expression of RFP in an operon . ( D ) Expression of a pflp-5 reporter in RMG , ASE , and M4 in different genotypes and at different O2 tensions . ( E ) Blocking exocytosis from RMG by targeted expression of a tetanus toxin strongly reduces expression of pflp-5::GFP in RMG . Expression of tetanus toxin can be tracked due to co-expression of RFP in an operon . ( F ) Expression of a pflp-21 reporter that includes 4 kb of upstream sequences in URA ( probably ) , RMH , ASJ , M2 , FLP , and RMG neurons in different genotypes and at different O2 tensions . ( G ) Expression of a pflp-21 reporter that includes 2 kb of upstream sequences in M2 , URA ( probably ) , ASH , ADL , and ASJ neurons in different genotypes and at different O2 tensions . ( H ) Blocking exocytosis from ASH by targeted expression of a tetanus toxin reduces expression of pflp-21 ( 2 kb ) ::GFP in ASH . Expression of tetanus toxin can be tracked due to co-expression of RFP in an operon . ( I–K ) Expression of ppdf-1::gfp ( I ) , pflp-17::gfp ( J ) or pflp-3::gfp ( K ) reporters in different neurons in different genotypes at different O2 concentrations . For ( A , D , F , G , I , J , and K ) , asterisks indicate comparisons to npr-1 animals kept at 21% O2 , *p<0 . 05; ***p<0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 04241 . 012 To test more explicitly if modulation of pflp-11::gfp expression levels was related to neuron depolarization state , we inhibited URX by expressing a constitutively active K+ channel related to Drosophila ether-a-go , EGL-2 ( GF ) ( Weinshenker et al . , 1999 ) . This transgene abolishes behavioral responses evoked by 21% O2 ( Cheung et al . , 2005 ) . EGL-2 ( GF ) expression strongly reduced pflp-11::gfp expression in URX ( Figure 5B ) . We next tested if feedback control of pflp-11::gfp expression was coupled to peptide release downstream of Ca2+ . To selectively block neurosecretion from URX , we cell-specifically expressed tetanus toxin , which cleaves synaptobrevin ( Schiavo et al . , 2000 ) . Tetanus toxin expression strongly reduced pflp-11::gfp expression in URX , and the effect was stronger when toxin expression was higher ( Figure 5C ) . These results suggest that pflp-11::gfp expression is coupled to peptide release in URX , and that peptide gene reporters may be useful surrogates to monitor long-term neurosecretory activity . To test this hypothesis further , we examined a pflp-5::GFP transgene expressed in the RMG , ASE , and M4 neurons ( Kim and Li , 2004 ) . pflp-5::gfp expression in RMG was higher in npr-1 animals grown at 21% O2 than in npr-1 animals grown at 7% O2 or gcy-35; npr-1 mutants grown at 21% O2 ( Figure 5D ) . These data suggest that pflp-5::gfp reporter expression is also coupled to neural activity . In contrast , expression of pflp-5::GFP in the M4 pharyngeal neuron , which is not known to respond to O2 , was not altered by O2 experience or by disrupting gcy-35 . To explore if pflp-5 expression in RMG , like pflp-11 expression in URX , was coupled to peptide release , we blocked RMG neurosecretion by cell-specifically expressing tetanus toxin . Animals expressing tetanus toxin in RMG showed significantly less pflp-5 expression in RMG than non-expressing siblings ( Figure 5E ) , consistent with neurosecretion feeding back to stimulate pflp-5 transcription . Expression of pflp-5::gfp in RMG was reduced in N2 animals compared to npr-1 mutants , both at 21% and 7% O2 . Since at 21% O2 RMG Ca2+ levels were not strikingly different between N2 and npr-1 mutants ( Figure 2E ) , our data support a model in which NPR-1 215V signaling acts either downstream of Ca2+ or locally to inhibit RMG neurosecretion . Interestingly , ocr-2; npr-1 animals kept at 21% O2 also showed reduced pflp-5::gfp expression compared to npr-1 worms kept at the same O2 levels ( Figure 5D ) , suggesting that disrupting ocr-2 reduced RMG neurosecretion in npr-1 animals . We next studied expression from a long version of the flp-21 promoter that drives GFP expression in URA , M4 , M2 , ASJ , RMH , and RMG ( Macosko et al . , 2009 ) . The RMH interneurons make gap junctions with RMG ( White et al . , 1986; wormwiring . org ) . Expression of the long pflp-21::GFP transgene in RMG and RMH was highest in npr-1 animals cultivated at 21% O2 , and reduced in gcy-35; npr-1 and ocr-2; npr-1 animals grown at 21% O2 , as well as npr-1 animals kept overnight at 7% O2 ( Figure 5F ) . pflp-21::GFP transgene expression in RMG and RMH was reduced in N2 animals compared to npr-1 mutants , both at 21% and 7% O2 , supporting a model in which NPR-1 215V signaling inhibits neurosecretion from these neurons ( Figure 5F ) . Thus , expression in RMG neurons driven from the long flp-21 promoter and the flp-5 promoter is regulated very similarly by O2 experience and genotype at the npr-1 and ocr-2 loci . Together , our data suggest that neurosecretion from RMG and RMH is tonically modulated by O2 levels and TRPV signaling . A shorter version of the flp-21 neuropeptide gene promoter drives GFP expression in the ADL , ASH , ASJ , URA , M4 , and M2 neurons ( Kim and Li , 2004 ) . We wondered if this promoter could provide readouts of ASH and ADL signaling . Disrupting the ocr-2 TRP channel subunit strongly reduced expression of the short pflp-21::gfp transgene in ASH and ADL in npr-1 animals ( Figure 5G ) . For ASH , we showed that expressing tetanus toxin also reduced pflp-21::gfp expression in this neuron ( Figure 5H ) . These results are consistent with tonic/chronic OCR-2-dependent neurosecretory activity in these neurons driving a feedback loop to promote transcription of neuropeptides . Changing O2 levels in npr-1 animals did not significantly alter expression of this shorter pflp-21::gfp in ASH and ADL , suggesting that O2 modulation of ASH and ADL secretion , if it occurs , is minor ( Figure 5G ) . Note that npr-1 is expressed in ASH neurons ( Figure 1—figure supplement 1 ) , which may account for differences in pflp-21::gfp expression in this neuron between npr-1 and N2 animals . These data are consistent with the ablation studies and the Ca2+ imaging data for ASH , which suggest that ASH neurons exhibit only weak O2-evoked changes in Ca2+ and are dispensable for the O2-evoked behavioral switch . Encouraged by our results , we looked for other neuropeptide genes that are reported to be expressed in neurons connected to the O2-sensing circuit by gap junctions or synapses . One such gene encodes the C . elegans ortholog of pigment dispersal factor , pdf-1 ( Barrios et al . , 2012; Meelkop et al . , 2012 ) . A ppdf-1::gfp transgene is expressed in PVP and AVB , two neurons heavily connected with O2-sensing neurons: PVP has gap junctions with AQR and PQR , whereas AVB has gap junctions with AQR and receives synaptic input from AQR , URX , and RMG . Expression of ppdf-1::gfp in both PVP and AVB was upregulated at 21% O2 compared to 7% O2 , suggesting that the activity of these neurons is also modulated by O2 ( Figure 5I and see below ) . The expression of ppdf-1::gfp in RMG from the transgene was too weak and variable to be assayed . Finally , we studied the expression of two additional peptide genes to demonstrate that not all neuropeptide promoters are stimulated by exposure to 21% O2 . The neuropeptide FLP-17 is expressed by BAG neurons and released when BAG is depolarized in vitro ( Ringstad and Horvitz , 2008; Smith et al . , 2013 ) . BAG neurons are stimulated by low O2 concentrations ( Zimmer et al . , 2009 ) . As expected , pflp-17::gfp expression in BAG was stimulated at 7% O2 compared to 21% O2 ( Figure 5J ) . The neuropeptide FLP-3 is expressed in the IL1 neurons ( Kim and Li , 2004 ) , which are part of a touch-responsive circuit , are poorly connected to known O2-modulated neurons , and do not express NPR-1 . pflp-3::gfp expression in IL1 was not regulated by O2 or npr-1 ( Figure 5K ) . Together , our data suggest that O2 levels regulate flp-11 , flp-5 , flp-21 , flp-17 , and pdf-1 peptide gene expression and peptide release in multiple neurons , including URX , AUA , RMG , RMH , BAG , PVP , and AVB , but not URA , SAB , IL1 , M2 , and M4 . They also suggest that peptide release from RMG is stimulated by O2-sensing neurons , depends on OCR-2 TRPV signaling , and is inhibited by NPR-1 215V signaling . Previous work has shown that RMG activity is necessary for high locomotory activity at 21% O2 ( Macosko et al . , 2009; Busch et al . , 2012 ) . We have shown that RMG activity is sufficient to drive high locomotory activity , and that ASH , ADL , and ASK are each dispensable for these effects . Does peptidergic signaling from RMG contribute to driving high locomotory activity at high O2 ? We first tested the effects of blocking all neurosecretion from RMG using cell-specific expression of a tetanus toxin transgene in npr-1 . To monitor expression , we used a polycistronic construct that co-expressed the toxin with RFP . Animals expressing the transgene lost any tonic response to high O2 and remained very poorly active at all O2 concentrations ( Figure 6A , B ) . This was expected from previous work ( Macosko et al . , 2009 ) , although that study did not explicitly examine O2-evoked responses . We next tested more specifically if neuropeptides secreted from RMG were required , using cell-specific RNAi to knock down the carboxypeptidase E ( CPE ) egl-21 in RMG . Carboxypeptidase E removes C-terminal lysine and arginine residues from pro-peptides during maturation , and analysis of peptide extracts from egl-21 mutants shows a deficit in the maturation of most C . elegans neuropeptides ( Husson et al . , 2007 ) . npr-1 animals with selective RNAi knockdown of egl-21 in RMG moved appropriately slowly in 7% O2 , but sped up much less than non-transgenic siblings when switched to 21% O2 ( Figure 6C ) and reversed more frequently ( Figure 6D ) . These results indicate that neuropeptide release from the RMG neurons plays a major role in evoking the highly active state of feeding C . elegans at 21% O2 . 10 . 7554/eLife . 04241 . 013Figure 6 . RMG neuropeptide secretion drives rapid movement at 21% O2 . ( A–D ) Selective expression of tetanus toxin ( A and B ) or RNAi knockdown of EGL-21 carboxypeptidase E ( C and D ) in RMG inhibits the O2-evoked switch in locomotory state . NS , not significant; ***p<0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 04241 . 013 Ultimately , changes in O2 levels modify behavioral state by altering motor circuits . We envisioned three ways by which O2 circuit output might alter downstream circuits . In one model , the downstream circuits would exhibit O2-evoked tonic changes in Ca2+ levels , much as we observe in the URX , AQR , PQR , AUA , and RMG neurons ( Busch et al . , 2012 ) . In a different model , behavioral state would be encoded across multiple neurons whose activity does not faithfully track O2 concentration , but which on average show O2-evoked changes in Ca2+ levels . To investigate these possibilities , we used Ca2+ indicators to image the activity of motoneurons , and of interneurons in layers upstream of motoneurons , at different O2 environments . The third possibility would not involve regulation of Ca2+ levels , but rather presynaptic effects that would only be observed using other reporters , such as the peptide promoter assay . We focused our studies on four sets of neurons: the A and B motoneurons , which are thought to mediate reverse and forward movement , respectively; the AVA ‘command’ interneurons , which promote backward movement by ensuring that the A motoneurons are more active than the B motoneurons; the AVB ‘command’ interneurons which promote forward movement by ensuring that the B motoneurons are more active than the A motoneurons; and the AIY interneurons which do not have anatomically defined connections with URX or RMG but are post-synaptic to multiple other sensory neurons ( White et al . , 1976 , 1986; Chalfie et al . , 1985; Kawano et al . , 2011; wormwiring . org; Figure 1—figure supplement 1 ) . As a control , we imaged the RMG neurons . We first measured spontaneous ( Figure 7A , B ) and O2-evoked ( Figure 7C , D ) Ca2+ changes in animals immobilized with 3 mM levamisole; levamisole immobilizes C . elegans by activating nicotinic acetylcholine receptors expressed in body wall muscle ( Lewis et al . , 1980 ) . As expected ( Kawano et al . , 2011 ) , Ca2+ levels in AVA and AVB neurons , and in the A and B neurons , were anti-correlated . As observed previously ( Schrödel et al . , 2013 ) , in unstimulated animals the bouts of activity in which neurons promoting backward movement ( the AVA and VA neurons ) were more active than neurons promoting forward movement ( the AVB and VB neurons ) often lasted for more than a minute before switching stochastically ( Figure 7A , B ) . This fictive behavior contrasted with the real behavior of animals moving on an agar plate , which only reverse for a few seconds . Nevertheless , the fictive behavior evoked by a rise in O2 recapitulated some of the expected response: the backward module ( VA neurons ) was reliably activated by O2 upsteps from 7% to 21% O2 ( Figure 7C ) . Moreover , when we exposed animals to a graded O2 series of 21–14–7–21% , we found that , despite frequent switches in activity , the average activity in AVB and AIY was modulated by O2 levels , with higher Ca2+ at 21% compared to 7% in both neurons ( Figure 7D ) . 10 . 7554/eLife . 04241 . 014Figure 7 . Spontaneous and O2-evoked Ca2+ responses in interneurons and motoneurons in immobilized animals . ( A and B ) Spontaneous bi-stable Ca2+ responses observed in VA and VB neurons ( A ) and in AVA and AVB neurons ( B ) in individual npr-1 animals immobilized with 3 mM levamisole and kept at 7% O2 . As expected , Ca2+ responses in VAs versus VBs ( A ) and AVA versus AVB ( B ) neurons were anti-correlated . Different colored shading and F and B indicate fictive forward ( F ) or backward ( B ) movement . ( C ) On average , an upstep from 7% to 21% O2 evoked a Ca2+ response in the VAs neurons of npr-1 animals immobilized with 3 mM levamisole . ( D ) Despite stochastic , high amplitude changes in Ca2+ levels of AIY and AVB neurons , on average , higher [O2] correlated with higher Ca2+ levels in AIY and AVB in animals immobilized with 3 mM levamisole . DOI: http://dx . doi . org/10 . 7554/eLife . 04241 . 014 The AVA and AVB interneurons , and their downstream targets the A and B motoneurons , respond to a variety of sensory cues transmitted by upstream circuits to control C . elegans' direction of movement . These upstream circuits include highly connected interneurons called AIA , AIB , and AIY that are post-synaptic to multiple sensory neurons ( White et al . , 1986; wormwiring . org ) . We speculated that the highly variable Ca2+ baseline we observed in AVA , AVB , and AIY reflected these diverse inputs , and that modulation by O2 levels acts on top of these other inputs . In levamisole-treated animals , the frequent switches observed in AVA , AVB , and AIY might reflect attempts by the animals to change their direction of movement . To examine this possibility without the caveats associated with levamisole treatment , and in a way that allows us to record information about the animals' behavior , we imaged neural activity in animals moving freely on food under a layer of PDMS , at 7% or 21% O2 . Under these conditions , animals modulated their speed of movement in response to O2 levels , although with reduced amplitude , but did not exhibit the high reversal state normally observed at 7% O2 , perhaps because of the effects of being under a PDMS layer ( Figure 8A; Figure 8—figure supplement 1 ) . 10 . 7554/eLife . 04241 . 015Figure 8 . Correlation of neural activity and behavior at 7% and 21% O2 . ( A ) Under our Ca2+ imaging conditions , freely moving npr-1 animals increase their speed but do not suppress reversals at 21% O2 . ( B–G ) RMG neurons respond to 21% O2 with a strong persistent increase in Ca2+ regardless of direction of travel ( B ) . By contrast , freely moving npr-1 animals display frequent brief Ca2+ changes in AIA ( C ) , AIB ( D ) , AVA ( E ) , AVB ( F ) , and AIY neurons ( G ) , at both 7% and 21% O2 . Most Ca2+ changes are associated with reversal events . DOI: http://dx . doi . org/10 . 7554/eLife . 04241 . 01510 . 7554/eLife . 04241 . 016Figure 8—figure supplement 1 . The experimental setup used to image freely moving animals . DOI: http://dx . doi . org/10 . 7554/eLife . 04241 . 016 As expected from our imaging of immobilized animals , RMG interneurons responded to a 7–21% increase in O2 with a large tonic rise in Ca2+ that lasted for as long as animals were at 21% O2 ( Figure 8B ) . Ca2+ levels in RMG did not change when animals reversed or stopped moving forward . By contrast , the AVB , AVA , AIY , AIA , and AIB interneurons each displayed frequent , large but brief changes in Ca2+ levels , at both 21% and 7% O2 ( Figure 8C–G ) . Superficially , these Ca2+ changes appeared to occur stochastically , as in immobilized animals , but careful analysis showed that they usually coincided with a switch in the direction of travel , and never lasted more than a few seconds ( Figure 8C–G ) . We hypothesized that the large changes in Ca2+ levels associated with executing reversal behaviors would obscure any smaller Ca2+ changes in these neurons that were evoked by a change in O2 concentration . To examine this possibility , we registered Ca2+ traces to the initiation of a reversal , and averaged responses across multiple animals kept at either 7% or 21% O2 . We then compared Ca2+ levels in each neuron as animals executed reversals at 7% or 21% O2 ( Figure 9A ) . As expected , we observed changes in Ca2+ associated with the reversal behavior sequence . Superimposed on this , our analysis revealed that Ca2+ levels in AVB and AIY , but not in AIA , AIB , and AVA neurons , differed significantly between animals kept at 21% and 7% O2 ( Figure 9B–F ) . AVB showed higher Ca2+ levels during forward locomotion at 21% O2 than at 7% O2 ( Figure 9B ) . AIY also showed significantly higher Ca2+ levels at 21% O2 than 7% O2 during forward locomotion ( Figure 9C ) , and only showed changes in activity correlating with reversal behavior at 7% O2 . As expected , the RMG interneurons showed strong modulation by O2 but no change in activity correlating with reversal state ( Figure 9G ) . Together , our data predict that modulation of AVB and AIY downstream of RMG mediates behavioral changes induced by different O2 environments . 10 . 7554/eLife . 04241 . 017Figure 9 . AVB and AIY interneurons integrate information about O2 levels with other input . ( A ) Animals showed a characteristic pattern of speed changes when traces were aligned according to the time of reversal initiations . ( B–G ) AVB ( B ) , AIY ( C ) , and RMG neurons ( G ) show increased Ca2+ at 21% O2 compared to 7% O2 during forward movement . By contrast , average Ca2+ in AIA ( D ) , AIB ( E ) , and AVA ( F ) was not significantly modulated by O2 . ( H and I ) Normalized Ca2+ traces in AVA , AVB , AIA , and AIB neurons aligned to the first frame of backward locomotion ( H ) , or to the first frame of forward movement ( I ) , during spontaneous reversals . Reversal initiation correlates with a rise in Ca2+ in AIB and AVA and a fall in Ca2+ in AIA and AVB . The converse pattern is observed when reversals are terminated . DOI: http://dx . doi . org/10 . 7554/eLife . 04241 . 017 Our data also suggest a sequence of changes in neural activity associated with the initiation and the termination of reversals at 7% and 21% O2 . In particular , we observed a fall in AIA Ca2+ coincident with a rise in AIB Ca2+ that immediately preceded reversal initiations , and was followed by a drop in AVB Ca2+ and a rise in AVA Ca2+ around the time reversals begin ( Figure 9H ) . Termination of reversals was preceded by a fall in AVA Ca2+ , and associated with a rise in Ca2+ in AVB and AIA , and a drop in Ca2+ in AIB ( Figure 9I ) . To test for physiological roles of AIY , AVB , and AIA interneurons in O2-evoked changes in behavioral state we turned to optogenetics and ablation experiments . We first expressed halorhodopsin ( eNpHR2 ) in the AIY , AVB , and AIA interneurons , or in AIA alone , in an npr-1 background , and used light to inhibit these neurons . At 7% O2 , acute inhibition of AIY , AVB and AIA together , or of AIA alone , only had minor effects on behavior ( Figure 10A ) . When the light was turned off , we observed a transient decrease in the rate of reversals that probably reflected disinhibition of AIA ( Figure 10A , compare AIA alone with AIA , AIY , and AVB ) . 10 . 7554/eLife . 04241 . 018Figure 10 . AIY and AVB interneurons contribute to the switch in locomotory activity evoked by 21% O2 . ( A and B ) Behavioral effects of inhibiting AIA or AIA , AIY and AVB using halorhodopsin in animals kept at 7% O2 ( A ) or 21% O2 ( B ) . ( C and D ) Behavioral effects of activating AIY neurons using channelrhodopsin in animals kept at 7% O2 ( C ) or 21% O2 ( D ) . ( E and F ) Ca2+ responses evoked in AVA interneurons by inhibition ( E ) or disinhibition ( F ) of AIA neurons using halorhodopsin . ( G and H ) Disrupting the ttx-3 homeobox transcription factor required to specify AIY cell fate attenuates the rapid movement evoked in npr-1 animals by 21% O2 but not the inhibition of reversals . ttx-3 ( ks5 ) and ttx-3 ( mg158 ) are different null alleles . ( I and J ) Ablating AIA neurons alone does not disrupt responses of npr-1 animals to 21% O2 . npr-1 ttx-3 mutants ablated for AIA neurons both move more slowly and reverse more frequently than controls . DOI: http://dx . doi . org/10 . 7554/eLife . 04241 . 018 When animals were kept at 21% O2 , acute inhibition of AIA caused a sustained increase in reversal rate compared to control animals , but did not significantly alter their speed of movement ( Figure 10B ) . Inhibition of AIA , AIY and AVB together at 21% O2 caused a sustained decrease in the speed of movement together with an increased rate of reversals compared to control animals ( Figure 10B , compare AIA alone with AIA , AIY , and AVB ) . Together , our data suggest that although we do not observe regulation of AIA Ca2+ by O2 ( Figure 9D ) , tonic AIA activity plays a role in inhibiting reversal rate at 21% O2 , while AIY and AVB promote rapid forward movement . Since inhibiting AIA at 21% O2 promoted reversal behavior ( and disinhibiting AIA suppressed reversals at 7% O2 ) , we examined how altering AIA activity modulated downstream circuits . We imaged Ca2+ in the AVA interneurons of freely moving npr-1 animals kept at 21% O2 while acutely inhibiting or disinhibiting AIA with light . Inhibiting AIA evoked a rise in AVA Ca2+ ( Figure 10E ) , while disinhibiting AIA caused a rapid fall in AVA Ca2+ ( Figure 10F ) . These results suggest that AIA has inhibitory control of AVA activity , and is consistent with AIA activity preceding AVA activity during spontaneous reversals . We next expressed ChR2 selectively in AIY in an npr-1 background , and used light to activate this interneuron at low and high O2 concentrations ( Figure 10C , D ) . Activating AIY caused a sustained decrease in reversal rate and an increase in speed , both at 7% and 21% O2 . Thus , consistent with the predictions made from our Ca2+ imaging experiments , optogenetics suggest that increasing AIY activity can increase the speed of locomotion and decrease reversal frequency . To test further for a role of AIY in the O2-evoked behavioral switch , we examined the O2 responses of npr-1 ttx-3 mutants . ttx-3 encodes a LIM homeodomain transcription factor required for specification of AIY ( Hobert et al . , 1997 ) . ttx-3 npr-1 animals sped up less than npr-1 animals upon being switched from 7% to 21% O2 , but inhibited reversal behavior to a similar extent ( Figure 10G , H ) . Ablating AIA interneurons in ttx-3 npr-1 animals did not alter speed further , but significantly increased reversal rate at 21% O2 ( Figure 10I , J ) . These results are consistent with AIY playing a physiological role in promoting rapid movement at 21% O2 , and suggest that AIA acts with AIY to inhibit reversal behavior at 21% O2 , although neither neuron appears to be essential .
To delineate long-term changes in the activity of neural populations , we sought simple readouts of neurosecretory activity that do not depend on Ca2+ sensors . Ca2+ imaging approaches , although powerful , are blind to signaling mechanisms that alter synaptic and neurosecretory activity without affecting Ca2+ levels . We were inspired by sporadic reports suggesting a positive correlation between neural activity and neuropeptide gene transcription in the mammalian brain . By studying multiple neuropeptide genes expressed in URX , RMG , and other neurons , we showed that transcription of these genes is positively coupled to the secretory activity of the neurons expressing them . The feedback loop appears to act downstream of Ca2+ , and to be linked to neurosecretion itself , since it can be disrupted by expressing tetanus toxin , which inhibits synaptic release by cleaving synaptobrevin . The effects of tetanus toxin on neuropeptide expression recapitulate the effects of manipulating circuit activity by disrupting signaling molecules or by altering O2 levels . If our observations are generalizable , neuropeptide transcriptional reporters provide a way to dissect tonic modulation of secretory activity in neural networks . The ability of RMG interneurons to alter behavioral state according to ambient O2 is inhibited by an NPY/RFamide-like neuropeptide receptor , NPR-1 . The npr-1 allele found in the N2 standard C . elegans laboratory strain , npr-1 215V , essentially abolishes RMG-mediated escape responses in feeding animals . Our data suggest npr-1 acts by inhibiting neurosecretion from RMG downstream of Ca2+ . First , we do not observe a striking difference in O2-evoked Ca2+ responses in RMG between npr-1 and npr-1 215V animals . Second , whereas ChR2 activation of RMG can induce npr-1 animals kept at 7% O2 to switch to rapid movement , it has little effect on animals encoding the NPR-1 215V receptor , even at 21% O2 when RMG Ca2+ levels are high . Third , transcriptional reporters of neuropeptide genes suggest that NPR-1 215V inhibits neurosecretion from RMG and other neurons . NPR-1 may inhibit synaptic release by altering the balance between Go and Gq signaling in favor of Go . Increasing Gq activity , by disrupting the EAT-16 RGS protein or its binding protein RSBP-1 , phenocopies the effects of disrupting npr-1 . EAT-16 and RSBP-1 activate the Gq GTPase . RMG lies at the center of an anatomically defined hub-and-spoke network connected by gap junctions that include the ASK , ADL , ASH , and URX sensory neurons ( Figure 11 ) . Previous work suggested that the RMG hub redistributes inputs across the spoke sensory neurons through gap junctions to direct behavioral responses ( Macosko et al . , 2009 ) . The importance of gap junctions in the circuit was inferred from the anatomy and genetic manipulations . Whether RMG could propagate electrical signals or Ca2+ across the network , and if so in which direction , or if gap junctions were rectifying or passive gates was unknown . Our data revise our understanding of how this circuit works , as detailed below . ASK are ‘OFF’ neurons that act analogously to vertebrate photoreceptors , responding to ascaroside pheromones or food stimuli with a decrease in Ca2+ ( Macosko et al . , 2009; Wakabayashi et al . , 2009 ) . ASK neurons were proposed to be a major output of the RMG hub-and-spoke network , inducing rapid movement at 21% O2 and promoting aggregation in response to pheromones ( Macosko et al . , 2009 ) . In this model , RMG stimulates ASK excitability via gap junctions in aggregating strains , but in solitary strains this is prevented because NPR-1 215V directly or indirectly inhibits gap junctional signaling in the network . We find that ablating ASK neurons or reducing their activity using halorhodopsin does not disrupt either O2-evoked changes in locomotory activity , or aggregation behavior . These data suggest ASK is not required for either of these responses , although we cannot exclude that it acts redundantly with other neurons to promote these behaviors . The ASK ablation result contrasts with both our own data ( Tremain , 2004 ) and those of others ( Macosko et al . , 2009 ) . Both these studies show that npr-1 animals defective in the cGMP gated ion channel TAX-4 fail to aggregate or to move rapidly in normoxia , and that these behaviors can be restored by expressing TAX-4 in URX and ASK ( Tremain , 2004; Macosko et al . , 2009 ) . Why does ablating ASK have different behavioral consequences from disrupting cGMP signaling in this neuron ? One possibility is that removing tax-4 reduces the basal excitability of ASK ( and other TAX-4-expressing neurons ) and shunts current from RMG , compromising its ability to direct O2 responses . However , since tax-4 mutations disrupt many sensory modalities ( de Bono and Maricq , 2005 ) , other explanations are also possible . Disrupting the OCR-2 TRPV-like cation channel attenuates the switch in locomotory state evoked in npr-1 animals by 21% O2 . Restoring ocr-2 to any one of the ASH , ADL , or ADF neurons restores O2 modulation of locomotion to ocr-2; npr-1 animals . Injecting current in ASH neurons using channelrhodopsin also restores strong modulation of locomotion by O2 to ocr-2; npr-1 mutants , consistent with ocr-2 mutations chronically reducing ASH activity . The neuroanatomy suggests that ASH and ADL are gap-junctionally connected to RMG . Although a rise in O2 evokes an increase in ASH Ca2+ , several findings argue against a simple ‘hub-to-spoke’ model in which ASH and ADL spokes are necessary to relay RMG hub activity outwards to evoke behavioral responses . The Ca2+ increase in ASH upon switch to 21% O2 is very small . Ablating ASH and ADL , or expressing tetanus toxin in these neurons , does not disrupt the O2 regulation of locomotion . Moreover , selective expression of ocr-2 in ADF neurons , which are not gap-junctionally connected to RMG , can partially rescue the ocr-2; npr-1 phenotype . Nevertheless , OCR-2 currents in ASH or ADL can influence the O2-evoked changes in locomotion . How ? One model is that tonic/chronic OCR-2 channel activity in ASH and ADL keeps these neurons depolarized and limits the current shunted locally from RMG to these neurons . Several observations support this possibility . First , expression of neuropeptide reporters in RMG is downregulated in ocr-2; npr-1 animals to levels found in npr-1 animals kept at 7% O2 . This suggests that loss of TRPV activity reduces neurosecretion from RMG . Second , the ability of ocr-2 expression in ASH or ADL neurons to rescue the ocr-2; npr-1 phenotype does not require neurotransmission , suggesting gap junctions are involved . Third , injecting current into ASH can restore to ocr-2; npr-1 animals the O2 regulated switch in high locomotory activity . Fourth , injecting current directly into RMG using channelrhodopsin can also rescue the ocr-2; npr-1 phenotype , indicating that increasing RMG Ca2+ is sufficient to bypass the lack of OCR-2-dependent activity . Unexpectedly , deleting ocr-2 did not affect the steady state O2-evoked Ca2+ responses in the RMG cell body , suggesting that any effects on RMG Ca2+ dynamics are local . The simplest model that explains our data is that OCR-2 signaling in ASH and ADL prevents shunting of current , or loss of a Ca2+-dependent second messenger , from RMG to ASH and ADL through gap junctions . Membrane potential and Ca2+ dynamics are not necessarily coupled , and a local current leak through gap junctions can be functionally important without necessarily altering Ca2+ at the soma . We show that none of the spoke neurons of the RMG ‘hub and spoke’ are individually necessary for RMG to alter the behavioral state in response to changes in O2 in npr-1 animals . This contrasts with the analysis of an anatomically similar circuit that mediates nose touch perception in C . elegans ( Chatzigeorgiou and Schafer , 2011 ) . In that circuit , OLQ and CEP mechanoreceptors that are coupled by gap junctions to each other and to the RIH hub interneuron act as coincidence detectors , pooling information through RIH , which in turn enables the high threshold FLP mechanoreceptors , also connected to RIH via gap junctions , to evoke a response to a gentle nose touch . In the RMG circuit , reducing RMG neurosecretory output by cell-selective expression of tetanus toxin or RNAi knockdown of the carboxypeptidase E egl-21 strongly reduces transmission of the high O2 information , suggesting that peptidergic release from RMG is a major output of the URX and RMG couple . Optogenetic experiments coupled with imaging experiments in freely moving animals provide insights into the information carried by different neurons in the O2 circuit . RMG tracks O2 levels , becoming more active at higher O2 levels . Increased RMG activity drives faster forward movement and inhibits short reversals . However , RMG activity does not change when animals slow down or execute a reversal . Thus , RMG provides modulatory input but is not actually part of the circuit executing the behavior . Like RMG , AVB and AIY have , on average , increased Ca2+ at 21% O2 . However , unlike RMG , these neurons do not show continuously high Ca2+ levels . Instead , they show switches in Ca2+ levels that correlate with reversal behavior . Most likely , the connectivity within the reversal modules ( AIA , AIB , AIY , AIZ , AVA , and AVB ) shapes a sequence of neuronal activation triggering/terminating the reversals . Modulation of the AVB and AIY interneurons by O2 acts on top of that patterned activity to control overall locomotory pattern . AIA , AIB , and AVA neurons do not appear to be modulated by O2 levels , and are involved in the reversal behavioral sequence irrespective of O2 concentration ( Figure 9D–F ) . For behavioral effectiveness , each module state should enhance a specific behavior while suppressing incompatible behaviors . Our results suggest a direct role for AVB and AIY in both promoting high speed and suppressing reversals at high [O2] . Such functions have been proposed for these neurons in other contexts ( Kawano et al . , 2011 ) . Interestingly , AIY was also proposed to be involved in the control of lifespan and metabolism , something also modulated by O2 ( Shen et al . , 2010 ) . Potentially , information about O2 concentration could flow from URX–RMG to AVB and AIY in part through neuropeptide secretion . Persistent changes in behavior and physiology require persistent changes in neural network activity . We identified several layers of neurons that display sustained changes in activity according to ambient O2 . However , the logic of the activity of different layers differs . In the O2-sensing neurons , URX , AQR , and PQR , Ca2+ sensors reported O2 levels , as expected ( Busch et al . , 2012 ) . In the first interneuronal layer , including the RMG hub interneuron and potentially the PVP interneurons , Ca2+ levels also tracked O2 concentration , although these neurons probably relay other information besides O2 concentration , such as activity of the TRPV expressing neurons . Moreover , the activity level of RMG controls behavioral state rather than directly commanding a specific action: RMG activity remains high at 21% O2 regardless of the animal's speed or direction of travel . By contrast , the second and third layers of interneurons modulated by O2 appear to be directly involved in generating specific behaviors . Ca2+ changes in these neurons anticipate or report the animal's behavior . The activity of some but not all of these neurons , including AVB and AIY , is modulated but not set by O2 input . These downstream interneurons are probably not dedicated to a small subset of sensory inputs , as RMG or PVP might be , but instead simultaneously translate multiple streams of information into the appropriate behaviors . Changing RMG activity is sufficient to evoke the behavioral states associated with high and low O2 . Downstream of URX–RMG we find widespread modulation of neurons and neuropeptide expression/secretion , highlighting the complexity of even a simple contextual cue , O2 . Neuropeptide signaling appears to be key to generating the behavioral switch . Similar to motivated states , organismic states re-organize the salience of different sensory cues and change the physiology of the animal . It is tempting to propose that peptides whose release is modulated by O2 re-organize sensory responses and modify the physiology of C . elegans . Studies of cross-modulation of CO2 avoidance by the O2 circuit support this scenario ( Carrillo et al . , 2013; Kodama-Namba et al . , 2013 ) . A similar example of how activation of a neuron and secretion of its associated peptides can coordinate behavioral and physiological responses is provided by mammalian tonic nociceptive neurons . In addition to amplifying the nociceptive signal in the spinal cord , the release of tachykinin and CGRP peptides from the nociceptive neurons alters gene expression in the surrounding tissues , producing neurogenic inflammation ( Carlton , 2014 ) . Behavioral states and emotions are under intense investigation in mammalian systems , but the circuits engaged in implementing these states are only partially mapped . Changes in the global state of mammals are proposed to involve altered activities in multiple brain areas . Here , we outline how a global organismic state is encoded in a system that can be more easily circumscribed and comprehensively dissected .
C . elegans was grown under standard conditions unless otherwise indicated ( Sulston and Hodgkin , 1988 ) . To cultivate animals in defined O2 environments , we used a Coy O2 control glove box ( Coy , Grass Lake , Michigan , USA ) . Sample preparation: To prevent aggregation behavior , which could confound our comparisons ( Andersen et al . , 2014 ) , we grew animals at very low density on a thin lawn of Escherichia coli OP50 spread to the edges of a 6 cm petri dish . To obtain a thin lawn , we grew bacteria on NGM containing only 5% of the regular amount of peptone . We picked five gravid hermaphrodites onto the lawn , let them lay eggs for 2–3 hr , and then removed them . The ∼40 eggs were allowed to hatch and grow to adult worms at the indicated O2 concentration . Animals from 10 such plates were rinsed off in M9 buffer and their RNA was extracted using RNeasy Mini Kits ( Qiagen , Germany ) . cDNA libraries were prepared using the TruSeq Stranded mRNA Sample Prep Kit ( Illumina ) and sequenced on an Illumina HiSeq 2500 platform . We generated independent libraries of biological replicates as follows: npr-1 , four libraries for 21% O2 and six for 7% O2; gcy-35; npr-1 , seven libraries for 21% O2 and seven for 7% O2 . Approximately 30 million 50 bp single-end reads were produced for each sample . RNA-seq data analysis: Reads were output in FASTQ format and their quality assessed using FastQC v0 . 11 . 2 ( http://www . bioinformatics . babraham . ac . uk/projects/fastqc/ ) . A small portion ( 3–4% ) of reads containing over-represented sequences ( e . g . , Illumina adapters ) identified by FastQC were removed with Trimmomatic v0 . 30 ( Bolger et al . , 2014 ) . The remaining reads were aligned to the C . elegans genome ( WBcel235 ) with TopHat v2 . 0 . 13 ( Kim et al . , 2013 ) . TopHat was run using default parameters with the following exceptions: coverage search was disabled using --no-coverage-search; library-type was changed to fr-firststrand; and the WBcel235 transcriptome annotations were provided via --transcriptome-index . In addition , since C . elegans has comparatively short introns ( Steijger et al . , 2013 ) , --min-intron-length and --min-segment-intron were both reduced to 30 . The aligned reads were then processed using the Cufflinks suite v2 . 2 . 1 ( Trapnell et al . , 2012 ) to assemble transcripts and ultimately compute differential gene expression between conditions . Transcripts were assembled for each sample with the Cufflinks tool ( Trapnell et al . , 2010 ) , again lowering --min-intron-length , as well as --overlap-radius , to 30 . The sample transcript assemblies were then merged with reference annotations ( WBcel235 ) using Cuffmerge to generate a single , overall transcript assembly . The Cuffquant tool , using this merged assembly , was then used to compute gene and transcript abundances for each sample . Finally , sample abundances were integrated with the merged assembly by Cuffdiff ( Trapnell et al . , 2013 ) to test for differential expression between all condition pairs . All the Cufflinks tools , aside from Cuffmerge , were used with the --library-type parameter again set to fr-firststrand and both --frag-bias-correct ( Roberts et al . , 2011 ) and --multi-read-correct enabled . A gene was considered as differentially expressed between two conditions if the q-value ( p-value after Benjamini-Hochberg correction for multiple testing ) for the given comparison was <0 . 05 . The gene-level differential expression output from Cuffdiff was used to generate the presented excel files , which were annotated with WormBase IDs ( WBcel235 ) and InterPro domains using custom Perl scripts . DNA cloning was carried out using standard methods ( Green and Sambrook , 2012 ) . Promoters used in this work included: gcy-32 ( Yu et al . , 1997 ) , sra-9 ( Troemel et al . , 1995 ) , pdf-1 ( Barrios et al . , 2012 ) , flp-8 ( Kim and Li , 2004 ) , flp-21 short ( Kim and Li , 2004 ) , flp-21 long ( Macosko et al . , 2009 ) , gpa-11 ( Jansen et al . , 1999 ) , srh-220 ADL ( F47C12 . 5 ) ( Troemel et al . , 1995 ) , srh-142 ADF ( T08G3 . 3 ) ( Troemel et al . , 1995 ) , ins-1 ( Tomioka et al . , 2006 ) , gcy-28 ( Tsunozaki et al . , 2008 ) , sra-11 ( Troemel et al . , 1995 ) , ttx-3 ( AIY fragment ) ( Wenick and Hobert , 2004 ) , and ncs-1 ( Macosko et al . , 2009 ) . Promoters were cloned into position 1 of the Multisite Gateway system ( Invitrogen ) . Expression clones used in the course of this work are listed in supplemental data . Microfluidic chambers for Ca2+ imaging and behavioral assays were cast as described previously ( Busch et al . , 2012 ) . For each assay , 20–25 adult hermaphrodites were transferred onto NGM plates seeded 12–14 hr earlier with 40 μl of E . coli OP50 . To control O2 levels experienced by the worms , we placed a 200 μm deep square PDMS chamber on top of the agar pad , with inlets connected to a PHD 2000 Infusion syringe pump ( Harvard Apparatus ) delivering humidified gas mixtures at a flow rate of 2 . 5 ml/min . We began pumping 5 min prior to the start of the recording to ensure that the initial environment was in a steady state . Movies were recorded using FlyCapture on a Leica M165FC dissecting microscope with a Point Grey Grasshopper camera running at two frames per second . Movies were analyzed using Zentracker , a custom-written Matlab software ( available at https://github . com/wormtracker/zentracker ) . Speed was calculated as instantaneous worm centroid displacement between successive frames . Frames in which the shape of the worm had an eccentricity value ( ratio of the two axes of an ellipse with the same second moments ) smaller than 0 . 8 or a compactness value ( perimeter2/area ) smaller than 30 , combined with a solidity value ( area/convex hull ) greater than 0 . 575 were identified as omega-turns . To detect reversals , sharp ( greater than 60° ) changes in direction were first identified . Each potential behavior pattern that could be assigned to each track was then examined using a heuristic algorithm based on penalizing any occurrences of continuous movements lasting for more than 7 . 5 s in the same direction that do not correspond to forward movements , any time an omega-turn is not followed by forward movement , and cases where a worm does not spend more time in forward than in backward movement . Using this information , directions of movement were determined for each worm in each frame according to the prospective behavioral pattern with the lowest overall penalty score . ChR2 codon-optimized for C . elegans and C-terminally tagged with mCitrine , or halorhodopsin tagged with mCherry , were expressed from the promoters of interest . Worms were grown on plates pre-seeded with 200 μl of the E . coli OP50 , with 15 μl of 5 mM all trans-Retinal ( Sigma ) dissolved in 100% ethanol added to the bacterial lawn prior to picking the worms onto the plates . Control worms were grown in parallel on seeded plates onto which 15 μl of 100% ethanol was added without the all trans-Retinal . Worms were preselected for fluorescence in the neurons of interest , and then assayed as described above . Light stimuli were delivered using a Leica EL6000 mercury lamp filtered for ChR2 or NpHR excitation using a Leica GFP2 or ET DSR filter , respectively . To avoid unwanted light-activation of the optogenetic actuators , we filtered the trans-illumination light using a 595 nm long-pass red optical cast plastic filter ( Edmund Optics ) for ChR2 , and a 705 nm long-pass colored glass filter ( Thorlabs ) for NpHR . We imaged immobilized animals as described previously ( Busch et al . , 2012 ) . The experimental setup to image freely moving animals on an agar pad is shown in Figure 8—figure supplement 1 . Emission light filtered for CFP and YFP wavelengths ( 460–495 nm and 525–580 nm , respectively ) , separated using the Tu-Cam system ( Andor ) with a 510 nm longpass , was relayed onto two identical Photometrics Cascade II 1024 EMCCD cameras running in frame-transfer modes with 60 ms exposure times . The imaged worm was kept in the field of view by moving the Prior stage manually using a joystick , with the acceleration rate set to its lowest value so as to disturb the worm as little as possible . To reconstruct the speed and trajectory of the worm , the actual stage position was continuously logged while acquiring image stacks using Micromanger ( Edelstein et al . , 2010 ) or custom-written software . To deliver gas stimuli to the worm , we placed a square PDMS chamber on top of the agar pad , with inlets connected to a PHD 2000 Infusion syringe pump ( Harvard Apparatus ) running at a flow rate of 50 μl/s . An electronic valve system placed between the syringes and the PDMS chamber allowed us to switch between two different gas mixtures in a controlled manner at pre-specified frames . Using a spot Optode ( PreSens ) , we monitored the O2 concentration within the chamber while switching between mixtures containing 7% and 21% O2 , and found that O2 levels could be switched reliably using this configuration ( Figure 8—figure supplement 1 ) . We used Neuron Analyzer , a custom-written Matlab program , to analyze the resulting image stacks ( available at https://github . com/neuronanalyser/neuronanalyser ) . In movies in which the neuron of interest provided the brightest signal in the field of view , we simply tracked the brightest points in each channel . Where this was not a viable option due to the complexity of the expression pattern , we used a semi-supervised heuristic tracking approach in which in each frame the area within a 20-by-20 pixel square centered around the location of the centroid of the region of interest ( ROI ) in the previous frame was examined for potential matches . Regions whose intensity differed by less than 1/3 from that of the ROI identified in the previous frame , and whose centroids were located within a 10-pixel radius of the ROI location in the previous frame , were identified as candidates for the ROI in the current frame . If no such candidates were detected , user input was requested in identifying the ROI in the current frame . If exactly one such candidate was detected , it was identified as the ROI in the current frame . If more than one such candidate was detected , match quality scores were calculated for each candidate based on their differences in intensity and centroid location from the ROI in the previous frame . If one candidate had a better quality score in both respects than all others , it was identified as the ROI in the current frame; user input was otherwise requested . In cases where the heuristic tracking method struggled to identify the ROI reliably in an automatic way , we used a user-supervised tracking method , in which a custom-sized moveable region was specified using the GUI , restricting the tracking of the brightest point within this regon of interest . Once the centroid of the neuron of interest was identified , we calculated the mean of the 20 brightest pixels within a 10-pixel radius of the centroids for both channels independently , and subtracted from this the median pixel intensity of the respective channels as the background . We then corrected for the crosstalk by subtracting 0 . 6 from the ratio values . To extract the behavior of the worm during the recording , first a more accurate measure of worm position was obtained based on not only the stage position , but also the location , within the field of view , of a successfully tracked neuron , effectively calculating the position , in stage coordinates , of the centroid of the ROI . The path was then smoothed using a moving average filter , and potential short gaps in the position data were identified , which may reflect the neuron temporarily leaving the field of view due to unexpected worm movements during the recordings . Gaps shorter than 30 frames were then filled using a spline interpolant , while gaps longer than this were excluded from the analysis . In order to identify reversals , the direction of movement , within the coordinate system of the stage , was calculated for each frame using a central differences method , and changes in direction between successive frames were obtained . Intervals in which the worm moved slower than 10 μm/s were flagged as stationary behavior , and were excluded from further analysis . For the remaining frames , the relative changes in direction between successive frames were then examined , and periods containing changes no larger than 7 . 5° were classified as consistent movement . Consistent movement lasting longer than 90 s was flagged automatically as forward movements . For consistent intervals lasting shorter than this , user input was requested about the direction of the movement . Statistics for both Ca2+ imaging and behavioral assays used the Mann–Whitney U-test . Where we sought to compare steady state values , we chose time intervals where we expected the fluorescence ratios , or behavioral parameters , to have plateaued , that is , with a delay with respect to the timing of the switch in O2 concentration or the presentation of blue or green light . Ca2+ imaging: when presenting time series over the duration of an experiment , n refers to the number of animals imaged . When presenting event-triggered averages ( e . g . , time relative to the onset of a reversal ) , n indicates the number of events . In this case , each animal was still only imaged once , but each animal could contribute to multiple events . Behavior: for the intervals of interest , we extracted independent per-subject means deriving from worms flagged as continuously valid for at least 10 s during the interval . A worm was considered valid at a time point if it was not in contact with another animal , it was on the food lawn , and it was located at least half a worm-length from the border . Following these criteria , each worm was sampled at most once per interval; n indicates the minimum number of samples obtained per interval for the two intervals being compared . | From humans to worms , animals must respond appropriately to environmental challenges to survive . Starving animals must conserve energy while they seek food; animals that encounter a predator must fight or flee . These responses involve the animals re-programming their bodies and behavior , and , in humans , are thought to coincide with feelings or emotions such as ‘hunger’ and ‘fear’ . Understanding these states in humans is difficult , but studies of simpler animals may provide some insights . The microscopic worm Caenorhabditis elegans offers a unique advantage to these studies because it has the most precisely described nervous system of any animal . The worm lives in rotting fruit , but it avoids the fruit's surface , perhaps because there is an increased risk of it drying out or being eaten by predators . Microbes that grow within the rotting fruit reduce the oxygen level below the 21% oxygen found in the surrounding air , and so one strategy that C . elegans uses to avoid surface exposure is to continuously monitor the oxygen concentration . If the worm senses that the oxygen level is approaching 21% , which suggests it is nearing the surface , it reverses and turns around . If it cannot find a lower-oxygen environment , the worm switches to continuous rapid movement until it locates such an environment , and adapts its body for surface exposure . Laurent , Soltesz et al . sought to understand the circuit of neurons that controls this switch . Monitoring gene expression in the worms revealed that specific oxygen-sensing neurons help generate the widespread changes that occur in the worm's body . These neurons also control the switch in the worm's behavior . Sensory neurons relay signals to downstream neurons that act on muscles to alter behavior . Neurons typically communicate with other neurons via specific connections; but neurons can also release signaling molecules , which act like ‘wireless’ signals and can affect many other cells . Laurent , Soltesz et al . showed that both kinds of signaling are needed to change the worm's behavior , and suggest that the release of signaling molecules may explain the widespread effects of 21% oxygen on the worm . Laurent , Soltesz et al . then monitored the activity of neurons in freely moving worms , and found that some neurons appear to encode and relay specific sensory information . Other neurons encode the behavior the animal is performing , and yet others can encode both kinds of information . To confirm which neurons control particular behavioral responses , Laurent , Soltesz et al . measured changes in the worm’s behavior after destroying or altering specific cells , or while they used light-based techniques to artificially excite or inhibit specific neurons . At a simple level the worm's response to 21% oxygen resembles the response of a mammal to a dangerous environment: both become more aroused , change how they respond to other sensory cues , and adapt both their bodies and behavior . As such , C . elegans provides a great model to explore at a small and accessible scale how changes in animals' states are generated . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"neuroscience"
] | 2015 | Decoding a neural circuit controlling global animal state in C. elegans |
Visual perception and imagery rely on similar representations in the visual cortex . During perception , visual activity is characterized by distinct processing stages , but the temporal dynamics underlying imagery remain unclear . Here , we investigated the dynamics of visual imagery in human participants using magnetoencephalography . Firstly , we show that , compared to perception , imagery decoding becomes significant later and representations at the start of imagery already overlap with later time points . This suggests that during imagery , the entire visual representation is activated at once or that there are large differences in the timing of imagery between trials . Secondly , we found consistent overlap between imagery and perceptual processing around 160 ms and from 300 ms after stimulus onset . This indicates that the N170 gets reactivated during imagery and that imagery does not rely on early perceptual representations . Together , these results provide important insights for our understanding of the neural mechanisms of visual imagery .
Twenty-five participants executed a retro-cue task in which they perceived and imagined faces and houses and rated their experienced imagery vividness on each trial using a sliding bar ( see Figure 1 ) . Prior to scanning , participants filled in the Vividness of Visual Imagery Questionnaire , which is a measure of people’s imagery ability ( Marks , 1973 ) . There was a significant correlation between VVIQ and averaged vividness ratings ( r = −0 . 45 , p=0 . 02 ) , which indicates that people with a higher imagery vividness as measured by the VVIQ also rated their imagery as more vivid on average during the experiment . Participants reported relatively high vividness on average ( 49 . 6 ± 26 . 6 on a scale between −150 and +150 , where 0 is the starting point of the slider and positive numbers indicate high vividness and negative numbers indicate low vividness ) . There was no significant difference in vividness ratings between faces ( 54 . 0 ± 29 . 7 ) and houses ( 48 . 7 ± 26 . 7; t ( 24 ) = 1 . 46 , p=0 . 16 ) . To ensure that participants were imagining the correct images , on 7% of the trials participants had to indicate which of four exemplars they imagined . The imagined exemplar was correctly identified in 89 . 8% ( ±5 . 4% ) of the catch trials , indicating that participants performed the task correctly . There was also no significant difference between the two stimulus categories in the percentage of correct catch trials ( faces: 90 . 9 ± 6 . 6 , houses: 88 . 8 ± 7 . 1; t ( 24 ) = −1 . 25 , p=0 . 22 ) . Furthermore , correctly identified catch trials were experienced as more vivid ( 52 . 06 ± 33 . 1 ) than incorrectly identified catch trials ( 24 . 31 ± 33 . 84; t ( 23 ) = 3 . 11 , p=0 . 0049 ) , giving support to the criterion validity of the identification task . We had to drop one subject for this comparison because this person did not have any incorrect trials . To uncover the temporal dynamics of category representations during perception and imagery , we decoded the category from the MEG signal over time . The results are shown in Figure 2 . Testing and training on the same time points revealed that during perception , significantly different patterns of activity for faces and houses were present from 73 ms after stimulus onset with the peak accuracy at 153 ms ( Figure 2A , left ) . During imagery , category information could be decoded significantly from 540 ms after retro-cue onset , with the peak at 1073 ms ( Figure 2A , right , Figure 2—figure supplement 1B ) . The generation of a visual representation from a cue thus seems to take longer than the activation via bottom-up sensory input . Furthermore , imagery decoding resulted in a much lower accuracy than perception decoding ( peak perception is at ~90% , peak imagery at ~60% ) . This is also observed in fMRI decoding studies ( Reddy et al . , 2010; Lee et al . , 2012 ) and is probably due to the fact that imagery is a less automatic process than perception , which leads to higher between trial variation in neural activations . Note that , to allow better comparison between perception and imagery , we only showed the first 1000 ms after cue onset during imagery ( see Figure 2—figure supplement 1 for the results throughout the entire imagery period ) . To reveal the generalization of representations over time , classifiers were trained on one time point and tested on all other time points ( King and Dehaene , 2014 ) ( Figure 2B ) . Furthermore , to investigate the temporal specificity of the representations at each time point , we calculated the proportion of off-diagonal classifiers that had a significantly lower accuracy than the diagonal classifier of that time point ( Astrand et al . , 2015 ) ( Figure 2C; see Materials and methods ) . During perception , distinct processing stages can be distinguished ( Figure 2B–C , left ) . During the first stage , between 70 and 120 ms , diagonal decoding was significant and there was very high temporal specificity . This indicates sequential processing with rapidly changing representations ( King and Dehaene , 2014 ) . During this time period , the classifier mostly relied on activity in posterior visual areas ( Figure 2D , left ) . Therefore , these results are consistent with initial feedforward stimulus processing . In the second stage , around 160 ms , the classifier generalized to neighboring points as well as testing points after 250 ms . The associated sources are spread out over the ventral visual stream ( Figure 2D , left ) , indicating that high-level representations are activated at this time . In the third stage , around 210 ms , we again observed high temporal specificity ( Figure 2C , left ) and a gap in generalization to 160 ms ( Figure 2B , left ) . This pattern could reflect feedback to low-level visual areas . Finally , from 300 ms onwards there is a broad off-diagonal generalization pattern that also generalizes to time points around 160 ms and an associated drop in temporal specificity ( Figure 2B–C , left ) . This broad off-diagonal pattern likely reflects stabilization of the visual representation . In contrast , during imagery , we did not observe any clear distinct processing stages . Instead , there was a broad off-diagonal generalization throughout the entire imagery period ( Figure 2B , right; Figure 2—figure supplement 1A ) . Already at the onset of imagery decoding , there was high generalization and low specificity ( Figure 2B–C , right ) . This indicates that the neural representation during imagery remains highly stable ( King and Dehaene , 2014 ) . The only change seems to be in decoding strength , which first increases and then decreases over time ( Fig . S1B ) , indicating that either representations at those times are weaker or that they are more variable over trials . The sources that contributed to classification were mostly located in the ventral visual stream and there was also some evidence for frontal and parietal contributions ( Figure 2D , right ) . Even though we attempted to remove eye movements from our data as well as possible ( see Materials and methods ) , it is theoretically possible that eye movements which systematically differed between the conditions caused part of the neural signal that was picked up by the decoding analyses ( Mostert et al . , 2017 ) . In order to investigate this possibility , we tried to decode the stimulus category from the X and Y position of the eyes as measured with an eye tracker . The results for this analysis are shown in Figure 2—figure supplement 2 . During imagery , eye tracker decoding was at chance level for all time points , indicating that there were no condition-specific eye movements during imagery ( Figure 2—figure supplement 2B ) . However , during perception , eye tracker decoding was significant from 316 ms onwards ( Figure 2—figure supplement 2A ) , indicating that differences in eye movements between the conditions may have driven ( part of ) the brain decoding . If this were the case , there would be a high , positive correlation between eye tracker decoding and brain decoding . Figure 2—figure supplement 2C however shows that there was no such correlation , suggesting that our perception decoding results for that time window were not driven by eye movements . To investigate when perceptual processing generalizes to imagery , we trained a classifier on one data segment and tested it on the other segment . We first trained a classifier during perception and then used this classifier to decode the neural signal during imagery ( Figure 3A–B ) . Already around 350 ms after imagery cue onset , classifiers trained on perception data from 160 , 700 and 960 ms after stimulus onset could significantly decode the imagined stimulus category ( Figure 3A ) . This is earlier than classification within imagery , which started at 540 ms after cue onset ( Figure 2A , right ) . Considering the increased decoding accuracy during perception compared to imagery ( Figure 2A , left versus right ) , this difference might reflect an increase in signal-to-noise ratio ( SNR ) by training on perception compared to imagery . Furthermore , the distinct processing stages found during perception ( Figure 2B , left ) were also reflected in the generalization to imagery ( Figure 3A–B ) . Perceptual processes around 160 ms and after 300 ms significantly overlapped with imagery ( Figure 3B , right plots ) . In contrast , processing at 90 ms did not generalize to any time point during imagery ( Figure 3B , top left ) . Perceptual processing at 210 ms showed intermittent generalization to imagery , with generalization at some time points and no generalization at other times ( Figure 3B , bottom left ) . Significant generalization at this time could also reflect the effects of smoothing over neighboring time points which are significant ( see Materials and methods ) . This would mean that there is no real overlap at 210 ms but that this overlap is caused by overlap from earlier or later time points . To further pinpoint when perception started to overlap with imagery , we performed an additional analysis in which we reversed the generalization: we trained classifiers on different time points during imagery and used these to classify perception data . This analysis revealed a similar pattern of high overlap with perception around 160 and after 300 ms and low overlap before 100 ms and around 210 ms ( Figure 3C–D ) . Note that this profile is stable throughout imagery and is already present at the start of imagery , albeit with lower accuracies ( Figure 3-D , bottom panel ) . Furthermore , the onset of perceptual overlap is highly consistent over the course of imagery: overlap starts around 130 ms , with the first peak at approximately 160 ms ( Figure 3C ) . In general , cross-classification accuracy was higher when training on imagery than when training on perception ( Figure 3C vs . Figure 3A ) . This is surprising , because training on high SNR data ( in our case , perception ) is reported to lead to higher classification accuracy than training on low SNR data ( King and Dehaene , 2014 ) ( imagery ) . This difference may reflect the fact that the perceptual representation contained more unique features than the imagery representation , leading to a lower generalization performance when training on perception . We also investigated whether the temporal dynamics were influenced by imagery vividness by investigating whether the results of previous analyses were different for participants with high or low vividness ( see Figure 2—figure supplement 3 for within perception and imagery decoding and Figure 3—figure supplement 1 for cross-decoding ) . Decoding accuracy seemed to be higher in the high vividness group , however , none of the differences were significant after correction for multiple comparisons .
We investigated the temporal dynamics of category representations during perception and imagery , as well as the overlap between the two . We first showed that , compared to perception , imagery decoding became significant later , indicating that it takes longer to generate a visual representation based on purely top-down processes . Furthermore , whereas perception was characterized by high temporal specificity and distinct processing stages , imagery showed wide generalization and low temporal specificity from the onset . Finally , cross-decoding between perception and imagery revealed a very clear temporal overlap profile which was consistent throughout the imagery period . We observed overlap between imagery and perceptual processing starting around 130 ms , decreasing around 210 ms and increasing again from 300 ms after stimulus onset . This pattern was already present at the onset of imagery . These findings cannot be explained by a clear cascading of activity up or down the visual hierarchy during imagery . If there was a clear order in activation of different areas , we would not have observed such wide temporal generalization at the start of imagery but instead a more diagonal pattern , as during the start of perception ( King and Dehaene , 2014 ) . Furthermore , we found that the complete overlap with perception was already present at the onset of imagery . One interpretation of our results is that during imagery the complete stimulus representation , including different levels of the hierarchy , is activated simultaneously . However , there was no overlap between imagery and perceptual processing until 130 ms after stimulus onset , when the feedforward sweep is presumably completed and high-level categorical information is activated for the first time ( Isik et al . , 2014; Carlson et al . , 2011; Thorpe et al . , 1996 ) . Overlap between perception and imagery in low-level visual cortex depends on the imagery task and experienced vividness ( Lee et al . , 2012; Albers et al . , 2013; Kosslyn and Thompson , 2003 ) . However , we did not observe a relationship between overlap at this time point and imagery vividness ( Figure 3—figure supplement 1 ) . This absence of early overlap seems to imply that , even though early visual cortex has been implicated in visual imagery , there is no consistent overlap between imagery and early perceptual processing . One explanation for this discrepancy is that representations in low-level visual areas first have to be sharpened by feedback connections ( Kok et al . , 2012 ) before they have a format that is accessible by top-down imagery . Alternatively , early perceptual activity during imagery may be more brief and variable over time than high-level activation , leading to a cancelling out when averaging over trials . The large temporal generalization at the onset of imagery might have been partly due to the nature of the task we used . Here , the start of imagery was based on the association between the cue and the items held in working memory , which could have led to an instantaneous initiation of the visual representation after the cue . It could be the case that visual representations are activated differently if imagery is operationalized as a more constructive process . A previous study has already showed that cueing imagery from long-term memory leads to less neural overlap with perception compared to cueing imagery from working memory ( Ishai et al . , 2002 ) . Perhaps the temporal dynamics of imagery from long-term memory are also different . It could be that if perceptual details have to be filled in from long-term memory , low-level areas are activated later , resulting in a more diagonal generalization pattern . Future studies comparing temporal generalization during imagery from short- and long-term memory are necessary to investigate this further . An alternative explanation for the broad temporal generalization during imagery is that , compared to perception , imagery is less time-locked . If the process during imagery is shifted in time between trials , averaging over trials per time point would obscure the underlying temporal dynamics . This temporal uncertainty will have a different effect on different underlying processes . For example , if the underlying process would be entirely sequential , meaning that independent representations are activated after each other ( like at the start of perception ) , temporal shifting between trials would smear out this sequence over time . This would result in a broader diagonal pattern , where the width of the diagonal is proportional to the temporal shift between trials . This means that the broad temporal generalization that we observed during imagery could represent a sequential process if there were temporal shifts of more than a second on average between trials . Alternatively , the underlying process could be only sequential in the onset , such that different areas become active after each other , but remain active throughout time ( or get reactivated later in time ) . In this case , temporal shifts between trials that are proportional to the difference in onset between the two processes would entirely obscure this dynamic . Note that this would mean that the start of significant imagery classification did not reflect the actual imagery onset , but the first point in time that the representations were consistent enough over trials to lead to above chance classification . Stimulus representations could actually be initiated before 350 ms after cue onset , but we would be unable to classify them at these early time points due to jitter in the onset . We cannot confidently rule out temporal uncertainty as an explanation for the broad temporal generalization at the onset of imagery . To fully resolve this issue , future research should systematically explore the effect of temporal uncertainty on different underlying processes and analysis tools need to be developed that can account for variation in temporal dynamics between trials . We observed clear overlap between imagery and perceptual processing around 160 ms after stimulus onset . The perceptual representation at this time likely reflects the face-specific N170 . This component has been shown to be involved in face processing and appears between 130 to 170 ms after stimulus onset ( Bentin et al . , 1996; Halgren et al . , 2000 ) , which corresponds well with the timing of overlap with imagery reported here . The sources of the N170 are thought to be face selective areas in the ventral stream ( Deffke et al . , 2007; Henson et al . , 2009 ) , which also corresponds to the location of our source reconstruction at this time point . A previous study showed an increase of the N170 after imagery of a face , indicating that imagery activates similar representations as the N170 ( Ganis and Schendan , 2008 ) . Here we confirm that idea and show that N170 representations are active during imagery throughout time . Furthermore , this time also showed long temporal generalization within perception , indicating that the N170 representations also remain active over time during perception . The lack of generalization between imagery and perceptual processing around 210 ms after stimulus onset was unexpected . This time window also showed an increase in temporal specificity during perception , indicating rapidly changing representations . One possible interpretation is that around this time feedback from higher areas arrives in low-level visual cortex ( Koivisto et al . , 2011; Roelfsema et al . , 1998 ) . If low-level representations are indeed more transient , this would explain the decrease in consistent generalization . Another possibility is that processing at this time reflects an unstable combination of feedback and feedforward processes , which is resolved around 300 ms when representations become more generalized and again start to generalize to imagery . In line with this idea , processing from 300 ms after stimulus onset has been associated with percept stabilization ( Bachmann , 2006; Carlson et al . , 2013; Kaiser et al . , 2016 ) . Future studies looking at changes in effective connectivity over time are needed to dissociate these interpretations . Surprisingly , we did not observe any influences of experienced imagery vividness on the overlap between perception and imagery over time ( Fig . S2 ) . One explanation for this is that we used whole-brain signals for decoding whereas the relationship between overlap and vividness has only been found for a specific set of brain regions ( Lee et al . , 2012; Dijkstra et al . , 2017a ) . Furthermore , if there is indeed strong temporal variability during imagery this would make it difficult to find any effect of vividness on specific time points . More studies on imagery vividness using MEG are necessary to explore this matter further . In conclusion , our findings suggest that , in contrast to perception , at the onset of imagery the entire visual representation is activated at once . This might partly be caused by the nature of our task , since visual representations were already present in working memory at the onset of imagery . However , more research is needed to fully explore the contribution of temporal uncertainty between trials to this broad temporal generalization . Furthermore , imagery consistently overlapped with perceptual processing around 160 ms and from 300 ms onwards . This reveals the temporal counterpart of the neural overlap between imagery and perception . The overlap around 160 ms points towards a re-activation of the N170 during imagery , whereas the lack of overlap with perceptual processes before 130 ms indicates that either imagery does not rely on early perceptual representations , or that these representations are more transient and variable over time . Together , these findings reveal important new insights into the neural mechanisms of visual imagery and its relation to perception .
We assumed a medium effect size ( d = 0 . 6 ) which , to reach a power of 0 . 8 , required twenty four participants . To take into account drop-out , thirty human volunteers with normal or corrected-to-normal vision gave written informed consent and participated in the study . Five participants were excluded: two because of movement in the scanner ( movement exceeded 15 mm ) , two due to incorrect execution of the task ( less than 50% correct on the catch trials , as described below ) and one due to technical problems . 25 participants ( mean age 28 . 6 , SD = 7 . 62 ) remained for the final analysis . The study was approved by the local ethics committee and conducted according to the corresponding ethical guidelines ( CMO Arnhem-Nijmegen ) . Prior to scanning , participants were asked to fill in the Vividness of Visual Imagery Questionnaire ( VVIQ ) : a 16-item questionnaire in which participants indicate their imagery vividness for a number of scenarios on a 5-point scale ( Marks , 1973 ) . The VVIQ has been used in many imagery studies and is a well-validated measure of general imagery ability ( Lee et al . , 2012; Albers et al . , 2013; Dijkstra et al . , 2017a; Cui et al . , 2007 ) . The score was summarized in a total between 16 and 80 ( low score indicates high vividness ) . Subsequently , the participants practiced the experimental task for ten trials outside the scanner , after which they were given the opportunity to ask clarification questions about the task paradigm . If they had difficulty with the task , they could practice a second time with ten different trials . The experimental task is depicted in Figure 1 . We adapted a retro-cue paradigm in which the cue was orthogonalized with respect to the stimulus identity ( Harrison and Tong , 2009 ) . Participants were shown two images after each other , a face and a house , or a house and a face , followed by a retro-cue indicating which of the images had to be imagined . After the cue , a frame was shown in which the participants had to imagine the cued stimulus as vividly as possible . After this , they had to indicate their experienced imagery vividness by moving a bar on a continuous scale . The size of the scale together with the screen resolution led to discretized vividness values between −150 and +150 . To prevent preparation of a motor response during imagery , which side ( left or right ) indicated high vividness , was pseudo-randomized over trials . The face stimuli were adapted from the multiracial face database ( courtesy of Michael J Tarr , Center for the Neural Basis of Cognition and Department of Psychology , Carnegie Mellon University ( Pittsburgh , Pennsylvania ) , http://www . tarrlab . org . Funding provided by NSF award 0339122 ) . The house stimuli were adapted from the Pasedena houses database collected by Helle and Perona ( California Institute of Technology , Pasadena , California ) . We chose faces and houses because these two categories elicit very different neural responses throughout the visual hierarchy , during both perception and imagery ( Ishai et al . , 2000; Epstein et al . , 2003; Kanwisher et al . , 1997 ) , and are therefore expected to allow for high-fidelity tracking of their corresponding neural representations . To ensure that participants were imagining the stimuli with great visual detail , both categories contained eight exemplars , and on 7% of the trials the participants had to indicate which of four exemplars they imagined ( Figure 1 , Catch trial ) . The exemplars were chosen to be highly similar in terms of low-level features to minimize within-class variability and increase between-class classification performance . We instructed participants to focus on vividness and not on correctness of the stimulus , to motivate them to generate a mental image including all visual features of the stimulus . The stimuli encompassed 2 . 7 × 2 . 7 visual degrees . A fixation bull’s-eye with a diameter of 0 . 1 visual degree was on screen throughout the trial , except during the vividness rating . In total , there were 240 trials , 120 per category , divided in ten blocks of 24 trials , lasting about 5 min each . After every block , the participant had the possibility to take a break . Data were recorded at 1200 Hz using a 275-channel MEG system with axial gradiometers ( VSM/CTF Systems , Coquitlam , BC , Canada ) . For technical reasons , data from five sensors ( MRF66 , MLC11 , MLC32 , MLF62 , MLO33 ) were not recorded . Subjects were seated upright in a magnetically shielded room . Head position was measured using three coils: one in each ear and one on the nasion . Throughout the experiment head motion was monitored using a real-time head localizer ( Stolk et al . , 2013 ) . If necessary , the experimenter instructed the participant back to the initial head position during the breaks . This way , head movement was kept below 8 mm in most participants . Furthermore , both horizontal and vertical electro-oculograms ( EOGs ) , as well as an electrocardiogram ( ECG ) were recorded for subsequent offline removal of eye- and heart-related artifacts . Eye position and pupil size were also measured for control analyses using an Eye Link 1000 Eye tracker ( SR Research ) . Data were analyzed with MATLAB version R2017a and FieldTrip ( Oostenveld et al . , 2011 ) ( RRID: SCR_004849 ) . Per trial , three events were defined . The first event was defined as 200 ms prior to onset of the first image until 200 ms after the offset of the first image . The second event was defined similarly for the second image . Further analyses focused only on the first event , because the neural response to the second image is contaminated by the neural response to the first image . Finally , the third event was defined as 200 ms prior to the onset of the retro-cue until 500 ms after the offset of the imagery frame . As a baseline correction , for each event , the activity during 300 ms from the onset of the initial fixation of that trial was averaged per channel and subtracted from the corresponding signals . The data were down-sampled to 300 Hz to reduce memory and CPU load . Line noise at 50 Hz was removed from the data using a DFT notch filter . To identify artifacts , the variance of each trial was calculated . Trials with high variance were visually inspected and removed if they contained excessive artifacts . After artifact rejection , on average 108 perception face trials ( ±11 ) , 107 perception house trials ( ±12 ) and 105 imagery face trials ( ±16 ) and 106 imagery house trials ( ±13 ) remained for analysis . To remove eye movement and heart rate artifacts , independent components of the MEG data were calculated and correlated with the EOG and ECG signals . Components with high correlations were manually inspected before removal . The eye tracker data was cleaned separately by inspecting trials with high variance and removing them if they contained blinks or other excessive artifacts . To track the neural representations within perception and imagery , we decoded the stimulus category from the preprocessed MEG signals during the first stimulus and after the retro-cue for every time point . To improve the signal-to-noise ratio , prior to classification , the data were averaged over a window of 30 ms centered on the time point of interest . We used a linear discriminant analysis ( LDA ) classifier with the activity from the 270 MEG sensors as features ( see Mostert et al . , 2015 for more details ) . A 5-fold cross-validation procedure was implemented where for each fold the classifier was trained on 80% of the trials and tested on the other 20% . To prevent a potential bias in the classifier , the number of trials per class was balanced per fold by randomly removing trials from the class with the most trials until the trial numbers were equal between the classes . By training a classifier on one time point and then testing it on other time points , we were able to investigate the stability of neural representations over time . The resulting temporal generalization pattern gives information about the underlying processing dynamics . For instance , a mostly diagonal pattern reflects sequential processing of specific representations , whereas generalization from one time point towards another reflects recurrent or sustained activity of a particular process ( King and Dehaene , 2014 ) . Here , we performed temporal generalization analyses during perception and during imagery to investigate the dynamics of the neural representations . Furthermore , to quantify the extent to which the representation at a given time point t was specific to that time point , we tested whether a classifier trained at time t and tested at time t ( i . e . diagonal decoding ) had a higher accuracy than a classifier trained at time t’ and tested at time t ( i . e . generalization ) . This shows whether there is more information at time t than can be extracted by the decoder t’ ( Astrand et al . , 2015; King et al . , 2014 ) . We subsequently calculated , for each training time point , the proportion of testing time points that were significantly lower than the diagonal decoding , giving a measure of specificity for each time point . To avoid overestimating the specificity , we only considered the time window during which the diagonal classifiers were significantly above chance . To investigate the overlap in neural representations between perception and imagery , a similar approach can be used . Here , we trained a classifier on different time points during perception and tested it on different time points during imagery and vice versa . This analysis shows when neural activity during perception contains information that can be used to dissociate mental representations during imagery and vice versa - that is which time points show representational overlap . For both the temporal generalization as well as the across condition generalization analysis , we also applied cross-validation to avoid overestimating generalization due to autocorrelation in the signals . It has been shown that representational overlap between imagery and perception , as measured by fMRI , is related to experienced imagery vividness ( Lee et al . , 2012; Albers et al . , 2013; Dijkstra et al . , 2017a ) . To investigate this in the current study , we performed a median split on the averaged vividness across trials on the group level , which yielded a high vividness ( N = 12 , vividness: 71 . 64 ± 12 . 44 ) and a low vividness ( N = 12 , vividness: 27 . 25 ± 17 . 69 ) group . We produced the accuracy maps for all previous analyses separately for the two groups and compared the decoding accuracies of the two groups using cluster based permutation testing ( see Statistical testing ) . Decoding accuracy was tested against chance using two-tailed cluster-based permutation testing with 1000 permutations ( Maris and Oostenveld , 2007 ) . In the first step of each permutation , clusters were defined by adjacent points that crossed a threshold of p<0 . 05 . The t-values were summed within each cluster , but separately for positive and negative clusters , and the largest of these were included in the permutation distributions . A cluster in the true data was considered significant if its p-value was less than 0 . 05 based on the permutations . Correlations with vividness were tested against zero on the group level using the same procedure . In order to identify the brain areas that were involved in making the dissociation between faces and houses during perception and imagery , we performed source reconstruction . In the case of LDA classifiers , the spatial pattern that underlies the classification reduces to the difference in magnetic fields between the two conditions ( see Haufe et al . , 2014 ) . Therefore , to infer the contributing brain areas , we performed source analysis on the difference ERF between the two conditions . For this purpose , T1-weighted structural MRI images were acquired using a Siemens 3T whole body scanner . Vitamin E markers in both ears indicated the locations of the head coils during the MEG measurements . The location of the fiducial at the nasion was estimated based on the anatomy of the ridge of the nose . The volume conduction model was created based on a single shell model of the inner surface of the skull . The source model was based on a reconstruction of the cortical surface created for each participant using FreeSurfer’s anatomical volumetric processing pipeline ( RRID: SCR_001847 ) . MNE-suite ( Version 2 . 7 . 0; RRID: SCR_005972 ) was subsequently used to infer the subject-specific source locations from the surface reconstruction . The resulting head model and source locations were co-registered to the MEG sensors . The lead fields were rank reduced for each grid point by removing the sensitivity to the direction perpendicular to the surface of the volume conduction model . Source activity was obtained by estimating linearly constrained minimum variance ( LCMV ) spatial filters ( Van Veen et al . , 1997 ) . The data covariance was calculated over the interval of 50 ms to 1 s after stimulus onset for perception and over the entire segment for imagery . The data covariance was subsequently regularized using shrinkage with a regularization parameter of 0 . 01 ( as described in Manahova et al . , 2017 ) . These filters were then applied to the axial gradiometer data , resulting in an estimated two-dimensional dipole moment for each grid point over time . For imagery , the data were low-pass filtered at 30 Hz prior to source analysis to increase signal to noise ratio . To facilitate interpretation and visualization , we reduced the two-dimensional dipole moments to a scalar value by taking the norm of the vector . This value reflects the degree to which a given source location contributes to activity measured at the sensor level . However , the norm is always a positive value and will therefore , due to noise , suffer from a positivity bias . To counter this bias , we employed a permutation procedure in order to estimate this bias . Specifically , in each permutation , the sign of half of the trials were flipped before averaging and projecting to source space . This way , we cancelled out the systematic stimulus-related part of the signal , leaving only the noise . Reducing this value by taking the norm thus provides an estimate of the positivity bias . This procedure was repeated 1000 times , resulting in a distribution of the noise . We took the mean of this distribution as providing the most likely estimate of the noise and subtracted this from the true , squared source signal . Furthermore , this estimate provides a direct estimate of the artificial amplification factor due to the depth bias . Hence , we also divided the data by the noise estimate to obtain a quantity that allowed visualization across cortical topography . For full details , see Manahova et al . ( 2017 ) . For each subject , the surface-based source points were divided into 74 atlas regions as extracted by FreeSurfer on the basis of the subject-specific anatomy ( Destrieux et al . , 2010 ) . To enable group-level estimates , the activation per atlas region was averaged over grid points for each participant . Group-level activations were then calculated by averaging the activity over participants per atlas region ( van de Nieuwenhuijzen et al . , 2016 ) . | If someone stops you in the street to ask for directions , you might find yourself picturing a particular crossing in your mind’s eye as you explain the route . This ability to mentally visualize things that we cannot see is known as visual imagery . Neuroscientists have shown that imagining an object activates some of the same brain regions as looking at that object . But do these regions also become active in the same order when we imagine rather than perceive ? Our ability to see the world around us depends on light bouncing off objects and entering the eye , which converts it into electrical signals . These signals travel to an area at the back of the brain that processes basic visual features , such as lines and angles . The electrical activity then spreads forward through the brain toward other visual areas , which perform more complex processing . Within a few hundred milliseconds of light entering the eye , the brain generates a percept of the object in front of us . So , does the brain perform these same steps when we mentally visualize an object ? Dijkstra et al . measured brain activity in healthy volunteers while they either imagined faces and houses , or looked at pictures of them . Electrical activity spread from visual areas at the back of the brain to visual areas nearer the front as the volunteers looked at the pictures . But this did not happen when the volunteers imagined the faces and houses . Contrary to perception , the different brain areas did not seem to become activated in any apparent order . Instead , the brain areas active during imagining were those that only became active during perception after 130 milliseconds . This is the time at which brain areas responsible for complex visual processing become active when we look at objects . These findings shed new light on how we see with our mind’s eye . They suggest that when we imagine an object , the brain activates the entire representation of that object at once rather than building it up in steps . Understanding how the brain forms a mental image in real time could help us develop new technologies , such as brain-computer interfaces . These devices aim to interpret patterns of brain activity and display the output on a computer . Such equipment could help people with paralysis to communicate . | [
"Abstract",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"neuroscience"
] | 2018 | Differential temporal dynamics during visual imagery and perception |
Phytophthora infestans , the cause of potato late blight , is infamous for having triggered the Irish Great Famine in the 1840s . Until the late 1970s , P . infestans diversity outside of its Mexican center of origin was low , and one scenario held that a single strain , US-1 , had dominated the global population for 150 years; this was later challenged based on DNA analysis of historical herbarium specimens . We have compared the genomes of 11 herbarium and 15 modern strains . We conclude that the 19th century epidemic was caused by a unique genotype , HERB-1 , that persisted for over 50 years . HERB-1 is distinct from all examined modern strains , but it is a close relative of US-1 , which replaced it outside of Mexico in the 20th century . We propose that HERB-1 and US-1 emerged from a metapopulation that was established in the early 1800s outside of the species' center of diversity .
Potato late blight's impact on humankind is rivaled by few other plant diseases . The Spanish introduced Europeans to the South American staple crop potato shortly after their conquest of the New World , but for three centuries Europe stayed free of P . infestans , the causal agent of late blight . In 1845 , the oomycete P . infestans finally reached Europe , spreading rapidly from Belgium to other countries of mainland Europe and then to Great Britain and Ireland . The impact of the epidemic reached catastrophic levels in Ireland , where the population was more dependent on potato for their subsistence than in other parts of Europe ( Bourke , 1964; Reader , 2009 ) . The subsequent Great Famine killed around 1 million people , and an additional million were forced to leave the island ( Turner , 2005 ) . Even today , the Irish population remains less than three quarters of what it was at the beginning of the 1840s . These dramatic consequences of the P . infestans epidemic were due to the absence of chemical and genetic methods to combat it; such means became available only several decades later . Ever since triggering the Irish famine , P . infestans has continued to wreak havoc on potato fields throughout the world . Late blight remains the most destructive disease of the third largest food crop , resulting in annual losses of potatoes that would be sufficient to feed anywhere from 80 to many hundreds of millions of people ( Fisher et al . , 2012 ) . Phytophthora infestans is an extraordinarily virulent and adaptable pathogen ( Fry , 2008; Haas et al . , 2009 ) . In agricultural systems , sexual reproduction may trigger explosive population shifts that are driven by the emergence and migration of asexual lineages ( Fry et al . , 1992 , 2009; Cooke et al . , 2012 ) . The species is thought to originate from Toluca Valley , Mexico , where it infects wild relatives of potato , frequently undergoes sexual reproduction and co-occurs with the two closely related species P . mirabilis and P . ipomoeae ( Tooley et al . , 1985; Goodwin et al . , 1994; Flier et al . , 2003; Grünwald and Flier , 2005 ) . In its center of origin , P . infestans is characterized by high levels of genetic and phenotypic diversity ( Grünwald and Flier , 2005 ) . The genomes of a few P . infestans strains have been described ( Haas et al . , 2009; Raffaele et al . , 2010a; Cooke et al . , 2012 ) . Compared to other species in the genus , the 240 Mb T30-4 reference genome of P . infestans is large , with three quarters of the genome consisting of repetitive DNA . A large number of genes codes for effector proteins , many of which are delivered inside plant cells to promote host colonization , for instance by suppressing plant immunity . RXLR proteins , the main class of host-translocated effectors , are encoded by about 550 genes in the P . infestans T30-4 genome . RXLR effectors that can be recognized by plant immune receptors , known as Resistance ( R ) proteins , are said to have ‘avirulence’ activity . Upon introduction of a cognate R gene into the host population , such avirulence effectors become a liability for the pathogen , and natural selection favors the spread of pseudogenized or mutated alleles ( Vleeshouwers et al . , 2011 ) . The detailed descriptions and drawings of Heinrich Anton de Bary and the reports of several other pioneers of plant pathology leave no doubt that the 19th century blight epidemic was triggered by P . infestans ( de Bary , 1876; Bourke , 1964 ) . What remains controversial is the relationship of the 19th century strains to modern isolates . The quest for understanding the origin of the strain that resulted in the Irish famine began with extant samples . Prior to the late 1970s , global P . infestans populations outside of South America and Mexico , the centers of diversity of the host and the pathogen , were dominated by a single clonal lineage that had the mitochondrial ( mtDNA ) haplotype Ib and was called US-1 ( Goodwin et al . , 1994 ) . It was therefore proposed that the US-1 lineage was a direct descendant of the strain that first caused widespread late blight in North America from 1843 on , and then triggered the Irish famine beginning in 1845 ( Bourke , 1964; Goodwin et al . , 1994 ) . This hypothesis was subsequently directly addressed by PCR analysis of infected 19th century potato leaves stored in herbaria . The conclusion from these studies was that the historic strains belonged to a mtDNA haplotype , Ia , that was distinct from that of the US-1 lineage ( Ristaino et al . , 2001; May and Ristaino , 2004 ) . Because Ia was at the time not only the predominant haplotype in the Toluca Valley in Mexico ( Gavino and Fry , 2002; Flier et al . , 2003 ) , but had also been found in South America ( Perez et al . , 2001 ) , May and Ristaino ( 2004 ) speculated that the 19th century and US-1 lineages represented two independent epidemics of divergent lineages that had both originated in South America and spread from there to North America and Europe . A caveat was that these far-reaching conclusions were based on only three mtDNA SNPs ( Ristaino et al . , 2001; May and Ristaino , 2004 ) . Since these first herbarium analyses , the retrieval and sequencing of DNA from museum specimens , fossil remains and archaeological samples—collectively known as ancient DNA ( aDNA ) ( Pääbo et al . , 2004 ) —have seen impressive advances thanks to the advent of high-throughput sequencing technologies . The combined analysis of modern and ancient genomes of human pathogens has begun to solve important questions about their history and evolution ( Bos et al . , 2011 , 2012 ) . Here we show that aDNA methods hold similar promise for plant pathology and that they can improve our understanding of historically important plant pathogen epidemics . To determine how the historic P . infestans strain ( s ) relate to extant isolates , we shotgun-sequenced 11 herbarium samples of infected potato and tomato leaves collected from continental Europe , Great Britain , Ireland , and North America in the period from 1845 to 1896 , and extracted information on P . infestans mitochondrial and nuclear genomes . To understand the subsequent evolution of the pathogen , we compared the historic P . infestans genomes to those of 15 modern 20th century strains that span the genetic diversity of the species , and to the two sister species P . ipomoeae and P . mirabilis . Our analyses revealed that the 19th century epidemic was caused by a single genotype , HERB-1 , that persisted for at least 50 years . While it is distinct from all examined modern strains , HERB-1 is closely related to the 20th century US-1 genotype , suggesting that these two pandemic genotypes may have emerged from a secondary metapopulation rather than from the species’ Mexican center of diversity .
Nineteenth-century samples of potato and tomato leaves with P . infestans lesions were obtained from the herbaria of the Botanische Staatssammlung München and the Kew Royal Botanical Gardens ( Table 1 and Figure 1 ) . DNA was extracted under clean room conditions and two genomic libraries were prepared from each sample for Illumina sequencing . The preparations were expected to comprise P . infestans DNA , host DNA from potato or tomato as well as DNA from microbes that had colonized either the living material at the time of its collection , or the dried material during its storage in the herbaria . 10 . 7554/eLife . 00731 . 003Table 1 . Provenance of P . infestans samplesDOI: http://dx . doi . org/10 . 7554/eLife . 00731 . 003IDCountry of originCollection yearHost speciesReference*Herbarium samplesKM177500England1845Solanum tuberosum1KM177513Ireland1846Solanum tuberosum1KM177502England1846Solanum tuberosum1KM177497England1846Solanum tuberosum1KM177514Ireland1847Solanum tuberosum1KM177548England1847Solanum tuberosum1KM177507England1856Petunia hybrida1M-0182898GermanyBefore 1863Solanum tuberosum2KM177509England1865Solanum tuberosum1M-0182900Germany1873Solanum lycopersicum2M-0182907Germany1875Solanum tuberosum1KM177517Wales1875Solanum tuberosum1M-0182897USA1876Solanum lycopersicum2M-0182906Germany1877Solanum tuberosum2M-0182896Germany1877Solanum tuberosum2M-0182904Austria1879Solanum tuberosum2M-0182903Canada1896Solanum tuberosum2KM177512EnglandNASolanum tuberosum1Modern samples06_3928AEngland2006Solanum tuberosum3DDR7602Germany1976Solanum tuberosum4P1362Mexico1979Solanum tuberosum5P6096Peru1984Solanum tuberosum5P7722 ( P . mirabilis ) USA1992Solanum lycopersicum5P9464USA1996Solanum tuberosum5P12204Scotland1996Solanum tuberosum5P13527Ecuador2002Solanum andreanum5P10127USA2002Solanum lycopersicum5P13626Ecuador2003Solanum tuberosum5P10650Mexico2004Solanum tuberosum5LBUS5South Africa2005Petunia hybrida6P11633Hungary2005Solanum lycopersicum5NL07434Netherlands2007Solanum tuberosum3P17777USA2009Solanum lycopersicum5P17721USA2009Solanum tuberosum5*1 , Kew Royal Botanical Gardens; 2 , Botanische Staatssammlung München; 3 , Cooke et al . ( 2012 ) ; 4 , Kamoun et al . ( 1999 ) ; 5 , World Oomycete Genetic Resource Collection at UC Riverside , CA; 6 , Dr Adele McLeod , Univ . of Stellenbosch , South Africa . 10 . 7554/eLife . 00731 . 004Figure 1 . Countries of origin of samples used in whole-genome , mtDNA genome or both analyses . Red indicates number of historic and blue of modern samples . More information on the samples is given in Tables 1 and 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 00731 . 004 The first set of libraries was used for verification of aDNA-like characteristics , and the second set was used for production . In this second set we used a repair protocol that removes aDNA-associated lesions , mainly characterized by cytosine deamination to uracil ( U ) , which is subsequently converted to and read as thymine ( T ) ( Hofreiter et al . , 2001; Briggs et al . , 2007 , 2010; Brotherton et al . , 2007 ) . The combination of uracil-DNA-glycosylase ( UDG ) and endonuclease VIII , which removes uracil residues and repairs abasic sites , reduces the overall per-base error rate to as low as one 20th of unrepaired aDNA ( Briggs et al . , 2010 ) . Ancient DNA fragments are typically shorter than 100 bp ( Pääbo , 1989 ) , and paired-end reads of 100 bases each will therefore substantially overlap . Forward and reverse reads from the unrepaired libraries ( Table 2 ) were merged , requiring at least 11 base overlap . Merging of short-insert libraries considerably decreases the error-rate and also generates sequences that reflect the original molecule length ( Kircher , 2012 ) . The median length of merged reads was in the range of approximately 50–85 bp ( Figure 2A , B ) . 10 . 7554/eLife . 00731 . 005Table 2 . Sequencing strategyDOI: http://dx . doi . org/10 . 7554/eLife . 00731 . 005IDInstrument and read typeSequencing centerCoverageM-0182896HiSeq 2000 ( 2 × 101 bp ) MPIHighM-0182897HiSeq 2000 ( 2 × 101 bp ) MPILow*M-0182898HiSeq 2000 ( 2 × 101 bp ) MPILowM-0182900HiSeq 2000 ( 2 × 101 bp ) MPILow†M-0182903HiSeq 2000 ( 2 × 101 bp ) MPILowM-0182904HiSeq 2000 ( 2 × 101 bp ) MPILow*M-0182906HiSeq 2000 ( 2 × 101 bp ) MPILow†M-0182907HiSeq 2000 ( 2 × 101 bp ) MPILowKM177497MiSeq ( 2 × 150 bp ) MPILowKM177500MiSeq ( 2 × 150 bp ) MPILow*KM177502AMiSeq ( 2 × 150 bp ) MPILow*KM177507MiSeq ( 2 × 150 bp ) MPILow*KM177509MiSeq ( 2 × 150 bp ) and HiSeq 2000 ( 2 × 101 bp ) MPILowKM177512MiSeq ( 2 × 150 bp ) and HiSeq 2000 ( 2 × 101 bp ) MPILowKM177513MiSeq ( 2 × 150 bp ) and HiSeq 2000 ( 2 × 101 bp ) MPILowKM177514MiSeq ( 2 × 150 bp ) and HiSeq 2000 ( 2 × 101 bp ) MPILowKM177517MiSeq ( 2 × 150 bp ) and HiSeq 2000 ( 2 × 101 bp ) MPILowKM177548MiSeq ( 2 × 150 bp ) and HiSeq 2000 ( 2 × 101 bp ) MPILow06_3928AGAIIX ( 2 × 76 bp ) TSLHighDDR7602GAIIX ( 2 × 76 bp ) TSLHighLBUS5GAIIX ( 2 × 76 bp ) TSLHighNL07434GAIIX ( 2 × 76 bp ) TSLHighP10127HiSeq 2000 ( 2 × 101 bp ) MPILowP10650HiSeq 2000 ( 2 × 101 bp ) MPILowP12204HiSeq 2000 ( 2 × 101 bp ) MPILowP13527GAIIX ( 2 × 76 bp ) TSLHighP1362HiSeq 2000 ( 2 × 101 bp ) MPILowP13626GAIIX ( 2 × 76 bp ) TSLHighP11633HiSeq 2000 ( 2 × 101 bp ) MPILowP17721HiSeq 2000 ( 2 × 101 bp ) MPILowP17777GAIIX ( 2 × 76 bp ) TSLHighP6096HiSeq 2000 ( 2 × 101 bp ) MPILowP7722HiSeq 2000 ( 2 × 101 bp ) MPILowP9464HiSeq 2000 ( 2 × 101 bp ) MPILow*PIC99114GAIIX ( 2 × 76 bp ) TSLHighPIC99167GAIIX ( 2 × 76 bp ) TSLHigh*Samples not included in any analysis due to extremely low coverage . †Samples used only in mtDNA analysis . 10 . 7554/eLife . 00731 . 006Figure 2 . Ancient DNA-like characteristic of historic samples . ( A ) Lengths of merged reads from historic sample M-0182898 . ( B ) Mean lengths of merged reads from historic samples . ( C ) Nucleotide mis-incorporation in reads from the historic sample M-0182898 . ( D ) Deamination at first 5′ end base in historic samples . ( E ) Percentage of merged reads that mapped to the P . infestans reference genome . DOI: http://dx . doi . org/10 . 7554/eLife . 00731 . 006 Merged sequences were mapped to the P . infestans T30-4 reference genome ( Haas et al . , 2009 ) . Deamination of C to U in aDNA is highest at the first base ( Briggs et al . , 2007 ) , and C-to-T was the predominant substitution at the 5′-end of molecules ( Figure 2C , D ) . Based on mapping against the reference genome , we estimated the fraction of P . infestans DNA in the samples to be between 1% and 20% ( Figure 2E ) . Most of the remaining reads could be mapped to the reference genomes for potato and tomato ( Potato Genome Sequencing Consortium , 2011; The Tomato Genome Consortium , 2012 ) . In addition to 11 historic samples from Ireland , Great Britain , continental Europe and North America ( Figure 1 and Table 1 ) , we shotgun sequenced 14 modern strains from Europe , the Americas and Africa ( Figure 1 and Table 1 ) . These had been selected based on preliminary mtDNA information to present a cross section of P . infestans diversity . Finally , we sequenced two strains of P . mirabilis , P7722 and PIC99114 , and a single strain of P . ipomoeae , PIC99167 . Both species are closely related to P . infestans and served as outgroups ( Kroon et al . , 2004; Raffaele et al . , 2010a ) . We considered genomes with mean-fold coverage of at least 20 as high coverage; one historic , seven modern and both outgroup genomes fulfilled this condition ( Figure 3A ) . We identified single nucleotide polymorphisms ( SNPs ) in each sample independently by comparison with the P . infestans reference T30-4 genome ( Figure 3B ) . Thresholds for calling homozygous and heterozygous SNPs were determined from simulated data from high- and low-coverage genomes ( Figure 3—figure supplement 1 ) . We accepted SNPs from low-coverage genomes if the variants had also been called in a high-coverage genome . Inverse cumulative coverage plots indicated how many high- or low-coverage samples were needed to cover different fractions of SNPs ( Figure 3C , D ) . A total of 4 . 5 million non-redundant SNPs were called . Eighty percent of all homozygous SNPs were found in at least eight samples , and only 20% of all SNPs were found in fewer than 10 strains . Thus , the great majority of polymorphic sites were shared by several strains and thus informative for phylogenetic analyses . 10 . 7554/eLife . 00731 . 007Figure 3 . Coverage and SNP statistics . ( A ) Mean nuclear genome coverage from historic ( red ) and modern ( blue ) samples . ( B ) Homo- and heterozygous SNPs in each sample . ( C ) Inverse cumulative coverage for all homozygous SNPs across all samples . ( D ) Same as ( C ) for homo- and heterozygous SNPs . DOI: http://dx . doi . org/10 . 7554/eLife . 00731 . 00710 . 7554/eLife . 00731 . 008Figure 3—figure supplement 1 . Accuracy and sensitivity of SNP calling at different cutoffs for SNP concordance based on 3- and 50-fold coverage of simulated data . Rescue cov . —minimum coverage required to accept SNP calls in low-coverage genomes based on these SNPs having been found in high-coverage genomes . The cutoffs enclosed in orange rectangles were used for the final analysis . DOI: http://dx . doi . org/10 . 7554/eLife . 00731 . 008 We reconstructed the mtDNA genomes from historic and modern strains using an iterative mapping assembler ( Green et al . , 2008 ) and estimated a phylogenetic tree from complete mtDNA genomes , with one of the P . mirabilis mtDNA genomes as outgroup . Previous studies have recognized four P . infestans mtDNA haplotype groups ( Ia , Ib , IIa and IIb ) , based on a small number of restriction fragment length polymorphisms ( RFLPs ) ( Carter et al . , 1990 ) . Surprisingly , a comparison of the complete mtDNA genomes revealed that the historic samples did not fit into any of these groups , and instead formed an independent clade , called HERB-1 from here on . Among the HERB-1 mtDNA genomes , there were very few differences , with a mean pair-wise difference of only 0 . 2 bp , compared to 3 . 9 bp for the modern haplotype I strains , and 9 . 0 bp for modern haplotype II strains . The origin of HERB-1 relative to haplotypes Ia and Ib could not be unequivocally resolved , and a polytomy was inferred for these three groups or support for branches were low ( Figure 4 and Figure 4—figure supplement 1 ) . 10 . 7554/eLife . 00731 . 009Figure 4 . Maximum-parsimony phylogenetic tree of complete mtDNA genomes . Sites with less than 90% information were not considered , leaving 24 , 560 sites in the final dataset . Numbers at branches indicate bootstrap support ( 100 replicates ) , and scale indicates changes . DOI: http://dx . doi . org/10 . 7554/eLife . 00731 . 00910 . 7554/eLife . 00731 . 010Figure 4—figure supplement 1 . Maximum-likelihood phylogenetic tree of complete mtDNA genomes . Sites with less than 90% information were not considered , leaving 24 , 560 sites in the final dataset . Numbers at branches indicate bootstrap support ( 100 replicates ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00731 . 01010 . 7554/eLife . 00731 . 011Figure 4—figure supplement 2 . mtDNA sequences around diagnostic Msp1 restriction site ( grey ) for reference haplotype modern strains ( blue ) and historic strains ( red ) . The Msp1 ( CCGG ) restriction site is only present in the Ib haplotype; all other strains have a C-to-T substitution ( CTGG ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00731 . 011 The clonal lineage US-1 , with the diagnostic mtDNA haplotype Ib , was the predominant genotype throughout the world until about 1980 ( Goodwin et al . , 1994 ) . The two US-1 representatives in our material , DDR7602 ( Germany ) and LBUS5 ( South Africa ) , clustered together with the Ib reference genome and samples P6096 ( Peru ) and P1362 ( Mexico ) ( Figure 4 and Figure 4—figure supplement 1 ) , even though these last two samples had not been classified before as US-1 isolates . Although the US-1 genotype is closely related to the herbarium strains , US-1 is not a derivative of HERB-1 . Rather , HERB-1 and US-1 are sister groups that share a common ancestor . There are three private substitutions that define the US-1 clade , and two that define the HERB-1 clade . In agreement with the previous report by Ristaino et al . ( 2001 ) , all historic samples had a T at the position diagnostic for haplotype Ib ( Figure 4—figure supplement 2 ) , which distinguishes them from the US-1 lineage , which carries instead a C at this position . In contrast to the previous report ( Ristaino et al . , 2001 ) , we found no other sequence differences around this diagnostic site . As the HERB-1 strains were sampled in the 19th century , their genomes should harbor fewer substitutions compared to modern strains , which have continued to evolve for over a hundred years . This can be exploited to directly calculate substitution rates and divergence times using the sampling age as tip calibration in a Bayesian framework analysis . Shorter evolutionary time usually translate into branch shortening in phylogenetic trees that include both modern and ancient pathogen strains ( Bos et al . , 2011 ) . By calculating the nucleotide distance to the outgroup P17777 , all HERB-1 strains were found to show significantly fewer mtDNA substitutions than modern strains with haplotype Ia or Ib ( p=0 . 0003 ) . Sampling age of the strain and the number of mtDNA substitutions were highly correlated ( r2 = 0 . 8; Figure 5 ) . 10 . 7554/eLife . 00731 . 012Figure 5 . Correlation between nucleotide distance of mtDNA genomes of HERB-1/haplotype Ia/haplotype Ib clade to the outgroup P17777 and sample age in calendar years before present . DOI: http://dx . doi . org/10 . 7554/eLife . 00731 . 012 Given the correlation between sample age and the number of mtDNA substitutions , a multiple sequence alignment of 12 nearly complete modern and the 13 HERB-1 mtDNA genomes was used as input for a Bayesian framework analysis using algorithms implemented in the software package Beast ( Drummond et al . , 2012 ) . The molecular clock hypothesis for the modern strains could not be rejected at a 5% significance level ( p=0 . 12 ) . Therefore , a strict molecular clock and a birth-death tree prior ( Stadler , 2010 ) were used for the Bayesian framework analysis . Collection dates for all herbaria samples and the isolation dates for all modern strains were used as tip calibration points , so that substitution rates per time interval could be calculated ( Table 1 ) . Three Markov Chain Monte Carlo ( MCMC ) runs with 147 million iterations were carried out . Stability of the estimated prior and posterior probability distributions ( ESS values >5000 ) and likelihood values ( ESS values >9000 ) were observed in the trace files throughout the merged iterations using the software Tracer ( Rambaut and Drummond , 2007 ) . From this procedure , we estimated the mtDNA substitution rate to be 2 . 4 × 10−6 per site and year ( 1 . 5–3 . 3 × 10−6 , 95% HPD ) . This rate resulted in a mean divergence time for P . infestans and P . mirabilis of 1318 years ago ( ya ) ( 853–1836 ya 95% HPD ) and for P . infestans type I and type II mtDNA haplotypes of 460 ya ( 300–643 ya 95% HPD ) . The origin of the 19th century herbarium clade was estimated to around 182 ya ( 168–201 ya 95% HPD ) ( Figure 6 and Table 3 ) . 10 . 7554/eLife . 00731 . 013Figure 6 . Divergence estimates of mtDNA genomes . Bayesian consensus tree from 147 , 000 inferred trees . Posterior probability support above 50% is shown next to each node . Blue horizontal bars represent the 95% HPD interval for the node height . Light yellow bars indicate major historical events discussed in the text . See Figure 5 and Table 3 for detailed estimates at the four main nodes in P . infestans . DOI: http://dx . doi . org/10 . 7554/eLife . 00731 . 01310 . 7554/eLife . 00731 . 014Table 3 . Inferred time to most recent common ancestor ( TMRCA ) for different splits in the mtDNA treeDOI: http://dx . doi . org/10 . 7554/eLife . 00731 . 014NodeTMRCA ( ya ) Best estimateLower 2 . 5%Upper 2 . 5%I/HERB-1 , II460300643Ia/Ib , HERB-1234187290HERB-1 strains182168201IIa , IIb14278214 To understand the evolutionary relationships between historical and modern strains in more detail , we also compared their nuclear genomes . We built phylogenetic trees with high-coverage genomes using maximum parsimony ( Figure 7A ) , maximum likelihood ( Figure 7B ) and neighbor-joining ( Figure 7—figure supplement 1 ) methods . We included in the analysis heterozygous biallelic SNPs by random sampling an allele from each of them . In all cases , the HERB-1 representative , M-0182896 , formed a distinct , isolated clade that appeared as a robust sister group to US-1 isolates DDR7602 and LBUS5 . As a caveat , the random sampling of SNPs at heterozygous positions will inflate bootstrap support . Nevertheless , a heat map ( Figure 7C ) highlights that the two US-1 strains are both most closely related to HERB-1 and the most distinct among modern strains . Phylogenetic analyses that included the low-coverage genomes ( Figure 7—figure supplement 1 ) were consistent with a close relationship between the HERB-1 and US-1 lineages . 10 . 7554/eLife . 00731 . 015Figure 7 . Phylogenetic trees of high-coverage nuclear genomes using both homozygous and heterozygous SNPs . ( A ) Maximum-parsimony tree , considering only sites with at least 95% information , leaving 4 , 498 , 351 sites in the final dataset . Numbers at branches indicate bootstrap support ( 100 replicates ) , and scale indicates genetic distance . ( B ) Maximum-likelihood tree . ( C ) Heat map of genetic differentiation ( color scale indicates SNP differences ) . US-1 strains DDR7062 and LBUS5 have the genomes sequences closest to M-0182896 ( asterisks ) . The two US-1 isolates in turn are outliers compared to all other modern strains ( highlighted by a gray box ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00731 . 01510 . 7554/eLife . 00731 . 016Figure 7—figure supplement 1 . Phylogenetic trees of high- and low-coverage nuclear genomes . ( A ) Neighbor-joining tree of high-coverage genomes using 4 , 595 , 012 homo- and heterozygous SNPs . Numbers at branches indicate bootstrap support ( 100 replicates ) , and scale indicates genetic distance . ( B ) Neighbor-joining tree of high- and low-coverage genomes using 2 , 101 , 039 homozygous and heterozygous SNPs . Numbers at branches indicate bootstrap support above 50 , from 100 replicates . Scale indicates genetic distance . ( C ) Maximum parsimony tree of high- and low-coverage genomes using 315 , 394 SNPs homozygous and heterozygous SNPs ( using only sites with at least 80% information ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00731 . 016 The independent diversification of the pandemic HERB-1 and US-1 lineages together with a very recent common ancestor are consistent with both lineages having originated from the same metapopulation . To test whether the global replacement of HERB-1 by US-1 in the 20th century was due to adaptive mutations , we searched for positively selected genes using PAML ( Yang , 2007 ) . We did not find any evidence for genes or sites that had experienced branch-specific positive selection in any of the lineages , only a mosaic pattern with potentially positively selected genes shared across lineages . Alternative scenarios could be that inactivating mutations were more important than non-synonymous substitutions , or that higher overall diversity and re-assortment of beneficial gene variants by recombination contributed to the success of US-1 . Genetic diversity can be increased by polyploidy , which has been reported in isolates of P . infestans ( e . g . , Daggett et al . , 1995; Catal et al . , 2010 ) , and which has major evolutionary implications for asexual organisms . To estimate ploidy level in our specimens , we investigated the distribution of read counts at biallelic SNPs for high-coverage genomes . In a diploid species , the mean of read counts at heterozygous positions should have a single mode at 0 . 5 , while there should be two modes , 0 . 33 and 0 . 67 , for triploid genomes , and three modes , 0 . 25 , 0 . 5 and 0 . 75 for tetraploid genomes . We compared the observed distributions of read counts with computational simulations of diploid , triploid and tetraploid genomes . Based on the shape and kurtosis of the distributions we concluded that the historic M-0182896 genome was apparently diploid . Of the modern genomes , only NL07434 and P17777 were diploid , whereas the majority , including the two US-1 strains DDR7602 and LBUS5 as well as P13527 and P13626 were triploid . One strain , 06_3928A , even seemed to be tetraploid ( Figure 8A , B and Figure 8—figure supplement 1 ) . This conclusion was supported by polyploid strains having evidence for triallelic polymorphism at many more sites than M-0182896 ( Figure 8C ) . 10 . 7554/eLife . 00731 . 017Figure 8 . Ploidy analysis . ( A ) Diagram of expected read frequencies of reads at biallelic SNPs for diploid , triploid and tetraploid genomes . ( B ) Reference read frequency at biallelic SNPs in gene dense regions ( GDRs ) for the historic sample M-0182896 , two modern samples , and simulated diploid , triploid and tetraploid genomes . The simulated tetraploid genome is assumed to have 20% of pattern 1 and 80% of pattern 3 shown in ( A ) . The shape and kurtosis of the observed distributions are similar to the corresponding simulated ones . ( C ) Polymorphic positions with more than one allele in the GDR . DOI: http://dx . doi . org/10 . 7554/eLife . 00731 . 01710 . 7554/eLife . 00731 . 018Figure 8—figure supplement 1 . Reference read frequency at biallelic SNPs in gene dense regions ( GDRs ) for five modern high-coverage samples . DOI: http://dx . doi . org/10 . 7554/eLife . 00731 . 018 To further confirm the ploidy inferences , we classified 40 , 352 SNPs as derived or ancestral based on information from the outgroup species P . mirabilis and P . ipomoeae . We then compared the rate of homozygosity at derived alleles in M-0182896 and DDR7602 . In agreement with the ploidy difference , M-0182896 had more than twice as many derived homozygous SNPs ( 8375 ) than DDR7602 ( 3440 ) , regardless of annotation as synonymous , non-synonymous and non-sense ( Figure 9A , B ) . 10 . 7554/eLife . 00731 . 019Figure 9 . Read allele frequencies of historic genome M-0182896 and US-1 isolate DDR7602 . Alleles were classified as ancestral or derived using outgroup species P . mirabilis and P . ipomoeae . There were 40 , 532 segregating sites . ( A ) Distributions of derived alleles at sites segregating between M-0182896 and DDR7602 . ( B ) Annotation of the different site classes . DOI: http://dx . doi . org/10 . 7554/eLife . 00731 . 019 Phytophthora infestans secretes a large repertoire of effector proteins , some of which are recognized by plant immune receptors encoded by R genes ( Haas et al . , 2009; Vleeshouwers et al . , 2011 ) . These R genes occur in wild potato ( Solanum ) species mostly originating from the pathogen center of diversity in Mexico , and have been introduced by breeding into cultivated potato since the beginning of the 20th century ( Hawkes , 1990 ) . The analysis of effector gene sequences in HERB-1 strains should reveal the effector repertoire prior to its disruption by the selective forces imposed by resistance gene breeding . Given that 19th century potato cultivars in North America and Europe were fully susceptible to late blight , we presume that they did not yet contain resistance genes that are effective against HERB-1 . Conversely , the first R genes for P . infestans resistance , introduced into cultivated potato only after the dates for our HERB-1 samples , should be effective against HERB-1 strains , which are predicted to carry matching avirulence effector genes . The R genes include in particular R1 to R4 from Solanum demissum ( Hawkes , 1990 ) . To date , 10 avirulence effector genes , recognized by 10 matching Solanum R genes , have been described in P . infestans ( Vleeshouwers et al . , 2011 ) . We first estimated the presence/absence profiles of these effector genes based on the fraction of gene length that was covered by reads ( ‘breadth of coverage’ ) in each high-coverage genome , and by merged reads from low-coverage genomes ( Table 4 ) . We deduced the amino acid sequences of these 10 effectors using both alignments of reads to the reference genome and de novo assemblies . All examined avirulence effector genes except Avr3b were present as full-length and intact coding sequences in the historic samples ( Table 4 ) , without any frame shift or nonsense mutations . The HERB-1 alleles of Avr1 , Avr2 , Avr3a and Avr4 were shared with those of the US-1 strain DDR7602 ( Table 5—source data 1 ) . In conclusion , the Avr1 , Avr2 , Avr3a and Avr4 alleles of HERB-1 are intact , presumably functional copies that are identical to ones that can be recognized by the matching R genes R1 , R2 , R3a and R4 ( Armstrong et al . , 2005; van Poppel et al . , 2008; Gilroy et al . , 2011; Vleeshouwers et al . , 2011 ) . This is consistent with the expectation that the HERB-1 genotype must have been avirulent on the first potato cultivars that acquired late blight resistance . 10 . 7554/eLife . 00731 . 020Table 4 . Presence or absence of avirulence effector genes in historic and modern samples , expressed as percentages of effector genes covered by readsDOI: http://dx . doi . org/10 . 7554/eLife . 00731 . 020Avr geneR geneHERB-1*US-1†20th century non-US-1OutgroupsEC3527EC3626P1777706_3928ANL07434MergedPm PIC99114Pip PIC99167Avr1R1100100001000010098100Avr2R2100100100100100811007797100Avr3aR3a100100100100100100100100028Avr3bR3b000010000100100100Avr4R41001001001009589100998592Avrblb1Rpi-blb110010010010010010010010000Avrblb2Rpi-blb21001001001009210010089880Avrvnt1Rpi-vnt1100100100100100100100100100100AvrSmira1Rpi-Smira110010010010010010010010097100AvrSmira2Rpi-Smira21001001001001001001001001000Sequences and polymorphisms are shown in Table 5 and Table 5—source data 1 . *Same sequences obtained for M-0182896 and merged sequences . †Same sequences obtained for DDR7602 and LBUS5 . 10 . 7554/eLife . 00731 . 021Table 5 . Amino acid differences in the avirulence effectors AVR1 , AVR2 , AVR3a and AVR4 encoded by the T30-4 reference genome , HERB-1 and DDR7602 ( US-1 ) DOI: http://dx . doi . org/10 . 7554/eLife . 00731 . 02110 . 7554/eLife . 00731 . 022Table 5—source data 1 . Full-length sequences of deduced amino acid sequences of HERB-1 AVR1 , AVR2 , AVR3a and AVR4 . DOI: http://dx . doi . org/10 . 7554/eLife . 00731 . 022PositionSubstitutionNoteT30-4HERB1DDR7602AVR1 ( PITG_16 , 663 ) 80TTT , SHERB-1 polymorphisms shared with T30-4 and DDR7602 . 142II , TT 154VV , AA 185III , VAVR2 ( PITG_22 , 870 ) 31NKKHERB-1 identical to DDR7602 . AVR3a ( PITG_14 , 371 ) 19SCCHERB-1 identical to DDR7602; both correspond to AVR3aKI isoform . 80EKK 103MII 139MLLAVR4 ( PITG_07 , 387 ) 19TT , ITHERB-1 polymorphisms shared with T30-4 and DDR7602 . 139LSL , S 221LVL , V 271VFV , FIDs in parentheses refer to gene models in reference genome . Full-length sequences of deduced amino acid sequences of HERB-1 AVR1 , AVR2 , AVR3a and AVR4 are provided in Table 5—source data 1 . We examined in more detail Avr3a , which is recognized by Solanum demissum R3a . The two major Avr3a alleles encode secreted proteins that differ in two amino acids in their effector domains: AVR3aKI and AVR3aEM ( Figure 10A , and Figure 10—figure supplement 1 ) . Only the AVR3aKI type triggers signaling by the resistance protein R3a ( Armstrong et al . , 2005 ) . The R3a gene was introduced into modern potato from S . demissum at the beginning of the 20th century , providing modern potato with resistance against the P . infestans strains prevalent at the time ( Hawkes , 1990; Gebhardt and Valkonen , 2001; Huang et al . , 2005 ) . Strains homozygous for Avr3aEM , which avoids R3a-mediated detection of the pathogen , appeared later; US-1 isolates lack Avr3aEM ( Armstrong et al . , 2005 ) . Examination of Avr3a SNPs in the historic samples only revealed the AVR3aKI allele , whereas both alleles were present in modern samples ( Figure 10B ) . To confirm that the potato hosts of the historic strains lacked the ability to recognize AVR3aKI , we assembled de novo short reads from the historic samples and aligned them against the R3a sequence from modern potato ( Huang et al . , 2005 ) . As expected , we only found R3 homologs that were distinct in sequence from R3a ( Figure 10C ) . 10 . 7554/eLife . 00731 . 023Figure 10 . The effector gene Avr3a and its cognate resistance gene R3a . ( A ) Diagram of AVR3A effector protein . ( B ) Frequency of Avr3a alleles in historic and modern P . infestans strains . ( C ) Neighbor-joining tree of R3a homologs from potato , based on 0 . 67 kb partial nucleotide sequences of S . tuberosum R3a ( blue , accession number AY849382 . 1 ) and homologs ( dark grey ) in GenBank , and de novo assembled contigs from M-0182896 ( red ) . Numbers at branches indicate bootstrap support with 500 replicates . Scale indicates changes . DOI: http://dx . doi . org/10 . 7554/eLife . 00731 . 02310 . 7554/eLife . 00731 . 024Figure 10—figure supplement 1 . Summary of de novo assembly of RXLR effector genes . TBLASTN query was performed with 549 RXLR proteins as a query and contigs as a database . When the High-scoring Segment Pair ( HSP ) and matched amino acids both covered ≥99% of the query length , we recorded a hit . Results with the optimal k-mer size are highlighted . DOI: http://dx . doi . org/10 . 7554/eLife . 00731 . 024 The absence of the Avr3b effector gene in HERB-1 could be viewed as puzzling , given that the S . demissum R3 locus was one of the first to be bred into potato . However , R3b , the matching R gene of Avr3b , is within 0 . 4 cM of R3a in the complex R3 locus ( Li et al . , 2011 ) . Based on the absence of an Avr3b gene in HERB-1 , we conclude that initial introgression of the R3 locus from S . demissum was based on the resistance phenotype conferred by the R3a gene . The R3 phenotype scored during the initial introgression must have been the recognition of Avr3a by R3a , and the presence of R3b must have been irrelevant until P . infestans strains carrying Avr3b emerged .
Historic strains from different geographic locations all carried a mtDNA haplotype , HERB-1 , that had not been recognized before ( Figure 4 ) . Although collected over a period of 50 years , the strains were distinguished from each other by few nuclear SNPs , indicating that the 19th century outbreak was a true pandemic of a rapidly spreading clonal genotype . That HERB-1 has so far not been found in any modern strain may point to its extinction after the 19th century pandemic , possibly associated with the onset of resistance gene breeding in the 20th century . We cannot , however , exclude that HERB-1 still infects some localized , genetically unimproved host populations , since we have explored only a fraction of current P . infestans genetic diversity . With the diagnostic variants we have discovered , one can now probe modern populations to determine whether or not HERB-1 still persists somewhere . Historic pathogen samples are molecular fossils that can be used as tip calibration points to estimate major divergence events in the evolution of a pathogen ( Bos et al . , 2011 ) . Using the collection dates of the herbarium samples and isolation dates of the modern P . infestans strains , we estimated that type I and type II mtDNA haplotypes diverged close to the beginning of the 16th century ( Figure 6 and Table 3 ) . This coincides with the first contact between Americans and Europeans in Mexico , which potentially fuelled P . infestans population migration and expansion outside its center of diversity . This major event in human history might thus have been responsible for wider dissemination of the P . infestans pathogen in the New World , several centuries before its introduction to Europe . In addition , the divergence estimates allowed us to date the split between P . mirabilis and P . infestans about 1300 ya . Even though this was firmly during the period of pre-Columbian civilization , what led to their relatively recent speciation remains unknown . To test the congruence of mtDNA and nuclear phylogenies , we reconstructed phylogenies with over 4 million nuclear SNPs from high-quality genomes ( Figure 7 ) . This confirmed the historic sample M-0182896 as a sister group to US-1 strains , a conclusion that was supported by a broader analysis that included the low-coverage historic samples ( Figure 7—figure supplement 1 ) . The private SNPs shared by the HERB-1 lineage ruled out that US-1 isolates are , as previously proposed ( Goodwin et al . , 1994 ) , direct descendants of the historic strains . Nevertheless , US-1 is more closely related to the historic strains than to the modern strains that have come to dominate the global population in the past two decades . We therefore propose a revision of the previous scenario , which posited that a single P . infestans lineage migrated around 1842 or 1843 from Mexico to North America , from where it was soon transferred to Europe , followed by global dissemination and persistence for over hundred years ( Goodwin et al . , 1994 ) . Our data make it likely that by the late 1970s , direct descendants of HERB-1 had either become rare or extinct . On the other hand , the close relationship between HERB-1 and US-1 suggests that the US-1 lineage originated from a similar source as HERB-1 , with our divergence estimates indicating that the two lineages separated only in the early 19th century . Given the much greater genetic diversity at the species’ likely origin in Mexico , it seems unlikely that HERB-1 and US-1 spread independently from this region . An alternative scenario is that a small P . infestans metapopulation was established at the periphery of its center of origin , or even outside Mexico , possibly in North America , some time before the first global P . infestans pandemic . The first lineage to spread from there was HERB-1 , which persisted globally for at least half a century . Subsequently , the US-1 lineage expanded and spread , replacing HERB-1 ( Figure 11 ) . 10 . 7554/eLife . 00731 . 025Figure 11 . Suggested paths of migration and diversification of P . infestans lineages HERB-1 and US-1 . The location of the metapopulation that gave rise to HERB-1 and US-1 remains uncertain; here it is proposed to have been in North America . DOI: http://dx . doi . org/10 . 7554/eLife . 00731 . 025 Host R genes that confer resistance to historic P . infestans strains , such as R3a , were bred into cultivated potato Solanum tuberosum from the wild species S . demissum at the beginning of the 20th century , years after our youngest historic sample had been collected in 1896 . In agreement with the products of these R genes being able to recognize HERB-1 effectors , HERB-1 strains seem to have only the Avr3aKI allele , which interacts with the product of the R gene R3a to trigger a host immune response ( Armstrong et al . , 2005; Huang et al . , 2005 ) . Moreover , de novo assembly of potato DNA did not provide evidence for the presence of R3a in the herbarium hosts , consistent with the narrative of potato breeding ( Figure 10C; Hawkes , 1990 ) . While it is uncertain when HERB-1 was displaced by the US-1 lineage , the US-1 lineage also carries only the Avr3aKI allele ( Armstrong et al . , 2005 ) . The origin of the Avr3aEM allele that emerged to high frequency after the breeding of R3a into cultivated potatoes remains unclear . A major genomic difference between the HERB-1 and US-1 lineages is the shift in ploidy , from diploid to triploid and even tetraploid ( Figure 8 and Figure 8—figure supplement 1 ) . Polyploidization could have provided an opportunity for US-1 isolates to enhance allelic diversity in the absence of frequent sexual reproduction , and could thus have contributed to their global success . Asexual reproduction leads to an increase of deleterious mutation in the population due to a lack of meiotic recombination ( Felsenstein , 1974 ) . Therefore , if the major selection pressure that led to the replacement of HERB-1 by US-1 was the introduction of resistance gene breeding , greater variation at effector genes in polyploid US-1 strains could have contributed to the replacement of HERB-1 soon after R genes from S . demissum and other wild species had been introduced into modern potato germplasm . We present the first genome-wide analyses of historic plant pathogen strains . The aDNA in the herbarium samples , which were about 150 years old , was remarkably well conserved , much better than most examples of aDNA from animals and humans , and only comparable with permafrost samples ( Miller et al . , 2008; Rasmussen et al . , 2010 ) . Our analyses not only highlight how knowledge of the genetics and geographic distribution of modern strains is insufficient to correctly infer the source of historic epidemics ( Goodwin et al . , 1994 ) , but they also reveal the shortcomings of inferences that are based on a very small number of genetic markers in historic strains ( Ristaino et al . , 2001; May and Ristaino , 2004 ) . With our much richer dataset , we could demonstrate that the late blight outbreaks during the 19th century were a pandemic caused by a single P . infestans lineage , but that this lineage was not the direct ancestor of the one that had come to dominate the global P . infestans population during much of the 20th century . Infected plant specimens stored in herbaria around the world are thus a largely untapped source to learn about events that affected millions of people during our recent history .
Plant specimens were sent to the Senckenberg Museum in Frankfurt am Main by the Botanische Staatssammlung München and the Kew Royal Botanical Gardens , where potato and tomato leaves with lesions indicative of P . infestans infection were sampled , retrieving both the lesions and healthy surrounding tissue . Sampling was carried out under sterile conditions in a laboratory with no prior exposure to P . infestans . Samples were subsequently sent to the Palaeogenetics laboratory at the University of Tübingen . Preamplification steps of historic samples were performed in clean room facilities with no prior exposure to P . infestans DNA . Samples were extracted following the protocol of ( Kistler , 2012 ) , using 380–500 µg of each sample . Tissue was crushed with mortal and pestle , 1 . 2 ml extraction buffer ( 1% SDS , 10 mM Tris pH 8 . 0 , 5 mM NaCl , 50 mM DTT , 0 . 4 mg/ml proteinase K , 10 mM EDTA , 2 . 5 mM N-phenacylthiazolium bromide ) was added , and samples were incubated over night at 37°C with constant agitation . A modified protocol with the Qiagen Plant DNEasy Mini kit ( Qiagen , Hilden , Germany ) was then used to purify the extracted DNA . Two independent Illumina sequencing libraries were created for each DNA extract . In the first library , C-to-T damage caused by deamination of cytosines ( Hofreiter et al . , 2001 ) was not repaired . 20 µl of each DNA extract , extraction blank control and water library blank control were converted into sequencing libraries as described ( Meyer and Kircher , 2010 ) with modifications for aDNA ( Meyer et al . , 2012 ) . To avoid potential sequencing artifacts caused by DNA damage , a second library was made from , 30 µl of each DNA extract , extraction blank control and water library blank control , and treated with uracil-DNA glycosylase ( UDG ) and Endonuclease VIII before conversion into sequencing libraries ( Briggs et al . , 2010 ) . Each library received sample-specific double indices after preparation via amplification with two ‘index’ PCR primers ( Meyer et al . , 2012 ) . Indexed libraries were individually amplified in 100 µl reactions containing 5 µl library template , 2 units of AccuPrime Pfx DNA polymerase ( Invitrogen , Karlsruhe , Germany ) , 1 unit of 10× PCR Mix and 0 . 3 µM primers spanning the index sequences of the libraries . The following thermal profile was used: 2-min initial denaturation at 95°C , two or three cycles consisting of 15 s denaturation at 95°C , a 30-s annealing at 60°C and a 2-min elongation at 68°C , and a 5-min final elongation at 68°C . Amplified products were purified and quantified on an Agilent 2100 Bioanalyzer DNA 1000 chip . DNA extracts of the modern P . infestans samples P17721 , P10650 , P6096 , P11633 , P10127 , P9464 , P12204 and P13626 and P . mirabilis P7722 were obtained from the World Phytophthora and Oomycete Genetic Resource Collection , fragmented to 300 bp using a S220 Covaris instrument according to the manufacturers’ protocol ( Duty cycle 10% , intensity 4 , cycles per burst 200 , time [in s] 120 ) , and converted into sequencing libraries following the above steps as described for the historic samples without UDG treatment ( Kircher , 2012; Meyer et al . , 2012 ) . For P . mirabilis PIC99114 and P . ipomoeae PIC99167 , genomic DNA used before ( Raffaele et al . , 2010b; Cooke et al . , 2012 ) was converted into Illumina sequencing libraries . Genomic DNA from the other modern strains was isolated as described ( Cooke et al . , 2012 ) . Libraries were sequenced on Illumina GAIIx , HiSeq 2000 or MiSeq instruments ( Table 2 ) . To estimate the fraction of P . infestans aDNA in the herbarium samples , we performed pilot sequencing . Once the samples with the highest amount of P . infestans were identified , production runs were carried out on an Illumina HiSeq 2000 instrument . For P . infestans 06_3928A analysis , we used publicly available short reads ( ENA ERP002420 ) . Reads for the herbarium samples were de-indexed as described ( Kircher , 2012 ) . Forward and reverse reads were merged after adapter trimming , requiring at least 11 nucleotides overlap ( Burbano et al . , 2010 ) . Only fragments that that allowed merging of reads were used in subsequent analyses . Merged reads were mapped to the P . infestans T30-4 reference genome ( Haas et al . , 2009 ) with BWA , allowing two gaps and without seeding ( Li and Durbin , 2009 ) . PCR duplicates were identified based on read start and end alignment coordinates . For each cluster of duplicates a consensus sequence was calculated as described ( Kircher , 2012 ) . Only reads with a Phred-like mapping quality score of at least 30 were used further . Alignments were converted to BAM files ( Li et al . , 2009 ) . For modern strains , single reads were mapped in a similar manner , and subsequent processing was performed as described ( Cooke et al . , 2012 ) . SNPs in the herbarium samples were called by independently comparing each strain with the P . infestans T30-4 genome . Raw allele counts for each position were obtained using pileup from SAMtools ( Li et al . , 2009 ) . To decide the cutoffs for SNP calling we resorted to simulations . Reads from two 50-fold and 3-fold coverage genomes were simulated using the pIRS software ( Hu et al . , 2012 ) , with empirical base-calling and GC%-depth profiles trained on five modern P . infestans genomes ( P13527 , P13626 , 06_3928A , NL07434 and P17777 ) . Based on the simulated data we called both homo and heterozygous SNPs employing different cutoffs for SNP concordance ( Figure 3—figure supplement 1 ) . Genotypes calls were classified as high quality if coverage was at least 10 . We also considered low-quality SNPs , if a high-quality SNP call had been made in a different strain , using specific coverage cutoffs for rescuing low-quality SNPs ( Figure 3—figure supplement 1 ) . We calculated sensitivity and accuracy of SNP calls for different combination of cutoffs and selected the following criteria:Minimum coverage of 10 for high quality calls . Concordance ≥80% for homozygous SNPs . Concordance between 20% and 80% for heterozygous SNPs . Minimum coverage of 3 to rescue low-quality SNPs . We called synonymous , nonsynonymous and nonsense polymorphisms employing snpEff ( Cingolani et al . , 2012 ) . Fragments that could be aligned to any of the four reference haplotypes ( Ia , IIa , Ib and IIb ) were used to assemble mtDNA genomes . For each strain four different assemblies ( one for each reference haplotype ) were built using an iterative mapping assembly program ( Green et al . , 2008; Burbano et al . , 2010 ) . Only positions with coverage of at least 3 were included in the assemblies . The four assemblies were aligned using Kalign ( Lassmann and Sonnhammer , 2005 ) with default parameters , and a consensus assembly was generated based on the alignment . The 1 . 8-kb insertion present in haplotype II was not considered for phylogenetic reconstruction . The mtDNA phylogeny was built using a maximum parsimony and a maximum likelihood tree using MEGA5 ( Tamura et al . , 2011 ) . For both , positions with less than 90% site coverage were eliminated . There were a total of 24 , 560 positions in the final dataset , compared to the multiple sequence alignment length ( 37 , 762 bp ) . For the maximum likelihood reconstruction we used the Hasegawa-Kishino-Yano ( HKY ) model assuming that a certain fraction of sites are evolutionarily invariable . The model was selected using MEGA5 ( Tamura et al . , 2011 ) . We reconstructed the nuclear phylogeny for the high-coverage samples alone and for all samples together independently , using maximum parsimony and maximum likelihood approaches . We built the neighbor-joining trees based on a genetic distance matrix calculated from both homo- and heterozygous SNPs ( Xu et al . , 2012 ) . For the high-coverage genomes we used only the SNPs positions with complete information in all samples . For the low-coverage genomes we used homo- and heterozygous SNPs , and allowing for missing data . So that we could include heterozygous SNPs in the analysis , we randomly sampled one allele at each site . The maximum parsimony trees were calculated with 100 replicates in MEGA5 using the Close-Neighbor-Interchange algorithm with search level 0 , in which the initial trees were obtained with the random addition of sequences ( 10 replicates ) . All positions with less than 95% site coverage were eliminated ( Tamura et al . , 2011 ) . For the high-coverage genomes-only analysis , all positions with less than 85% site coverage were eliminated . For the all-sample analysis the threshold was lowered to 80% . Maximum likelihood trees were built using RaxML 7 . 0 . 4 with 100 replicates using the rapid bootstrap algorithm ( Stamatakis , 2006 ) . To address presence and absence polymorphisms of effectors , we used a previously published pipeline ( Raffaele et al . , 2010a ) . We calculated the fraction of each gene that was covered by reads ( ‘breadth of coverage’ ) for each strain . We regarded zero breadth of coverage as absence of the gene . For herbarium and modern samples with genome-wide coverage depth less than 20× , we merged BAM files from each strain into a single BAM file , and used this to estimated breadth of coverage . We also tested for presence/absence polymorphisms of RXLR effector genes between herbarium samples and modern strains using de novo assembly of short reads . First , we extracted 140 bp fragments from M-0182896 merged reads , and assembled these with SOAPdenovo v1 . 05 ( Luo et al . , 2012 ) . We aligned the six-frame translation of resulting contigs to a non-redundant protein database using blastx ( Altschul et al . , 1990 ) . Most contigs matched proteins from either potato , Solanum tuberosum , or to microbial species P . infestans , Pantoea vagans and Pseudomonas sp . To focus on P . infestans , we aligned fragments that were at least 140 bp to the genomes of P . infestans , S . tuberosum , P . vagans , P . syringae pv . syringae , and P . fluorescens with blastn . We extracted fragments that aligned the P . infestans genomic regions encoding RXLR effector genes , but over at most 90 bp . These and unmapped fragments were de novo assembled with SOAPdenovo v1 . 05 . A k-mer size of 67 was deemed optimal , because it resulted in the highest coverage of Avr1 , Avr2 and Avr3a , and resulted in the largest number of RXLR proteins with TBLASTN hits ( Figure 10—figure supplement 1 ) . We obtained partial sequences of Avr4 and Avrblb1 . We visually evaluated BWA alignments of M-0182896 in the Avr4 and Avrblb1 genomic regions and identified T30-4 sequences uncovered by alignments using Integrative Genomics Viewer ( Robinson et al . , 2011 ) . We then identified T30-4 genomic regions with at least 99% similarity to these uncovered regions . In BWA , if reads match several genomic regions , one genomic location is randomly chosen as default ( Li and Durbin , 2009 ) . Thus , it is possible that BWA alignment distributes reads coming from the same gene across several , closely related genes in the target genome . We assembled such reads that mapped to closely related sequences in the reference genome together with the partial sequences of Avr4 and Avrblb1 using Geneious Pro 5 . 6 . 3 to obtain full-length sequences of these Avr genes . Homozygous SNPs from modern P . infestans strains EC3527 , EC3626 , NL07434 , 06_3928A , DDR7602 , LBUS5 , P17777 and the historic strain M-0182896 were used for selection tests . Gene sequences were converted into amino acid sequences using EMBOSS tools ( Rice et al . , 2000 ) , and Pla2Nal v14 ( Suyama et al . , 2006 ) was used to convert protein alignments to codon alignments . The codeml module of PAML package v4 . 6 ( Yang , 2007 ) was used for positive selection studies with site models M7 ( parameters NSsites = 7 , fix_omega = 0 , omega = 2 and kappa = 3 ) and M8 ( NSsites = 8 , fix_omega = 0 , omega = 2 and kappa = 3 ) . A 5% level of significance was established with Likelihood ratio test . Genes were considered to be under positive selection if at least one site was found to be under selection with a Bayes Empirical Bayes confidence >95% . To estimate ploidy levels , we assessed the distributions of read counts at biallelic SNPs . For diploid species , the mean frequency of reads for each allele at non-homozygous sites is 1/2 , while we expect two modes for triploid genomes , at 1/3 and 2/3 , and four modes for tetraploid genomes , at 1/4 , 1/2 and 3/4 ( Figure 8A ) . We simulated genomes with different ploidy levels using pIRS ( Hu et al . , 2012 ) , based on two strains , P . infestans T30-4 and EC3527 . The SNPs used for the construction of two simulated chromosomes were determined with SAMtools v0 . 1 . 8 mpileup and bcftools v0 . 1 . 17 ( Li et al . , 2009 ) . For the diploid genome , we simulated 10x coverage reads for each of two different chromosomes . For the triploid genome , we merged simulated 5× and 15× coverage reads from two different chromosomes . For the tetraploid genome , we merged simulated 10× coverage reads from two different chromosomes ( Figure 8B ) . Next , we aligned the simulated reads to the P . infestans T30-4 reference genome with BWA and called heterozygous SNPs under the following criteria: minimum coverage of 10 for high-quality calls , and concordance between 20% and 80% for heterozygous SNPs . Since tetraploid species are considered to be a mixtures of two ratio , we mixed SNPs from the 20× coverage diploid reads and the 20× coverage tetraploid reads in following ratios: 0:100 , 10:90 , 20:80 , 30:70 , 40:60 , 50:50 , 60:40 , 70:30 , 80:20 , 90:10 and 100:0 . Finally , we estimated frequency of reads assigning each allele at each SNP position . Based on shapes , standard deviation , skewedness and kurtosis of the observed distributions and comparison with the simulated distributions , we classified the tested P . infestans genomes as diploid , triploid and tetraploid . In order to test whether we can detect a temporal signal in the ancient P . infestans mtDNA sequences compared to modern strains , that is , shorter branches in the ancient strains compared to the modern ones , we calculated the nucleotide distance as the number of substitutions between HERB-1 , haplotype Ia and haplotype Ib mtDNA genomes to the outgroup P17777 . The analysis involved 19 nucleotide sequences . All positions with less than 90% site coverage were eliminated , resulting in 34 , 174 informative positions . The samples were subsequently grouped into ancient and modern strains . The ancient and modern nucleotide distances were significantly different ( Mann–Whitney U-test , p=0 . 0003 ) . We furthermore correlated the nucleotide distance of HERB-1 , haplotype Ia and haplotype Ib mtDNA genomes to the outgroup P17777 with the tip age of each sample . To estimate divergence times of P . infestans strains , substitution rates were calculated in a Bayesian framework analysis using the software package BEAST 1 . 7 . 5 ( Drummond et al . , 2012 ) . A multiple sequence alignment that included all 12 nearly complete modern P . infestans mtDNA sequences plus all 13 herbaria samples was used as input . In order to test if the mtDNAs evolved clock like a likelihood ratio test was performed in MEGA5 ( Tamura et al . , 2011 ) by comparing the maximum likelihood ( ML ) value for the given topology using only the modern strains with and without the molecular clock constraints . The null hypothesis of equal evolutionary rate throughout the tree was not rejected at a 5% significance level ( p=0 . 115 ) . All positions containing gaps and missing data were eliminated resulting in a total of 22 , 591 positions in the final dataset . As a result a strict molecular clock and the HKY sequence evolution model were used for the Bayesian framework . For the tree prior , five different models were tested including four coalescence models: constant size , expansion growth , exponential growth , logistic growth and a epidemiology birth–death model ( Stadler , 2010 ) . For each tree prior , three MCMC runs were carried out with 10 , 000 , 000 iterations each and subsequently merged using LogCombiner 1 . 7 . 5 from the BEAST package . Resulting ESS values and overall posterior likelihoods were compared using the software Tracer ( Rambaut and Drummond , 2007 ) . The birth-death model gave the highest ESS values and posterior likelihood and was therefore chosen for the subsequent dating analysis . The collection dates for all herbaria samples as well as the isolation dates for all modern strains were used as tip calibration points ( Table 1 ) . Three MCMC runs were carried out with 50 , 000 , 000 iterations each , sampling every 10 , 000 steps . The first 1 , 000 , 000 iterations were discarded as burn-in resulting in a total of 147 , 000 , 000 iterations . | Few crop failures have been as devastating as those caused by potato late blight in the 1840s . This disease is caused by a filamentous microbe called Phytophthora infestans , which spread from North America to Europe in 1845 , leading to the Great Famine in Ireland and to severe crop losses in the rest of Europe . Phytophthora is thought to have originated in the Toluca valley of Mexico , where many different strains evolve alongside wild potato relatives , but the exact strain that caused the Great Famine , and how it is related to modern strains of the pathogen , has remained a mystery . Yoshida et al . have used a technique call ‘shotgun’ sequencing to map the genomes of 11 historical strains of P . infestans and 15 modern strains . The historical strains were extracted from the leaves of potato and tomato plants that were collected in North America and Europe , including Ireland and Great Britain , from 1845 onwards and stored in herbaria for future research . By comparing the genomes of the historical and modern samples , Yoshida et al . found that the historical strains all belonged to a single lineage that shows very little genetic diversity . Previously it has been proposed that this lineage was the same as US-1 , which was the dominant strain of potato blight in the world until the end of the 1970s , or that it was more closely related to modern strains than to US-1 . Yoshida et al . now rule out both of these possibilities and show that the lineage that caused the great famine , which they call HERB-1 , is clearly distinct from US-1 , although they are closely related , and they conclude that both HERB-1 and US-1 might have dispersed from a common ancestor that existed outside of Mexico in the early 1800s . Why US-1 later replaced HERB-1 as the dominant strain in the world is an important question for future studies . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Material",
"and",
"methods"
] | [
"genetics",
"and",
"genomics"
] | 2013 | The rise and fall of the Phytophthora infestans lineage that triggered the Irish potato famine |
Virulence in the ubiquitous intracellular protozoon Toxoplasma gondii for its natural intermediate host , the mouse , appears paradoxical from an evolutionary standpoint because death of the mouse before encystment interrupts the parasite life cycle . Virulent T . gondii strains secrete kinases and pseudokinases that inactivate the immunity-related GTPases ( IRG proteins ) responsible for mouse resistance to avirulent strains . Such considerations stimulated a search for IRG alleles unknown in laboratory mice that might confer resistance to virulent strains of T . gondii . We report that the mouse IRG system shows extraordinary polymorphic complexity in the wild . We describe an IRG haplotype from a wild-derived mouse strain that confers resistance against virulent parasites by interference with the virulent kinase complex . In such hosts virulent strains can encyst , hinting at an explanation for the evolution of virulence polymorphism in T . gondii .
A virulent parasite that overcomes the immune system and kills its host may seem to have won the confrontation , but it is a Pyrrhic victory when the early death of the host reduces the probability of parasite transmission . Indeed it is in the interests of all hosts and most parasites to prolong the encounter . In this sense , virulence is a failure of co-adaptation . Haldane’s conjecture ( Haldane , 1949 ) that intense and fluctuating selection imposed by parasites will generate host protein polymorphism is widely accepted ( Woolhouse et al . , 2002; Clark et al . , 2007; Kosiol et al . , 2008; Fumagalli et al . , 2011 ) . The presence in a population of multiple host resistance alleles confronting multiple parasite virulence alleles may reflect a dynamic equilibrium permissive for the persistence of both parties . However this equilibrium is achieved only at the expense of individual interactions fatal for either the parasite or the host , as a consequence of confrontations of inappropriate alleles . In mammals , however , the ability of the adaptive immune system to respond within the time scale of an individual infection and to remember for a lifetime , buffers individuals against dangerous genetic novelty arising from parasites . As a result , life or death outcomes for common infectious diseases in mammals are not generally determined by single , highly penetrant , polymorphic genes . We here report such a case , involving infection of the house mouse , Mus musculus , with the ubiquitous intracellular protozoan parasite , Toxoplasma gondii . T . gondii has a complex life cycle ( Dubey , 1998 ) ( Figure 1 ) . The sexual process occurs in true cats ( Felidae ) and intermediate hosts become infected by ingesting oocysts spread in cat faeces . A phase of fast intracellular replication and spread ( tachyzoite phase ) stimulates immunity in the intermediate host , and this in turn induces parasite encystment in brain and muscle cells and lifelong persistence . Predation of the infected host by a cat completes the life cycle . If immunity fails , tachyzoite replication continues uninterrupted , killing the infected host within a few days ( Deckert-Schlüter et al . , 1996 ) . Thus the probability that T . gondii completes its life cycle , which is roughly linear with duration of infection of the intermediate host , depends on early immune control . 10 . 7554/eLife . 01298 . 003Figure 1 . The life cycle of T . gondii . All warm blooded animals may serve as intermediate hosts , which are infected by ingestion of food or water contaminated by oocysts . Felids , the definitive hosts , are infected by ingesting tissue cysts from their prey . The intermediate phase of the life cycle may be prolonged by carnivory between intermediate hosts ( not shown ) . Modified from free-license pictures . DOI: http://dx . doi . org/10 . 7554/eLife . 01298 . 003 Mus musculus is probably the evolutionarily most important intermediate host for T . gondii , because it is very abundant worldwide and sympatric with a uniquely abundant felid , the domestic cat . Early immune control of T . gondii in mice depends on a family of IFNγ-inducible cytoplasmic effector proteins , the 47 kDa immunity-related GTPases ( IRG proteins; for nomenclature of IRG genes and proteins see Bekpen et al . ( 2005 ) ; Martens and Howard ( 2006 ) and ‘Materials and methods’ ) ( Taylor et al . , 2000; Collazo et al . , 2001; Liesenfeld et al . , 2011 ) . These assemble on the cytosolic face of the parasitophorous vacuole membrane ( PVM ) , causing its rupture and killing the included parasite ( Martens et al . , 2005; Zhao et al . , 2009b ) . In the C57BL/6 ( BL/6 ) laboratory mouse strain about 20 IRG genes occur in two adjacent clusters on chromosome 11 and one cluster on chromosome 18 ( Bekpen et al . , 2005 ) ( Figure 2A ) . The whole 47 kDa sequences of IRG proteins are typically translated from a single exon . Exceptional are certain ‘tandem’ IRGB proteins with a molecular weight of about 94 kDa . These genes are transcribed across two chromosomally adjacent IRG coding units and the intergenic spacer is spliced out as an intron resulting in a single open reading frame ( Bekpen et al . , 2005 . See also ‘Materials and methods’ for a note on nomenclature of the tandem genes and proteins ) . IRG proteins fall into two major functional and sequence sub-families , the GKS group ( IRGA , IRGB and IRGD proteins ) that are effector proteins at the PVM , and the GMS group ( Irgm1 , Irgm2 and Irgm3 ) that are negative regulators of the GKS proteins ( Hunn et al . , 2008 ) . Many strains of T . gondii ( e . g . , the abundant Eurasian strains designated types II and III ) are well controlled by the IRG system in laboratory mice , encyst , and are considered avirulent . But others ( e . g . , type I ) are highly virulent ( Sibley and Boothroyd , 1992 ) , killing the mouse host during the tachyzoite phase of infection . It has very recently been shown that virulence differences between T . gondii strains are largely due to polymorphic variation in ROP18 and ROP5 ( Saeij et al . , 2006; Taylor et al . , 2006; Khan et al . , 2009; Behnke et al . , 2011 ) , members of a family of kinases and pseudokinases ( El Hajj et al . , 2006 ) . These proteins are secreted during parasite entry and accumulate on the cytosolic face of the PVM ( Boothroyd and Dubremetz , 2008 ) . Virulent ROP allotypes inactivate IRG effector proteins by phosphorylating essential threonines in the nucleotide-binding domain ( Fentress et al . , 2010; Steinfeldt et al . , 2010 ) . In view of the unique importance of Mus musculus for the transmission of T . gondii , parasite strains acutely lethal for mice should be counterselected . Their presence at a significant frequency therefore demands an explanation . In this paper , we reveal that the IRG resistance system in mice has a scale of polymorphic complexity that rivals the MHC , and that a resistant IRG haplotype in the mouse can counter polymorphic virulence factors in the parasite , generating a host phenotype that is permissive for encystment , and thus for the propagation , of virulent strains . 10 . 7554/eLife . 01298 . 004Figure 2 . IRG protein polymorphism in inbred mouse strains . ( A ) Linear order of IRG gene clusters on Chr 11 and Chr 18 of mouse strain BL/6 . ( B ) Polymorphism at the protein level in the IRG genes of Chr 11 and Chr 18 . Irga9–Irga20 are absent from BL/6 and are inferred from resequencing data . Each colour block represents one IRG open reading frame and shows the number of amino acid substitutions/indels relative to the BL/6 allele . The colours of the blocks indicate their phylogenetic relationship ( Figure 2—figure supplement 1 ) . Open blocks in strains Czech II and CIM indicate homologues expected but not yet found . ‘ψ’ indicates probable pseudogenes . ( C ) Dot plots of the longer IRG gene clusters on Chr 11 in CAST/Ei and MSM/Ms against the BL/6 genomic sequence . Small blue squares show the positions of homologous coding units , blue lines indicate the positions of genes in BL/6 that are absent from the other genomes . DOI: http://dx . doi . org/10 . 7554/eLife . 01298 . 00410 . 7554/eLife . 01298 . 005Figure 2—figure supplement 1 . Unrooted phylogenetic trees of the indicated IRG genes were overlapped with RGB colour wheels . In each case the branch containing the taxon of BL/6 was set to 6 o'clock on the colour wheel . The centre of the colour wheel in some cases represents one of the putative roots of the phylogenetic tree . In other cases , however , the position of the wheel centre was adjusted for a better resolution of sequence differences ( e . g . , for Irga7 ) . Taxa with similar colours ( e . g . , red and orange ) have a relatively closer sequence relationship than taxa with contrasting colours ( e . g . , red and green ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01298 . 00510 . 7554/eLife . 01298 . 006Figure 2—figure supplement 2 . Chr 11 ( shorter contig containing Irgb10 , Irgm2 and Irgm3 ) of CAST/Ei and MSM/Ms against the BL/6 genomic sequence ( Top ) . Chr 18 contigs of CAST/Ei and MSM/Ms against the BL/6 genomic sequence ( Bottom ) . Small blue squares show the positions of homologous coding units , blue lines indicate the positions of genes in BL/6 that are absent from the other genomes . DOI: http://dx . doi . org/10 . 7554/eLife . 01298 . 006
We compared the IRG genes from a number of mouse strains , mostly those re-sequenced in the Mouse Genomes Project ( Keane et al . , 2011 ) with the canonical BL/6 sequences ( Figure 2B ) . The re-sequenced mice are largely from established laboratory strains . Others are recently derived from the wild and not admixed with laboratory mice . Mus spretus is a wild species distinct from Mus musculus with a divergent history of 1–3 million years ( Suzuki et al . , 2004 ) . Within laboratory strains we found relatively little IRG protein polymorphism on Chr 11 . In contrast , the wild-derived strains showed variation in IRG gene number and remarkable protein polymorphism . For example Irgb6 , a protein of 406 residues , had an allele in the CAST/EiJ strain with 47 amino acid substitutions relative to BL/6 . On Chr 18 , IRG protein sequence variation between laboratory strains was more apparent , and again , wild-derived strains showed extensive sequence polymorphism and copy number variation . The IRG proteins of the outgroup , M . spretus , showed considerable protein divergence from M . musculus sequences . Pseudogenes occurred in every haplotype and some ( e . g . Irga5 and Irgb7 ) were preserved between M . musculus and M . spretus . Other IRG sequences were pseudogenes in some haplotypes and apparently intact in others , for example Irga3 and Irga8 . Even within this limited group of strains we can distinguish eleven distinct IRG gene haplotypes on Chr 11 and thirteen on Chr 18 , already yielding a theoretical population diversity of several hundred IRG genotypes . Diagonal dot-plot comparisons between IRG gene clusters of wild-derived CAST/Ei and MSM/Ms with BL/6 showed IRG genes within tracts of duplication and deletion , associated with numerous repeats in both orientations , a genomic configuration that would be expected to be dynamic even on short time-scales , and doubtless responsible for the two very different dot-plots ( Figure 2C and Figure 2—figure supplement 2 ) . We added sequences amplified from genomic DNA of wild mice from several Eurasian sites ( Figure 3—figure supplement 1 ) to the data from inbred strains to generate nearest neighbour phylograms ( Figure 3A ) . We analysed five IRG genes , namely Irgm1 , a regulatory IRG protein of the GMS class ( Hunn et al . , 2008 ) , and Irga6 , Irgb2 , Irgb6 , and Irgb10 , all effector GKS proteins localizing to the T . gondii PVM during infection ( Khaminets et al . , 2010 ) . Irgb2 forms the N-terminal half of the tandem protein , Irgb2-b1 . The nearest-neighbour phylograms of Irgm1 , Irgb10 and Irga6 are shallow and the M . spretus sequences fall into outgroups . Thus most of the sparse polymorphic variation in these sequences has been acquired since the divergence of M . musculus from M . spretus . This conclusion is supported by a tendency for individual minor variants to be concentrated in one or other of the three recent , geographically separated and partially isolated subspecies , Mus musculus musculus , Mus musculus domesticus and Mus musculus castaneus ( Figure 3A , Figure 3—figure supplements 2–4 ) . In contrast , the phylograms for Irgb2 and Irgb6 have a depth similar to the mouse-rat divergence ( about 20 million years ) ( Gibbs et al . , 2004 ) , the M . spretus sequences are embedded in the M . musculus trees , and any tendency to correlation with subspecies is seen only in the outermost branches . Thus the polymorphism in these two genes is ancient and has persisted through a number of speciation events . The scale of polymorphism in the IRG system estimated by Tajima’s π from 7 mouse strains resembles that of classical MHC genes ( Figure 3B ) . That there is also considerable local polymorphism within geographic ranges was confirmed by the identification of numerous heterozygotes at all loci examined by PCR among the wild mouse captures . 10 . 7554/eLife . 01298 . 007Figure 3 . Polymorphism of five IRG genes . ( A ) Phylogenetic trees of five IRG genes sequenced from DNA of wild mice collected from various sites in Eurasia . Green , blue and red taxa represent M . m . domesticus , M . m . musculus and M . m . castaneus samples respectively . The black taxon represents Mus spretus . Alleles found in heterozygous condition in certain mice are indicated by numbers appended to individual mouse identifiers ( some haplotypes contain 2 Irgb6 paralogous genes ( Figure 2B ) , hence potentially up to 4 alleles ) . Bootstrap values are shown if >90 . The sequences are avaliable in ( Figure 3—source data 2-6 ) . ( B ) The nucleotide pairwise diversities ( π ) of genes across seven laboratory and wild-derived inbred mouse strains ( BL/6 , AKR/J , MSM/Ms , CAST/Ei , PWK/PhJ , WSB/Ei and Spretus/EiJ ) . Grey bars indicate the distribution of π from 50 ‘random’ genes ( Figure 3—source data 1 ) . The π values of individual IRG and MHC genes are indicated by arrows . DOI: http://dx . doi . org/10 . 7554/eLife . 01298 . 00710 . 7554/eLife . 01298 . 008Figure 3—source data 1 . Nucleotide diversities of 50 random genes in seven mouse strains . Seven laboratory and wild-derived inbred mouse strains were analysed . The ORFs of 50 random genes were selected based on their position in the C57BL/6 genome ( NCBI reference assembly build 37 ) . In addition , selected IRG and MHC members were assembled , and Tajima's π values were calculated . Klra4 is closest to 130M in Chr 6 , but lost in many mouse strains ( Cutler and Kassner , 2008 ) . The adjacent gene Klra5 was used instead . DOI: http://dx . doi . org/10 . 7554/eLife . 01298 . 00810 . 7554/eLife . 01298 . 009Figure 3—source data 2 . Alignment of Irgm1 alleles , in FASTA format . DOI: http://dx . doi . org/10 . 7554/eLife . 01298 . 00910 . 7554/eLife . 01298 . 010Figure 3—source data 3 . Alignment of Irga6 alleles , in FASTA format . DOI: http://dx . doi . org/10 . 7554/eLife . 01298 . 01010 . 7554/eLife . 01298 . 011Figure 3—source data 4 . Alignment of Irgb2 alleles , in FASTA format . DOI: http://dx . doi . org/10 . 7554/eLife . 01298 . 01110 . 7554/eLife . 01298 . 012Figure 3—source data 5 . Alignment of Irgb6 alleles , in FASTA format . DOI: http://dx . doi . org/10 . 7554/eLife . 01298 . 01210 . 7554/eLife . 01298 . 013Figure 3—source data 6 . Alignment of Irgb10 alleles , in FASTA format . DOI: http://dx . doi . org/10 . 7554/eLife . 01298 . 01310 . 7554/eLife . 01298 . 014Figure 3—figure supplement 1 . Mouse samples collected for this study . Colour code relates to the individual subspecies ( green—M . m . domesticus , blue—M . m . musculus , red—M . m . castaneus ) and corresponds to the text colours for the phylograms displayed in Figure 3 and Figure 3—figure supplement 2–4 . Purple colour indicates the hybrid zone of M . m . musculus and M . m . castaneus . The definition of M . m . castaneus in the Indian subcontinent is complex . Modified from Din et al . ( 1996 ) ; Guenet and Bonhomme , 2003 . DOI: http://dx . doi . org/10 . 7554/eLife . 01298 . 01410 . 7554/eLife . 01298 . 015Figure 3—figure supplement 2 . Phylogenetic trees of Irga6 , Irgm1 and Irgb10 in mouse strains and wild mice . DOI: http://dx . doi . org/10 . 7554/eLife . 01298 . 01510 . 7554/eLife . 01298 . 016Figure 3—figure supplement 3 . Phylogenetic tree of Irgb2 in mouse strains and wild mice . DOI: http://dx . doi . org/10 . 7554/eLife . 01298 . 01610 . 7554/eLife . 01298 . 017Figure 3—figure supplement 4 . Phylogenetic tree Irgb6 in mouse strains and wild mice . DOI: http://dx . doi . org/10 . 7554/eLife . 01298 . 017 With such striking polymorphic diversity in the sequences of proteins known to be involved in resistance against T . gondii , it was appropriate to assay the ability of different IRG genotypes to resist virulent T . gondii . The wild-derived Indian strain CIM , which has a number of divergent IRG alleles ( Figure 2B ) , proved to be remarkably resistant to the type I virulent T . gondii strain , GT-1 ( Figure 4A ) . All CIM mice survived intraperitoneal infection with GT-1 tachyzoites while all laboratory mice ( NMRI strain ) died within 15 days . In vitro , tachyzoite proliferation of avirulent strains in mouse cells is inhibited by induction of IRG proteins with IFNγ ( Könen-Waisman and Howard , 2007 ) . In contrast , proliferation of type I virulent strains is not inhibited ( Zhao et al . , 2009c ) . By this assay , IFNγ-induced CIM diaphragm-derived cells ( DDC see ‘Materials and methods’ ) inhibited proliferation of type I virulent RH-YFP strain tachyzoites as efficiently as they inhibited the proliferation of avirulent strains , while BL/6 DDC inhibited only the avirulent strains ( Figure 4B ) . Cells from two other M . m . castaneus strains , CAST/EiJ and CTP were almost as resistant as CIM . Additionally , IFNγ-induced CIM cells died by reactive cell death after infection with both virulent RH-YFP and avirulent ME49 , while BL/6 cells died only after infection with avirulent strains ( Figure 4C ) . Reactive death of T . gondii-infected mouse cells after induction with IFNγ is associated with IRG protein-mediated host resistance , as previously reported ( Zhao et al . , 2009b ) . 10 . 7554/eLife . 01298 . 018Figure 4 . Resistance of wild-derived mouse strains to virulent T . gondii . ( A ) Cumulative mortality of NMRI and CIM mice infected with 100 or 300 ( data pooled ) tachyzoites of the indicated T . gondii strains . ( B ) IFNγ-mediated growth inhibition of virulent ( type I RH-YFP , BK ) and avirulent ( type II ME49 , type III CTG ) T . gondii strains in DDC of laboratory ( BL/6 ) and wild-derived , inbred mice ( CAST/Ei , CIM , CTP ) . Proliferation of parasites was measured by 3H-uracil incorporation and is displayed as percentage of residual T . gondii proliferation , as described in ‘Materials and methods’ . Error bars show standard deviations of quadruplicate values . ( C ) IFNγ-dependent reactive cell death of mouse DDC cell lines infected with T . gondii . DDC were either stimulated with 100 U/ml of IFNγ 24 hr prior to infection or left unstimulated . Cells were infected with type II strain ME49 or type I strain RH-YFP at the indicated MOIs for 8 hr . Cell viabilities were measured as described in ‘Materials and methods’ and expressed as percentages of those recorded for uninfected cells ( MOI = 0 ) . Error bars show standard deviations of quadruplicate values . DOI: http://dx . doi . org/10 . 7554/eLife . 01298 . 018 Resistance of CIM mice to the type I virulent RH-YFP strain in vivo was exploited to test linkage of resistance to the IRG system in a ( CIM×BL/6 ) F1×BL/6 backcross . For typing purposes the IRG gene clusters from CIM and BL/6 were differentiated by RFLPs in Irga1 ( Chr 18 ) and Irgb6 ( Chr 11 ) . Fluorescent-tagged tachyzoites ( RH-YFP ) were injected intraperitoneally into mice and the frequency of infected peritoneal cells measured by FACS after 5 days . In five independent experiments involving a total of 65 typed animals ( Figure 5A ) , resistance by this assay was almost completely dominant and was tightly linked to the IRGCIM gene cluster on Chr 11 ( p<<10−6 ) . There was no detectable association of virulence with the IRGCIM cluster on Chr 18 . In a similar analysis of 53 F2 progeny , mice typed as homozygous IRGCIM at the Chr 11 gene cluster were as resistant as wild-type CIM mice . Thus the reduced resistance of some mice heterozygous for IRGCIM on Chr 11 may be due to a gene dosage effect at the IRG locus . Assayed directly by survival , 51 backcross and F2 mice infected with RH-YFP followed the same pattern as in the peritoneal cell assay ( Figure 5B ) , with the Chr 11 IRGCIM homozygotes ( cc in Figure 5B ) showing complete resistance , the heterozygotes ( bc ) substantial but incomplete resistance and the IRGBL/6 homozygotes ( bb ) complete and acute susceptibility . 10 . 7554/eLife . 01298 . 019Figure 5 . Resistance of CIM mice to virulent T . gondii is dependent on the Chr 11 IRG locus . ( A ) Infected CD45+ peritoneal cells in ( BL/6×CIM ) F1×BL/6 backcross ( 5 experiments , total 83 mice ) and ( BL/6×CIM ) F2 mice ( 4 experiments , total 69 mice ) 5 days after i . p . injection of 500 RH-YFP tachyzoites . Genotypes at IRG loci and at the Nalp1 locus for backcross and F2 mice are shown ( see key ) as bb ( homozygous BL/6 ) , bc ( heterozygous BL/6/CIM ) or cc ( homozygous CIM ) . Elimination of infected cells is linked to the CIM haplotype on Chr 11 ( n . b . the y-axis is logarithmic below 5% ) . ( B ) Cumulative mortality of ( BL/6×CIM ) F1×BL/6 and ( BL/6×CIM ) F2 mice infected with 500 RH-YFP tachyzoites . Irgb6 genotypes are shown as in ( A ) . ( C ) Cysts of type I T . gondii strains in brain homogenates of CIM mice infected 6–8 weeks earlier ( quantitation in Table 1 ) . Bar = 20 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 01298 . 019 Recent results have implicated the inflammasome core component Nlrp1 ( NLR family , pyrin domain containing 1 ) in resistance to T . gondii in human ( Witola et al . , 2011 ) and rat ( Sergent et al . , 2005; Cavailles et al . , 2006 ) . Since the Nlrp1 complex locus is about 20 Mb telomeric to the IRG system on Chr 11 it was possible that polymorphism at this locus ( Boyden and Dietrich , 2006 ) was responsible for the differential resistance apparently linked to the IRG complex . We therefore typed all the backcross progeny shown in Figure 5A by PCR designed to distinguish the BL/6 and CIM allotypes of the Nlrp1b locus ( see ‘Materials and methods’ ) . We obtained 17% recombinants between Irgb6 and Nlrp1 . There was no correlation between Nlrp1 genotype and the ability to clear virulent T . gondii ( Figure 5A ) . It is anyway unlikely that Nlrp1 plays a role in the resistance polymorphism we describe since it is not detectably expressed in the IFNγ-induced DDC transcriptomes ( data not shown ) . The existence of mouse genotypes resistant to virulent T . gondii strains is consistent with a co-evolutionary explanation for the evolution of virulence . To sustain the argument , however , it would be necessary to show that T . gondii strains that are lethal in laboratory mice , and thereby suffer a major cost , can form functional cysts in resistant CIM mice , permitting their propagation . We therefore searched for brain cysts in CIM mice infected 6–8 weeks earlier with virulent type I or avirulent type III strains of T . gondii . Two type I virulent strains , GT-1 and BK , formed cysts in CIM mice ( Figure 5C , Table 1 ) . Thus the resistance of the CIM mouse provides an adaptive niche for highly virulent T . gondii strains . The avirulent type III strain , NED , encysted in the laboratory mice >20× more efficiently than in CIM mice ( Table 1 ) . 10 . 7554/eLife . 01298 . 020Table 1 . Cyst counts in T . gondii infected miceDOI: http://dx . doi . org/10 . 7554/eLife . 01298 . 020MouseUIC*CIMCIMCIMCIMCIMCIMCIMNMRINMRINMRINMRIT . gondii ( # injected ) –GT-1 500GT-1 1000BK 5000BK 10 , 000NED 10 , 000NED 10 , 000NED 10 , 000GT-1 500BK 500NED 10 , 000NED 10 , 000Q-PCR ( cycle ) >3522 . 924 . 218 . 722 . 026 . 926 . 228 . 8DeadDead22 . 620 . 3Cysts per brain01005013015022090157204800Antibody test–+++++++++*Uninfected control . Mice were sacrificed 5 weeks ( NED ) or 6–8 weeks ( BK and GT-1 ) after tachyzoite injection . Infection was verified by serum antibody . Cysts were evaluated by direct counting in homogenised brains and by quantitative PCR of a repeat element of T . gondii ( Reischl et al . , 2003 ) in genomic DNA samples isolated from mouse brains . The IRG resistance mechanism operates cell-autonomously , and in BL/6 cells the loading of IRG proteins onto the parasitophorous vacuole occurs only with avirulent strains even in cells infected simultaneously with virulent and avirulent strains ( Zhao et al . , 2009a; Khaminets et al . , 2010 ) . Consistently , the resistance of the CIM strain against virulent T . gondii was reflected in the behaviour of IRG proteins in IFNγ-induced CIM-derived cells infected with virulent RH-YFP strain T . gondii . In BL/6 cells Irgb6BL/6 loaded onto only about 8% of vacuoles , while in CIM cells the highly divergent Irgb6CIM protein loaded onto more than 50% of vacuoles ( Figure 6A ) . Even more extreme , the tandem protein , Irgb2-b1BL/6 , which is poorly expressed in BL/6 cells ( Figure 6B ) , was not detectable on the RH-YFP PVM , while the highly divergent Irgb2-b1CIM was well-expressed in CIM cells and loaded onto over 95% of vacuoles ( Figure 6A ) . 10 . 7554/eLife . 01298 . 021Figure 6 . Irgb2-b1CIM protects other IRG members from inactivation . ( A ) Immunofluorescent quantitation of loading of Irgb6 and Irgb2-b1 on to RH-YFP vacuoles in IFNγ-induced BL/6 and CIM DDC . ( B ) Strain dependence of expression levels of Irgb-tandem proteins . ( C ) Reduced phosphorylation of Irga6CIM on T108 by RH-YFP in IFNγ-induced CIM DDC ( chopped western blot for calnexin , phosphorylated Irga6 and total Irga6 ) . n . b . Irga6CIM characteristically runs at a higher apparent molecular weight than Irga6BL/6 ( D ) Transfected Irga6CIM and Irga6BL/6 are both phosphorylated in IFNγ-induced L929 cells infected with RH-YFP as shown in western blot of detergent lysates . Phosphorylation of Irga6 is indicated by a size-shift ( black arrowhead ) for both Irga6BL/6 and Irga6CIM in infected cells . The lower band in the two transfected/infected tracks is the endogenous Irga6BL/6 . ( E ) Irgb2-b1CIM ( yellow ) transfected into IFNγ-induced BL/6 MEFs inhibits phosphorylation of Irga6 ( red ) by RH-YFP seen in untransfected cells serving as control . Bar = 10 µm . ( F ) Immunofluorescent quantitation of phosphorylated Irga6 on the PVM of RH-YFP in Irgb2-b1CIM-transfected cells and untransfected cells prepared in ( E ) . ( G ) Immunofluorescent quantitation of total Irga6 on the PVM of RH-YFP in Irgb2-b1CIM transfected and untransfected cells prepared in ( E ) . ( H ) Enumeration of Irgb6-positive vacuoles in BL/6 MEFs induced by IFNγ and transfected with Irgb2-b1CIM or Irgb2-b1BL/6 . DOI: http://dx . doi . org/10 . 7554/eLife . 01298 . 021 The loading of the effector IRG protein Irga6 onto the parasitophorous vacuole of virulent T . gondii strains is prevented by a parasite-derived kinase complex ( ROP5/ROP18 ) that phosphorylates two threonines that are essential for IRG protein function ( Steinfeldt et al . , 2010 ) . Remarkably , in CIM cells infected with RH-YFP , phosphorylated Irga6CIM was barely detectable with an antiserum specific for Irga6BL/6 phosphorylated at T108 ( Figure 6C ) . Irga6CIM differs from Irga6BL/6 at only two residues , both distant from the phosphorylation sites and was phosphorylated normally when transfected into IFNγ-induced L929 cells infected with virulent RH-YFP ( Figure 6D ) . Thus Irga6CIM apparently remains unphosphorylated in CIM cells as a result of active inhibition of the parasite kinase complex . The following experiment showed that the highly polymorphic tandem protein , Irgb2-b1CIM , is largely responsible . Irgb2-b1CIM was transfected into IFNγ-induced BL/6 mouse embryonic fibroblasts ( MEFs ) infected with RH-YFP virulent strain T . gondii . Phosphorylated Irga6 was measured at the PVs in transfected cells expressing Irgb2-b1CIM and in untransfected cells . Figure 6E , F show that the amount of phosphorylated Irga6 was strikingly reduced on vacuoles loaded with Irgb2-b1CIM while the amount of total Irga6 loaded onto Irgb2-b1CIM positive vacuoles was increased ( Figure 6G ) . Thus the decreased signal of phosphorylated Irga6 on the PVM was not caused by competition for loading between Irga6 and Irgb2-b1 , but rather by inhibition of phosphorylation . Transfected Irgb2-b1CIM also stimulated the loading of endogenous Irgb6BL/6 onto RH-YFP vacuoles ( Figure 6H ) . Transfected Irgb2-b1BL/6 , which is well expressed unlike the endogenous protein , had little or no effect on the loading of Irgb6BL/6 ( Figure 6H ) . Thus the essential difference between Irgb2-b1BL/6 and Irgb2-b1CIM in determining resistance lies in the amino acid sequence polymorphism rather than in the protein expression level . Preliminary results suggest that Irgb2-b1CIM may bind directly to the protein product of the virulent allele of ROP5 . Thus loading of Irgb2-b1CIM onto the PVM of ROP5-deficient RH strain parasites in CIM cells was greatly reduced , and consistently , the amount of loading on to different RH-related strains correlated with strain-specific variation in the amount of ROP5 ( Figure 7A ) . Irgb2-b1CIM itself becomes phosphorylated during infection with virulent T . gondii ( Figure 7B ) , thus is also a target for the active ROP5-ROP18 kinase complex , suggesting that Irgb2-b1CIM may block phosphorylation of Irga6 by binding ROP5 pseudokinase at the vacuole , thereby distracting rather than inhibiting ROP18 kinase . ROP5 binds to Irga6 via helix 4 ( H4 ) of the Irga6 nucleotide binding domain ( Fleckenstein et al . , 2012 ) , so a homologous structure on Irgb2-b1 may be involved in ROP5 interaction . Indeed the putative H4 and αD structural domains of the Irgb2 subunit are highly polymorphic and show recent divergent selection ( Figure 7C ) , indicating possible co-evolution with ROP kinases and pseudokinases . The high polymorphism of H4 in Irgb2-b1 is also consistent with a direct interaction with a polymorphic component of the pathogen . If Irgb2-b1 interacts with a host protein to bridge to ROP5 the interaction surface between the two host proteins would not be expected to evolve rapidly under divergent selection . 10 . 7554/eLife . 01298 . 022Figure 7 . IRG-tandem proteins interact with virulence factors . ( A ) Loading of Irgb2-b1CIM on to vacuoles of RH variants in IFNγ-induced CIM DDC . Irgb2-b1CIM loading ( dot plots , upper image ) is positively correlated with ROP5 expression level in the parasite ( western blot , lower image ) . ( B ) Autoradiogram ( 33P ) of immunoprecipitated IRG proteins in DDC infected with T . gondii . Irga6 was phosphorylated ( open arrow head ) by virulence factors of T . gondii in BL/6 DDC , but not in CIM DDC . Irgb-tandem proteins were phosphorylated only in CIM DDC ( filled arrow head ) in a ROP5/ROP18-dependent manner . UIC = uninfected control . Asterisk ( * ) indicates non-specific phosphorylated proteins ( C ) Ribbon model of Irgb2 predicted based on the structure of Irga6 , showing diversifying selection associated with H4 and αD of the G-domain . The colours of the ribbons are based on the π and πa/πs values among 40 alleles sequenced from mouse strains and wild mice with a 120 bp slide window and 10 bp step . The colours indicate the πa/πs value , from purifying selection ( blue ) to significantly diversifying selection ( red ) . The saturations of colours are defined by overall π value in the sliding window , indicate conserved regions of the protein ( low saturation ) to highly polymorphic regions ( high saturation ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01298 . 022
We have shown that the IRG protein system essential for resistance against T . gondii infection in the mouse has a complex polymorphism on the scale of the MHC , and that at least one IRG haplotype , found in the wild-derived CIM strain mouse , is strikingly resistant to T . gondii strains that are highly virulent for laboratory mice . We also provide a mechanistic explanation for the resistance of the CIM mouse against type I virulent T . gondii strains . Resistance is determined by the presence of the polymorphic tandem IRG protein , Irgb2-b1CIM encoded on Chr 11 , which blocks the ROP5/ROP18 kinase complex of the virulent parasite , preventing phosphorylation and consequent inactivation of IRG effector proteins . Taken together , our results suggest a selective explanation for the evolution of T . gondii strains that are highly virulent for certain mice . If the mouse is an evolutionarily significant host for T . gondii the parasite must balance its virulence against mouse resistance , in order to allow encystment . To minimise the cost of infection , selection on the mouse favours the evolution of strong resistance alleles . This in turn leads to the selection of parasite strains able to counteract heightened resistance sufficiently to allow encystment in these mouse genotypes , as type I T . gondii strains in CIM mice . However such virulent T . gondii strains are counterselected by acute lethality in less resistant mice , while less resistant mice are better hosts for less virulent T . gondii strains . Why , however , are not all mice highly resistant ? Loss of the entire IRG system in several vertebrate groups , for example higher primates and birds ( Bekpen et al . , 2005 ) , suggests that possession of the IRG system may be costly . Highly resistant genotypes may be more costly than less resistant ones . Alternatively , molecular specificity in interactions between polymorphic parasite virulence factors and mouse IRG proteins may favour IRG genotypes that are highly resistant to some T . gondii genotypes but more susceptible to others . Such an allele-specific system is familiar from polymorphic plant disease resistance ( R ) genes and strain-specific pathogen virulence and avirulence ( Flor 1971; Dodds et al . , 2006 ) , where exact molecular matches or mismatches determine the outcome of a given infection . These alternative models can now be analysed formally and tested experimentally . Much of the argument in this paper has focused on the importance of the house mouse as an intermediate host and vector for T . gondii , resting on the global abundance of the species and its sympatry with the cat , and obviously supported by the intimate antagonistic relationship between T . gondii virulence factors and mouse resistance factors . However , although many bacteria and protozoal parasites are not resisted by the IRG system , T . gondii is not the only organism that is . At least two members of the Chlamydia species complex are resisted by the IRG system in mice , and polymorphic variation on Chr 11 , possibly associated with Irgb10 , affects the level of resistance ( Bernstein-Hanley et al . , 2006; Miyairi et al . , 2007 ) . There will surely be other organisms that may contribute to the genomic complexity and polymorphism of the IRG system . For example , the massive and ancient polymorphism of Irgb6 is not directly accounted for by our experiments . In in vitro experiments in BL/6 cells transfected Irgb2-b1CIM protects Irga6 from phosphorylation in the absence of the CIM allotypes of the other IRG proteins including Irgb6 ( Figure 6 ) and in the CAST/Ei strain , which is almost as resistant as CIM against T . gondii , Irgb6 is barely expressed ( unpublished results ) . It may be that Irgb6 polymorphism reflects a pattern of resistance against another abundant mouse pathogen . However it is also not excluded that Irgb6 polymorphisms relate to further virulence polymorphism in T . gondii not expressed in the limited range of strains used in the present study . Likewise , mice are not the only intermediate hosts for T . gondii . The parasite infects all known mammals and birds , and many other species than M . musculus are prey for domestic cats . Furthermore , individuals of other species , for example rats ( Jacobs and Jones , 1950 ) and American deer mice ( Peromyscus ) ( Frenkel , 1953 ) have been shown to resist infection with strains virulent for laboratory mice . Thus some of the genomic complexity of the ROP system and other virulence-associated T . gondii secretory proteins may be relevant to immunity modification in other host species . Nevertheless , despite these distractions , mice and T . gondii have clearly had a large selective impact on each other . The close association of cat and mouse with humans has led to large-scale transformations in the ecology of the parasite over the last 10 , 000 years . The impact of the dramatic narrowing of the parasite’s effective host range on the genetics of the virulence-resistance relationship may help to understand better the co-evolutionary forces at work in this system . The patterns of genetic resistance against T . gondii in its most important intermediate hosts will determine the relative abundance of strains contaminating the environment , and thereby the strains available for human infection .
The systematic , phylogenetically-based nomenclature of IRG genes and proteins used in this paper was published in Bekpen et al . ( 2005 ) and replaced the non-systematic naming of some of the first-discovered members of the family . A partial synonymy relating the new nomenclature to other names that have appeared in the literature is given in Martens and Howard ( 2006 ) . The root name for members of the IRG family is IRG . This may followed by single letters designating phylogenetically distinct sequence sub-families , as IRGA , IRGB , IRGC , IRGD , IRGM . Individual gene or protein names are given in the standard mouse nomenclatural form , as Irga6 ( protein ) and Irga6 ( gene ) or Irgb6 ( protein ) and Irgb6 ( gene ) . A subset of IRGB genes consist of two adjacent IRG coding units that are transcribed together from single promoters and spliced to form ‘tandem’ IRG proteins with a molecular weight twice that of the individual IRG coding units . In the original study it was not clear whether the individual coding units could also be expressed separately and they were therefore given separate names ( as Irgb1 and Irgb2 ) ( Bekpen et al . , 2005 ) . It is now clear that these coding units are expressed only as tandems . In the present study we have therefore named the tandem genes and proteins with the names of their constituent coding units in N-C order , thus Irgb2-b1 is a tandem IRG protein with the Irgb2 coding unit N-terminal to the Irgb1 coding unit . The Mus musculus genomes sequenced by the Mouse Genomes Project ( MGP ) ( Keane et al . , 2011 ) , including 129P2/OlaHsd , 129S1/SvImJ , 129S5/SvEvBrd , A/J , AKR/J , BALB/cJ , C3H/HeJ , C57BL/6NJ , CAST/EiJ , CBA/J , DBA/2J , FVB/N , LP/J , NOD/ShiLtJ , NZO/HlLtJ , PWK/PhJ , WSB/EiJ and the Mus spretus genome , SPRET/EiJ , were accessed by LookSeq ( http://www . sanger . ac . uk/resources/mouse/genomes/ ) , and the sequences of IRG genes were assembled ( for details see below , ‘Bioinformatics’ ) . Three 129 strains ( 129P2/OlaHsd , 129S1/SvImJ and 129S5/SvEvBrd ) were identical in all IRG coding units , so they were combined into one virtual strain , 129 . Mouse strain C57BL/6NJ was confirmed identical for all IRG genes with C57BL/6J , so the results were combined into one strain , C57BL/6 . The sequences of IRG genes of mouse strains Czech II , NMRI and JYG were acquired from the NCBI database and are listed in Table 2 . The CAST/Ei BAC library CHORI-26 was screened with 40bp synthetic probes for Irga1 , Irga3 , Irga4 , Irga6 , Irgm1 , Irgm2 , Irgb1 , Irgb6 , Irgb8 and Irgb10 . Eight out of 61 positive BAC clones were sent to the Wellcome Trust Sanger Institute for shotgun sequencing as follows: 226N16 ( NCBI accession number CU695224 ) , 243M20 ( CU695226 ) , 445J9 ( CU695230 ) , 332A4 ( CU695228 ) , 333C17 ( CU695229 ) , 240F21 ( CU695225 ) , 316B17 ( CU695227 ) and 76B7 ( CU695231 ) . The sequences of IRG genes were extracted from these results and cross-checked with the data from the MGP . Clones containing IRG genes from the mouse strain MSM/Ms ( Abe et al . , 2004 ) were chosen based on the BAC end sequences from RIKEN BRC ( http://www . brc . riken . jp/lab/dna/en/MSMBACen . html ) . Seven clones were sent to the Beijing Genomics Institute ( BGI ) for Illumina sequencing: 329H21 , 362F04 , 544P17 , 494M12 , 419B05 , 148E20 and 355D01 . The sequencing results were assembled and uploaded to the NCBI GenBank with accession number KF705682 , KF705684 , KF705686 , KF705680 , KF705685 , KF705681 , KF705683 . IRG genes of the CIM mouse strain were sequenced via full transcriptome Illumina sequencing of IFNγ-induced diaphragm-derived cells ( DDC , see below ) in the Cologne Centre of Genomics ( CCG ) . For wild mouse samples , genomic DNA was acquired from a variety of sources . From this material 5 key IRG members were amplified by PCR with appropriate primers ( Table 3 ) . The full ORFs of Irgm1 , Irgb2 , Irgb6 and Irgb10 , and a partial sequence of Irga6 ( 964 bp in length ) were amplified . PCR products were cloned into the pGEM-T vector and insertions in individual positive clones were sequenced . The sequences are attached as supplementary files in FASTA format . 10 . 7554/eLife . 01298 . 023Table 2 . IRG sequences from NCBI databaseDOI: http://dx . doi . org/10 . 7554/eLife . 01298 . 023GeneStrainTypeAccess numberIrgm1Czech IIESTsBI150356 , BF161711 , BF168437 , BF164781 , BE367794Irgb2-b1Czech IImRNABC022776JYGmRNAAK145236Irgb6Czech IImRNABC093522Czech IImRNABC034256JYGmRNAAK166353NMRImRNABC085259C . D2mRNAU15636IrgdCzech IImRNABC001986 , BC009131Irgm2Czech IIESTsBG518498 , BF137080 , BE284209 , BF168033 , BI149246 , BI414397 , BE283352 , BE306442Irgm3Czech IIESTsBI414397 , BI149246 , BF163420 , BE283352 , BF168033 , BI153387 , BE281683 , BE306442 , BI106672 , BI150745 , BF225799 , BF168273Irga6Czech IIESTsBF143764 , BI150692 , BF163277 , BE369870 , BI152144 , BF168743 , BF022265 , BI105027 , BE306549 , BF140175NMRIESTsBG862486 , BI654967 , BI854263 , BI654186 , BG974278 , BI662561 , BI853679 , BG864306 , BI658908Irga8Czech IImRNABC023105Irga9Czech IImRNABC040796Irga10Czech IImRNABC02011810 . 7554/eLife . 01298 . 024Table 3 . Primer listDOI: http://dx . doi . org/10 . 7554/eLife . 01298 . 024NameSequence 5′ to 3′FunctionIrga6_56B_fwCTACTATGAATGGTATATGTAGCATTGTGIrga6 amplificationIrga6_56B_bwCAGGACTTCAGCTTAATTAGAAGGCIrga6 amplificationIrgb2_66F_fCTGGACTCTGCGCTTTTATTGGIrgb2 amplificationIrgb2_66F_bCTGGAAACACTTTGCCCACGIrgb2 amplificationIrgb6_67Y_fwCCTCTCTTCTCCATTCAGCTTCIrgb6 amplificationIrgb6_67Y_bwCCAAGGTGAAGCTAAGAGTGAACIrgb6 amplificationIrgb10_682_fwCTCCAGTGTCCTGTGTGCCCIrgb10 amplificationIrgb10_682_bwCAGGAATGCCCTCAGTCGTCIrgb10 amplificationIrgm1_655_fwCTGCCGATTCGATTCATAAACIrgm1 amplificationIrgm1_655_bwCCTCTCAGAGAATCTAAAACCCIrgm1 amplificationIrgm1_66F_bwGAGACAGGGGAGATGAGTGATIrgm1 amplificationIrga1_221_fwATCGATAGTTCCCTTGTCAATGTGGbackcross and F2 mice genotyping , Chr 18Irga1_221_bwTTTGTAGAGTTTGGCTAGGGCCTGbackcross and F2 mice genotyping , Chr 18Irgb6_21D_fwATGGCTTGGGCCTCCAGCTTbackcross and F2 mice genotyping , Chr 11Irgb6_614_bwCCACCATTCCACTTGGTGGbackcross and F2 mice genotyping , Chr 11Tox-9AGGAGAGATATCAGGACTGTAGT . gondii qPCR primerTox-11GCGTCGTCTCGTCTAGATCGT . gondii qPCR primerNlrp1_FWAACTTATCTCAGGTCTCTGTGATTNlrp1b genotyping , forwardNlrp1_BL6GATATAGGTCAGGACCAATGCNlrp1b backward , BL/6 specificNlrp1_CIMGATATAGGTCAGGACCATCAANlrp1b backward , CIM specific Neighbor-joining trees of IRG genes and proteins were built with MEGA5 ( Tamura et al . , 2011 ) . The MSM/Ms BAC Illumina data were assembled de novo with the assistance of Geneious Pro 5 . 5 . 6 ( Biomatters Ltd . ) . Dot plots of genomic IRG gene clusters of BL/6 vs CIM and BL/6 vs MSM/Ms were calculated by LBDOT 1 . 0 ( Lynnon Corporation ) with a sliding window of 20 bp and a maximum mismatch of 2 bp . The raw Illumina reads from inbred mouse strains were accessed by LookSeq and all reads were grouped based on SNPs and manually aligned to their BL/6 homologues . Individual IRG gene sequences are available through Genbank . For analysis of the average diversity between house mouse strains , the sequences of 50 random functional genes were acquired from MGP and the National Institute of Genetics ( NIG ) , Japan ( http://molossinus . lab . nig . ac . jp/msmdb/ ) . The analysis covered seven mouse strains: two laboratory inbred strains , BL/6 ( MHC haplotype b ) and AKR/J ( MHC haplotype k ) ; four wild-derived inbred house mouse strains , MSM/Ms , CAST/EiJ , PWK/PhJ and WSB/EiJ; the M . spretus strain Spretus/EiJ . As shown in Figure 3—source data 1 , genes were arbitrarily selected based on numerical position in the NCBI reference assembly build 37 . Potentially functional genes closest to the designated genome position with EST evidence on the NCBI database were chosen . If the ORF of the gene was longer than 1500 bp , only 1500 bp were considered . Tajima’s π and πa/πs values were calculated with DnaSP5 ( Librado and Rozas , 2009 ) . The secondary structure of Irgb2-b1 was predicted by PSIpred ( UCL department of computer science , http://bioinf . cs . ucl . ac . uk/psipred/ ) . T . gondii strains ( Table 4 ) were maintained by serial passage in confluent monolayers of Hs27 cells . When Hs27 cells were lysed by T . gondii tachyzoites , parasites were harvested from the supernatant and purified from host cell debris by differential centrifugation ( 5 min at 100×g , 15 min at 500×g ) . The pelleted parasites were resuspended in IMDM , 5% FCS supplemented with 100 U/ml penicillin , 100 μg/ml streptomycin ( PAA , Pasching , Austria ) , counted and immediately used for infection of mice , cells or lysed for subsequent immunoblot . 10 . 7554/eLife . 01298 . 025Table 4 . T . gondii strains used in this studyDOI: http://dx . doi . org/10 . 7554/eLife . 01298 . 025TypeStrain nameReferenceNoteI ( virulent ) RH ( Albert and Sabin , 1941 ) RH-YFP ( Gubbels et al . , 2003 ) transgenic RH strain expressing YFPRHΔrop5 ( Behnke et al . , 2011 ) transgenic RH strain , the ROP5 locus has been deletedRHΔrop18 ( Reese et al . , 2011 ) transgenic RH strain , the ROP18 locus has been deletedBK ( Winsser et al . , 1948 ) GT-1 ( Dubey 1980 ) canonical type I strain , full sequence in ToxoDB DatabaseII ( avirulent ) ME49 ( Lunde and Jacobs , 1983 ) III ( avirulent ) NED ( Darde et al . , 1992 ) The IRG proteins all have distinctive properties . Irga6 and Irgb6 are the most highly expressed ( in lab mice ) and there are excellent serological reagents available for them . We therefore used Irga6 and Irgb6 for experiments that monitor loading at the vacuole . In addition , we have antisera against the specific phosphoserines on Irga6 so we can measure phosphorylation directly . Irgb6 is however more drastically affected by the genetic difference between avirulent types II and III T . gondii strains and virulent type I strains , dropping from up to 90% loading of vacuoles to around 10% . Irga6 is also greatly affected but the effect is seen more conspicuously as a reduction in the intensity of loading rather than in a reduction in the percent of vacuoles detectably loaded . These and other distinctive properties , are described elsewhere ( Khaminets et al . , 2010; Steinfeldt et al . , 2010; Fleckenstein et al . , 2012 ) . The tandem IRG protein , Irgb2-b1 , became a focus of attention because of its high expression in resistant CIM mice , its large polymorphic variation , and evidence for being under recent divergent selection ( see below ) . The origins of mice used in this study are listed in Table 5 . Cells were prepared from diaphragm tissue by a modification of the technique described by Antony et al . , ( 1989 ) . Diaphragm-derived cells are easy to prepare and have the advantage over MEFs that they can be prepared from a single adult mouse , enabling individual genetically different animals to be studied genetically and functionally at the cellular level . One mouse from each of BL/6 , CAST/EiJ , CTP and CIM strains was sacrificed and the diaphragm removed under sterile conditions . The diaphragm was washed with PBS , chopped up , incubated with collagenase/dispase ( 1 mg/ml , Roche , Mannheim , Germany ) for 1 hr at 37°C and then centrifuged for 15 s at 100×g . The supernatant was collected , centrifuged for 5 min at 500×g and the pellet plated in DMEM , 10% FCS supplemented with 4 mM L-glutamine , 2 mM non-essential amino acids , 1 mM sodium pyruvate , 1× MEM non-essential amino acids , 100 U/ml penicillin , 100 μg/ml streptomycin ( all PAA , Pasching , Austria ) . The remaining cell debris after collagenase/dispase-incubation was further incubated in 1× trypsin ( Gibco , Grand Island , New York , USA ) for 1 hr at 37°C , and then centrifuged for 15 s at 100×g . The supernatant was collected , centrifuged for 5 min at 500×g and the pellet plated ( see above ) . Primary diaphragm-derived cells ( DDC ) were grown until they had reached ∼50% confluence and then transfected with 2 µg of psv3-neo ( Southern and Berg , 1982 ) using the FuGENE HD transfection reagent ( Roche , Mannheim , Germany ) according to the manufacturer’s protocol . Cells were put under selection with G418 ( Geneticin , PAA , Pasching , Austria ) at a concentration of 150 μg/ml until immortalised clones had overgrown the culture . DDC isolated from mice and the mouse cell line L929 ( from mouse strain C3H ) were maintained in supplemented DMEM ( see above ) without G418 . 10 . 7554/eLife . 01298 . 026Table 5 . Origin of mouse samples and mouse genomic DNADOI: http://dx . doi . org/10 . 7554/eLife . 01298 . 026Sample nameSubspeciesOriginLocation of collectionProvided byD9 , D18 , D12 , D22 , D31 , D34M . m . domesticusGermany50°50′N 6°45′EGenomic DNA provided by B Harr Max Planck Institute for Evolutionary Biology , GermanyMC8 , MC4 , MC6 , MC52 , MC13 , MC27 , MC58M . m . domesticusFrance44°20′N 3°0′EW1 . 1 , W3 . 1 , W3 . 2 , W4 . 1 , W7 . 1M . m . musculusAustria48°12′N 16°22′EAL12 , AL21 , AL24 , AL30 , AL32 , AL41M . m . musculusKazakhstan43°N 77°EMW2 , MW4M . m . musculusInner Mongolia China41°5′N 108°9′ECaught by J Lilue for this study . Institute for Genetics , University of Cologne , GermanyMT1 , MT240°47′N 111°1′EJH4 , JH6 , JH11 , JH12M . m . musculus or Hybrid zoneHebei Province China37°37′N 115°19′EYX3 , YX5 , YX11M . m . castaneusHenan Province China32°4′N 115°3′EMIB3 , MIB4 , MIB6 , M . m . castaneusIndia13°3′N 77°34′ECaught by UB Müller for this study . Institute for Genetics , University of Cologne , GermanyMIB23 , MIB24 , MIB2513°6′N 77°34′EMIB35 , MIB3612°54′N 77°29′ECTP ( living mice ) M . m . castaneusThailandMouse strainF Bonhomme , Institut de Science de l’Evolution , Montpellier , FranceCIM ( living mice ) M . m . castaneusIndiaCAST/Ei ( living mice ) M . m . castaneusThailandInbred strainThe Jackson Laboratory , Bar Harbor , Maine , USAC57BL/6 ( living mice ) M . m . domesticusLab mouseInbred strainCentre for Mouse Genetics , University of Cologne , GermanyNMRI ( living mice ) M . m . domesticusLab mouseInbred strainCharles River Laboratories , Sulzfeld , Germany Cells were induced for 24 hr with 200 U/ml of IFNγ ( Peprotech , Rocky Hill , New York , USA ) unless indicated otherwise . Irga6BL/6 and Irga6CIM were cloned into the pGW1H vector with C-terminal ctag1 tags ( Martens et al . , 2005 ) ; full length Irgb2-b1BL/6 and Irgb2-b1CIM were cloned into the pGW1H vector with C-terminal Flag tags . Constructs were transfected using FuGENE HD Transfection Reagent ( Roche , Mannheim , Germany ) according to the manufacturer’s protocol . The multiplicities of infection ( MOI ) were 1 for the 3H-uracil incorporation assay , 2–5 for immunofluorescence microscopy , 2 . 5 , 5 and 10 for the cell viability assay , and ∼10 for in-cell phosphorylation experiments . Cells were either fixed for immunofluorescence or lysed for western blot 2 hr after infection . Cells were fixed with PBS/3% paraformaldehyde ( PFA ) for 20 min at room temperature ( RT ) , washed three times with PBS and then permeabilized with 100% methanol on ice ( for stainings including serum 87 , 558 , below ) or PBS/0 . 1% saponin at RT ( all other antibodies ) for 10 min followed by blocking with PBS/3% bovine serum albumin ( BSA ) for 1 hr . Cells were incubated with primary antibodies diluted in PBS/3% BSA for 1 hr and subsequently incubated with secondary antibodies for 30 min at RT . Antibodies against Irgb6 ( 141/1 ) and against the conserved Irgb-tandem C-terminal peptide CLSDLPEYWETGMEL ( 954/1-C15A ) shared by Irgb-tandem proteins of both BL/6 and CIM mice raised at Innovagen AB ( Lund , Sweden ) . Rabbit polyclonal anti-Irga6 phosphorylated at T108 has been described ( serum 87 , 558 ) ( Steinfeldt et al . , 2010 ) . Other primary immunoreagents were anti-recombinant Irga6 antiserum 165/3 ( Martens et al . , 2004 ) , mouse anti-FLAG ( M2 , Sigma-Aldrich , St . Louis , Missouri , USA ) and rabbit anti-calnexin ( Calbiochem , Darmstadt , Germany ) . Second-stage antibodies were: Alexa 488 and Alexa 555 labelled donkey anti-mouse and anti-rabbit sera ( Molecular Probes , Eugene , Oregon , USA ) . Images were taken with a Zeiss Axioplan II fluorescence microscope equipped with an AxioCam MRm camera ( Zeiss , Jena , Germany ) . Images were processed with Axiovision 4 . 7 ( Zeiss , Jena , Germany ) . Quantification of IRG protein signal intensity at the T . gondii PVM was performed as described before ( Khaminets et al . , 2010 ) . All quantification of microscopical images was performed double blind . Error bars in Figure 6A , G represent standard deviations of repeated measurements . 4×105 cells were seeded to individual wells of a six-well plate and induced with IFNγ for 24 hr . Cell lysis and western blot analysis was performed essentially as described elsewhere ( Steinfeldt et al . , 2010 ) . 6×105 BL/6 and CIM DDC were seeded in 6-cm dishes and induced with IFNγ for 24 hr . Metabolic labelling with 33P-phosphoric acid ( Hartmann Analytic , Braunschweig , Germany ) , cell lysis and immunoprecipitation was performed essentially as described elsewhere ( Steinfeldt et al . , 2010 ) . T . gondii proliferation was measured using the 3H-uracil incorporation assay ( Pfefferkorn and Guyre , 1984 ) . DDC were seeded on 96-well plates ( 6500 cells/well ) and induced with IFNγ ( 100 U/ml ) or left untreated . 24 hr after induction cells were infected for a further 24 hr with specified T . gondii strains at different multiplicities of infection , or left uninfected . The cultures were labeled with 0 . 3 μCi/well of 3H-uracil ( 3HU , Hartmann Analytic , Braunschweig , Germany ) for 24 hr and then frozen at −20°C . The amount of radioactivity incorporated into proliferating parasites was determined by a MatrixTM 9600 β-counter ( Packard , Meriden , Connecticut , USA ) . Data are shown for MOI = 1 and presented as the percentage of residual parasite proliferation under IFNγ treatment ( Figure 4B ) . Residual parasite proliferation was defined as follows: 100— ( [3HU counts—background in infected , IFNγ-treated culture/mean 3HU counts—background in infected , non-treated cultures] ×100 ) where background is 3HU ( mean ) counts of uninfected , non-induced cultures . DDC were seeded and induced as described for the 3H-uracil incorporation assay ( see above ) . Cells were infected with type I RH-YFP or type II ME49 strains of T . gondii with indicated MOIs for 8 hr . Viable cells were quantified by the CellTiter 96 AQueous non-radioactive cell proliferation assay ( Promega , Madison , Wisconsin , USA ) according to the manufacturer’s protocol . The absorption of a bioreduced formazan of the tetrazolium compound MTS , which is generated by metabolically active cells during incubation at 37°C for 2–4 hr , was measured in an ELISA reader ( Molecular Devices , Menlo Park , California , USA ) at 490 nm . The quantity of formazan product is proportional to the number of living cells in the culture . Mice were infected i . p . with 500 RH-YFP tachyzoites in 200 µl of PBS and tail samples were taken when animals succumbed during the acute phase of infection . Survivors were sacrificed 60 days post infection , tested for sero-conversion using the Toxocell Latex Kit ( biokit , Barcelona , Spain ) and tail biopsies taken . Biopsies were digested in 500 µl of buffer ( 100 mM Tris-HCl [pH 8 . 5] , 5 mM EDTA , 200 mM NaCl , 0 . 2% SDS , 150 µg/ml proteinase K ) and genomic DNA precipitated with isopropanol . An 804 bp fragment of Irga1 and an 857 bp fragment of Irgb6 were amplified from genomic DNA using the primers listed in Table 3 . PCR products were digested with restriction enzymes AccI ( Irga1 ) or FokI ( Irga6 , both New England BioLabs , Ipswich , Massachusetts , USA ) for 45 min at 37°C , followed by 20 min at 60°C . DNA fragments were separated on a 2% agarose gel . Nlrp1b fragments were amplified with a universal forward primer and strain-specific backward primers ( BL/6 or CIM , see Table 3 ) . Mice were infected i . p . with 500 RH-YFP tachyzoites , sacrificed on day 5 post infection and subsequently subjected to peritoneal lavage with 6 ml of PBS . Lavage suspension ( 1 ml ) was centrifuged in microtubes for 5 min at 500×g , the supernatant was discarded , the cell pellet resuspended with 30 µl of PBS/0 . 5% BSA containing PE-conjugated rat anti-mouse CD45 antibody ( BD Biosciences , San Jose , California , USA ) and subsequently incubated for 20 min on 4°C in the dark . The microtubes were then filled with PBS/0 . 5% BSA , centrifuged for 5 min at 500×g , the supernatant discarded and stained cells resuspended in 300 µl of PBS/0 . 5% BSA . Cells were analysed using a BD FACS Calibur flow cytometer ( BD Biosciences , San Jose , California , USA ) and the percentage of infected CD45-positive cells was calculated as percentage of events positive for YFP over 5×104 CD45-positive cells using WinMDI 2 . 9 . Mice were infected with T . gondii tachyzoites in 200 µl of PBS as indicated in Table 1 . 6–8 weeks ( GT-1 and BK ) or 5 weeks ( NED ) post infection mice were sacrificed , the brains removed and triturated in 1 ml of PBS . Cysts were counted in 15–20 drops of 10 µl per brain homogenate to estimate the total number of cysts per brain . Additionally , homogenised mouse brains were digested with proteinase K ( final concentration 100 μg/ml ) overnight , and total genomic DNA was isolated with DNeasy Blood & Tissue Kit ( Qiagen , Hilden , Germany ) according to the manufacturer’s protocol . The presence of T . gondii DNA was detected with a Taqman qPCR method ( 7900fast; Applied Biosystems , Foster City , California , USA ) by primer Tox-9 and Tox-11 as described before ( Reischl et al . , 2003 ) . Differences were tested for statistical significance using the unpaired two-tailed Student’s t test . | The parasite Toxoplasma gondii is one of the most common parasites worldwide and is known for its unusual life cycle . It reproduces sexually inside its primary host—the cat—and produces eggs that are released in faeces . Other animals , most often rodents , can then become infected when they unknowingly eat the eggs while foraging . Once inside its new host , the parasite reproduces asexually until the rodent’s immune system begins to fight back . It then becomes semi-dormant and forms cysts within the brain and muscle cells of its host . In an added twist , the parasite also causes rodents to lose their fear of cats . This increases their chances of being caught and eaten , thereby helping the parasite to return to its primary host and complete its life cycle . Previous work has shown that virulent strains of T . gondii can evade the host immune system in mice by secreting enzymes that inactivate immune-related proteins called IRG proteins . This prevents the infection being cleared and leads to death of the host within a few days . The existence of these virulent strains is intriguing because parasites that kill their host , and thus prevent their own reproduction , should be eliminated from the population . The fact that they are fairly common suggests that there must be a hitherto unknown mechanism that allows rodents to survive these virulent strains . Lilue et al . now report the existence of such a mechanism in strains of mice found in the wild . In contrast to laboratory mice , wild mice produce IRG proteins that inhibit the enzymes secreted by the virulent strains of T . gondii . Moreover , the IRG genes in wild mice are highly variable , whereas laboratory mice all have virtually identical IRG genes . By uncovering the complexity and variability of IRG genes in wild mice—complexity that has been lost from laboratory strains—Lilue et al . solve the conundrum of how highly virulent T . gondii strains can persist in the mouse population , and offer an explanation for the evolution of parasitic strains with differing levels of virulence . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"microbiology",
"and",
"infectious",
"disease"
] | 2013 | Reciprocal virulence and resistance polymorphism in the relationship between Toxoplasma gondii and the house mouse |
Trained neural network models , which exhibit features of neural activity recorded from behaving animals , may provide insights into the circuit mechanisms of cognitive functions through systematic analysis of network activity and connectivity . However , in contrast to the graded error signals commonly used to train networks through supervised learning , animals learn from reward feedback on definite actions through reinforcement learning . Reward maximization is particularly relevant when optimal behavior depends on an animal’s internal judgment of confidence or subjective preferences . Here , we implement reward-based training of recurrent neural networks in which a value network guides learning by using the activity of the decision network to predict future reward . We show that such models capture behavioral and electrophysiological findings from well-known experimental paradigms . Our work provides a unified framework for investigating diverse cognitive and value-based computations , and predicts a role for value representation that is essential for learning , but not executing , a task .
A major challenge in uncovering the neural mechanisms underlying complex behavior is our incomplete access to relevant circuits in the brain . Recent work has shown that model neural networks optimized for a wide range of tasks , including visual object recognition ( Cadieu et al . , 2014; Yamins et al . , 2014; Hong et al . , 2016 ) , perceptual decision-making and working memory ( Mante et al . , 2013; Barak et al . , 2013; Carnevale et al . , 2015; Song et al . , 2016; Miconi , 2016 ) , timing and sequence generation ( Laje and Buonomano , 2013; Rajan et al . , 2015 ) , and motor reach ( Hennequin et al . , 2014; Sussillo et al . , 2015 ) , can reproduce important features of neural activity recorded in numerous cortical areas of behaving animals . The analysis of such circuits , whose activity and connectivity are fully known , has therefore re-emerged as a promising tool for understanding neural computation ( Zipser and Andersen , 1988; Sussillo , 2014; Gao and Ganguli , 2015 ) . Constraining network training with tasks for which detailed neural recordings are available may also provide insights into the principles that govern learning in biological circuits ( Sussillo et al . , 2015; Song et al . , 2016; Brea and Gerstner , 2016 ) . Previous applications of this approach to 'cognitive-type' behavior such as perceptual decision-making and working memory have focused on supervised learning from graded error signals . Animals , however , learn to perform specific tasks from reward feedback provided by the experimentalist in response to definite actions , i . e . , through reinforcement learning ( Sutton and Barto , 1998 ) . Unlike in supervised learning where the network is given the correct response on each trial in the form of a continuous target output to be followed , reinforcement learning provides evaluative feedback to the network on whether each selected action was 'good' or 'bad . ' This form of feedback allows for a graded notion of behavioral correctness that is distinct from the graded difference between the network’s output and the target output in supervised learning . For the purposes of using model networks to generate hypotheses about neural mechanisms , this is particularly relevant in tasks where the optimal behavior depends on an animal’s internal state or subjective preferences . In a perceptual decision-making task with postdecision wagering , for example , on a random half of the trials the animal can opt for a sure choice that results in a small ( compared to the correct choice ) but certain reward ( Kiani and Shadlen , 2009 ) . The optimal decision regarding whether or not to select the sure choice depends not only on the task condition , such as the proportion of coherently moving dots , but also on the animal’s own confidence in its decision during the trial . Learning to make this judgment cannot be reduced to reproducing a predetermined target output without providing the full probabilistic solution to the network . It can be learned in a natural , ethologically relevant way , however , by choosing the actions that result in greatest overall reward; through training , the network learns from the reward contingencies alone to condition its output on its internal estimate of the probability that its answer is correct . Meanwhile , supervised learning is often not appropriate for value-based , or economic , decision-making where the 'correct' judgment depends explicitly on rewards associated with different actions , even for identical sensory inputs ( Padoa-Schioppa and Assad , 2006 ) . Although such tasks can be transformed into a perceptual decision-making task by providing the associated rewards as inputs , this sheds little light on how value-based decision-making is learned by the animal because it conflates external with 'internal , ' learned inputs . More fundamentally , reward plays a central role in all types of animal learning ( Sugrue et al . , 2005 ) . Explicitly incorporating reward into network training is therefore a necessary step toward elucidating the biological substrates of learning , in particular reward-dependent synaptic plasticity ( Seung , 2003; Soltani et al . , 2006; Izhikevich , 2007; Urbanczik and Senn , 2009; Frémaux et al . , 2010; Soltani and Wang , 2010; Hoerzer et al . , 2014; Brosch et al . , 2015; Friedrich and Lengyel , 2016 ) and the role of different brain structures in learning ( Frank and Claus , 2006 ) . In this work , we build on advances in recurrent policy gradient reinforcement learning , specifically the application of the REINFORCE algorithm ( Williams , 1992; Baird and Moore , 1999; Sutton et al . , 2000; Baxter and Bartlett , 2001; Peters and Schaal , 2008 ) to recurrent neural networks ( RNNs ) ( Wierstra et al . , 2009 ) , to demonstrate reward-based training of RNNs for several well-known experimental paradigms in systems neuroscience . The networks consist of two modules in an 'actor-critic' architecture ( Barto et al . , 1983; Grondman et al . , 2012 ) , in which a decision network uses inputs provided by the environment to select actions that maximize reward , while a value network uses the selected actions and activity of the decision network to predict future reward and guide learning . We first present networks trained for tasks that have been studied previously using various forms of supervised learning ( Mante et al . , 2013; Barak et al . , 2013; Song et al . , 2016 ) ; they are characterized by 'simple' input-output mappings in which the correct response for each trial depends only on the task condition , and include perceptual decision-making , context-dependent integration , multisensory integration , and parametric working memory tasks . We then show results for tasks in which the optimal behavior depends on the animal’s internal judgment of confidence or subjective preferences , specifically a perceptual decision-making task with postdecision wagering ( Kiani and Shadlen , 2009 ) and a value-based economic choice task ( Padoa-Schioppa and Assad , 2006 ) . Interestingly , unlike for the other tasks where we focus on comparing the activity of units in the decision network to neural recordings in the dorsolateral prefrontal and posterior parietal cortex of animals performing the same tasks , for the economic choice task we show that the activity of the value network exhibits a striking resemblance to neural recordings from the orbitofrontal cortex ( OFC ) , which has long been implicated in the representation of reward-related signals ( Wallis , 2007 ) . An interesting feature of our REINFORCE-based model is that a reward baseline—in this case , the output of a recurrently connected value network ( Wierstra et al . , 2009 ) —is essential for learning , but not for executing the task , because the latter depends only on the decision network . Importantly , learning can sometimes still occur without the value network but is much more unreliable . It is sometimes observed in experiments that reward-modulated structures in the brain such as the basal ganglia or OFC are necessary for learning or adapting to a changing environment , but not for executing a previously learned skill ( Turner and Desmurget , 2010; Schoenbaum et al . , 2011; Stalnaker et al . , 2015 ) . This suggests that one possible role for such circuits may be representing an accurate baseline to guide learning . Moreover , since confidence is closely related to expected reward in many cognitive tasks , the explicit computation of expected reward by the value network provides a concrete , learning-based rationale for confidence estimation as a ubiquitous component of decision-making ( Kepecs et al . , 2008; Wei and Wang , 2015 ) , even when it is not strictly required for performing the task . Conceptually , the formulation of behavioral tasks in the language of reinforcement learning presented here is closely related to the solution of partially observable Markov decision processes ( POMDPs ) ( Kaelbling et al . , 1998 ) using either model-based belief states ( Rao , 2010 ) or model-free working memory ( Todd et al . , 2008 ) . Indeed , as in Dayan and Daw , ( 2008 ) one of the goals of this work is to unify related computations into a common language that is applicable to a wide range of tasks in systems neuroscience . Such policies can also be compared more directly to behaviorally 'optimal' solutions when they are known , for instance to the signal detection theory account of perceptual decision-making ( Gold and Shadlen , 2007 ) . Thus , in addition to expanding the range of tasks and neural mechanisms that can be studied with trained RNNs , our work provides a convenient framework for the study of cognitive and value-based computations in the brain , which have often been viewed from distinct perspectives but in fact arise from the same reinforcement learning paradigm .
For concreteness , we illustrate the following in the context of a simplified perceptual decision-making task based on the random dots motion ( RDM ) discrimination task as described in Kiani et al . ( 2008 ) ( Figure 1A ) . In its simplest form , in an RDM task the monkey must maintain fixation until a 'go' cue instructs the monkey to indicate its decision regarding the direction of coherently moving dots on the screen . Thus the three possible actions available to the monkey at any given time are fixate , choose left , or choose right . The true direction of motion , which can be considered a state of the environment , is not known to the monkey with certainty , i . e . , is partially observable . The monkey must therefore use the noisy sensory evidence to infer the direction in order to select the correct response at the end of the trial . Breaking fixation early results in a negative reward in the form of a timeout , while giving the correct response after the fixation cue is extinguished results in a positive reward in the form of juice . Typically , there is neither a timeout nor juice for an incorrect response during the decision period , corresponding to a 'neutral' reward of zero . The goal of this section is to give a general description of such tasks and how an RNN can learn a behavioral policy for choosing actions at each time to maximize its cumulative reward . 10 . 7554/eLife . 21492 . 003Figure 1 . Recurrent neural networks for reinforcement learning . ( A ) Task structure for a simple perceptual decision-making task with variable stimulus duration . The agent must maintain fixation ( at=F ) until the go cue , which indicates the start of a decision period during which choosing the correct response ( at=L or at=R ) results in a positive reward . The agent receives zero reward for responding incorrectly , while breaking fixation early results in an aborted trial and negative reward . ( B ) At each time t the agent selects action at according to the output of the decision network πθ , which implements a policy that can depend on all past and current inputs 𝐮1:t provided by the environment . In response , the environment transitions to a new state and provides reward ρt+1 to the agent . The value network vϕ uses the selected action and the activity of the decision network 𝐫tπ to predict future rewards . All the weights shown are plastic , i . e . , trained by gradient descent . ( C ) Performance of the network trained for the task in ( A ) , showing the percent correct by stimulus duration , for different coherences ( the difference in strength of evidence for L and R ) . ( D ) Neural activity of an example decision network unit , sorted by coherence and aligned to the time of stimulus onset . Solid lines are for positive coherence , dashed for negative coherence . ( E ) Output of the value network ( expected return ) aligned to stimulus onset . Expected return is computed by performing an 'absolute value'-like operation on the accumulated evidence . DOI: http://dx . doi . org/10 . 7554/eLife . 21492 . 00310 . 7554/eLife . 21492 . 004Figure 1—figure supplement 1 . Learning curves for the simple perceptual decision-making task . ( A ) Average reward per trial . Black indicates the network realization shown in the main text , gray additional realizations , i . e . , trained with different random number generator seeds . ( B ) Percent correct , for trials on which the network made a decision ( ≥99% required for termination ) . Red: target performance ( when training was terminated ) . DOI: http://dx . doi . org/10 . 7554/eLife . 21492 . 00410 . 7554/eLife . 21492 . 005Figure 1—figure supplement 2 . Reaction-time version of the perceptual decision-making task , in which the go cue coincides with the onset of stimulus , allowing the agent to choose when to respond . ( A ) Task structure for the reaction-time version of the simple perceptual decision-making task , in which the agent can choose to respond any time after the onset of stimulus . ( B ) Reaction time as a function of coherence for correct ( solid circles ) and error ( open circles ) trials . ( C ) Neural activity of an example decision network unit , sorted by the coherence ( the difference in strength of evidence for L and R ) and aligned to the time of stimulus onset . Each trial ends when the network breaks fixation . DOI: http://dx . doi . org/10 . 7554/eLife . 21492 . 00510 . 7554/eLife . 21492 . 006Figure 1—figure supplement 3 . Learning curves for the reaction-time version of the simple perceptual decision-making task . ( A ) Average reward per trial . Black indicates the network realization shown in the main text , gray additional realizations , i . e . , trained with different random number generator seeds . ( B ) Percent correct , for trials on which the network made a decision ( ≥99% required for termination ) . Red: target performance ( when training was terminated ) . DOI: http://dx . doi . org/10 . 7554/eLife . 21492 . 00610 . 7554/eLife . 21492 . 007Figure 1—figure supplement 4 . Learning curves for the simple perceptual decision-making task with a linear readout of the decision network as the baseline . ( A ) Average reward per trial . Black indicates the network realization shown in the main text , gray additional realizations , i . e . , trained with different random number generator seeds . ( B ) Percent correct , for trials on which the network made a decision ( ≥99% required for termination ) . Red: target performance ( when training was terminated ) . DOI: http://dx . doi . org/10 . 7554/eLife . 21492 . 007 Consider a typical interaction between an experimentalist and animal , which we more generally call the environment ℰ and agent 𝒜 , respectively ( Figure 1B ) . At each time t the agent chooses to perform actions 𝐚t after observing inputs 𝐮t provided by the environment , and the probability of choosing actions 𝐚t is given by the agent’s policy πθ ( 𝐚t|𝐮1:t ) with parameters θ . Here the policy is implemented as the output of an RNN , so that θ comprises the connection weights , biases , and initial state of the decision network . The policy at time t can depend on all past and current inputs 𝐮1:t= ( 𝐮1 , 𝐮2 , … , 𝐮t ) , allowing the agent to integrate sensory evidence or use working memory to perform the task . The exception is at t=0 , when the agent has yet to interact with the environment and selects its actions 'spontaneously' according to πθ ( 𝐚0 ) . We note that , if the inputs give exact information about the environmental state 𝐬t , i . e . , if 𝐮t=𝐬t , then the environment can be described by a Markov decision process . In general , however , the inputs only provide partial information about the environmental states , requiring the network to accumulate evidence over time to determine the state of the world . In this work we only consider cases where the agent chooses one out of Na possible actions at each time , so that πθ ( 𝐚t|𝐮1:t ) for each t is a discrete , normalized probability distribution over the possible actions a1 , … , aNa . More generally , 𝐚t can implement several distinct actions or even continuous actions by representing , for example , the means of Gaussian distributions ( Peters and Schaal , 2008; Wierstra et al . , 2009 ) . After each set of actions by the agent at time t the environment provides a reward ( or special observable ) ρt+1 at time t+1 , which the agent attempts to maximize in the sense described below . In the case of the example RDM task above ( Figure 1A ) , the environment provides ( and the agent receives ) as inputs a fixation cue and noisy evidence for two choices L ( eft ) and R ( ight ) during a variable-length stimulus presentation period . The strength of evidence , or the difference between the evidence for L and R , is called the coherence , and in the actual RDM experiment corresponds to the percentage of dots moving coherently in one direction on the screen . The agent chooses to perform one of Na=3 actions at each time: fixate ( at=F ) , choose L ( at=L ) , or choose R ( at=R ) . Here , the agent must choose F as long as the fixation cue is on , and then , when the fixation cue is turned off to indicate that the agent should make a decision , correctly choose L or R depending on the sensory evidence . Indeed , for all tasks in this work we required that the network 'make a decision' ( i . e . , break fixation to indicate a choice at the appropriate time ) on at least 99% of the trials , whether the response was correct or not . A trial ends when the agent chooses L or R regardless of the task epoch: breaking fixation early before the go cue results in an aborted trial and a negative reward ρt=-1 , while a correct decision is rewarded with ρt=+1 . Making the wrong decision results in no reward , ρt=0 . For the zero-coherence condition the agent is rewarded randomly on half the trials regardless of its choice . Otherwise the reward is always ρt=0 . Formally , a trial proceeds as follows . At time t=0 , the environment is in state 𝐬0 with probability ℰ ( 𝐬0 ) . The state 𝐬0 can be considered the starting time ( i . e . , t=0 ) and 'task condition , ' which in the RDM example consists of the direction of motion of the dots ( i . e . , whether the correct response is L or R ) and the coherence of the dots ( the difference between evidence for L and R ) . The time component of the state , which is updated at each step , allows the environment to present different inputs to the agent depending on the task epoch . The true state 𝐬0 ( such as the direction of the dots ) is only partially observable to the agent , so that the agent must instead infer the state through inputs 𝐮t provided by the environment during the course of the trial . As noted previously , the agent initially chooses actions 𝐚0 with probability πθ ( 𝐚0 ) . The networks in this work almost always begin by choosing F , or fixation . At time t=1 , the environment , depending on its previous state 𝐬0 and the agent’s action 𝐚0 , transitions to state 𝐬1 with probability ℰ ( 𝐬1|𝐬0 , 𝐚0 ) and generates reward ρ1 . In the perceptual decision-making example , only the time advances since the trial condition remains constant throughout . From this state the environment generates observable 𝐮1 with a distribution given by ℰ ( 𝐮1|𝐬1 ) . If t=1 were in the stimulus presentation period , for example , 𝐮1 would provide noisy evidence for L or R , as well as the fixation cue . In response , the agent , depending on the inputs 𝐮1 it receives from the environment , chooses actions 𝐚1 with probability πθ ( 𝐚1|𝐮1:1 ) =πθ ( 𝐚1|𝐮1 ) . The environment , depending on its previous states 𝐬0:1= ( 𝐬0 , 𝐬1 ) and the agent’s previous actions a0:1= ( 𝐚0 , 𝐚1 ) , then transitions to state 𝐬2 with probability ℰ ( 𝐬2|𝐬0:1 , 𝐚0:1 ) and generates reward ρ2 . These steps are repeated until the end of the trial at time T . Trials can terminate at different times ( e . g . , for breaking fixation early or because of variable stimulus durations ) , so that T in the following represents the maximum length of a trial . In order to emphasize that rewards follow actions , we adopt the convention in which the agent performs actions at t=0 , … , T and receives rewards at t=1 , … , T+1 . The goal of the agent is to maximize the sum of expected future rewards at time t=0 , or expected return ( 1 ) J ( θ ) =𝔼H[∑t=0Tρt+1] , where the expectation 𝔼H is taken over all possible trial histories H= ( 𝐬0:T+1 , 𝐮1:T , 𝐚0:T ) consisting of the states of the environment , the inputs given to the agent , and the actions of the agent . In practice , the expectation value in Equation 1 is estimated by performing Ntrials trials for each policy update , i . e . , with a Monte Carlo approximation . The expected return depends on the policy and hence parameters θ , and we use Adam stochastic gradient descent ( SGD ) ( Kingma and Ba , 2015 ) with gradient clipping ( Graves , 2013; Pascanu et al . , 2013b ) to find the parameters that maximize this reward ( Materials and methods ) . More specifically , after every Ntrials trials the decision network uses gradient descent to update its parameters in a direction that minimizes an objective function ℒπ of the form ( 2 ) ℒπ ( θ ) =1Ntrials∑n=1Ntrials[−Jn ( θ ) +Ωnπ ( θ ) ] with respect to the connection weights , biases , and initial state of the decision network , which we collectively denote as θ . Here Ωnπ ( θ ) can contain any regularization terms for the decision network , for instance an entropy term to control the degree of exploration ( Xu et al . , 2015 ) . The key gradient ∇θJn ( θ ) is given for each trial n by the REINFORCE algorithm ( Williams , 1992; Baird and Moore , 1999; Sutton et al . , 2000; Baxter and Bartlett , 2001; Peters and Schaal , 2008; Wierstra et al . , 2009 ) as ( 3 ) ∇θJn ( θ ) =∑t=0T[∇θlogπθ ( at|u1:t ) ][∑τ=tTρτ+1−vϕ ( a1:t , r1:tπ ) ] , where 𝐫1:tπ are the firing rates of the decision network units up to time t , vϕ denotes the value function as described below , and the gradient ∇θlogπθ ( at|u1:t ) , known as the eligibility , [and likewise ∇θΩnπ ( θ ) ] is computed by backpropagation through time ( BPTT ) ( Rumelhart et al . , 1986 ) for the selected actions 𝐚t . The sum over rewards in large brackets only runs over τ=t , … , T , which reflects the fact that present actions do not affect past rewards . In this form the terms in the gradient have the intuitive property that they are nonzero only if the actual return deviates from what was predicted by the baseline . It is worth noting that this form of the value function ( with access to the selected action ) can , in principle , lead to suboptimal policies if the value network’s predictions become perfect before the optimal decision policy is learned; we did not find this to be the case in our simulations . The reward baseline is an important feature in the success of almost all REINFORCE-based algorithms , and is here represented by a second RNN vϕ with parameters ϕ in addition to the decision network πθ ( to be precise , the value function is the readout of the value network ) . This baseline network , which we call the value network , uses the selected actions 𝐚1:t and activity of the decision network 𝐫1:tπ to predict the expected return at each time t=1 , … , T; the value network also predicts the expected return at t=0 based on its own initial states , with the understanding that 𝐚1:0=∅ and 𝐫1:0π=∅ are empty sets . The value network is trained by minimizing a second objective function ( 4 ) ℒv ( ϕ ) =1Ntrials∑n=1Ntrials[En ( ϕ ) +Ωnv ( ϕ ) ] , ( 5 ) En ( ϕ ) =1T+1∑t=0T[∑τ=tTρτ+1−vϕ ( a1:t , r1:tπ ) ]2 every Ntrials trials , where Ωnv ( ϕ ) denotes any regularization terms for the value network . The necessary gradient ∇ϕEn ( ϕ ) [and likewise ∇ϕΩnv ( ϕ ) ] is again computed by BPTT . The policy probability distribution over actions πθ ( 𝐚t|𝐮1:t ) and scalar baseline vϕ ( 𝐚1:t , 𝐫1:tπ ) are each represented by an RNN of N firing-rate units 𝐫π and 𝐫v , respectively , where we interpret each unit as the mean firing rate of a group of neurons . In the case where the agent chooses a single action at each time t , the activity of the decision network determines πθ ( 𝐚t|𝐮1:t ) through a linear readout followed by softmax normalization: ( 6 ) zt=Woutπrtπ+boutπ , ( 7 ) πθ ( at=k|u1:t ) =e ( zt ) k∑ℓ=1Nae ( zt ) ℓ for k=1 , … , Na . Here Woutπ is an Na×N matrix of connection weights from the units of the decision network to the Na linear readouts 𝐳t , and 𝐛outπ are Na biases . Action selection is implemented by randomly sampling from the probability distribution in Equation 7 , and constitutes an important difference from previous approaches to training RNNs for cognitive tasks ( Mante et al . , 2013; Carnevale et al . , 2015; Song et al . , 2016; Miconi , 2016 ) , namely , here the final output of the network ( during training ) is a specific action , not a graded decision variable . We consider this sampling as an abstract representation of the downstream action selection mechanisms present in the brain , including the role of noise in implicitly realizing stochastic choices with deterministic outputs ( Wang , 2002 , 2008 ) . Meanwhile , the activity of the value network predicts future returns through a linear readout ( 8 ) vϕ ( 𝐚1:t , 𝐫1:tπ ) =Woutv𝐫tv+boutv , where Woutv is an 1×N matrix of connection weights from the units of the value network to the single linear readout vϕ , and boutv is a bias term . In order to take advantage of recent developments in training RNNs [in particular , addressing the problem of vanishing gradients ( Bengio et al . , 1994 ) ] while retaining intepretability , we use a modified form of Gated Recurrent Units ( GRUs ) ( Cho et al . , 2014; Chung et al . , 2014 ) with a threshold-linear 'f-I' curve [x]+=max ( 0 , x ) to obtain positive , non-saturating firing rates . Since firing rates in cortex rarely operate in the saturating regime , previous work ( Sussillo et al . , 2015 ) used an additional regularization term to prevent saturation in common nonlinearities such as the hyperbolic tangent; the threshold-linear activation function obviates such a need . These units are thus leaky , threshold-linear units with dynamic time constants and gated recurrent inputs . The equations that describe their dynamics can be derived by a naïve discretization of the following continuous-time equations for the N currents 𝐱 and corresponding rectified-linear firing rates 𝐫: ( 9 ) λ=sigmoid ( Wrecλr+Winλu+bλ ) , ( 10 ) γ=sigmoid ( Wrecγr+Winγu+bγ ) , ( 11 ) τλ⊙x . =−x+Wrec ( γ⊙r ) +Winu+b+2τσrec2ξ , ( 12 ) r=[x]+ . Here x . =dx/dt is the derivative of 𝐱 with respect to time , ⊙ denotes elementwise multiplication , sigmoid ( x ) =[1+e−x]−1 is the logistic sigmoid , 𝐛λ , 𝐛γ , and 𝐛 are biases , 𝝃 are N independent Gaussian white noise processes with zero mean and unit variance , and σrec2 controls the size of this noise . The multiplicative gates 𝝀 dynamically modulate the overall time constant τ for network units , while the 𝜸 control the recurrent inputs . The N×N matrices Wrec , Wrecλ , and Wrecγ are the recurrent weight matrices , while the N×Nin matrices Win , Winλ , and Winγ are connection weights from the Nin inputs 𝐮 to the N units of the network . We note that in the case where 𝝀→1 and 𝜸→1 the equations reduce to 'simple' leaky threshold-linear units without the modulation of the time constants or gating of inputs . We constrain the recurrent connection weights ( Song et al . , 2016 ) so that the overall connection probability is pc; specifically , the number of incoming connections for each unit , or in-degree K , was set to K=pcN ( see Table 1 for a list of all parameters ) . 10 . 7554/eLife . 21492 . 008Table 1 . Parameters for reward-based recurrent neural network training . Unless noted otherwise in the text , networks were trained and run with the parameters listed here . DOI: http://dx . doi . org/10 . 7554/eLife . 21492 . 008ParameterSymbolDefault valueLearning rateη0 . 004Maximum gradient normΓ1Size of decision/value networkN100Connection probability ( decision network ) pcπ0 . 1Connection probability ( value network ) pcv1Time stepΔt10 msUnit time constantτ100 msRecurrent noiseσrec20 . 01Initial spectral radius for recurrent weightsρ02Number of trials per gradient updateNtrials# of task conditions The result of discretizing Equations 9–12 , as well as details on initializing the network parameters , are given in Materials and methods . We successfully trained networks with time steps Δt=1 ms , but for computational convenience all of the networks in this work were trained and run with Δt=10 ms . We note that , for typical tasks in systems neuroscience lasting on the order of several seconds , this already implies trials lasting hundreds of time steps . Unless noted otherwise in the text , all networks were trained using the parameters listed in Table 1 . While the inputs to the decision network πθ are determined by the environment , the value network always receives as inputs the activity of the decision network 𝐫π , together with information about which actions were actually selected at each time step ( Figure 1B ) . The value network serves two purposes: first , the output of the value network is used as the baseline in the REINFORCE gradient , Equation 3 , to reduce the variance of the gradient estimate ( Williams , 1992; Baird and Moore , 1999; Baxter and Bartlett , 2001; Peters and Schaal , 2008 ) ; second , since policy gradient reinforcement learning does not explicitly use a value function but value information is nevertheless implicitly contained in the policy , the value network serves as an explicit and potentially nonlinear readout of this information . In situations where expected reward is closely related to confidence , this may explain , for example , certain disassociations between perceptual decisions and reports of the associated confidence ( Lak et al . , 2014 ) . A reward baseline , which allows the decision network to update its parameters based on a relative quantity akin to prediction error ( Schultz et al . , 1997; Bayer and Glimcher , 2005 ) rather than absolute reward magnitude , is essential to many learning schemes , especially those based on REINFORCE . Indeed , it has been suggested that in general such a baseline should be not only task-specific but stimulus ( task-condition ) -specific ( Frémaux et al . , 2010; Engel et al . , 2015; Miconi , 2016 ) , and that this information may be represented in OFC ( Wallis , 2007 ) or basal ganglia ( Doya , 2000 ) . Previous schemes , however , did not propose how this baseline critic may be instantiated , instead implementing it algorithmically . Here we use a simple neural implementation of the baseline that automatically depends on the stimulus and thus does not require the learning system to have access to the true trial type , which in general is not known with certainty to the agent . The training procedure described in the previous section can be used for a variety of tasks , and results in networks that qualitatively reproduce both behavioral and electrophysiological findings from experiments with behaving animals . For the example perceptual decision-making task above , the trained network learns to integrate the sensory evidence to make the correct decision about which of two noisy inputs is larger ( Figure 1C ) . This and additional networks trained for the same task were able to reach the target performance in ∼7000 trials starting from completely random connection weights , and moreover the networks learned the 'core' task after ∼2000 trials ( Figure 1—figure supplement 1 ) . As with monkeys performing the task , longer stimulus durations allow the network to improve its performance by continuing to integrate the incoming sensory evidence ( Wang , 2002; Kiani et al . , 2008 ) . Indeed , the output of the value network shows that the expected reward ( in this case equivalent to confidence ) is modulated by stimulus difficulty ( Figure 1E ) . Prior to the onset of the stimulus , the expected reward is the same for all trial conditions and approximates the overall reward rate; incoming sensory evidence then allows the network to distinguish its chances of success . Sorting the activity of individual units in the network by the signed coherence ( the strength of the evidence , with negative values indicating evidence for L and positive for R ) also reveals coherence-dependent ramping activity ( Figure 1D ) as observed in neural recordings from numerous perceptual decision-making experiments , e . g . , Roitman and Shadlen ( 2002 ) . This pattern of activity illustrates why a nonlinear readout by the value network is useful: expected return is computed by performing an 'absolute value'-like operation on the accumulated evidence ( plus shifts ) , as illustrated by the overlap of the expected return for positive and negative-coherence trials ( Figure 1E ) . The reaction time as a function of coherence in the reaction-time version of the same task , in which the go cue coincides with the time of stimulus onset , is also shown in Figure 1—figure supplement 2 and may be compared , e . g . , to Wang ( 2002 ) ; Mazurek et al . ( 2003 ) ; Wong and Wang ( 2006 ) . We note that in many neural models [e . g . , Wang ( 2002 ) ; Wong and Wang ( 2006 ) ] a 'decision' is made when the output reaches a fixed threshold . Indeed , when networks are trained using supervised learning ( Song et al . , 2016 ) , the decision threshold is imposed retroactively and has no meaning during training; since the outputs are continuous , the speed-accuracy tradeoff is also learned in the space of continuous error signals . Here , the time at which the network commits to a decision is unambiguously given by the time at which the selected action is L or R . Thus the appropriate speed-accuracy tradeoff is learned in the space of concrete actions , illustrating the desirability of using reward-based training of RNNs when modeling reaction-time tasks . Learning curves for this and additional networks trained for the same reaction-time task are shown in Figure 1—figure supplement 3 . In addition to the example task from the previous section , we trained networks for three well-known behavioral paradigms in which the correct , or optimal , behavior is ( pre- ) determined on each trial by the task condition alone . Similar tasks have previously been addressed with several different forms of supervised learning , including FORCE ( Sussillo and Abbott , 2009; Carnevale et al . , 2015 ) , Hessian-free ( Martens and Sutskever , 2011; Mante et al . , 2013; Barak et al . , 2013 ) , and stochastic gradient descent ( Pascanu et al . , 2013b; Song et al . , 2016 ) , so that the results shown in Figure 2 are presented as confirmation that the same tasks can also be learned using reward feedback on definite actions alone . For all three tasks the pre-stimulus fixation period was 750 ms; the networks had to maintain fixation until the start of a 500 ms 'decision' period , which was indicated by the extinction of the fixation cue . At this time the network was required to choose one of two alternatives to indicate its decision and receive a reward of +1 for a correct response and 0 for an incorrect response; otherwise , the networks received a reward of −1 . 10 . 7554/eLife . 21492 . 009Figure 2 . Performance and neural activity of RNNs trained for 'simple' cognitive tasks in which the correct response depends only on the task condition . Left column shows behavioral performance , right column shows mixed selectivity for task parameters of example units in the decision network . ( A ) Context-dependent integration task ( Mante et al . , 2013 ) . Left: Psychometric curves show the percentage of R choices as a function of the signed 'motion' and 'color' coherences in the motion ( black ) and color ( blue ) contexts . Right: Normalized firing rates of examples units sorted by different combinations of task parameters exhibit mixed selectivity . Firing rates were normalized by mean and standard deviation computed over the responses of all units , times , and trials . Solid and dashed lines indicate choice 1 ( same as preferred direction of unit ) and choice 2 ( non-preferred ) , respectively . For motion and choice and color and choice , dark to light corresponds to high to low motion and color coherence , respectively . ( B ) Multisensory integration task ( Raposo et al . , 2012 , 2014 ) . Left: Psychometric curves show the percentage of high choices as a function of the event rate , for visual only ( blue ) , auditory only ( green ) , and multisensory ( orange ) trials . Improved performance on multisensory trials shows that the network learns to combine the two sources of information in accordance with Equation 13 . Right: Sorted activity on visual only and auditory only trials for units selective for choice ( high vs . low , left ) , modality [visual ( vis ) vs . auditory ( aud ) , middle] , and both ( right ) . Error trials were excluded . ( C ) Parametric working memory task ( Romo et al . , 1999 ) . Left: Percentage of correct responses for different combinations of f1 and f2 . The conditions are colored here and in the right panels according to the first stimulus ( base frequency ) f1; due to the overlap in the values of f1 , the 10 task conditions are represented by seven distinct colors . Right: Activity of example decision network units sorted by f1 . The first two units are positively tuned to f1 during the delay period , while the third unit is negatively tuned . DOI: http://dx . doi . org/10 . 7554/eLife . 21492 . 00910 . 7554/eLife . 21492 . 010Figure 2—figure supplement 1 . Learning curves for the context-dependent integration task . ( A ) Average reward per trial . Black is for the network realization in the main text , gray for additional realizations , i . e . , trained with different random number generator seeds . ( B ) Percent correct , for trials on which the network made a decision ( ≥99% required for termination ) . Red: target performance ( when training was terminated ) . DOI: http://dx . doi . org/10 . 7554/eLife . 21492 . 01010 . 7554/eLife . 21492 . 011Figure 2—figure supplement 2 . Learning curves for the multisensory integration task . ( A ) Average reward per trial . Black indicates the network realization shown in the main text , gray additional realizations , i . e . , trained with different random number generator seeds . ( B ) Percent correct , for trials on which the network made a decision ( ≥99% required for termination ) . Red: target performance ( when training was terminated ) . DOI: http://dx . doi . org/10 . 7554/eLife . 21492 . 01110 . 7554/eLife . 21492 . 012Figure 2—figure supplement 3 . Learning curves for the parametric working memory task . ( A ) Average reward per trial . Black indicates the network realization shown in the main text , gray additional realizations , i . e . , trained with different random number generator seeds . ( B ) Percent correct , for trials on which the network made a decision ( ≥99% required for termination ) . Red: target performance ( when training was terminated ) . DOI: http://dx . doi . org/10 . 7554/eLife . 21492 . 012 The context-dependent integration task ( Figure 2A ) is based on Mante et al . ( 2013 ) , in which monkeys were required to integrate one type of stimulus ( the motion or color of the presented dots ) while ignoring the other depending on a context cue . In training the network , we included both the 750 ms stimulus period and 300–1500 ms delay period following stimulus presentation . The delay consisted of 300 ms followed by a variable duration drawn from an exponential distribution with mean 300 ms and truncated at a maximum of 1200 ms . The network successfully learned to perform the task , which is reflected in the psychometric functions showing the percentage of trials on which the network chose R as a function of the signed motion and color coherences , where motion and color indicate the two sources of noisy information and the sign is positive for R and negative for L ( Figure 2A , left ) . As in electrophysiological recordings , units in the decision network show mixed selectivity when sorted by different combinations of task variables ( Figure 2A , right ) . Learning curves for this and additional networks trained for the task are shown in Figure 2—figure supplement 1 . The multisensory integration task ( Figure 2B ) is based on Raposo et al . ( 2012 , 2014 ) , in which rats used visual flashes and auditory clicks to determine whether the event rate was higher or lower than a learned threshold of 12 . 5 events per second . When both modalities were presented , they were congruent , which implied that the rats could improve their performance by combining information from both sources . As in the experiment , the network was trained with a 1000 ms stimulus period , with inputs whose magnitudes were proportional ( both positively and negatively ) to the event rate . For this task the input connection weights Win , Winλ , and Winγ were initialized so that a third of the N=150 decision network units received visual inputs only , another third auditory inputs only , and the remaining third received neither . As shown in the psychometric function ( percentage of high choices as a function of event rate , Figure 2B , left ) , the trained network exhibits multisensory enhancement in which performance on multisensory trials was better than on single-modality trials . Indeed , as for rats , the results are consistent with optimal combination of the two modalities , ( 13 ) 1σvisual2+1σauditory2≈1σmultisensory2 , where σvisual2 , σauditory2 , and σmultisensory2 are the variances obtained from fits of the psychometric functions to cumulative Gaussian functions for visual only , auditory only , and multisensory ( both visual and auditory ) trials , respectively ( Table 2 ) . As observed in electrophysiological recordings , moreover , decision network units exhibit a range of tuning to task parameters , with some selective to choice and others to modality , while many units showed mixed selectivity to all task variables ( Figure 2B , right ) . Learning curves for this and additional networks trained for the task are shown in Figure 2—figure supplement 2 . 10 . 7554/eLife . 21492 . 013Table 2 . Psychophysical thresholds σvisual , σauditory , and σmultisensory obtained from fits of cumulative Gaussian functions to the psychometric curves in visual only , auditory only , and multisensory trials in the multisensory integration task , for six networks trained from different random initializations ( first row , bold: network from main text , cf . Figure 2B ) . The last two columns show evidence of 'optimal' multisensory integration according to Equation 13 ( Raposo et al . , 2012 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 21492 . 013σvisualσauditoryσmultisensory1σvisual2+1σauditory21σmultisensory22 . 1242 . 0991 . 4510 . 4490 . 4752 . 1072 . 0861 . 4480 . 4550 . 4772 . 2762 . 1281 . 5520 . 4140 . 4152 . 1182 . 1551 . 5080 . 4380 . 4402 . 0772 . 1711 . 5820 . 4440 . 4002 . 0882 . 1491 . 4800 . 4460 . 457 The parametric working memory task ( Figure 2C ) is based on the vibrotactile frequency discrimination task of Romo et al . ( 1999 ) , in which monkeys were required to compare the frequencies of two temporally separated stimuli to determine which was higher . For network training , the task epochs consisted of a 500 ms base stimulus with 'frequency' f1 , a 2700–3300 ms delay , and a 500 ms comparison stimulus with frequency f2; for the trials shown in Figure 2C the delay was always 3000 ms as in the experiment . During the decision period , the network had to indicate which stimulus was higher by choosing f1<f2 or f1>f2 . The stimuli were constant inputs with amplitudes proportional ( both positively and negatively ) to the frequency . For this task we set the learning rate to η=0 . 002; the network successfully learned to perform the task ( Figure 2C , left ) , and the individual units of the network , when sorted by the first stimulus ( base frequency ) f1 , exhibit highly heterogeneous activity ( Figure 2C , right ) characteristic of neurons recorded in the prefrontal cortex of monkeys performing the task ( Machens et al . , 2010 ) . Learning curves for this and additional networks trained for the task are shown in Figure 2—figure supplement 3 . Additional comparisons can be made between the model networks shown in Figure 2 and the neural activity observed in behaving animals , for example state-space analyses as in Mante et al . ( 2013 ) , Carnevale et al . ( 2015 ) , or Song et al . ( 2016 ) . Such comparisons reveal that , as found previously in studies such as Barak et al . 2013 ) , the model networks exhibit many , but not all , features present in electrophysiological recordings . Figure 2 and the following make clear , however , that RNNs trained with reward feedback alone can already reproduce the mixed selectivity characteristic of neural populations in higher cortical areas ( Rigotti et al . , 2010 , 2013 ) , thereby providing a valuable platform for future investigations of how such complex representations are learned . All of the tasks in the previous section have the property that the correct response on any single trial is a function only of the task condition , and , in particular , does not depend on the network’s state during the trial . In a postdecision wager task ( Kiani and Shadlen , 2009 ) , however , the optimal decision depends on the animal’s ( agent’s ) estimate of the probability that its decision is correct , i . e . , its confidence . As can be seen from the results , on a trial-by-trial basis this is not the same as simply determining the stimulus difficulty ( a combination of stimulus duration and coherence ) ; this makes it difficult to train with standard supervised learning , which requires a pre-determined target output for the network to reproduce; instead , we trained an RNN to perform the task by maximizing overall reward . This task extends the simple perceptual decision-making task ( Figure 1A ) by introducing a 'sure' option that is presented during a 1200–1800 ms delay period on a random half of the trials; selecting this option results in a reward that is 0 . 7 times the size of the reward obtained when correctly choosing L or R . As in the monkey experiment , the network receives no information indicating whether or not a given trial will contain a sure option until the middle of the delay period after stimulus offset , thus ensuring that the network makes a decision about the stimulus on all trials ( Figure 3A ) . For this task the input connection weights Win , Winλ , and Winγ were initialized so that half the units received information about the sure target while the other half received evidence for L and R . All units initially received fixation input . 10 . 7554/eLife . 21492 . 014Figure 3 . Perceptual decision-making task with postdecision wagering , based on Kiani and Shadlen ( 2009 ) . ( A ) Task structure . On a random half of the trials , a sure option is presented during the delay period , and on these trials the network has the option of receiving a smaller ( compared to correctly choosing L or R ) but certain reward by choosing the sure option ( S ) . The stimulus duration , delay , and sure target onset time are the same as in Kiani and Shadlen 2009 ) . ( B ) Probability of choosing the sure option ( left ) and probability correct ( right ) as a function of stimulus duration , for different coherences . Performance is higher for trials on which the sure option was offered but waived in favor of L or R ( filled circles , solid ) , compared to trials on which the sure option was not offered ( open circles , dashed ) . ( C ) Activity of an example decision network unit for non-wager ( left ) and wager ( right ) trials , sorted by whether the presented evidence was toward the unit’s preferred ( black ) or nonpreferred ( gray ) target as determined by activity during the stimulus period on all trials . Dashed lines show activity for trials in which the sure option was chosen . DOI: http://dx . doi . org/10 . 7554/eLife . 21492 . 01410 . 7554/eLife . 21492 . 015Figure 3—figure supplement 1 . Learning curves for the postdecision wager task . ( A ) Average reward per trial . Black indicates the network realization shown in the main text , gray additional realizations , i . e . , trained with different random number generator seeds . ( B ) Percent correct , for trials on which the network made a decision ( ≥99% required for termination ) . Red: target performance when the sure bet was accepted between 40–50% of the time . DOI: http://dx . doi . org/10 . 7554/eLife . 21492 . 015 The key behavioral features found in Kiani and Shadlen ( 2009 ) ; Wei and Wang ( 2015 ) are reproduced in the trained network , namely the network opted for the sure option more frequently when the coherence was low or stimulus duration short ( Figure 3B , left ) ; and when the network was presented with a sure option but waived it in favor of choosing L or R , the performance was better than on trials when the sure option was not presented ( Figure 3B , right ) . The latter observation is taken as indication that neither monkeys nor trained networks choose the sure target on the basis of stimulus difficulty alone but based on their internal sense of uncertainty on each trial . Figure 3C shows the activity of an example network unit , sorted by whether the decision was the unit’s preferred or nonpreferred target ( as determined by firing rates during the stimulus period on all trials ) , for both non-wager and wager trials . In particular , on trials in which the sure option was chosen , the firing rate is intermediate compared to trials on which the network made a decision by choosing L or R . Learning curves for this and additional networks trained for the task are shown in Figure 3—figure supplement 1 . We also trained networks to perform the simple economic choice task of Padoa-Schioppa and Assad ( 2006 ) and examined the activity of the value , rather than decision , network . The choice patterns of the networks were modulated only by varying the reward contingencies ( Figure 4A , upper and lower ) . We note that , on each trial there is a 'correct' answer in the sense that there is a choice which results in greater reward . In contrast to the previous tasks , however , information regarding whether an answer is correct in this sense is not contained in the inputs but rather in the association between inputs and rewards . This distinguishes the task from the cognitive tasks discussed in previous sections: although the task can be transformed into a cognitive-type task by providing the associated rewards as inputs , training in this manner conflates external with 'internal , ' learned inputs . 10 . 7554/eLife . 21492 . 016Figure 4 . Value-based economic choice task ( Padoa-Schioppa and Assad , 2006 ) . ( A ) Choice pattern when the reward contingencies are indifferent for roughly 1 'juice' of A and 2 'juices' of B ( upper ) or 1 juice of A and 4 juices of B ( lower ) . ( B ) Mean activity of example value network units during the pre-choice period , defined here as the period 500 ms before the decision , for the 1A = 2B case . Units in the value network exhibit diverse selectivity as observed in the monkey orbitofrontal cortex . For 'choice' ( last panel ) , trials were separated into choice A ( red diamonds ) and choice B ( blue circles ) . DOI: http://dx . doi . org/10 . 7554/eLife . 21492 . 01610 . 7554/eLife . 21492 . 017Figure 4—figure supplement 1 . Fit of cumulative Gaussian with parameters μ , σ to the choice pattern in Figure 4 ( upper ) , and the deduced indifference point nB*/nA*= ( 1+μ ) / ( 1-μ ) . DOI: http://dx . doi . org/10 . 7554/eLife . 21492 . 01710 . 7554/eLife . 21492 . 018Figure 4—figure supplement 2 . Fit of cumulative Gaussian with parameters μ , σ to the choice pattern in Figure 4A ( lower ) , and the deduced indifference point nB*/nA*= ( 1+μ ) / ( 1-μ ) . DOI: http://dx . doi . org/10 . 7554/eLife . 21492 . 01810 . 7554/eLife . 21492 . 019Figure 4—figure supplement 3 . Learning curves for the value-based economic choice task . ( A ) Average reward per trial . Black indicates the network realization shown in the main text , gray additional realizations , i . e . , trained with different random number generator seeds . ( B ) Percentage of trials on which the network chose the option that resulted in greater ( or equal ) reward , for trials where the network made a decision ( ≥99% required for termination ) . Note this is conceptually different from the previous tasks , where 'correct' depends on the sensory inputs , not the rewards . Red: target performance ( when training was terminated ) . DOI: http://dx . doi . org/10 . 7554/eLife . 21492 . 019 Each trial began with a 750 ms fixation period; the offer , which indicated the 'juice' type and amount for the left and right choices , was presented for 1000–2000 ms , followed by a 750 ms decision period during which the network was required to indicate its decision . In the upper panel of Figure 4A the indifference point was set to 1A = 2 . 2B during training , which resulted in 1A = 2 . 0B when fit to a cumulative Gaussian ( Figure 4—figure supplement 1 ) , while in the lower panel it was set to 1A = 4 . 1B during training and resulted in 1A = 4 . 0B ( Figure 4—figure supplement 2 ) . The basic unit of reward , i . e . , 1B , was 0 . 1 . For this task we increased the initial value of the value network’s input weights , Winv , by a factor of 10 to drive the value network more strongly . Strikingly , the activity of units in the value network vϕ exhibits similar types of tuning to task variables as observed in the orbitofrontal cortex of monkeys , with some units ( roughly 20% of active units ) selective to chosen value , others ( roughly 60% , for both A and B ) to offer value , and still others ( roughly 20% ) to choice alone as defined in Padoa-Schioppa and Assad ( 2006 ) ( Figure 4B ) . The decision network also contained units with a diversity of tuning . Learning curves for this and additional networks trained for the task are shown in Figure 4—figure supplement 3 . We emphasize that no changes were made to the network architecture for this value-based economic choice task . Instead , the same scheme shown in Figure 1B , in which the value network is responsible for predicting future rewards to guide learning but is not involved in the execution of the policy , gave rise to the pattern of neural activity shown in Figure 4B .
In this work we have demonstrated reward-based training of recurrent neural networks for both cognitive and value-based tasks . Our main contributions are twofold: first , our work expands the range of tasks and corresponding neural mechanisms that can be studied by analyzing model recurrent neural networks , providing a unified setting in which to study diverse computations and compare to electrophysiological recordings from behaving animals; second , by explicitly incorporating reward into network training , our work makes it possible in the future to more directly address the question of reward-related processes in the brain , for instance the role of value representation that is essential for learning , but not executing , a task . To our knowledge , the specific form of the baseline network inputs used in this work has not been used previously in the context of recurrent policy gradients; it combines ideas from Wierstra et al . ( 2009 ) where the baseline network received the same inputs as the decision network in addition to the selected actions , and Ranzato et al . ( 2016 ) , where the baseline was implemented as a simple linear regressor of the activity of the decision network , so that the decision and value networks effectively shared the same recurrent units . Indeed , the latter architecture is quite common in machine learning applications ( Mnih et al . , 2016 ) , and likewise , for some of the simpler tasks considered here , models with a baseline consisting of a linear readout of the selected actions and decision network activity could be trained in comparable ( but slightly longer ) time ( Figure 1—figure supplement 4 ) . The question of whether the decision and value networks ought to share the same recurrent network parallels ongoing debate over whether choice and confidence are computed together or if certain areas such as OFC compute confidence signals locally , though it is clear that such 'meta-cognitive' representations can be found widely in the brain ( Lak et al . , 2014 ) . Computationally , the distinction is expected to be important when there are nonlinear computations required to determine expected return that are not needed to implement the policy , as illustrated in the perceptual decision-making task ( Figure 1 ) . Interestingly , a separate value network to represent the baseline suggests an explicit role for value representation in the brain that is essential for learning a task ( equivalently , when the environment is changing ) , but not for executing an already learned task , as is sometimes found in experiments ( Turner and Desmurget , 2010; Schoenbaum et al . , 2011; Stalnaker et al . , 2015 ) . Since an accurate baseline dramatically improves learning but is not required—the algorithm is less reliable and takes many samples to converge with a constant baseline , for instance—this baseline network hypothesis for the role of value representation may account for some of the subtle yet broad learning deficits observed in OFC-lesioned animals ( Wallis , 2007 ) . Moreover , since expected reward is closely related to decision confidence in many of the tasks considered , a value network that nonlinearly reads out confidence information from the decision network is consistent with experimental findings in which OFC inactivation affects the ability to report confidence but not decision accuracy ( Lak et al . , 2014 ) . Our results thus support the actor-critic picture for reward-based learning , in which one circuit directly computes the policy to be followed , while a second structure , receiving projections from the decision network as well as information about the selected actions , computes expected future reward to guide learning . Actor-critic models have a rich history in neuroscience , particularly in studies of the basal ganglia ( Houk et al . , 1995; Dayan and Balleine , 2002; Joel et al . , 2002; O'Doherty et al . , 2004; Takahashi et al . , 2008; Maia , 2010 ) , and it is interesting to note that there is some experimental evidence that signals in the striatum are more suitable for direct policy search rather than for updating action values as an intermediate step , as would be the case for purely value function-based approaches to computing the decision policy ( Li and Daw , 2011; Niv and Langdon , 2016 ) . Moreover , although we have used a single RNN each to represent the decision and value modules , using 'deep , ' multilayer RNNs may increase the representational power of each module ( Pascanu et al . , 2013a ) . For instance , more complex tasks than considered in this work may require hierarchical feature representation in the decision network , and likewise value networks can use a combination of the different features [including raw sensory inputs ( Wierstra et al . , 2009 ) ] to predict future reward . Anatomically , the decision networks may correspond to circuits in dorsolateral prefrontal cortex , while the value networks may correspond to circuits in OFC ( Schultz et al . , 2000; Takahashi et al . , 2011 ) or basal ganglia ( Hikosaka et al . , 2014 ) . This architecture also provides a useful example of the hypothesis that various areas of the brain effectively optimize different cost functions ( Marblestone et al . , 2016 ) : in this case , the decision network maximizes reward , while the value network minimizes the prediction error for future reward . As in many other supervised learning approaches used previously to train RNNs ( Mante et al . , 2013; Song et al . , 2016 ) , the use of BPTT to compute the gradients ( in particular , the eligibility ) make our 'plasticity rule' not biologically plausible . As noted previously ( Zipser and Andersen , 1988 ) , it is indeed somewhat surprising that the activity of the resulting networks nevertheless exhibit many features found in neural activity recorded from behaving animals . Thus our focus has been on learning from realistic feedback signals provided by the environment but not on its physiological implementation . Still , recent work suggests that exact backpropagation is not necessary and can even be implemented in 'spiking' stochastic units ( Lillicrap et al . , 2016 ) , and that approximate forms of backpropagation and SGD can be implemented in a biologically plausible manner ( Scellier and Bengio , 2016 ) , including both spatially and temporally asynchronous updates in RNNs ( Jaderberg et al . , 2016 ) . Such ideas require further investigation and may lead to effective yet more neurally plausible methods for training model neural networks . Recently , Miconi ( 2016 ) used a 'node perturbation'-based ( Fiete and Seung , 2006; Fiete et al . , 2007; Hoerzer et al . , 2014 ) algorithm with an error signal at the end of each trial to train RNNs for several cognitive tasks , and indeed , node perturbation is closely related to the REINFORCE algorithm used in this work . On one hand , the method described in Miconi ( 2016 ) is more biologically plausible in the sense of not requiring gradients computed via backpropagation through time as in our approach; on the other hand , in contrast to the networks in this work , those in Miconi ( 2016 ) did not 'commit' to a discrete action and thus the error signal was a graded quantity . In this and other works ( Frémaux et al . , 2010 ) , moreover , the prediction error was computed by algorithmically keeping track of a stimulus ( task condition ) -specific running average of rewards . Here we used a concrete scheme ( namely a value network ) for approximating the average that automatically depends on the stimulus , without requiring an external learning system to maintain a separate record for each ( true ) trial type , which is not known by the agent with certainty . One of the advantages of the REINFORCE algorithm for policy gradient reinforcement learning is that direct supervised learning can also be mixed with reward-based learning , by including only the eligibility term in Equation 3 without modulating by reward ( Mnih et al . , 2014 ) , i . e . , by maximizing the log-likelihood of the desired actions . Although all of the networks in this work were trained from reward feedback only , it will be interesting to investigate this feature of the REINFORCE algorithm . Another advantage , which we have not exploited here , is the possibility of learning policies for continuous action spaces ( Peters and Schaal , 2008; Wierstra et al . , 2009 ) ; this would allow us , for example , to model arbitrary saccade targets in the perceptual decision-making task , rather than limiting the network to discrete choices . We have previously emphasized the importance of incorporating biological constraints in the training of neural networks ( Song et al . , 2016 ) . For instance , neurons in the mammalian cortex have purely excitatory or inhibitory effects on other neurons , which is a consequence of Dale’s Principle for neurotransmitters ( Eccles et al . , 1954 ) . In this work we did not include such constraints due to the more complex nature of our rectified GRUs ( Equations 9–12 ) ; in particular , the units we used are capable of dynamically modulating their time constants and gating their recurrent inputs , and we therefore interpreted the firing rate units as a mixture of both excitatory and inhibitory populations . Indeed , these may implement the 'reservoir of time constants' observed experimentally ( Bernacchia et al . , 2011 ) . In the future , however , comparison to both model spiking networks and electrophysiological recordings will be facilitated by including more biological realism , by explicitly separating the roles of excitatory and inhibitory units ( Mastrogiuseppe and Ostojic , 2016 ) . Moreover , since both the decision and value networks are obtained by minimizing an objective function , additional regularization terms can be easily included to obtain networks whose activity is more similar to neural recordings ( Sussillo et al . , 2015; Song et al . , 2016 ) . Finally , one of the most appealing features of RNNs trained to perform many tasks is their ability to provide insights into neural computation in the brain . However , methods for revealing neural mechanisms in such networks remain limited to state-space analysis ( Sussillo and Barak , 2013 ) , which in particular does not reveal how the synaptic connectivity leads to the dynamics responsible for implementing the higher-level decision policy . General and systematic methods for analyzing trained networks are still needed and are the subject of ongoing investigation . Nevertheless , reward-based training of RNNs makes it more likely that the resulting networks will correspond closely to biological networks observed in experiments with behaving animals . We expect that the continuing development of tools for training model neural networks in neuroscience will thus contribute novel insights into the neural basis of animal cognition .
Here we review the application of the REINFORCE algorithm for policy gradient reinforcement learning to recurrent neural networks ( Williams , 1992; Baird and Moore , 1999; Sutton et al . , 2000; Baxter and Bartlett , 2001; Peters and Schaal , 2008; Wierstra et al . , 2009 ) . In particular , we provide a careful derivation of Equation 3 following , in part , the exposition in Zaremba and Sutskever ( 2016 ) . Let Hμ:t be the sequence of interactions between the environment and agent ( i . e . , the environmental states , observables , and agent actions ) that results in the environment being in state 𝐬t+1 at time t+1 starting from state 𝐬μ at time μ: ( 14 ) Hμ:t= ( 𝐬μ+1:t+1 , 𝐮μ:t , 𝐚μ:t ) . For notational convenience in the following , we adopt the convention that , for the special case of μ=0 , the history H0:t includes the initial state 𝐬0 and excludes the meaningless inputs 𝐮0 , which are not seen by the agent: ( 15 ) H0:t= ( 𝐬0:t+1 , 𝐮1:t , 𝐚0:t ) . When t=0 , it is also understood that 𝐮1:0=∅ , the empty set . A full history , or a trial , is thus denoted as ( 16 ) H≡H0:T= ( 𝐬0:T+1 , 𝐮1:T , 𝐚0:T ) , where T is the end of the trial . Here we only consider the episodic , 'finite-horizon' case where T is finite , and since different trials can have different durations , we take T to be the maximum length of a trial in the task . The reward ρt+1 at time t+1 following actions 𝐚t ( we use ρ to distinguish it from the firing rates 𝐫 of the RNNs ) is determined by this history , which we sometimes indicate explicitly by writing ( 17 ) ρt+1=ρt+1 ( H0:t ) . As noted in the main text , we adopt the convention that the agent performs actions at t=0 , … , T and receives rewards at t=1 , … , T+1 to emphasize that rewards follow the actions and are jointly determined with the next state ( Sutton and Barto , 1998 ) . For notational simplicity , here and elsewhere we assume that any discount factor is already included in ρt+1 , i . e . , in all places where the reward appears we consider ρt+1→e-t/τrewardρt+1 , where τreward is the time constant for discounting future rewards ( Doya , 2000 ) ; we included temporal discounting only for the reaction-time version of the simple perceptual decision-making task ( Figure 1—figure supplement 2 ) , where we set τreward=10s . For the remaining tasks , τreward=∞ . Explicitly , a trial H0:T comprises the following . At time t=0 , the environment is in state 𝐬0 with probability ℰ ( 𝐬0 ) . The agent initially chooses a set of actions 𝐚0 with probability πθ ( 𝐚0 ) , which is determined by the parameters of the decision network , in particular the initial conditions 𝐱0 and readout weights Woutπ and biases 𝐛outπ ( Equation 6 ) . At time t=1 , the environment , depending on its previous state 𝐬0 and the agent’s actions 𝐚0 , transitions to state 𝐬1 with probability ℰ ( 𝐬1|𝐬0 , 𝐚0 ) . The history up to this point is H0:0= ( 𝐬0:1 , ∅ , 𝐚0:0 ) , where ∅ indicates that no inputs have yet been seen by the network . The environment also generates reward ρ1 , which depends on this history , ρ1=ρ1 ( H0:0 ) . From state 𝐬1 the environment generates observables ( inputs to the agent ) 𝐮1 with a distribution given by ℰ ( 𝐮1|𝐬1 ) . In response , the agent , depending on the inputs 𝐮1 it receives from the environment , chooses the set of actions 𝐚1 according to the distribution πθ ( 𝐚1|𝐮1:1 ) =πθ ( 𝐚1|𝐮1 ) . The environment , depending on its previous states 𝐬0:1 and the agent’s previous actions 𝐚0:1 , then transitions to state 𝐬2 with probability ℰ ( 𝐬2|𝐬0:1 , 𝐚0:1 ) . Thus H0:1= ( 𝐬0:2 , 𝐮1:1 , 𝐚0:1 ) . Iterating these steps , the history at time t is therefore given by Equation 15 , while a full history is given by Equation 16 . The probability pθ ( H0:τ ) of a particular sub-history H0:τ up to time τ occurring , under the policy πθ parametrized by θ , is given by ( 18 ) pθ ( H0:τ ) =[∏t=1τℰ ( 𝐬t+1|𝐬0:t , 𝐚0:t ) πθ ( 𝐚t|𝐮1:t ) ℰ ( 𝐮t|𝐬t ) ]ℰ ( 𝐬1|𝐬0 , 𝐚0 ) πθ ( 𝐚0 ) ℰ ( 𝐬0 ) . In particular , the probability pθ ( H ) of a history H=H0:T occurring is ( 19 ) pθ ( H ) =[∏t=1Tℰ ( 𝐬t+1|𝐬0:t , 𝐚0:t ) πθ ( 𝐚t|𝐮1:t ) ℰ ( 𝐮t|𝐬t ) ]ℰ ( 𝐬1|𝐬0 , 𝐚0 ) πθ ( 𝐚0 ) ℰ ( 𝐬0 ) . A key ingredient of the REINFORCE algorithm is that the policy parameters only indirectly affect the environment through the agent’s actions . The logarithmic derivatives of Equation 18 with respect to the parameters θ therefore do not depend on the unknown ( to the agent ) environmental dynamics contained in ℰ , i . e . , ( 20 ) ∇θlogpθ ( H0:τ ) =∑t=0τ∇θlogπθ ( at|u1:t ) , with the understanding that 𝐮1:0=∅ ( the empty set ) and therefore πθ ( 𝐚0|𝐮1:0 ) =πθ ( 𝐚0 ) . The goal of the agent is to maximize the expected return at time t=0 ( Equation 1 , reproduced here ) ( 21 ) J ( θ ) =𝔼H[∑τ=0Tρτ+1 ( H0:τ ) ] , where we have used the time index τ for notational consistency with the following and made the history-dependence of the rewards explicit . In terms of the probability of each history H occurring , Equation 19 , we have ( 22 ) J ( θ ) =∑Hpθ ( H ) [∑τ=0Tρτ+1 ( H0:τ ) ] , where the generic sum over H may include both sums over discrete variables and integrals over continuous variables . Since , for any τ=0 , … , T , ( 23 ) pθ ( H ) =pθ ( H0:T ) =pθ ( Hτ+1:T|H0:τ ) pθ ( H0:τ ) ( cf . Equation 18 ) , we can simplify Equation 22 to ( 24 ) J ( θ ) =∑τ=0T∑Hpθ ( H ) ρτ+1 ( H0:τ ) ( 25 ) =∑τ=0T∑H0:τpθ ( H0:τ ) ρτ+1 ( H0:τ ) ∑Hτ+1:Tpθ ( Hτ+1:T|H0:τ ) ( 26 ) =∑τ=0T∑H0:τpθ ( H0:τ ) ρτ+1 ( H0:τ ) . This simplification is used below to formalize the intuition that present actions do not influence past rewards . Using the 'likelihood-ratio trick' ( 27 ) ∇θf ( θ ) =f ( θ ) ∇θf ( θ ) f ( θ ) =f ( θ ) ∇θlogf ( θ ) , we can write ( 28 ) ∇θJ ( θ ) =∑τ=0T∑H0:τ[∇θpθ ( H0:τ ) ]ρτ+1 ( H0:τ ) ( 29 ) =∑τ=0T∑H0:τpθ ( H0:τ ) [∇θlogpθ ( H0:τ ) ]ρτ+1 ( H0:τ ) . From Equation 20 we therefore have ( 30 ) ∇θJ ( θ ) =∑τ=0T∑H0:τpθ ( H0:τ ) [∑t=0τ∇θlogπθ ( at|u1:t ) ]ρτ+1 ( H0:τ ) ( 31 ) =∑Hpθ ( H ) ∑τ=0T∑t=0τ[∇θlogπθ ( at|u1:t ) ]ρτ+1 ( H0:τ ) ( 32 ) =∑Hpθ ( H ) ∑t=0T∑τ=tT[∇θlogπθ ( at|u1:t ) ]ρτ+1 ( H0:τ ) ( 33 ) =∑Hpθ ( H ) ∑t=0T∇θlogπθ ( at|u1:t ) [∑τ=tTρτ+1 ( H0:τ ) ] , where we have 'undone' Equation 23 to recover the sum over the full histories H in going from Equation 30 to Equation 31 . We then obtain the first terms of Equation 2 and Equation 3 by estimating the sum over all H by Ntrials samples from the agent’s experience . In Equation 22 it is evident that , while subtracting any constant b from the reward J ( θ ) will not affect the gradient with respect to θ , it can reduce the variance of the stochastic estimate ( Equation 33 ) from a finite number of trials . Indeed , it is possible to use this invariance to find an 'optimal' value of the constant baseline that minimizes the variance of the gradient estimate ( Peters and Schaal , 2008 ) . In practice , however , it is more useful to have a history-dependent baseline that attempts to predict the future return at every time ( Wierstra et al . , 2009; Mnih et al . , 2014; Zaremba and Sutskever , 2016 ) . We therefore introduce a second network , called the value network , that uses the selected actions 𝐚1:t and the activity of the decision network 𝐫1:tπ to predict the future return ∑τ=tTρτ+1 by minimizing the squared error ( Equations 4–5 ) . Intuitively , such a baseline is appealing because the terms in the gradient of Equation 3 are nonzero only if the actual return deviates from what was predicted by the value network . Carrying out the discretization of Equations 9–12 in time steps of Δt , we obtain ( 34 ) λt=sigmoid ( Wrecλrt−1+Winλut+bλ ) , ( 35 ) γt=sigmoid ( Wrecγrt−1+Winγut+bγ ) , ( 36 ) xt= ( 1−αλt ) ⊙xt−1+αλt⊙[Wrec ( γt⊙rt−1 ) +Winut+b+2α−1σrec2N ( 0 , 1 ) ] , ( 37 ) rt=[xt]+ for t=1 , … , T , where α=Δt/τ and 𝐍 ( 0 , 1 ) are normally distributed random numbers with zero mean and unit variance . We note that the rectified-linear activation function appears in different positions compared to standard GRUs , which merely reflects the choice of using 'synaptic currents' as the dynamical variable rather than directly using firing rates as the dynamical variable . One small advantage of this choice is that we can train the unconstrained initial conditions 𝐱0 rather than the non-negatively constrained firing rates 𝐫0 . The biases 𝐛λ , 𝐛γ , and 𝐛 , as well as the readout weights Woutπ and Woutv , were initialized to zero . The biases for the policy readout 𝐛outπ were initially set to zero , while the value network bias boutv was initially set to the 'reward' for an aborted trial , −1 . The entries of the input weight matrices Winγ , Winλ , and Win for both decision and value networks were drawn from a zero-mean Gaussian distribution with variance K/Nin2 . For the recurrent weight matrices Wrec , Wrecλ , and Wrecγ , the K nonzero entries in each row were initialized from a gamma distribution Γ ( α , β ) with α=β=4 , with each entry multiplied randomly by ±1; the entire matrix was then scaled such that the spectral radius—the largest absolute value of the eigenvalues—was exactly ρ0 . Although we also successfully trained networks starting from normally distributed weights , we found it convenient to control the sign and magnitude of the weights independently . The initial conditions 𝐱0 , which are also trained , were set to 0 . 5 for all units before the start of training . We implemented the networks in the Python machine learning library Theano ( The Theano Development Team , 2016 ) . We used a recently developed version of stochastic gradient descent known as Adam , for adaptive moment estimation ( Kingma and Ba , 2015 ) , together with gradient clipping to prevent exploding gradients ( Graves , 2013; Pascanu et al . , 2013b ) . For clarity , in this section we use vector notation 𝜽 to indicate the set of all parameters being optimized and the subscript k to indicate a specific parameter θk . At each iteration i>0 , let ( 38 ) 𝐠 ( i ) =∂ℒ∂𝜽|𝜽=𝜽 ( i-1 ) be the gradient of the objective function ℒ with respect to the parameters 𝜽 . We first clip the gradient if its norm |𝐠 ( i ) | exceeds a maximum Γ ( see Table 1 ) , i . e . , ( 39 ) g^ ( i ) =g ( i ) ×min ( 1 , Γ|g ( i ) | ) . Each parameter θk is then updated according to ( 40 ) θk ( i ) =θk ( i-1 ) -η1-β2i1-β1imk ( i ) vk ( i ) +ε , where η is the base learning rate and the moving averages ( 41 ) m ( i ) =β1m ( i−1 ) + ( 1−β1 ) g^ ( i ) , ( 42 ) v ( i ) =β2v ( i−1 ) + ( 1−β2 ) [g^ ( i ) ]2 estimate the first and second ( uncentered ) moments of the gradient . Initially , 𝐦 ( 0 ) =𝐯 ( 0 ) =0 . These moments allow each parameter to be updated in Equation 40 according to adaptive learning rates , such that parameters whose gradients exhibit high uncertainty and hence small 'signal-to-noise ratio' lead to smaller learning rates . Except for the base learning rate η ( see Table 1 ) , we used the parameter values suggested in Kingma and Ba ( 2015 ) :β1=0 . 9 , β2=0 . 999 , ε=10−8 . All code used in this work , including code for generating the figures , is available at http://github . com/xjwanglab/pyrl . | A major goal in neuroscience is to understand the relationship between an animal’s behavior and how this is encoded in the brain . Therefore , a typical experiment involves training an animal to perform a task and recording the activity of its neurons – brain cells – while the animal carries out the task . To complement these experimental results , researchers “train” artificial neural networks – simplified mathematical models of the brain that consist of simple neuron-like units – to simulate the same tasks on a computer . Unlike real brains , artificial neural networks provide complete access to the “neural circuits” responsible for a behavior , offering a way to study and manipulate the behavior in the circuit . One open issue about this approach has been the way in which the artificial networks are trained . In a process known as reinforcement learning , animals learn from rewards ( such as juice ) that they receive when they choose actions that lead to the successful completion of a task . By contrast , the artificial networks are explicitly told the correct action . In addition to differing from how animals learn , this limits the types of behavior that can be studied using artificial neural networks . Recent advances in the field of machine learning that combine reinforcement learning with artificial neural networks have now allowed Song et al . to train artificial networks to perform tasks in a way that mimics the way that animals learn . The networks consisted of two parts: a “decision network” that uses sensory information to select actions that lead to the greatest reward , and a “value network” that predicts how rewarding an action will be . Song et al . found that the resulting artificial “brain activity” closely resembled the activity found in the brains of animals , confirming that this method of training artificial neural networks may be a useful tool for neuroscientists who study the relationship between brains and behavior . The training method explored by Song et al . represents only one step forward in developing artificial neural networks that resemble the real brain . In particular , neural networks modify connections between units in a vastly different way to the methods used by biological brains to alter the connections between neurons . Future work will be needed to bridge this gap . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"neuroscience"
] | 2017 | Reward-based training of recurrent neural networks for cognitive and value-based tasks |
During sleep , the thalamus generates a characteristic pattern of transient , 11-15 Hz sleep spindle oscillations , which synchronize the cortex through large-scale thalamocortical loops . Spindles have been increasingly demonstrated to be critical for sleep-dependent consolidation of memory , but the specific neural mechanism for this process remains unclear . We show here that cortical spindles are spatiotemporally organized into circular wave-like patterns , organizing neuronal activity over tens of milliseconds , within the timescale for storing memories in large-scale networks across the cortex via spike-time dependent plasticity . These circular patterns repeat over hours of sleep with millisecond temporal precision , allowing reinforcement of the activity patterns through hundreds of reverberations . These results provide a novel mechanistic account for how global sleep oscillations and synaptic plasticity could strengthen networks distributed across the cortex to store coherent and integrated memories .
Memories are stored in distributed networks across the cortex . In the two-stage model of memory consolidation ( McClelland et al . , 1995; Rasch and Born , 2007 ) , memories are integrated in the hippocampus and then linked in the neocortex for long-term storage , where information represented in visual , auditory , somatosensory , or cognitive regions must be bound into a coherent whole ( Wheeler et al . , 2000; Horner et al . , 2015 ) . It is well established that sleep oscillations actively contribute to this process: during stage 2 sleep spindles , the thalamus generates a rhythmic activity pattern that becomes widespread through large-scale thalamocortical loops ( Contreras et al . , 1996 ) , and spindles are critical to sleep-dependent memory consolidation ( Gais et al . , 2002; Mednick et al . , 2013 ) . Long-range connections in cortex result primarily from excitatory pyramidal cells ( Sholl , 1956; Schüz et al . , 2002 ) , but precisely how sleep oscillations aid strengthening of these excitatory connections between distributed cortical networks through spike-time dependent plasticity ( STDP ) remains unclear , particularly in the presence of long axonal conduction delays ( Lubenov and Siapas , 2008 ) . Here , we identify a global activity pattern repeatedly observed during sleep spindle oscillations in human neocortex that could serve this role . We study intracranial electrocorticogram ( ECoG ) recordings of five clinical patients in stage 2 sleep and apply recently developed computational methods ( Muller et al . , 2014 ) to classify spatiotemporal dynamics at the level of individual oscillation cycles . ECoG arrays were implanted in subjects undergoing evaluation for resective surgery of epileptogenic cortex ( Figure 1A , left ) . Over several days of recording , these subjects exhibit long periods without major epileptic events . During that time , subjects express a relatively normal sleep architecture , with well-defined sleep oscillations . Stage 2 sleep epochs were then manually identified by an expert rater , and sleep spindles recorded on the ECoG were isolated using automated techniques ( Hagler et al . , 2016 ) . These spindles appeared physiologically normal and well-isolated from background noise ( Figure 1A , right and Figure 1—figure supplement 1 ) . Our algorithmic approach classifies spatiotemporal patterns as expanding waves , defined as a significant linear increase in phase offset with distance from a point source ( Figure 1—figure supplement 2; see Materials and methods – Spatiotemporal dynamics ) , or rotating waves , defined as a significant increase in phase offset with rotation about a wave center ( Figure 1—figure supplement 3 ) . In 41 , 860 spindle oscillation cycles tested across subjects , a large proportion ( 50 . 8% ) was classified as rotating waves , along with a smaller subset ( 15 . 6% ) as expanding . After inspecting these results , we observed further that the rotating waves exhibited a clear bias towards travel in the temporal → parietal → frontal ( TPF ) direction ( 69 . 5% , p<10−10 , one-tail binomial test against equal occurrence , 14 , 796 TPF cycles , 21 , 272 total; for each individual subject p<10−3 , see Figure 1—source data 1 for individual wave totals ) ( Figure 1B and Video 1 ) . Propagation speed distributions peaked between 2–5 m/s ( Figure 1C ) , varying within a narrow range from the 20th to the 80th percentiles ( 3–9 m/s for the full distribution; 4–10 , 3–8 , 3–10 , 2–3 , and 4–13 m/s for individual subjects , respectively ) , within the range of conduction speeds for the short ( Girard et al . , 2001 ) and long ( Schüz et al . , 2002; Swadlow and Waxman , 2012 ) white matter association fibers . Further , this rotating TPF organization occurred consistently across subjects and implantation hemispheres ( Figure 1D; see also Figure 1—figure supplements 5–7 ) . 10 . 7554/eLife . 17267 . 003Figure 1 . Rotating waves during spindles . ( A ) Electrode placement for subject 1 ( left ) , with a stereotypical spindling epoch observed on the array ( right ) . The right panel depicts the average over channels ( black ) together with the individual channels ( gray ) . ( B ) When visualized on the cortex , individual spindle cycles are often organized as rotating waves traveling from temporal ( +0 ms , top ) to parietal ( +20 ms , middle ) to frontal ( +40 ms , bottom ) lobes . ( C ) Phase speed distributions across subjects . Plotted is the kernel smoothing density estimate for individual subjects ( gray dotted lines ) and for the full distribution ( black line ) . ( D ) The field of propagation directions , aligned on the putative rotation center and averaged across oscillation cycles and across subjects , shows a consistent flow in the temporal → parietal → frontal ( TPF ) direction . The center point is marked in red . DOI: http://dx . doi . org/10 . 7554/eLife . 17267 . 00310 . 7554/eLife . 17267 . 004Figure 1—source data 1 . Patient information and wave classification totals . This CSV file provides patient age , sex , and the number of spindle oscillation cycles isolated in each class ( total , expanding , rotating , and rotating TPF ) . DOI: http://dx . doi . org/10 . 7554/eLife . 17267 . 00410 . 7554/eLife . 17267 . 005Figure 1—figure supplement 1 . Power spectral density and spatial correlation analysis . ( A ) Average power spectral density estimate for spindle epochs ( black ) and matched non-spindle epochs ( red ) , illustrating the nearly 10 dB increase within the 11–15 Hz spindle band during algorithmically selected epochs ( divisive normalization , inset ) . Power at line noise frequency omitted for clarity . ( B ) Average spatial correlation values for spindle ( black ) and non-spindle ( red ) epochs illustrate increased global correlations during spindles . DOI: http://dx . doi . org/10 . 7554/eLife . 17267 . 00510 . 7554/eLife . 17267 . 006Figure 1—figure supplement 2 . Method for isolating expanding spatiotemporal patterns . The algorithmic approach to detecting expanding waves in noisy multichannel data proceeds in three steps: ( 1 ) phase extraction , through the analytic signal ( Feldman , 2011 ) representation , followed by ( 2 ) center localization by taking the divergence of the phase gradient , and ( 3 ) quantification of the spatiotemporal pattern by calculating the circular-linear correlation of phase with distance from the center point . By finding an anchor point ( in step 2 ) and then quantifying the spatial pattern of activity through the appropriate correlation metric ( in step 3 ) , we can capture the spatiotemporal pattern of expanding activity on the array in a single number ( ρϕ , δ ) . DOI: http://dx . doi . org/10 . 7554/eLife . 17267 . 00610 . 7554/eLife . 17267 . 007Figure 1—figure supplement 3 . Method for isolating rotating spatiotemporal patterns . The approach to detecting rotating waves in noisy multichannel data proceeds similarly to that for expanding waves: ( 1 ) phase extraction , followed by ( 2 ) center localization by taking the curl of the phase gradient , and ( 3 ) quantification of the spatiotemporal pattern by calculating the circular-circular correlation of phase with rotation about the center point . By finding an anchor point ( in step 2 ) and then quantifying the spatial pattern of activity through an appropriate correlation metric ( in step 3 ) , we can capture the spatiotemporal pattern of rotating activity on the array in a single number ( ρϕ , θ ) . DOI: http://dx . doi . org/10 . 7554/eLife . 17267 . 00710 . 7554/eLife . 17267 . 008Figure 1—figure supplement 4 . Analysis of 1528 spindles in Subject 1 . ( A ) Cumulative distribution function for expanding ( blue ) and rotational ( red ) correlation coefficients . Dotted lines indicate the 99th percentile from a random shuffling permutation test . Rotational waves exceeding the determined threshold exhibit a marked preference for the temporal → parietal → frontal ( TPF ) direction ( inset ) . ( B ) Phase speed distributions were similar for both expanding and rotational waves . Plotted is the kernel smoothing density estimate with the constraint of positive support . DOI: http://dx . doi . org/10 . 7554/eLife . 17267 . 00810 . 7554/eLife . 17267 . 009Figure 1—figure supplement 5 . Vector field averaging controls . ( A ) Average re-centered phase gradient directions over identified rotational cycles traveling in the TPF direction , illustrating the robustness of the multichannel detection algorithm . Center point marked in red . ( B ) Average re-centered phase gradient directions over expanding waves . DOI: http://dx . doi . org/10 . 7554/eLife . 17267 . 00910 . 7554/eLife . 17267 . 010Figure 1—figure supplement 6 . Vector field distribution control . Distributions of phase gradient directions ( gray lines ) and circular mean direction of each distribution ( black lines ) are plotted relative to the rotation center ( red dot ) for all TPF wave cycles across subjects . The sharp distributions of phase gradient directions extend out to the edge of the electrode array , illustrating the global extent of the detected rotating waves . DOI: http://dx . doi . org/10 . 7554/eLife . 17267 . 01010 . 7554/eLife . 17267 . 011Figure 1—figure supplement 7 . Vector field averages for each subject . Average re-centered phase gradient directions over all spindle cycles in subjects 1–5 . In each case , a rotational vector field indicates the robust spatiotemporal organization of rotating waves in a preferred direction ( temporal → parietal → frontal cortex ) , in both right and left hemisphere electrode implantations ( average ) . The two right columns for each subject depict average re-centered phase gradient directions for identified rotating waves in the TPF and TFP direction , respectively . DOI: http://dx . doi . org/10 . 7554/eLife . 17267 . 01110 . 7554/eLife . 17267 . 012Figure 1—figure supplement 8 . Distribution of rotation center . ( A ) The distribution of rotation center is plotted on the cortical surface of Subject 1 , for 10 , 944 rotating TPF waves . The distribution is concentrated around the dorsal temporal lobe , and centered on the Sylvian fissure . For this panel , a Gaussian smoothing with a standard deviation of 1 pixel has been applied . ( B ) The distribution of change in wave center for consecutive rotating TPF wave cycles in Subjects 1–5 illustrates that the rotating wave center changes slowly , if at all ( cf . Video 1 ) . ( C–F ) The distribution of rotation center for Subjects 2–5 , plotted as in panel ( A ) , illustrates approximately consistent anatomical localization of the TPF rotation center across subjects . DOI: http://dx . doi . org/10 . 7554/eLife . 17267 . 01210 . 7554/eLife . 17267 . 013Figure 1—figure supplement 9 . Summary statistics across subjects . The proportion of expanding and rotational waves is given for subjects 1–5 ( left ) , in addition to cycles unclassified by the detection algorithm . In each case , a rotational preference for the temporal → parietal → frontal ( TPF ) direction is present at the level of individual spindle oscillation cycles ( right ) . DOI: http://dx . doi . org/10 . 7554/eLife . 17267 . 01310 . 7554/eLife . 17267 . 014Figure 1—figure supplement 10 . Consistent , coherent phase flow during spindles occurs uniquely in the 9–18 Hz frequency band . Average re-centered phase gradient directions over all spindle cycles in three frequency bands ( 1–4 Hz , left; 9–18 Hz , center; 40–80 Hz , right ) for subjects 1 ( top row ) and 2 ( bottom row ) . In the 40–80 Hz frequency band , notch filtering in time was applied to remove the 60 Hz line noise . A clear pattern in the average directions emerges for the spindle frequency band , while the phase gradient direction maps in the other frequency bands are less clear . DOI: http://dx . doi . org/10 . 7554/eLife . 17267 . 01410 . 7554/eLife . 17267 . 015Figure 1—figure supplement 11 . Robustness to noise and center position . Using simulated expanding and rotating waves , we verified the robustness of our detection approach under Gaussian white noise of varying amplitude . ( A ) Snapshots of simulated expanding waves in one oscillation cycle ( at 85 , 105 , and 125 ms ) , under varying levels of noise ( standard deviation 0 , 0 . 1 , and 0 . 2 ) . ( B ) Simulated rotating waves , as in ( A ) . ( C ) The performance of the algorithm in detecting expanding and rotational waves ( blue and red solid lines , mean ± SEM ) , as compared to the thresholds derived from the permutation test ( dotted lines ) . Every point represents 25 simulated spindles with 14 oscillation cycles each . Even when the noise standard deviation is unity ( at parity with oscillation amplitude , highly obscuring visual detection ) , the algorithm recovers a correlation magnitude around 0 . 8 in each case , far above the determined permutation thresholds . In contrast , with only Gaussian white noise as input , the algorithm returns a low correlation value for rotational detection , below the determined permutation threshold ( gray line ) . ( D ) Robustness of the algorithm to center position . This panel illustrates the performance of the algorithm for expanding ( blue ) and rotating ( red ) waves at different points on the electrode array , under Gaussian white noise ( standard deviation 0 . 2 ) . Note that performance drops for rotational detection at the border of the array , but this is small in comparison to the threshold determined from the permutation control ( dotted red line ) . DOI: http://dx . doi . org/10 . 7554/eLife . 17267 . 01510 . 7554/eLife . 17267 . 016Figure 1—figure supplement 12 . Local versus global simulated rotating waves . To test whether highly local spindles organized as traveling waves would be detected , we ran our algorithm on simulated global ( 64 electrodes ) and local ( 6 electrodes , or 9 . 4% of the simulated array ) rotating waves . ( A ) Each channel in the simulated waves had added noise similar to that observed in the data ( top trace ) . Distal channels in the local waves had only noise ( bottom trace ) . Note 100 ms scale bar in gray . ( B ) The correlation magnitude for the global wave ( black line , mean + SEM ) is high above the determined permutation threshold ( dotted red line ) , while the correlation magnitude for the highly local wave ( gray line ) is well below . This control illustrates that finding high ρϕ , θ is indicative of a distributed and global phase pattern on the electrode array . DOI: http://dx . doi . org/10 . 7554/eLife . 17267 . 01610 . 7554/eLife . 17267 . 017Video 1 . Rotating waves over five spindle oscillation cycles . Normalized activity for bandpass filtered timeseries is plotted in falsecolor at electrode positions on the cortical surface of Subject 1 . The cortical electrode marked with a red dot ( bottom ) corresponds to the black timecourse in the inset ( top ) . The other ECoG channels are plotted in gray . The time period visualized corresponds to approximately 300 milliseconds , or five cycles of the spindle oscillation . Note that no spatial smoothing is applied in these data . DOI: http://dx . doi . org/10 . 7554/eLife . 17267 . 017 Spike-time dependent plasticity is a well-studied mechanism for regulating synaptic strengths that depends on the relative timing of presynaptic inputs and postsynaptic spikes ( Markram et al . , 1997; Bi and Poo , 1998 ) , but for establishing large-scale neural assemblies during sleep oscillations through synaptic plasticity , axonal conduction delays pose a specific problem ( Lubenov and Siapas , 2008 ) . For example , cortical white matter association fibers have conduction delays up to 50 milliseconds across the cortex ( Figure 2A , left ) ( Girard et al . , 2001; Schüz et al . , 2002; Swadlow and Waxman , 2012 ) . 10 . 7554/eLife . 17267 . 018Figure 2 . Schematic of spindles and axonal delays . ( A ) Spikes emitted from region A will arrive at B with a temporal delay of 20 milliseconds ( left ) . If spindle oscillations were perfectly synchronized across the cortex , EPSPs from region A would occur after the spikes in region B , within the window for long-term depression ( right ) . ( B ) In contrast , if spindles are spatiotemporally organized with stereotyped trajectories ( left ) , then EPSPs from region A would align with population spiking in region B , allowing for synaptic strengthening to occur . DOI: http://dx . doi . org/10 . 7554/eLife . 17267 . 018 It is well established that spindles cause pyramidal cells and interneurons in cortex to fire preferentially at the peak of the surface-positive ( depth-negative ) LFP oscillation , both in intracellular ( Contreras and Steriade , 1995 , 1996; Kandel and Buzsáki , 1997 ) and extracellular ( Peyrache et al . , 2011 ) recordings . If cortical spindles were perfectly synchronized , spikes emitted during one cycle of the spindle oscillation would arrive at their post-synaptic targets with this temporal delay , leading to a pairing within the window for persistent long-term depression ( LTD ) that would progressively weaken long-range connections ( Figure 2A , right ) . If , however , spindles are self-organized into large-scale wave-like activity patterns , with phase speeds matching those of the underlying fiber networks and stereotyped , precisely repeating trajectories ( Figure 2B , left ) , then EPSPs caused by spikes traveling along pyramidal axons to distant regions in the cortex would align with the local burst of population activity ( Figure 2B , right ) , creating the conditions necessary for synaptic strengthening to occur . Next , we wanted to understand whether these population activity patterns repeat with the temporal precision required for strengthening of large-scale assemblies . We defined the correlation magnitude over phase values on the electrode array between individual oscillation cycles to be a pairwise similarity index ( see Materials and methods ) , in order to detect similar spatiotemporal patterns across oscillation cycles ( Video 2 ) . By calculating this metric over all cycle pairs in different wave classes ( all cycles , expanding , rotational ) , we can directly compare the temporal precision mediated by each type . The cumulative distribution function ( CDF ) of similarity indices among identified rotational waves is highly shifted to the right ( Figure 3A , black ) compared to the CDF for all cycles ( 5 subjects , 42/54 sleep epochs , 77 . 8% significant , one-tailed two-sample Kolmogorov-Smirnov test , α=0 . 01 , Bonferroni correction ) , indicating higher intra-class similarity between these cycles than for other wave types . Note that this is not simply a consequence of the rotational phase pattern itself , as expanding waves emanating from a consistent point source could certainly exhibit higher intra-class similarity than rotational waves with a varying center . Further , the median similarity index consistently increases in individual subjects when rotational waves of progressively increasing strength are considered ( Figure 3B ) . This indicates the observed rotating waves strongly modulate temporal precision in repeated patterns of population activity . To be specific , by utilizing the average temporal frequency for these spindle oscillations ( 13 . 5 Hz ) , we can estimate that in two cycles whose similarity index falls into the highest bin in Figure 3C ( 0 . 9–1 . 0 ) , 50% of electrodes will experience an alignment of the spatiotemporal activity pattern within a 5 millisecond temporal window . Recent experiments have shown a tight temporal link between field potentials and synaptic currents ( both EPSPs and IPSPs; Haider et al . , 2016 ) . By detecting these precisely recurring activity patterns in ECoG recordings , we can infer that distributed networks composed of local excitatory and inhibitory groups , whose firing is modulated by thalamocortical fibers during the sleep spindle , are repeatedly activated with a millisecond accuracy that is well within the temporal precision required for STDP . 10 . 7554/eLife . 17267 . 019Video 2 . Rotating waves with high spatiotemporal similarity . Two rotating waves with high phase similarity on the ECoG array , separated by 5 . 62 min of stage 2 sleep . Bandpass filtered timeseries are normalized to their maximum within the interval and plotted in falsecolor ( bottom panels ) . Activity for each channel is plotted as a function of time ( top panels ) , with an indication of temporal progression ( red dotted line ) . DOI: http://dx . doi . org/10 . 7554/eLife . 17267 . 01910 . 7554/eLife . 17267 . 020Figure 3 . Phase pattern analysis . ( A ) Rotating waves exhibit higher intra-class similarity . Cumulative distribution functions ( CDFs ) for shuffled data ( purple ) , expanding waves ( red ) , all cycles ( blue ) , and rotating waves ( black ) are given for an example 15 min epoch of stage 2 sleep ( subject 5 ) . ( B ) Spindle cycles exhibiting stronger rotating patterns also express greater intra-class similarity . Gray lines indicate the median similarity index ( ordinate ) for the population of oscillation cycles expressing rotational waves above a threshold strength ( abscissa ) , averaged over individual sleep epochs . Red dots and error bars indicate the median and median absolute deviation for the full distribution , respectively . ( C ) Spindle cycles exhibiting high similarity index are temporally precise . The distribution of phase difference at each electrode across spindle cycles is given as a function of the similarity index ( indicated by colors , inset ) . By utilizing the mean spindle oscillation frequency ( 13 . 5 Hz ) , the midspread ( interquartile range ) of each distribution is given in units of time ( inset ) . DOI: http://dx . doi . org/10 . 7554/eLife . 17267 . 020 If this millisecond precision in fact mediates formation and maintenance of corticocortical assemblies , we would then expect spiking associated with these synaptic currents to drive increased reverberation throughout the night , as excitatory connections between local groups of pyramidal cells and interneurons are strengthened and in turn promote more replay of the expressed activity pattern . High gamma-band power ( HGP , 80–120 Hz ) , a reliable electrophysiological correlate of spiking activity ( Ray et al . , 2008; Ray and Maunsell , 2011; Ray , 2015; similar in nature to the 'broadband power shift' described in Manning et al . , 2009 ) , consistently increases around spindles ( Figure 4A ) . Further , HGP is modulated by spindle phase ( Figure 4B ) , increasing towards the surface-positive ( depth-negative ) peak , consistent with previous animal ( Peyrache et al . , 2011 ) and human ( Andrillon et al . , 2011 ) recordings . Finally , by studying repeats of rotating waves over 2 . 5 hr of continuous sleep recording in Subject 1 , we observe a preliminary indication of increased reverberation consistent with strengthening of distributed excitatory networks: similarity in the next identified rotating wave is highly predictive of the number of strong reverberations throughout the night ( black dots , Figure 4C ) . Randomizing the relationship between the next rotating wave and the rest of the sleep recording eliminates this effect ( shuffling control , Figure 4C ) , and such increased reverberation is not observed for expanding waves under similar conditions ( Figure 4—figure supplement 1 ) . Similar observations are consistent across subjects ( Figure 4—figure supplement 2 ) . These results support the hypothesis that precisely repeating rotating waves may enable strengthening of large-scale corticocortical assemblies throughout the night . 10 . 7554/eLife . 17267 . 021Figure 4 . Spiking activity and increased reverberation . ( A ) High gamma-band power ( HGP ) consistently increases around spindle onset . Plotted are the normalized amplitude envelopes for spindles ( black ) and HGP ( red ) , averaged over 186 spindles in Subject 1 . ( B ) HGP is modulated by spindle phase . Plotted is the mean high gamma-band power at each phase of the spindle oscillation ( 20 bins ) , for varying amplitudes of the spindle oscillation ( see colorbar ) , each normalized by the mean HGP in matched non-spindle epochs . ( C ) Strength of the first repeat predicts the number of strong reverberations . The number of similar rotational patterns ( above similarity index 0 . 7 ) following a spindle oscillation cycle is given as a function of the next cycle’s similarity index ( black dots , mean + SEM ) over 2 . 5 hr of sleep in Subject 1 . Error bars are obscured by markers . Light blue lines indicate results from a shuffling permutation test ( 10 iterations ) . DOI: http://dx . doi . org/10 . 7554/eLife . 17267 . 02110 . 7554/eLife . 17267 . 022Figure 4—figure supplement 1 . Strong increase of reverberation observed for rotating , but not expanding , waves . ( A ) The increase in the number of strong repeats is given for expanding ( gray dots , mean + SEM ) and rotational ( black dots ) waves . Error bars are obscured behind the markers . Plotted also are the shuffling controls for rotating ( light blue ) and expanding ( light gray ) waves ( 25 iterations at each point ) . ( B ) The numbers of strong repeats in panel A are plotted as z-scores relative to the shuffling control distribution , for expanding ( gray ) and rotating ( black ) waves . This panel illustrates that the substantial , marked increase in reverberation is observed for rotating , but not expanding , waves . DOI: http://dx . doi . org/10 . 7554/eLife . 17267 . 02210 . 7554/eLife . 17267 . 023Figure 4—figure supplement 2 . Increase in reverberation observed across subjects . By aggregating data from shorter recordings in Subjects 2–5 , we replicated the reverberation analysis ( Figure 4C ) using fewer bins ( Low , less than 0 . 5 similarity index; high , greater than 0 . 5 ) . Rotating wave patterns with a high-strength first repeat showed increased reverberation relative to the shuffling control ( light blue dotted lines ) . DOI: http://dx . doi . org/10 . 7554/eLife . 17267 . 02310 . 7554/eLife . 17267 . 024Figure 4—figure supplement 3 . Modulation of high-gamma power ( HGP ) by spindle phase across subjects . ( Left ) Mean HGP as a function of spindle phase is plotted for example stage 2 sleep epochs in Subjects 1–5 . For this plot , spindle amplitudes above a given threshold ( 5 z-score units ) were considered . As in Figure 4B , each plot is normalized by mean HGP in matched non-spindle epochs . ( Right ) Plotting each HGP modulation ratio normalized to its maximum ( higher values in white ) shows the consistent increase in HGP near the surface-positive ( depth-negative ) peak in each subject . DOI: http://dx . doi . org/10 . 7554/eLife . 17267 . 024 Early animal sleep spindle studies , using up to 8 electrodes in a linear array ( Andersen et al . , 1967; Kim et al . , 1995; Contreras et al . , 1996 , 1997 ) , in addition to preliminary EEG evidence in the human ( Achermann and Borbély , 1998 ) , proposed that spindles involve global synchronization of cortical circuits , raising the possibility that this sleep oscillation places neocortex into a specialized state for consolidation of long-term memories . In recent years , several studies have reported a mixture of 'local' and 'global' spindles using amplitude-duration thresholding approaches ( Nir et al . , 2011; Andrillon et al . , 2011 ) . By carefully studying the phase information in the spindle frequency band recorded on large-scale ECoG arrays , we have uncovered that a substantial number of spindle oscillation cycles are organized into global , hemisphere-spanning patterns of rotating and expanding waves ( Figure 1—figure supplement 7 ) . These patterns most likely represent the characteristic spatiotemporal organization of the 'global' spindles observed in Andrillon et al . ( Andrillon et al . , 2011 ) ( ≥40% involvement , cf . their Figure 5C ) , with more localized patterns left unclassified by our detection approach ( Figure 1—figure supplements 9 and 12 ) . These global patterns are likely established through widespread thalamocortical loops , placing the cortex into a state of large-scale coherence ( Contreras et al . , 1996 ) , shaped into rotating and expanding waves through corticocortical white matter connections with axonal conduction speeds consistent with the observed propagation speeds ( Figure 1C ) . Future computational modeling work will address in detail the role of thalamocortical , corticocortical , and corticothalamic connections in generating the spatiotemporal activity patterns reported here . Spindles have recently been specifically and causally implicated in the sleep-dependent consolidation of long-term memories ( Mednick et al . , 2013; Hennies et al . , 2016 ) . While some memories integrate content from single sensory modalities , requiring consolidation in only single cortical regions ( such as motor cortex , Khazipov et al . , 2004 ) , many memories integrate multimodal sensory and cognitive information ( Gibson and Maunsell , 1997 ) , and require 'global' integration of distributed networks across the cortex . In this work , we have identified a novel mechanism by which this process could occur: the stereotyped activity patterns reported here may enable STDP to establish large-scale neuronal assemblies at scales where axonal conduction delays are long relative to the oscillation cycle ( Fries , 2005; Lubenov and Siapas , 2008 ) , and repeat many times throughout sleep with millisecond accuracy . While the schema illustrated in Figure 2 is a highly simplified view of the microscale interactions between long-range excitatory projections and local networks during spindle oscillations , computational and theoretical studies have previously obtained a detailed understanding of STDP dynamics with neurons receiving sequenced ( Rao and Sejnowski , 2001 , 2003 ) , bursting ( Song et al . , 2000 ) , and oscillating inputs ( Muller et al . , 2011; Luz and Shamir , 2016 ) . This theoretical understanding of the interplay between STDP and population activity can allow in future work a precise account of how microscale synaptic interactions are shaped by global oscillation patterns , and how variability in these patterns ( e . g . variation in wave speed , Figure 1C ) will affect this mechanism . Taken together , these results provide insight into how distributed information stored across cortical regions may be bound into a coherent , integrated , but specific memory through spike-time dependent synaptic plasticity .
Patients with longstanding pharmacologically resistant complex seizures gave fully informed consent according to NIH guidelines as monitored by the local Institutional Review Board ( Massachusetts General Hospital ) . Electrocorticogram ( ECoG ) recordings during natural sleep were made over the course of clinical monitoring for spontaneous seizures . Electrode placement was determined solely by clinical criteria , with electrode grids usually spanning the Sylvian fissure and multiple lobes of the cerebral cortex ( frontal , parietal , and temporal ) . Patients were informed that participation in the research would not alter their clinical treatment in any way , and that they may withdraw their consent at any time without jeopardizing clinical care . ECoG contacts ( Ad-Tech Medical Instrument Corp . , Racine , WI ) were 3 mm platinum-iridium ( 90% platinum ) discs arranged in a two-dimensional grid ( 8 rows and 8 columns , Subjects 1 , 3 , and 4; 8 rows and 12 columns , Subject 2; 8 rows and 6 columns , Subject 5 ) implanted semi-chronically on the pial surface in an effort to localize the seizure origin . Within the grid , electrodes were spaced 10 mm apart . In some patients , linear ECoG arrays provided additional spatial coverage; application of our multichannel detection approach , however , focused on the two-dimensional electrode grid . One strip of electrodes positioned over the pial surface and facing the skull served as the reference during the recordings; results were additionally verified using an average reference . Note that due to reference artifacts , an average reference was employed for the recordings in Subject 4 . We note as well that the temporal extent of the waves , over tens of milliseconds , makes electrophysiological artifacts such as volume conduction an unlikely explanation for the observations reported here . Recordings were performed with clinical EEG monitoring equipment ( XLTEK , Natus Medical Inc . , Pleasanton , CA ) and sampled at 500 or 512 Hz . Post-implantation electrode localization utilized coregistration of preoperative magnetic resonance imaging ( MRI ) with postoperative computed tomography ( CT ) , as described by Dykstra et al . ( Dykstra et al . , 2012 ) . Cortical surfaces were computed with FreeSurfer ( Dale et al . , 1999; Fischl et al . , 1999 ) . To account for the misalignment between the MRI and CT due to the craniotomy , the locations of the grid electrodes were projected onto the cortical surface ( Dykstra et al . , 2012 ) . Geodesic electrode distances , which take into account the folded geometry of the cortical surface and were used in some calculations ( e . g . estimation of spatial correlation values ) , were estimated using a shortest paths approach on the cortical surface mesh . During the monitoring period for spontaneous seizures , the subjects slept in the clinical environment and expressed relatively normal sleep patterns . ECoG recordings that did not have a seizure in the preceding or following 12 hr were scored visually by an expert rater following the standard sleep stage classification ( Silber et al . , 2007 ) . For each patient , we obtained from 15 to 101 . 5 min of NREM stage 2 sleep , when the spindles are most prevalent . Individual sleep spindles were then detected during stage 2 sleep using one of several complementary methods , either based on amplitude-duration thresholding ( Gais et al . , 2002; Warby et al . , 2014 ) or a similar wavelet-based approach with additional verification steps ( Hagler et al . , 2016 ) . The number of spindles detected was in agreement with previous reports of spindle density ( Gais et al . , 2002; Warby et al . , 2014 ) . The results were additionally verified using a novel approach quantifying the signal-to-noise ratio ( SNR ) of power in the bandpass ( 9–18 Hz , 8th-order Butterworth filter ) versus the bandstop ( 1–100 Hz bandpass , with 9–18 Hz bandstop ) signal . In this approach , the SNR metric is calculated on short ( 500 ms ) sliding windows in each channel . When the SNR metric reaches 0 dB , the signal and noise power are at parity , corresponding to a sharp , narrowband epoch in the recording . Picking a constant SNR threshold ( 5 dB ) corresponds roughly to the constant false alarm rate ( CFAR ) technique in radar . This approach yields a conservative but approximately amplitude-invariant method for detecting arbitrary narrowband epochs in multichannel data . Following spindle detection , we made a verification analysis by calculating the average power spectral density ( PSD ) over isolated spindle epochs in each subject . Data were initially filtered to remove line noise artifacts , and PSDs were then calculated in 1 s intervals during the spindle and matched non-spindle epochs . PSDs for individual channel and spindle epochs were concatenated into a large array and averaged in each case . A clear peak in the 11–15 Hz frequency band for the spindle epochs can be seen , while no peak is observed in the matched non-spindle epochs ( Figure 1—figure supplement 1A ) . Divisive normalization is calculated by dividing the power at each frequency in the spindle epochs by the power in the matched non-spindle epochs , and expressing the result in dB ( Figure 1—figure supplement 1A , inset ) . If the divisive normalization over an epoch of stage 2 sleep reached 5 dB , then the spindles were taken to be well-isolated and possessing the spectral characteristics necessary for an accurate phase representation , and were then included in further analysis . Based on this calculation , 54 individual epochs of stage 2 sleep , varying from 30 s to 35 min in duration , were selected in five clinical subjects . Temporal filtering of stage 2 sleep recordings was carried out with an 8th-order digital Butterworth bandpass filter ( 9–18 Hz ) , forward-reverse in time to prevent phase distortion ( see MATLAB function filtfilt ) . All results were checked with multiple cutoff frequencies to ensure against parametric sensitivity . To assess spatial correlation during spindle oscillations as a function of distance in the cortex , we adapted standard methods ( Destexhe et al . , 1999 ) with a Monte Carlo implementation more suited for sampling correlations on two-dimensional electrode arrays . To calculate this metric , one electrode is first selected at random , and a second is then selected from the set of electrodes within a binned distance di from the first . The temporal correlation between these electrode pairs is then computed in the bandpass timeseries between the start and end points of the spindle . This process is repeated for a given number of iterations Nk at each distance bin di , and the average spatial correlation is computed as the mean of the correlation values for the individual epochs ( Figure 1—figure supplement 1B , black ) . The spatial correlation values were computed for non-spindle epochs matched to the temporal extent of the individual tested spindle epochs ( Figure 1—figure supplement 1B , red ) . The average spatial correlation values are elevated during spindle oscillations with respect to the matched non-spindle periods of stage 2 sleep , indicating that a coherent , large-scale increase in global activity occurs during spindles , in agreement with previous studies ( Destexhe et al . , 1999 ) . To study spatiotemporal dynamics in these neural recordings , we adapted our previously introduced method for detecting arbitrarily shaped traveling waves ( Muller et al . , 2014 ) to multisite ECoG arrays ( Figure 1—figure supplements 2 and 3 ) . This approach allowed us to characterize and classify the spatiotemporal dynamics during thousands of episodes of spindling activity in many hours of sleep recordings . The method proceeds in three steps: ( 1 ) analytic signal representation for characterization of instantaneous signal characteristics at each electrode , ( 2 ) center localization at each individual oscillation cycle , and ( 3 ) quantification of the spatiotemporal pattern in each oscillation cycle as a function of distance from ( or rotation about ) the isolated center point . In the following , we describe in detail the method for isolating expanding and rotating waves in multichannel data . To estimate instantaneous signal characteristics , we employ the well-known analytic signal representation . This approach entails transforming a real-valued timeseries into a complex phasor , whose modulus ( length ) and argument ( angle ) in the complex plane represent the signal instantaneous amplitude and phase , respectively . Specifically , if vx , y , t is a real-valued , narrowband timeseries at a point ( x , y ) , x∈[1 , Nc] , y∈[1 , Nr] , where Nc and Nr denote the number of rows and columns , and t∈[1 , Nt] is the sample number , then its analytic signal representation is ( 1 ) Vx , y , t=vx , y , t+iv^x , y , t where i is the complex unit and f^ denotes the Hilbert transform of a signal f . The instantaneous phase of vx , y , t is then the argument at each point in this complex sequence ( 2 ) ϕx , y , t=Arg ( Vx , y , t ) , and instantaneous amplitude is the modulus . At several points in the analysis , results were confirmed with an FIR implementation of the Hilbert transform , in addition to the standard FFT-based approach ( Marple , 1999 ) . We evaluated phase values at a set of time points 𝒯={t1 , t2 , … , tK} in each spindle near the positive oscillation peaks . The phase fields were then smoothed using a robust approach ( Garcia , 2010 ) for center localization to reduce noise and interpolate values from missing electrodes; note that the smoothed values were not used in the calculations for detecting expanding and rotational wave patterns . These phase values are then used to capture spatiotemporal dynamics in the multichannel data . To isolate putative expanding or rotating wave centers in each oscillation cycle , we first assess the spatial gradient of phase ( 3 ) g→x , y , tj≡-∇ϕx , y , tj with tj∈𝒯 . For the spatial gradient , derivatives are taken across the two dimensions of space and are approximated by the appropriate forward and centered finite differences . As in previous work , phase derivatives were implemented as multiplications in the complex plane ( Feldman , 2011; Muller et al . , 2014 ) . To detect expanding waves , we assess the divergence of the phase gradient field ( 4 ) dx , y , tj=∇⋅g→x , y , tj , and define the putative wave source to be that point which satisfies the arg max over space in each cycle ( 5 ) 𝒮≡ ( x , y;tj ) =argmaxx , ydx , y , tj , where argmaxf ( a , b ) ≡{a , b|∀p , q:f ( p , q ) ≤f ( a , b ) } . This step allows us to find the source for a possible expanding wave in each cycle ( step 2 , Figure 1—figure supplement 2 ) , about which the phase field is then evaluated to quantify the evidence for an expanding wave spatiotemporal organization ( step 3 , Figure 1—figure supplement 2 ) . For this next step , we calculate the circular-linear correlation coefficient ρϕ , δ ( Jammalakadaka and Sengupta , 2001; Berens , 2009 ) between signal phase ϕ and radial distance δ from the source point in the original , unsmoothed phase field ( 6 ) ρϕ , δ=rcδ2+rsδ2−2rcδrsδrcs1−rcs2 , where rcδ represents the Pearson correlation between the cosine of the circular variable ϕ and the linear variable δ , rsδ between the sine of ϕ and the variable δ , and rcs between the cosine and sine of ϕ . This approach allows us to quantify the strength of the spatiotemporal pattern of activity on the array in a single number , which is then compared to the value produced by repeating the calculations many times under random shuffling of the data ( blue dotted line , Figure 1—figure supplement 4A and Materials and methods – Shuffling Controls ) . Analogous to the above case , we start by assessing the curl of the phase gradient field ( 7 ) c→x , y , tj=∇×g→x , y , tj , and defining the putative center to be that point which satisfies the arg max over space ( 8 ) 𝒞≡ ( x , y;tj ) =argmaxx , y||c→x , y , tj|| . This center point then defines an anchor about which we can pass into a polar coordinate system , describing the distance δ and rotation angle θ about that point ( step 2 , Figure 1—figure supplement 3 ) . With the putative rotation center isolated in each oscillation cycle , we then proceed to calculate the circular-circular correlation coefficient ρϕ , θ between signal phase ϕ and rotation angle θ ( Fisher , 1993; Berens , 2009 ) in the original , unsmoothed phase field ( 9 ) ρϕ , θ=-∑xysin ( ϕxy-ϕ¯ ) sin ( θxy-θ¯ ) ∑xysin2 ( ϕxy-ϕ¯ ) sin2 ( θxy-θ¯ ) , where overbar indicates circular mean ( 10 ) ϕ¯=Arg[∑xyeiϕxy] . Similar to the previous case , this number ρϕ , θ quantifies the evidence for a rotational wave organization on the array , which is then compared to the value derived from a random-shuffling permutation test ( red dotted line , Figure 1—figure supplement 4A and Materials and methods – Shuffling Controls ) . For each case , additional control analyses with simulated rotating waves embedded in noise were used to verify the robustness of our approach ( Figure 1—figure supplement 11 ) . Finally , in the case that both expanding and rotational elements are detected , the pattern is classified as rotational , because sub-patterns of rotational waves tend to be detected as expanding elements ( verified in Figure 1—figure supplement 5; see also Materials and methods – Average vector field controls ) . To quantify the level of spatiotemporal phase flow expected in the data by chance , we implemented a shuffling procedure to establish a permutation-based threshold for both the expanding and rotating wave measures . To do this , we shuffled the phase values in each oscillation cycle randomly across space a number of times ( 100 or 1000 times in initial tests , then reduced to 25 without changing results ) , repeating each time the same calculation as for the un-shuffled data . The 99th percentile of the resulting distribution then determines a threshold above which the value for the correlation metric ( either for expanding or rotational waves , considered separately ) exceeds chance , with the spatial autocorrelation erased . A possible confound resulting from this shuffling procedure is that the data intrinsically possess some spatial autocorrelation ( Figure 1—figure supplement 1B ) , which is ignored by the so-constructed permutation test . To address this point in the context of rotational wave detection , we conceived an additional permutation test control . In this second control , we considered the set of points at a Chebyshev ( i . e . King’s chessboard ) distance di∈[1 , dm] from the putative rotation center , where dm indicates the maximum distance on the electrode array from that point . We then shuffled channels at distances di for all i , and repeated the calculation for the rotational wave detection . The resulting permuted data have a spatial correlation function identical to that in the un-shuffled data , but with the rotational structure fully destroyed . The 99th percentile cutoff determined from this second control analysis fell within 0 . 01 of the originally estimated value ( 3% difference ) , validating the original shuffling permutation test employed above . Using simulated expanding waves of the form ( 11 ) f ( t , δ ) =Aei ( ωt-κδ ) +ση ( t ) , and simulated rotating waves ( 12 ) f ( t , θ ) =Aei ( ωt-γθ ) +ση ( t ) , where A is the oscillation amplitude , ω is the oscillation angular frequency , κ is the wavenumber , γ is the polar wavenumber , and η ( t ) is a Gaussian white noise term , we verified the robustness of our detection approach under noise of varying amplitudes . Note that δ and θ are defined with respect to the wave center , left unspecified for simplicity . Oscillation amplitude was set to unity , without loss of generality , and other oscillation parameters were matched to those observed during stage 2 sleep spindles . Oscillation frequency ω was set to the average instantaneous frequency estimated from 702 spindles in Subject 1 ( 13 . 5 Hz ) , and wavenumbers were adjusted to approximate the wavelengths observed in the data . Varying systematically the level of added noise , we ran the algorithms for detecting expanding and rotating waves described above for 25 trials at each point and recorded the algorithm’s detection performance in each case ( mean ± SEM , Figure 1—figure supplement 11C ) . These results illustrate the approximate invariance of our computational approach to random noise . In another test , we systematically varied the position of the wave center on the simulated 64 electrode array , for both expanding and rotational waves ( Figure 1—figure supplement 11D ) . Parameters were set as above , and simulations were again run over 25 trials at each point . This test probed the sensitivity of the rotational detection approach to border effects , which is expected to be negligible at the encountered noise levels in comparison to the thresholds established by the permutation controls ( dotted lines , Figure 1—figure supplement 11D ) . In a third test , we verified that the spatiotemporal patterns observed here are not due to variations in spindle frequency , which are known to occur along the rostro-caudal axis ( Peter-Derex et al . , 2012 ) . To do this , we re-ran our analysis on one stage 2 sleep session containing 179 spindles in Subject 1 , generating surrogate data as follows . Each electrode evolved in time according to its mean instantaneous frequency during the spindle , but with a randomized initial phase angle . These surrogate data thus possessed the same frequency content on average as in the original data , but with their spatial organization of phase removed . In this control , both rotating and expanding wave patterns were highly decreased ( 3 . 5% and 0 . 8% of cycles , respectively , compared to 64% and 14% in the original data ) . The algorithmic classification of wave patterns in individual oscillation cycles involves several steps , and we wanted to make an independent check to verify these results . To do this , we adopted a re-centered averaging approach , shifting the vector field of propagation directions from the smoothed phase fields at each oscillation cycle to the putative rotation center ( red dots , Figure 1—figure supplement 5 ) , and taking the circular mean ( Fisher , 1993; Berens , 2009 ) of propagation direction at each point . Performing the calculation in this way prevents regions with noise or high phase gradient magnitude from dominating the result . The obtained vector fields for rotational TPF ( Figure 1—figure supplement 5A ) and expanding ( Figure 1—figure supplement 5B ) waves illustrate the accuracy of the algorithm and the general validity of our classification approach in separating waves into expanding and rotational groups . To quantify the precision of repeated spatiotemporal patterns during across spindle oscillations over several minutes of data , we calculated the circular-circular correlation between phase values in individual oscillation cycles . For two phase maps αx , y and βx , y , the circular correlation is defined as above ( Fisher , 1993; Berens , 2009 ) ( 13 ) ρα , β=∑xysin ( αxy-α¯ ) sin ( βxy-β¯ ) ∑xysin2 ( αxy-α¯ ) sin2 ( βxy-β¯ ) . This correlation value defined between individual phase maps then constitutes elements of an M×M matrix Cij , where M is the number of isolated oscillation cycles in question:Cij=[1ρ1 , 2ρ1 , 3ρ1 , 4⋯ρ1 , Mρ2 , 11ρ2 , 3ρ2 , 4⋯ρ2 , Mρ3 , 1ρ3 , 21ρ3 , 4⋯ρ3 , Mρ4 , 1ρ4 , 2ρ4 , 31⋯ρ4 , M⋮⋮⋮⋮⋱⋮ρM , 1ρM , 2ρM , 3ρM , 4⋯1] where all elements in the diagonal and lower triangle ( Cij∀i≥j ) are not considered , without loss of generality . We then construct this matrix for all cycles in an individual wave classification ( all cycles , expanding , rotational ) , and consider the cumulative distribution function ( CDF ) of values in the upper triangle ( Figure 3A ) . To construct the CDF in the permutation case ( Figure 3A , purple line ) , we first randomly shuffled the phase maps in all spindle cycles and then proceeded with the calculation as normal . A MATLAB toolbox for analysis of traveling waves and complex spatiotemporal dynamics in noisy multisite data is available as an open-source release on BitBucket: http://bitbucket . org/lylemuller/wave-matlab | When you wake up in the morning after a good night's sleep you feel refreshed . You can also think more clearly because your memory has been re-organized , a process called memory consolidation . The problem that the brain has to solve during sleep is how to integrate memories of experiences that happened during the day with old memories , without losing the older memories . Scientists know that waves of electrical activity , referred to as spindles , help to consolidate and integrate memories during sleep . Spindles are active in the cerebral cortex , the part of your brain used for thinking , in the time between dream sleep and deep sleep . Yet it is not known exactly how these bursting patterns of electrical activity help to strengthen memories . Now , Muller et al . explored how the spindles could strengthen and connect parts of memories stored in distant parts of the brain . First , a computer algorithm analyzed electrical recordings of brain activity taken while five patients with epilepsy slept . The patients were being monitored to help with their seizures , and the recordings showed that spindles do not occur at the same time throughout the cortex as previously thought . Instead , the spindle is a wave that begins in portion of the cortex near the ear , spirals through the cortex toward the top of back of the head and then on to the forehead area before circling back . These repeated circular waves of electrical activity strengthen connections between brain cells in distant parts of the brain . For example , these waves may help strengthen connections between the cells of the cortex that separately store memories of the sound , sight and feel of an event during the day , whether that’s being bitten by a dog or talking with a friend . Next , Muller et al . plan to develop computer models of the spindles and verify whether their models make accurate predictions by studying spindles in sleeping mice and rats . | [
"Abstract",
"Introduction",
"Materials",
"and",
"methods"
] | [
"short",
"report",
"neuroscience"
] | 2016 | Rotating waves during human sleep spindles organize global patterns of activity that repeat precisely through the night |
The energy required to fuse synaptic vesicles with the plasma membrane ( ‘activation energy’ ) is considered a major determinant in synaptic efficacy . From reaction rate theory , we predict that a class of modulations exists , which utilize linear modulation of the energy barrier for fusion to achieve supralinear effects on the fusion rate . To test this prediction experimentally , we developed a method to assess the number of releasable vesicles , rate constants for vesicle priming , unpriming , and fusion , and the activation energy for fusion by fitting a vesicle state model to synaptic responses induced by hypertonic solutions . We show that complexinI/II deficiency or phorbol ester stimulation indeed affects responses to hypertonic solution in a supralinear manner . An additive vs multiplicative relationship between activation energy and fusion rate provides a novel explanation for previously observed non-linear effects of genetic/pharmacological perturbations on synaptic transmission and a novel interpretation of the cooperative nature of Ca2+-dependent release .
Regulation of synaptic efficacy is an essential aspect of information processing in neuronal networks . The energy barrier for vesicle fusion is considered to be a main contributing factor . To release neurotransmitters , synaptic vesicles ( SVs ) need to fuse with the neuronal plasma membrane , which requires substantial energy . Local membrane deformation , dehydration of lipid head groups , neutralization of opposite membrane charges , lipid splaying , and the creation of a lipid stalk all contribute to the energy barrier that needs to be overcome before neurotransmitters are released ( Kuzmin et al . , 2001; Kozlovsky and Kozlov , 2002; Markin and Albanesi , 2002; Sorensen , 2009 ) . Reaction rate theory suggests that specifically modulation of the fusion energy barrier is a powerful way to regulate synaptic efficacy . According to the Arrhenius equation , reaction rates change exponentially with changes in the activation energy , which is the minimum energy required for a reaction ( e . g . , vesicle fusion ) ( Arrhenius , 1889a , 1889b ) . Thus , we predict that a set of modulations of the release rate may exist , which act by lowering the activation energy for fusion . If this is the case , they will have a supra-linear effect on the fusion rates , and converting rates to energies ( by inverting the Arrhenius equation ) should reveal additive effects on the fusion barrier . This is highly relevant since many presynaptic factors may act on the activation energy for fusion simultaneously and potentially independently during synaptic stimulation . Much of the energy required for SV fusion is likely provided by the SNARE proteins , synaptobrevin/VAMP , syntaxin , and SNAP25 , whose assembly into a trimeric SNARE-complex drives the fusion reaction ( Sorensen , 2009; Jahn and Fasshauer , 2012 ) . However , several other proteins likely contribute to the efficient and fast reduction of the activation energy for SV fusion that is required for fast synaptic transmission . During action potential ( AP ) stimulation , for example , SV fusion rates increase by several orders of magnitude within a few milliseconds due to the rapid activation of Ca2+ sensors of the synaptotagmin-family , which control SNARE-mediated fusion ( Rhee et al . , 2005; Xu et al . , 2007; Walter et al . , 2010; Weber et al . , 2010; Kochubey and Schneggenburger , 2011; Arancillo et al . , 2013; Sudhof , 2013 ) . Other proteins , such as Munc18 and Munc13 , might also support synaptic transmission by reducing the activation energy for SV fusion , either through their established roles in SNARE-complex assembly ( Basu et al . , 2007; Wierda et al . , 2007; Weber et al . , 2010; Xue et al . , 2010; Arancillo et al . , 2013 ) or through independent actions . Direct measurements of the exact contributions of different molecular events inside living nerve terminals to the activation energy for SV fusion are not possible . However , the predicted supra-linear modulation of release rates can be measured experimentally . This can be interpreted as changes in the activation energy under certain assumptions ( e . g . , a constant empirical prefactor A , see below ) . SV release kinetics has been intensively studied using flash photolysis of caged Ca2+ ( Schneggenburger and Neher , 2000; Lou et al . , 2005; Sakaba et al . , 2005; Korogod et al . , 2007; Sun et al . , 2007; Wolfel et al . , 2007; Kochubey and Schneggenburger , 2011; Burgalossi et al . , 2012 ) . However , synaptic responses to Ca2+ elevation ( either triggered by natural stimulations by APs or by Ca2+ uncaging ) are caused by a rapid synaptotagmin/Ca2+-induced lowering of the energy barrier for vesicle fusion . This mechanism might be modified by several factors that interact with synaptotagmin . Therefore , to assess changes in the energy barrier per se , caused by other factors , we must use a different , Ca2+-independent method to assess changes in release kinetics . In this regard , hypertonic solutions have been used widely as they cause SV release from the same readily releasable SV pool ( RRP ) as APs do , but by a Ca2+-independent stimulus ( Fatt and Katz , 1952; Stevens and Tsujimoto , 1995; Rosenmund and Stevens , 1996 ) . Correspondingly , changes in the kinetics of synaptic responses to hypertonicity-induced SV fusion have been interpreted as changes in the intrinsic ‘release willingness’ or ‘fusogenicity’ of SVs , which may represent an inverse measure for the activation energy for SV fusion ( Basu et al . , 2007; Wierda et al . , 2007 ) . Here , we introduce a method to quantify vesicle fusion rate constants and RRP-pool size by fitting a kinetic model to synaptic responses triggered by hypertonicity-induced SV fusion . Using this approach , we show that independent osmotic , genetic , and biochemical perturbations modulate SV release in a multiplicative/supralinear manner . The fact that linear ( additive ) effects on the energy barrier ( activation energy ) produce supralinear ( multiplicative ) effects on the release rate , helps to explain previously unexplained effects of genetic/pharmacological perturbations on synaptic transmission and provides a novel interpretation of the previously identified cooperative nature of Ca2+-dependent release .
Fusion of the lipid bilayer of synaptic vesicles with the plasma membrane involves deformation of membranes , dehydration of lipid head groups , neutralization of opposite membrane charges , and lipid splaying ( Kuzmin et al . , 2001; Kozlovsky and Kozlov , 2002; Markin and Albanesi , 2002; Sorensen , 2009 ) , which together requires substantial energy . Vesicle priming and fusion can be represented in terms of an energy landscape , with energy barriers separating non-primed , primed , and fused states ( Figure 1A ) ( Sorensen , 2009; Walter et al . , 2013 ) . The Arrhenius equation predicts an exponential relation between the rate constants of transitions between these states and the activation energies for these reactions , which correspond to the relative heights of these energy barriers ( Figure 1B ) ( Arrhenius , 1889a , 1889b; Jahn and Grubmuller , 2002 ) . Hence , for transition from the primed to the fused state , the vesicle fusion-rate constant is given by ( 1 ) k2=Ae−EaR¯T , with T the absolute temperature , R¯ the gas constant , and Ea the activation energy for synaptic vesicle fusion ( Figure 1A ) . Since the speed of the reaction is determined by Ea and not by the absolute height of the energy barrier for fusion , we use Ea in the rest of this paper to explain effects on release kinetics . The prefactor A is an empirical prefactor that takes into account the probability of collisions between reactants . For reactions in which the activation energy is low , this factor can limit release rates ( diffusion limited reactions ) . Since SV fusion from the RRP proceeds from primed states where reactants are already positioned in close proximity and since fusion involves high-energy intermediate states , we assume that SV-release rates are predominantly governed by the activation energy and not by the number of collisions . Hence , we assume that changes in release rates most likely reflect changes in Ea with constant A . In that case , if the activation energy for fusion at rest Ea , 0 is reduced by an amount ΔE1 ( Figure 1C ) , the corresponding new release rate constant is given by ( 2 ) k2 , new=Ae− ( Ea , 0−ΔE1 ) R¯T=Ae−Ea , 0R¯TeΔE1R¯T=k2 , 0·m1 , with m1=eΔE1R¯T a multiplication factor and k2 , 0=Ae−Ea , 0R¯T the rate constant for the Ca2+-independent part of spontaneous release ( Xu et al . , 2009; Ermolyuk et al . , 2013 ) . Similarly , a further reduction of the activation energy with an amount ΔE2 by a second ( independent ) process ( Figure 1D ) leads to multiplication of the fusion-rate constant with an additional multiplication factor m2=eΔE2R¯T , ( 3 ) k2 , new=Ae− ( Ea , 0−ΔE1−ΔE2 ) R¯T=k2 , 0·m1·m2 . 10 . 7554/eLife . 05531 . 003Figure 1 . Supralinear modulation of synaptic efficacy through additive effects on the activation energy for fusion . ( A ) Schematic of the energy landscape for synaptic vesicle priming and fusion , with Ea the activation energy for vesicle fusion , and ( B ) the corresponding vesicle-state model . ( C ) Reduction of the fusion activation energy at rest Ea , 0 by an amount ΔE1 , or ( D ) by a combined effect of ΔE1 and ΔE2 . ( E ) Additive effect of ΔE2 causes a constant shift of the effective activation energy for fusion ΔEa for different values of ΔE1 , but a ( F ) multiplicative effect on the release rate constant k2 . DOI: http://dx . doi . org/10 . 7554/eLife . 05531 . 003 This generalizes to ( 4 ) k2 , new=Ae− ( Ea , 0−∑i=1NΔEi ) R¯T=k2 , 0·∏i=1Nmi , for N independent reductions ΔEi ( −ΔEi for enhancements ) of the activation energy with corresponding multiplication factors mi=eΔEiR¯T . Equation ( 4 ) implies that additive effects on the activation energy for SV fusion result in multiplicative effects on the fusion rate ( Figure 1E , F ) , which renders it a powerful way to modulate synaptic strength . In comparison , additive effects on the number of readily releasable vesicles cause additive effects on the fusion rate . We developed a method to quantify fusion rate constants from synaptic responses to hypertonic stimulation and tested whether osmotic , genetic , and biochemical perturbations modulate synaptic vesicle fusion rate in a supralinear manner . Exposing neurons to hypertonic solution induces vesicle fusion selectively from the readily releasable pool ( primed state ) ( Rosenmund and Stevens , 1996 ) . This occurs by a mechanism that is not mediated by Ca2+ , as hypertonic sucrose ( HS ) -induced excitatory postsynaptic currents ( EPSCs ) are not changed when intracellular Ca2+ is buffered by BAPTA , or when Ca2+ influx through voltage gated Ca2+ channels is blocked by CdCl2 ( Rosenmund and Stevens , 1996 ) . HS-induced EPSCs display concentration-dependent changes in release kinetics , with higher degrees of hypertonicity leading to faster release , causing a decrease in time-to-peak and an increase in peak release rate ( Basu et al . , 2007 ) ( Figure 2A ) . We applied a minimal vesicle state model , similar to Weis et al . ( 1999 ) ( Figure 1B ) , and extended this with a time dependent description of the sucrose action on the release rate constant ( Figure 2B , see ‘Materials and methods’ for mathematical description ) to describe these release kinetics at various sucrose concentrations . EPSCs were simulated by modelling sucrose induced SV release rates and convolving them with a canonical miniature EPSC ( see ‘Materials and methods’ ) . We found that—by varying only the maximal fusion rate constant k2 , max—our model reproduced all features in the experimental traces: a decrease in time-to-peak , an increase in peak release rate , and more release for increasing sucrose concentrations ( Figure 2B–C ) ( Figure 2—source data 1 ) . Above a given stimulus strength ( 0 . 5M sucrose in WT cells ) , the total amount of release remained constant , because the complete RRP was depleted , but peaks became larger and narrower when k2 , max kept increasing . Latter features were also present in a reduced version of the model that neglects vesicle replenishment , which could be solved analytically ( Figure 2—figure supplement 1 ) . Hence , selective modulation of the fusion rate constant by HS stimulation in a simple vesicle state model is sufficient to describe characteristic features of synaptic responses to different levels of hypertonicity . 10 . 7554/eLife . 05531 . 004Figure 2 . Modelling HS-induced EPSCs . ( A ) Concentration dependence of HS-induced release kinetics . ( B ) Model simulations of time courses of k2 , for different values of k2 , max and ( C ) corresponding synaptic responses ( −k2R ) . DOI: http://dx . doi . org/10 . 7554/eLife . 05531 . 00410 . 7554/eLife . 05531 . 005Figure 2—source data 1 . Parameter values for Figure 2—figure supplements 1 and , 3 . DOI: http://dx . doi . org/10 . 7554/eLife . 05531 . 00510 . 7554/eLife . 05531 . 006Figure 2—figure supplement 1 . Analytical solution for hypertonic sucrose-induced release from a RRP without replenishment . Current responses obtained from Equation ( 14 ) after convolution with a typical mEPSC . The magenta line corresponds to k2 , max = 0 . 5 s−1 , blue to k2 , max = 3 s−1 , red to k2 , max = 5 s−1 , and black to k2 , max = 10 s−1 . DOI: http://dx . doi . org/10 . 7554/eLife . 05531 . 00610 . 7554/eLife . 05531 . 007Figure 2—figure supplement 2 . Open tip experiments show rapid solution exchange . Solution exchange was measured by the change in holding current when switching from normal ( 0 . 3M ) extracellular solution to 10 times diluted ( 0 . 03M ) extracellular solution with 0 . 5 or 1M sucrose . Green curves are the average responses for 6 recordings , corrected for baseline and inverted for displaying purposes . Blue curves represent postsynaptic current responses to different sucrose concentrations which show a delayed response with respect to the sucrose stimulus . DOI: http://dx . doi . org/10 . 7554/eLife . 05531 . 00710 . 7554/eLife . 05531 . 008Figure 2—figure supplement 3 . Effect of different model parameters on simulated HS-induced EPSCs . The default parameter set , represented by the black traces , is [k1 , k−1 , k2 , max , tdel , τ , D]=[0 . 09 , 0 . 16 , 3 . 5 , 0 . 60 , 0 . 20 , 1000] . In each subpanel , one of these parameters is either multiplied by 2 ( dark blue ) or divided by 2 ( light blue ) . The Gaussian white noise added to these curves was generated using the MATLAB ‘randn ( ) ’ function , with µ = 0 pA and σ = 10 pA . ( A ) Absolute traces . ( B ) Traces scaled and aligned to peak . DOI: http://dx . doi . org/10 . 7554/eLife . 05531 . 008 Next , we set out to fit HS-induced responses with our vesicle state model to assess synaptic release parameters including RRP , and rate constants for priming , unpriming , and fusion . Cultured autaptic neurons between DIV13-18 were challenged with HS concentrations ranging from 0 . 25–1M using a fast application system to establish a rapid transition from normal extracellular solution to hypertonic solution . In addition , spontaneous release was measured before cells were exposed to HS to quantify the release rate at 0M sucrose . The model accurately fitted synaptic responses induced by RRP depleting concentrations of 0 . 5M and higher , providing estimates for all model parameters ( Rosenmund and Stevens , 1996; Basu et al . , 2007 ) ( Figure 3A–C and Figure 3—figure supplement 1 ) . For 0 . 5M , we found a priming rate k1D of 0 . 132 ± 0 . 031 nC/s , which corresponded to 0 . 10 pool-units/s given an average pool size of 1 . 31 nC ( see below ) and was of the same order of magnitude as the 0 . 20 ± 0 . 03 pool-units/s at 25°C reported by Pyott et al . ( 2002 ) . The unpriming rate constant k−1 at 0 . 5M was 0 . 11 ± 0 . 01 s−1 , corresponding to a RRP recovery time constant of 1/k−1 = 9 . 1 s ( see Equation ( 21 ) , ‘Materials and methods’ ) , which was of the same order of magnitude as recovery time constants reported in previous studies ( 10 s at 36°C ( Stevens and Tsujimoto , 1995 ) , 2 . 9 s at 32°C ( Toonen et al . , 2006 ) , and 13 s ( slow phase ) at 25°C ( Pyott and Rosenmund , 2002 ) ) . Priming and unpriming rates were not significantly different between different concentrations suggesting that these processes are not affected by hypertonic stimulation ( Figure 3—figure supplement 1 ) . We used estimations of the priming and unpriming parameters k1D , and k−1 to calculate RRP size from the steady state solution of the model given by Equation ( 9 ) , neglecting the value of k2 before stimulation , which is three orders of magnitude smaller than k−1 ( compare Figure 3C and Figure 3—figure supplement 1B , Figure 3—source data 1 ) . For stimulation with 0 . 5M , this yielded a RRP of 1 . 31 ± 0 . 23 nC , corresponding to 11 . 9 ± 2 . 4·103 ( n = 12 ) vesicles , which was in the same range as reported for wild-type autaptic neurons by other studies ( 15 . 9 ± 2 . 9·103 ( Altrock et al . , 2003 ) , 2 . 5 ± 1 . 1·103 ( Augustin et al . , 1999 ) , 5 . 36 ± 0 . 87·103 ( Priller et al . , 2006 ) , 24 . 7 ± 5 . 6·103 ( Priller et al . , 2007 ) , 17 . 2 ± 3 . 0·103 ( Priller et al . , 2009 ) , 6 . 35 ± 0 . 9·103 ( Reim et al . , 2001 ) , 11 . 0 ± 1 . 2·103 ( Rhee et al . , 2002 ) ) . RRP sizes were similar for the RRP depleting concentrations of 0 . 5M and higher ( Figure 3B ) . Our fit method yielded a more accurate estimate of the RRP size compared to quantification methods that use the charge transfer during the peak of the sucrose response and need to correct for on-going vesicle replenishment , either by subtracting the steady state current at the end of the response as a baseline ( Basu et al . , 2007; Arancillo et al . , 2013 ) ( Figure 3—figure supplement 2A ) , or by integrating the current to an arbitrary time-point after the peak ( Reim et al . , 2001; Rosenmund et al . , 2002; Toonen et al . , 2006; Ikeda and Bekkers , 2009 ) ( Figure 3—figure supplement 2B ) . In addition , the rate constant for vesicle replenishment k1 is one of the fitted model parameters , which allows the reconstruction of vesicle recruitment during sucrose application ( see ‘Materials and methods’ and Figure 3—figure supplement 2C ) . We noticed that responses to 1M sucrose tended to have lower noise levels ( Figure 3A1 ) , which might point to an effect of receptor saturation and/or desensitization that was shown to be absent at 0 . 5M ( Pyott and Rosenmund , 2002 ) but might play a role at higher concentrations . We confirmed that kinetics of responses to 0 . 5M were identical in the absence or presence of competitive AMPA receptor antagonist kynurenic acid ( KYN ) , but found faster kinetics of 0 . 75M responses in the presence of KYN , suggesting that quantifications of model parameters obtained for concentrations higher than 0 . 5M should be interpreted with caution ( Figure 3—figure supplement 3 ) . 10 . 7554/eLife . 05531 . 009Figure 3 . Probing the energy barrier for synaptic vesicle fusion . ( A1 ) HS induced EPSCs ( black ) with model fits ( red ) superimposed . ( A2 ) Spontaneous vesicle release at 0M sucrose . ( B ) RRP size obtained from model fits using Equation ( 9 ) . ( C ) Fitted maximal release rate constants k2 , max at different sucrose concentrations . ( D ) Changes in activation energy ( at 293 K ) obtained from values for k2 , max in C using Equation ( 5 ) . Data for 0 . 25M and higher were fitted with a monoexponential function , which was transformed into the dose–response curve in C using the equations given in Figure 3—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 05531 . 00910 . 7554/eLife . 05531 . 010Figure 3—source data 1 . Parameter values for Figure 3B–D , bootstrap analysis Figure 3 , Figure 3—figure supplement 1A–C , bootstrap analysis Figure 3—figure supplement 1 , Figure 3—figure supplement 3B–E , and Figure 3—figure supplement 4B–E . DOI: http://dx . doi . org/10 . 7554/eLife . 05531 . 01010 . 7554/eLife . 05531 . 011Figure 3—source data 2 . Parameter values for Figure 3—figure supplement 5A and , C . DOI: http://dx . doi . org/10 . 7554/eLife . 05531 . 01110 . 7554/eLife . 05531 . 012Figure 3—figure supplement 1 . Higher concentrations of hypertonic do not significantly affect upstream parameters but reduce the delay of sucrose action onset with respect to time of switching of the application barrel . ( A ) Priming rate k1D , ( B ) Unpriming rate constant k−1 , and ( C ) Delay of sucrose onset , tdel . DOI: http://dx . doi . org/10 . 7554/eLife . 05531 . 01210 . 7554/eLife . 05531 . 013Figure 3—figure supplement 2 . Different methods to estimate RRP size from HS responses . Red line represents a typical current response in a . u . induced by hypertonic stimulation . ( A ) HS induced current response is corrected for vesicle replenishment by taking the steady state current at the end of the response as baseline and subtracting this from the total current . Integration of the corrected current response yields the RRP size in nC , or in vesicles , after dividing total charge by the quantal content of a single mEPSC ( green area ) ( Basu et al . , 2007; Arancillo et al . , 2013 ) . This gives an underestimation of the RRP since vesicle replenishment does not start at the maximal rate at the onset of the response but grows gradually during the stimulation . ( B ) RRP size is estimated from integration of the total charge transfer from the beginning of the response to an arbitrary timepoint after the peak ( green area ) , neglecting any contribution from vesicle replenishment ( grey area ) ( Reim et al . , 2001; Rosenmund et al . , 2002; Toonen et al . , 2006; Ikeda and Bekkers , 2009 ) . This usually leads to an overestimation . ( C ) In this paper , the definition of the steady state RRP in Equation ( 9 ) is used to infer the RRP size from the fitted model parameters . Effectively , in comparison to methods shown in A and B , we correct for vesicle replenishment by subtracting the calculated vesicle replenishment using Equation ( 20 ) ( black line ) from the total current . Integration of the corrected HS induced current response yields an accurate estimation of the RRP ( green area ) . DOI: http://dx . doi . org/10 . 7554/eLife . 05531 . 01310 . 7554/eLife . 05531 . 014Figure 3—figure supplement 3 . Effect of the non-selective glutamate receptor antagonist kynurenic acid ( KYN ) on release kinetics . ( A ) Current traces induced by 0 . 5 or 0 . 75M sucrose in the presence or absence of 0 . 2 mM KYN ( measured in the same neuron ) . Shown are raw and scaled traces . Insets show zoom of 0 . 75M peak . ( B–D ) KYN induced changes in ( B ) release rate constant k2 , max ( C ) RRP size , ( D ) priming rate k1Dk1D , ( E ) unpriming rate constant k−1 . Parameters are obtained from unscaled raw data and normalized to the condition without KYN . Since KYN reduced the measured current , RRP size and priming rates are reduced . The maximal release rate is unaffected in 0 . 5M sucrose , but increased by KYN in 0 . 75M sucrose . This suggests that post-synaptic receptor saturation might play a role in sucrose concentrations of 0 . 75M or higher . DOI: http://dx . doi . org/10 . 7554/eLife . 05531 . 01410 . 7554/eLife . 05531 . 015Figure 3—figure supplement 4 . Subtraction of non-receptor current does not affect fitted model parameters . ( A ) Example trace of postsynaptic response evoked by 0 . 5M sucrose ( black ) . Green trace is corrected for the non-receptor current induced by 0 . 5M in the presence of AMPA and NMDA blockers DNQX ( 10 µM ) and APV ( 50 µM ) ( grey ) . ( B ) Priming rate k1D . ( C ) Unpriming rate constant k−1 . ( D ) Release rate constant k2 , max . ( E ) RRP size . DOI: http://dx . doi . org/10 . 7554/eLife . 05531 . 01510 . 7554/eLife . 05531 . 016Figure 3—figure supplement 5 . Fitting HS-induced EPSCs . ( A ) The default parameter set is as in Figure 2—figure supplement 3 . Each panel shows the first 4 s of the simulated trace per parameter setting in black . Traces are overlaid with results of 10 independent fits starting at different initial conditions , shown in red ( best fit ) , green ( accepted fit upon visual inspection ) and grey ( rejected fit upon visual inspection ) . With the exception of the results for 2k−1 , the same scale holds for all curves . ( B ) Key features encircled in red to judge quality of the fit by visual inspection: ( 1 ) Late onset of fit , ( 2 ) wrong peak amplitude and/or time-to-peak , ( 3 ) too slow decay towards steady state phase , ( 4 ) Steady-state phase ( refill ) is fitted incorrectly . ( C ) Fit method robustly discriminates between different model parameters . Graphs display fitted model parameters , obtained from fits approved after visual inspection in ( A ) ( red and green curves ) , as a function of the adapted model parameter . Strong linear correlation is found for the adapted model parameter , whereas the other parameters are invariant . DOI: http://dx . doi . org/10 . 7554/eLife . 05531 . 016 Maximal release rate constants k2 , max were obtained from fits of responses to 0 . 25–1M sucrose . For non-depleting hypertonic stimulation ( e . g . , 0 . 25M ) , k2 , max can be overestimated due to an underestimate of the RRP . Therefore , we fitted such current responses simultaneously with the response to a maximal depleting stimulation ( e . g . , 0 . 5M ) from the same cell , keeping all the model parameters the same between two stimulations , except k2 , max , tdel , and τ . The release rate constant at 0M was obtained by dividing the frequency of spontaneously released events ( mEPSCs ) by the number of vesicles in the RRP ( calculated by dividing the total RRP charge by the average mEPSC charge ) . However , this was probably an overestimation since the majority ( >95% ) of spontaneous release is Ca2+-dependent , and intracellular Ca2+ was not buffered in these experiments ( Xu et al . , 2009; Groffen et al . , 2010 ) . Ca2+-dependent mEPSCs are most likely triggered by rapid spontaneous Ca2+ fluctuations ( SCFs ) in the synaptic terminals , either caused by stochastic opening of voltage gated Ca2+ channels ( ∼50% ) ( Goswami et al . , 2012; Ermolyuk et al . , 2013 ) or release from intracellular calcium stores ( ∼50% ) ( Emptage et al . , 2001 ) . This suggests that the frequency of these SCFs contributes with a constant k2 , SCFs ( ∼2–4 10−4 s−1 ) to the calculated release rate constant k2 , max , which dominates at 0M sucrose but is negligible compared to fusion rate constants induced with higher concentrations ( Figure 3—source data 1 ) . In contrast to the other fitted model parameters , k2 , max was significantly different between different concentrations and showed a sigmoidal dependence on sucrose concentration ( Figure 3C ) . The values for k2 , max at 0 . 75 and 1M might be underestimated due to receptor saturation as discussed above ( Figure 3—figure supplement 3 ) . As we argued above , Ca2+-triggered exocytosis belongs to a class of reactions that are likely to be limited by activation energy , rather than by the frequency of collisions between reactants . This follows from the preassembly of a fusion machinery during vesicle priming , and from the expected existence of high-energy intermediates . During stimulation with hypertonic solution , drawing water from the cell will increase the concentration of reactants . This might increase collision rates proportional with the increased concentration , but this is unlikely to account for the 104-fold increase in k2 , max . Moreover , the ( moderate ) increase in reactant concentration might be counteracted by molecular crowding effects and increases in viscosity ( Miermont et al . , 2013 ) . Consistent with this notion , we observed that upstream steps in the exocytotic cascade , which are in fact more likely to be collision limited ( such as vesicle docking and priming , reflected in the overall priming rate k1D ) , showed a tendency to decrease with high osmolarity ( Figure 3—figure supplement 1 ) , indicating that molecular crowding/viscosity dominates the effect of increased reactant concentration . Overall , we conclude that a HS challenge is most likely to change fusion through a change of the activation energy for fusion ( i . e . , the exponential factor in the Arrhenius equation ) , rather than the pre-exponential factor A . Changes in activation energy for fusion follow from changes in k2 , max using Equation ( 1 ) assuming A is constant , ( 5 ) ΔEa=Ea , 1−Ea , 2=R¯T ( ln ( A ) −ln ( k2 , max , 1 ) ) −R¯T ( ln ( A ) −ln ( k2 , max , 2 ) ) =R¯T ( ln ( k2 , max , 2 ) −ln ( k2 , max , 1 ) ) . Figure 3D depicts the calculated changes in activation energies corresponding to the changes in k2 , max for different sucrose concentrations in Figure 3C . We find that the maximal reduction in the activation energy for fusion by 1M sucrose is 9 . 3R¯T . This value is probably about 3R¯T too low since ( as discussed above ) k2 , max is overestimated at 0M ( up to 20 fold ) , but not at higher sucrose concentrations . Expressed in units of kCal/mol , the HS-induced change in activation energy corresponds to 5 . 4 kCal/mol , which is comparable to the estimated reduction of 5 . 9 kCal/mol during the action potential ( Rhee et al . , 2005 ) . Hence , fusion rate constants obtained from fitting HS-induced synaptic responses to a minimal vesicle-state model can be used to calculate changes in activation energy for fusion , which enables to study this parameter under different experimental conditions . The extent of RRP depletion upon application of submaximal sucrose has been used as a measure of ‘release willingness’ or ‘fusiogenicity’ of vesicles , which is proposed to be inversely related to the energy barrier for fusion ( Basu et al . , 2007; Gerber et al . , 2008; Xue et al . , 2010; Rost et al . , 2011 ) . To investigate whether changes in the activation energy for fusion can explain changes in the depleted RRP fraction at submaximal sucrose , we analyzed the relation between release kinetics ( k2 , max ) and RRP depletion in the model and compared this with experimental data . The depleted RRP fraction was defined as the fraction of the RRP depleted by a submaximal HS stimulus relative to a maximal depleting stimulus ( 0 . 5M sucrose ) . Simulations applying 7 s HS-stimulations for different values of k2 , max yielded a linear relation for low values of k2 , max , which levels off and saturates to 1 ( complete depletion ) at high k2 , max . This relation transforms into a sigmoidal curve when k2 , max is plotted on a log10 scale ( black line in Figure 4B ) and can be approximated by an analytically derived function ( see ‘Materials and methods’ and Figure 4—figure supplement 1 ) ( Figure 4—source data 1 ) . The value for k2 , max , that we experimentally find with 0 . 5M stimulation , predicts only a 94% depletion of the RRP implying that up to 6% more release is expected with higher concentrations . However , in practice , these slightly larger responses might be difficult to detect because of receptor saturation and desensitization effects at these concentrations . We experimentally confirmed the predicted relation with data points from submaximal 0 . 25M responses being distributed along the steep phase of the curve ( Figure 4A , B ) . As expected , 0 . 75 and 1M responses yielded high values for k2 , max and complete RRP depletion . These results show that a change in k2 , max only is sufficient to explain changes in the depleted RRP fraction: with slow release kinetics ( low k2 , max ) , the RRP is not effectively depleted , because of on-going refilling ( priming ) , whereas from a certain value of k2 , max the amount of RRP depletion is maximal , but depletion occurs with faster kinetics . Hence , with this relation the extent of RRP depletion in response to different sucrose concentrations can be used to discriminate between effects on release kinetics and priming . Maximally depleting stimuli report the RRP , while changes in the depleted RRP-fraction at submaximal ( e . g . , 0 . 25M ) stimuli are an indication of changes in k2 , max , indicative of changes in the activation energy for fusion . 10 . 7554/eLife . 05531 . 017Figure 4 . Relation between depleted RRP fraction and release kinetics . ( A ) Examples of submaximal responses in different cells . 0 . 25M responses ( black ) , scaled to 0 . 5M responses ( grey ) in the same cell , display faster kinetics when a larger fraction of the RRP is depleted . ( B ) Fitted data overlayed on the predicted curve . Data points corresponding to the examples in A are indicated . Data points for 0 . 50M , 0 . 75M , and 1 . 0M are shown as mean ± SEM . Note that whereas the model predicts a 94% depletion of the RRP with 0 . 5M the y-axis value at 0 . 5M is one per definition since the RRP size at this concentration was used as a reference to calculate the depleted RRP fraction . DOI: http://dx . doi . org/10 . 7554/eLife . 05531 . 01710 . 7554/eLife . 05531 . 018Figure 4—source data 1 . Parameter values for Figure 4B and Figure 4—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 05531 . 01810 . 7554/eLife . 05531 . 019Figure 4—figure supplement 1 . Comparison of analytical approximation and model predictions of the relation between release kinetics and RRP depletion . For small k2 , max , the duration of the sucrose pulse dictates the depleted RRP fraction: 7 s stimuli deplete a smaller fraction than stimuli of 20 s and longer . For large k2 , max , the blue curve ( D depletable ) exceeds the others , because the steady-state RRP at the end of the stimulus is smaller when D is depletable . This is due to Equation ( 24 ) : Rf = k1Df/ ( k−1 + k2 , max ) . A smaller upstream pool at the end of the stimulus ( Df ) thus yields a smaller Rf and hence a larger depleted RRP fraction ( Ri − Rf ) /Ri . DOI: http://dx . doi . org/10 . 7554/eLife . 05531 . 019 Next , we investigated the additivity between osmotic and genetic or biochemical perturbations on release kinetics and RRP depletion . We extracted data from literature on genetic and/or biochemical perturbations with an effect on the release willingness of vesicles . Interestingly , changes in release willingness were reported for proteins with distinct presynaptic functions , including the priming factor Munc13 , the tSNARE Syntaxin , the SNARE-complex binding protein Complexin , and the metabotropic GABAB receptor ( Basu et al . , 2007; Gerber et al . , 2008; Xue et al . , 2010; Rost et al . , 2011 ) . We retrieved for different types of perturbations , the reported depleted RRP fractions , and corresponding peak release rates , defined as the release rate at the peak of the HS-induced response ( Basu et al . , 2007 ) . Plotting these data points in one graph showed the same non-linear relation between release kinetics and RRP depletion for the four different data sets ( Figure 5 ) . To compare this experimentally observed relation with our model prediction , we simulated sucrose responses for different values of k2 , max , keeping all other parameters constant , and calculated peak release rates and corresponding depleted RRP fractions from the simulated traces in the same way as was done for the experimental traces ( Figure 3—figure supplement 2A ) ( Figure 5—source data 1 ) . The model prediction of the relation between release kinetics and RRP depletion was in good accordance with the experimental data ( Figure 5 ) . Hence , this non-linear dependence can be explained by changes in the release rate constant k2 , max only . 10 . 7554/eLife . 05531 . 020Figure 5 . Model predicts relation between peak release rate , defined as the release rate at the peak of a HS-induced response , and depleted RRP fraction for different combinations of HS stimulations and genetic or biochemical manipulations of the activation energy for fusion . Data are taken from ( Basu et al . , 2007; Gerber et al . , 2008; Xue et al . , 2010; Rost et al . , 2011 ) Model prediction is obtained from peak release rates and depleted RRP fractions extracted from model simulations where parameter k2 , max is varied keeping other model parameters constant . Note that beyond 0 . 5M the predicted curve and some data points overshoot the value of one because 0 . 5M was used as a reference to calculate the depleted RRP fraction at the other concentrations , assuming complete depletion at 0 . 5M , whereas the model predicts only 94% depletion at this point . DOI: http://dx . doi . org/10 . 7554/eLife . 05531 . 02010 . 7554/eLife . 05531 . 021Figure 5—source data 1 . Parameter values for Figure 5 . DOI: http://dx . doi . org/10 . 7554/eLife . 05531 . 021 Next , we tested whether these biochemical and genetic perturbations modulate release kinetics in a supralinear manner , measuring release rate constants at different sucrose concentrations between 0 and 0 . 5M to avoid effects of receptor saturation and desensitization . Phorbol ester is known to potentiate synaptic release in a number of systems ( Searl and Silinsky , 1998; Rhee et al . , 2002; Basu et al . , 2007; Wierda et al . , 2007; Lou et al . , 2008 ) . First , we recorded spontaneous release and responses to 0 . 2–0 . 5M hypertonic stimulations , before and after PDBu application ( 1 μM ) ( Figure 6—figure supplements 1 , 2 ) . We observed potentiation of the spontaneous release and submaximal ( 0 . 2–0 . 4M ) responses as well as faster kinetics for the 0 . 5M response , but no effect on RRP size or priming and unpriming rate constants ( Figure 6A , Figure 6—figure supplement 3 ) . When comparing the effect of PDBu on release kinetics between different sucrose concentrations , indeed a supralinear increase in k2 , max was found , with the increase in k2 , max being three orders of magnitude larger at 0 . 5M than at 0M ( Figure 6B , Figure 6—source data 1 ) . Next , we calculated the activation energies from the changes in k2 , max , using Equation ( 5 ) , which were reduced with a similar ΔEa for all sucrose concentrations ( Figure 6C , Figure 6—source data 1 ) . This multiplicative effect on release kinetics , but additive effect in the activation energy domain , became more evident when absolute changes in these variables were plotted , with an exponential increase in k2 , max and a ∼−0 . 3 R¯T shift in the fusion-activation energy for 0 . 2–0 . 5M sucrose ( Figure 6D–E ) . The almost twofold higher decrease at 0M was probably an overestimation because of the increased sensitivity to spontaneous Ca2+ fluctuations after PDBu , which will increase the contribution of k2 , SCFs to k2 , max , again dominating k2 , max at 0M but being negligible at higher concentrations . 10 . 7554/eLife . 05531 . 022Figure 6 . Additive effect on the activation energy for fusion induced by PDBu causes supralinear effect on release kinetics . ( A ) Current traces , ( B ) release rate constants k2 , max , and ( C ) activation energies for fusion at different sucrose concentrations in the absence and presence of PDBu . PDBu-induced changes in k2 , max and ΔEa , obtained by subtraction of the data curves in B and C before and after PDBu application , show ( D ) an exponential increase in k2 , max for increasing sucrose concentrations whereas ( E ) the changes in the energy domain are in the same order of magnitude ( reduction at 0M is probably an overestimation due to Ca2+ depenence of the spontaneous release , [see text] ) . Mean values of k2 , max displayed are all within the 95% confidence interval as determined by Bootstrap analysis . DOI: http://dx . doi . org/10 . 7554/eLife . 05531 . 02210 . 7554/eLife . 05531 . 023Figure 6—source data 1 . Parameter values for Figure 6B–E , bootstrap analysis Figure 6 , Figure 6—figure supplement 3A–D , and Figure 6—figure supplement 3 . DOI: http://dx . doi . org/10 . 7554/eLife . 05531 . 02310 . 7554/eLife . 05531 . 024Figure 6—figure supplement 1 . Random examples of individual HS-evoked EPSCs ( black ) in the absence of PDBu , overlaid with their best fit ( red ) . ( A ) Responses to 0 . 5M . ( B ) Responses to 0 . 3M . DOI: http://dx . doi . org/10 . 7554/eLife . 05531 . 02410 . 7554/eLife . 05531 . 025Figure 6—figure supplement 2 . Random examples of individual HS-evoked EPSCs ( blue ) in the presence of PDBu , overlaid with their best fit ( red ) . ( A ) Responses to 0 . 5M . ( B ) Responses to 0 . 3M . DOI: http://dx . doi . org/10 . 7554/eLife . 05531 . 02510 . 7554/eLife . 05531 . 026Figure 6—figure supplement 3 . Upstream parameters and RRP size are not affected by PDBu application . ( A ) Priming rate k1D . ( B ) Unpriming rate constant k−1 . ( C ) RRP size . ( D ) Relation between k2 , max and depleted RRP is maintained in the presence of PDBu , but synaptic responses to submaximal HS-stimulation display faster kinetics and more RRP depletion . DOI: http://dx . doi . org/10 . 7554/eLife . 05531 . 026 Next , we reanalysed the raw responses to 0 , 0 . 25 , and 0 . 5M sucrose in complexinI/II deficient neurons and their controls from a study by Xue et al . ( 2010 ) . Whereas responses to 0 . 5M did not differ in released RRP size , and priming and unpriming were not affected ( Figure 7A , Figure 7—figure supplement 1 ) , a markedly reduced fraction of the RRP was released by 0 . 25M stimuli in the null mutants , suggesting an increased activation energy for fusion in the absence of complexins . Indeed , release kinetics were slowed down as predicted by the relation between k2 , max and depleted RRP fraction ( Figure 7A , Figure 7—figure supplement 1D ) . This effect of complexin deletion on release kinetics was supralinear with an eightfold larger reduction of k2 , max at 0 . 5M than at 0 . 25M , whereas the corresponding activation energies shifted with 0 . 4 and 0 . 8 R¯T at these concentrations ( Figure 7B–E ) . The overall supralinearity is in line with an activating role of complexin in exocytosis by a reduction of the activation energy for fusion ( Figure 7B–C , Figure 7—source data 1 ) . However , the reduction of the activation energy was less at 0M , and also seemed less at 0 . 5M than at 0 . 25M ( Figure 7E ) , possibly indicating that complexins exert several effects , for instance clamping a secondary Ca2+ sensor for spontaneous and asynchronous release , rendering the synapse more sensitive to spontaneous Ca2+ fluctuations ( Yang et al . , 2010; Ermolyuk et al . , 2013 ) . Another possibility is that complexin also affects the frequency factor , for example , because the absence of complexin changes the cooperativity of exocytosis . 10 . 7554/eLife . 05531 . 027Figure 7 . Additive effect on the activation energy for fusion induced by Cpx deletion causes supralinear effect on release kinetics . ( A ) Current traces , ( B ) release rate constants k2 , max , and ( C ) fusion energy barrier heights at different sucrose concentrations for control and CpxKO cells . Cpx deletion-induced changes in k2 , max and ΔEa , obtained by subtraction of the data curves for control and CpxKO in B and C , show ( D ) an exponential increase in k2 , max for increasing sucrose concentrations whereas ( E ) the changes in the energy domain are in the same order of magnitude . Mean values of k2 , max displayed are all within the 95% confidence interval as determined by Bootstrap analysis . Cpx data were published before in ( Xue et al . , 2010 ) and reanalysed here . DOI: http://dx . doi . org/10 . 7554/eLife . 05531 . 02710 . 7554/eLife . 05531 . 028Figure 7—source data 1 . Parameter values for , bootstrap analysis Figure 7B–E , bootstrap analysis Figure 7 , Figure 7—figure supplement 1A–D , and bootstrap analysis Figure 7—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 05531 . 02810 . 7554/eLife . 05531 . 029Figure 7—figure supplement 1 . Upstream parameters and RRP size are not affected in Cpx KO . ( A ) Priming rate k1D . ( B ) Unpriming rate constant k−1 . ( C ) RRP size . ( D ) Relation between k2 , max and depleted RRP is maintained in Cpx KO synapses , but synaptic responses to submaximal HS-stimulation display slower kinetics and less RRP depletion . DOI: http://dx . doi . org/10 . 7554/eLife . 05531 . 029
Exploiting the kinetic model presented here to assess essential release parameters like RRP-size and fusion kinetics from HS-induced responses has advantages over existing methods . Firstly , this model uses the steady state solution ( Equation ( 9 ) ) to calculate the RRP size . This circumvents the necessity to correct post-hoc for RRP replenishment during the stimulus as in other RRP estimation methods ( Schneggenburger et al . , 1999; Moulder and Mennerick , 2005 ) ( Figure 3—figure supplement 2A , B ) . Secondly , the relation between release kinetics and RRP depletion can be used to predict changes in k2 , max from changes in the depleted RRP fraction . This makes it possible to discriminate between changes in the activation energy ( indicated by changes in the depleted RRP fraction tested with submaximal HS stimuli ( Xue et al . , 2010; Arancillo et al . , 2013 ) ) and priming effects ( indicated by changes in the response to maximal depleting HS stimuli ) . An important consequence is that in situations where the activation energy is increased ( e . g . , by genetic deletion of a gene that reduces the energy barrier for fusion ) , 0 . 5M sucrose might not be enough to fully deplete the RRP . This could be erroneously interpreted as a priming defect . Thirdly , our model also quantifies priming- and unpriming-rate constants ( k1 and k−1 ) , which for instance allows reconstruction of the time course of replenishment during HS stimulation at resting Ca2+ levels . Finally , all model parameters mentioned above are quantified using a Ca2+-independent stimulus , which to a large extent excludes differences in Ca2+ signalling or Ca2+ sensitivity as confounding factors . Since activation energies cannot be directly measured in synapses , we used the Arrhenius equation to infer these from HS-induced release rate constants . Four arguments suggest that the effect of hypertonic solution ( HS ) on synaptic release is primarily due to a reduction in activation energy , and not by an increase in the number of collisions as a result of shrinkage ( accounted for by the Arrhenius pre-exponential factor A ) . First , exocytosis is expected to take place via a sequence of high-energy intermediates , together determining the activation energy for fusion ( see ‘Discussion’ below ) . Therefore , modulation of the fusion activation energy is a plausible efficient route to regulate vesicle fusion . Second , HS specifically releases primed vesicles ( Rosenmund and Stevens , 1996 ) , which are bound to the plasma membrane with the fusion machinery preassembled . Thus , fusion is unlikely to be diffusion limited . Third , rapid cell shrinking can have opposite effects on the number of collisions , which are expected to affect priming/unpriming and fusion rates similarly . It can either increase the collision frequency due to an increase in the concentrations of reactants or ( given the already high protein concentrations in synapses ( Wilhelm et al . , 2014 ) ) decrease collision frequency because of molecular crowding and viscocity effects ( Miermont et al . , 2013 ) . Since upstream docking/priming steps displayed a trend towards a decrease upon higher HS application , molecular crowding seems to offset any effect on reactant concentration , and therefore , the drastic increase in fusion rate cannot be attributed to A via an increased collision rate . Finally , the reduction in activation energy identified here ( 6 . 1 R¯T for 0 . 25M ) ( Figure 3D ) is comparable to the reduction expected by HS stimulation ( 0 . 2M ) of liposome fusion on theoretical grounds ( ∼7 R¯T ( Malinin and Lentz , 2004 ) ) . Nevertheless , manipulations that change the pre-exponential factor will also contribute to changes in the fusion rate of vesicles in the presence of HS . Many factors influence synaptic release probability , such as RRP size , modulation of Ca2+-and K+-channel properties , Ca2+-buffering/diffusion , and the sensitivity of Ca2+ sensors ( Neher and Sakaba , 2008; Fioravante and Regehr , 2011 ) . Changes in the activation energy are suggested to affect release probability by rendering vesicles more/less fusogenic ( Basu et al . , 2007; Wierda et al . , 2007; Gerber et al . , 2008; Xue et al . , 2010 ) . This is a powerful way to regulate synaptic transmission because of its exponential effect on the fusion rate , whereas RRP size modulation affects synaptic transmission in a proportional fashion ( Sakaba and Neher , 2001; Rhee et al . , 2002; Lipstein et al . , 2013; Walter et al . , 2013 ) . A well-studied example is the facilitatory effect of diacylglycerol ( DAG ) analogues such as phorbol esters on AP induced release . DAG activates two interdependent pathways: direct activation of Munc13 via its C1 domain and PKC dependent phosphorylation of Munc18 . Together , these events reduce the energy barrier for fusion , potentiate vesicular release probability after high frequency stimulation , and produce faster synaptic depression ( Rhee et al . , 2002; Basu et al . , 2007; Wierda et al . , 2007; Garcia-Perez and Wesseling , 2008; de Jong and Verhage , 2009; Genc et al . , 2014 ) . Other presynaptic proteins may also contribute to activation energy reductions ( Gerber et al . , 2008; Weber et al . , 2010; Xue et al . , 2010; Rost et al . , 2011 ) . This suggests that there are either multiple ways by which proteins can modulate the activation energy for fusion or that they all converge onto the same process ( e . g . , SNARE formation/stabilization ) controlling the activation energy . Interestingly , a model of additive modulation of the activation energy implies that molecules can exert their effect independently and do not necessarily need to interact physically to produce complex supra-linear effects on synaptic transmission . Ca2+ controls vesicle fusion in a cooperative fashion ( Dodge and Rahamimoff , 1967 ) . This has been extensively studied in the Calyx of Held showing that a 3 orders of magnitude increase in Ca2+ give rise to a 6 orders of magnitude increase in the vesicle fusion rate ( Schneggenburger and Neher , 2000; Lou et al . , 2005; Neher and Sakaba , 2008 ) . This supra-linear relationship can be well described by a phenomenological model for ‘allosteric’ modulation of the presynaptic Ca2+ sensor ( Lou et al . , 2005 ) , which captures the low cooperativity ( <1 ) for triggering vesicle fusion at basal Ca2+ and high Ca2+ cooperativity ( ∼4 ) at Ca2+ concentrations beyond 5 μM ( Figure 8A ) . However , we note that the exact same model follows from Equation ( 4 ) when assuming that the Ca2+ sensor reduces the activation energy with an amount ΔECa for each Ca2+-ion binding . In this model ( as in the previous model ( Lou et al . , 2005 ) ) , a vesicle can be in one of six different states depending on how much Ca2+ ions are bound to the Ca2+ sensor associated with the vesicle . From each state , release will occur with a specific fusion rate constant ( 6 ) k2 , n=l+fn , with l+ = k2 , 0 the basal fusion rate constant , f=eΔECaR¯T a multiplication factor , and n the number of Ca2+ ions bound to the Ca2+ sensor ( Figure 8B ) . In line with our findings here , the fusion promoting effect of PDBu , described in Lou et al . by the increase of the spontaneous release rate constant l+ ( Lou et al . , 2005 ) , corresponds to a ΔEPDBu reduction of the activation energy resulting in a new rate constant l+ , new=l+eΔEPDBuR¯T . 10 . 7554/eLife . 05531 . 030Figure 8 . Supralinear Ca2+ dependency of release can be explained by additive modulation of the activation energy for fusion by the Ca2+ sensor . ( A ) Non-linear relation between Ca2+ and release rate in the Calyx of Held as predicted by the allosteric model of Lou et al . ( 2005 ) . Allosteric model with 6 different vesicle states ( V , VCa , · · · , V5Ca ) is depicted in inset . ( B ) . Reinterpretation of this allosteric model in terms of additive effects on the activation energy of the binding of Ca2+ to the Ca2+ sensor: each Ca2+ ion that binds reduces the activation energy Ea , 0 by an amount ΔECa . From Equation ( 4 ) it follows that for each vesicle state the release rate constant krelease is given by Equation ( 6 ) , with l+=Ae−Ea , 0R¯T the spontaneous release rate constant and f=eΔECaR¯T a multiplication factor . This is mathematically equivalent to the release rate constants depicted for the different vesicle states in the allosteric model in A and thus yields the same prediction of the non-linear relation between Ca2+ and release rate . DOI: http://dx . doi . org/10 . 7554/eLife . 05531 . 030 All together , this suggests that the Ca2+ sensor modulates fusion supralinearly through additive effects on the fusion activation energy . As a consequence , other factors ( such as PDBu ) do not necessarily need to interact directly with the sensor to modulate the Ca2+ sensitivity of release , but can exert their effect on the activation energy independently . Membrane fusion is a complex process assumed to proceed via a stalk intermediate , with many steps contributing to the activation energy for fusion ( Jahn and Grubmuller , 2002; Kozlovsky and Kozlov , 2002 ) . A state immediately preceding stalk formation may consist of ‘splayed’ lipids , which have left their native leaflet and form a high-energy intermediate ( Risselada and Grubmuller , 2012 ) . Formation and zippering of the SNARE-complex allows the membranes to approach closely ( Lindau et al . , 2012 ) and might also induce or support lipid splaying directly along the linker regions of syntaxin and synaptobrevin/VAMP ( Risselada et al . , 2011 ) . Molecular changes in these proteins , changes in their number or stoichiometry , and/or association/dissociation of additional factors such as complexins , Munc13 , or Munc18 may all lower the activation energy ( Gerber et al . , 2008; Li et al . , 2011; Ma et al . , 2011 ) . Whether or not SNARE-complexes are already ( partly ) assembled at the time when APs open Ca2+ channels is a matter of intense debate ( Jahn and Fasshauer , 2012 ) . The energy released during the formation of a SNARE-complex has been estimated to range between 20 and 35 R¯T ( Mohrmann and Sorensen , 2012 ) , which is 2–3 times higher than what we find for 1M sucrose . However , in case , SNARE-complexes are partly preassembled , only part of the estimated energy would become available for fusion when HS would promote full assembly ( see review ( Sorensen , 2009 ) ) . Furthermore , the similar values of HS-induced reduction in activation energy , identified here and in a theoretical study of protein-free liposome fusion ( Malinin and Lentz , 2004 ) , indicate that the effect of hypertonicity might be on the lipids themselves , by helping to fill energetically expensive ‘voids’ that form during fusion ( Malinin and Lentz , 2004 ) . If this is the case , several other molecules might act in similar ways , including Ca2+-bound synaptotagmin and SNAREs , and several accessory proteins that also interact directly with lipids ( Seiler et al . , 2009; Shin et al . , 2010 ) . The actions of a small number of accessory proteins like complexin , Munc13 , CAPS , and Munc18 , and the proposed stoichiometry of SNARE-complexes per vesicle ( Sinha et al . , 2011; van den Bogaart et al . , 2010; Mohrmann et al . , 2010 ) provide all the necessary input for molecular-dynamic models ( Lindau et al . , 2012 ) to resolve the exact nature of the synaptic vesicle fusion process . Kinetic analysis of HS induced synaptic responses will be highly instrumental to test predictions from such models .
Autaptic hippocampal neurons from wild-type mice were grown for 13–18 days on glia island cultures before measuring . Whole-cell voltage-clamp recordings ( Vm = −70 mV ) were performed at room temperature ( 20–24°C ) with borosilicate glass pipettes ( 2 . 5–4 . 5 MOhm ) filled with 125 mM K+-gluconic acid , 10 mM NaCl , 4 . 6 mM MgCl2 , 4 mM K2-ATP , 15 mM creatine phosphate , 10U/ml phosphocreatine kinase , and 1 mM EGTA ( pH 7 . 30 ) . External solution contained the following ( in mM ) :10 HEPES , 10 glucose , 140 NaCl , 2 . 4 KCl , 4 MgCl2 , and 4 CaCl2 ( pH = 7 . 30 , 300 mOsmol ) . Recordings were acquired with an Axopatch 200A amplifier ( Molecular Devices , Sunnyvale CA ) , Digidata 1322A , and Clampex 9 . 0 software ( Molecular Devices ) . After whole cell mode was established , only cells with a leak current of <250 pA were accepted for analysis . Ca2+-independent vesicle release was evoked by hypertonic solutions consisting of external solution containing 0 . 25 , 0 . 5 , 0 . 75 , or 1M sucrose . Gravity infused external solution was alternated with 7 s of perfusion with hypertonic solution by rapidly switching between barrels within a custom-made tubing system ( FSS standard polyamine coated fused silica capillary tubing , ID 430 µm , OD550 µm , Postnova analytics , Landsberg am Lech , Germany ) attached to a perfusion Fast-Step delivery system ( SF-77B , Warner instruments corporation , Hamden CT ) and directed at the neuron . Solution flow was controlled with an Exadrop precision flow rate regulator ( B Braun , Melsungen , Germany ) to assure all sucrose solutions flowed with a rate of 0 . 5 ml/min irrespective of differences in viscosity . Using this system , solution exchange was complete within 0 . 4 s as measured by the change in holding current after switching from normal ( 0 . 3M ) to 10 times diluted ( 0 . 03M ) extracellular solution containing 0 . 5 or 1M sucrose in an open-tip experiment ( Figure 2—figure supplement 2 ) . Therefore , solution exchange can be considered instantaneous compared to the induced postsynaptic currents , which respond with a delay of 1 . 1 ( 1M ) –1 . 6 s ( 0 . 25M ) ( Figure 3—figure supplement 1C ) . Multiple sucrose solutions with various concentrations were applied to the same cell , taking a 1–2 min rest period in between solutions to accommodate complete recovery of RRP size . In between protocols , a constant flow of external solution was applied to the cells . For PDBu experiments , sucrose applications were performed as usual , after which neurons were incubated with 1 µM PDBu ( Merck Millipore , Darmstadt , Germany ) , and sucrose applications were repeated . The order of sucrose solutions was alternated between neurons to avoid systematic errors due to possible rundown of RRP size after multiple applications . Other sources for systematic errors were investigated and , when experimentally assessable , found to be small for 0 . 5M and lower: sucrose responses were compared in the absence and presence of 0 . 2 mM kynurenic acid ( Sigma , St . Louis MO ) , and no effect of receptor saturation on release kinetics was found for sucrose concentrations of 0 . 5M ( Figure 3—figure supplement 3 ) . Receptor desensitization did not affect RRP size measurements with 0 . 5M sucrose in a previous study ( Pyott and Rosenmund , 2002 ) . However , we could not investigate its effect on release kinetics , since cyclothiazide ( CTZ ) , next to blocking AMPA receptor desensitization , also stimulates the presynaptic release machinery ( Diamond and Jahr , 1995; Bellingham and Walmsley , 1999; Ishikawa and Takahashi , 2001 ) . We did not detect any contribution of HS-induced non-receptor currents , since subtracting the small current remaining after blocking NMDA and AMPA currents by 50 μM AP5 ( Ascent ) and 10 μM DNQX ( Tocris , Bristol , UK ) had a negligible effect on the fitted model rates ( Figure 3—figure supplement 4 ) . Offline analysis of electrophysiology was performed using Clampfit v9 . 0 ( Molecular Devices ) , Mini Analysis Program v6 . 0 ( Synaptosoft , Decatur GA ) , Axograph X ( Axograph Scientific , Berkeley CA ) , and custom-written software routines ( Source code 1 ) in Matlab 7 . 10 . 0 or R2010a ( Mathworks , Natick MA ) . We used a minimal vesicle state model with a similar scheme as proposed by Weis et al . ( 1999 ) for Ca2+-dependent vesicle pool dynamics in the Calyx of Held , consisting of a depot pool of non-primed vesicles D , RRP with primed vesicles R and a fused pool F . Our model differs from the Weis-model on three aspects: ( 1 ) we model fusion as an continuous process during hypertonic stimulation , whereas in the Weis-model this is modelled as a discrete event during action potential stimulation , ( 2 ) in our model the rate constant for priming k1 is constant , and not Ca2+ dependent as in the Weis-model , since we use Ca2+-independent stimuli to evoke release , and ( 3 ) opposed to Weis-model our model has a finite D pool . This allowed us , in contrast to other pool models , to model synaptic responses to hypertonic sucrose , the relation between RRP depletion and release kinetics , and RRP replenishment during HS-stimulation . Vesicle dynamics for the vesicles in the depot pool D and the readily releasable pool R are described by two-coupled differential equations ( 7 ) dDdt=−k1D+k−1R , ( 8 ) dRdt=k1D− ( k−1+k2 ) R , with k−1 and k2 the rate constants for unpriming and fusion , respectively ( Figure 1B ) . To compensate for leak of vesicles from the system due to spontaneous release , we would need an extra term in Equation ( 7 ) to refill D . However , since we assume the spontaneous release rate before sucrose stimulation to be negligibly small compared to the other rates , we can neglect the refill term in Equation ( 7 ) . Equation ( 7 ) was included to account for depletion of the depot pool during long or repetitive HS stimulation . However , for the durations of the HS stimulations used in this paper , depletion of D was small and responses could be fitted with the priming rate k1D being treated as a constant ( see fitting procedures ) . For convenience , the pool sizes are expressed in nC instead of vesicles . In this version of the model , we did not include release sites since this would introduce an extra fit parameter , whereas such an extended model is mathematically equivalent ( if immediate availability and recycling of release sites is assumed; see below ) . The RRP size at steady state is the result of a dynamic equilibrium between priming , unpriming , and fusion ( Weis et al . , 1999 ) , and can be obtained from Equation ( 8 ) under the assumption of dR/dt = 0 , ( 9 ) R∞=k1Dk−1+k2 . As mentioned above , for the purpose of determining the RRP size before stimulation , we assumed that k2 was zero . For simulation of synaptic responses to hypertonic stimulation , we assume that this form of stimulation selectively reduces the activation energy for fusion , and thus increases the release rate constant k2 according to Equation ( 4 ) , without affecting upstream processes of fusion . Although solution exchange is very rapid ( <0 . 5 s ) , the onset of a HS-evoked synaptic response starts with a delay with respect to the rise in hypertonicity , most likely due to compensatory mechanisms that initially successfully counteract this osmotic perturbation ( see Figure 2—figure supplement 2 ) . In addition , after the delay there is a smooth , rather than an abrupt transition to the evoked inward current . To capture these features , the time course of k2 in response to sucrose is modelled as an expo-exponential ( 10 ) k2 ( t ) =k2 , maxe−e− ( t−t0−tdel ) /τ ( t≥t0 ) , with t0 the time point of sucrose application , tdel a constant which determines the delay of the onset of k2 with respect to t0 , τ a time constant that sets the steepness of the rising phase , and k2 , max the maximal value of k2 ( t ) ( Figure 2B ) . Each model parameter constrains the simulated HS-response in a specific way as shown in Figure 2—figure supplement 3A ( absolute traces ) and Figure 2—figure supplement 3B ( traces scaled and aligned to peak ) . An increase in the priming rate constant k1 or the depot pool D both increases the total RRP and steady-state priming phase at the end of the response without affecting release kinetics . Decreasing the unpriming rate constant k−1 increases the RRP , but without an effect on the steady-state priming phase . Increase of tdel further delays the response but does not change its shape . Increase of the maximal fusion rate constant k2 , max produces features that are typically observed experimentally when evoking post-synaptic responses with increasing levels of hypertonicity ( Figure 2A ) , such as increase in peak amplitude , shorter the time to peak , and speed-up of the decay phase after the peak . Finally , decrease of τ speeds up the rise phase , increases the peak amplitude , but only mildly affects the decay phase after the peak . These characteristic effects allow the accurate estimation of the individual model parameters by fitting the vesicle state model to experimental HS-induced traces ( see fitting procedures below ) . By ignoring vesicle replenishment during HS-stimulation and the delayed onset of the HS-induced response , our vesicle state model can be simplified such that an analytical solution can be obtained that qualitatively captures the main features of HS-induced release . Release from a readily releasable pool R without replenishment is given by ( 11 ) dRdt=−k2 ( t ) R , with k2 ( t ) a release rate parameter that changes over time during the application of hypertonic sucrose with a time-course as described in Equation ( 10 ) . When neglecting the delayed onset of sucrose action , the time dependence of k2 ( t ) can be approximated with a single exponential ( 12 ) k2 ( t ) =k2 , max ( 1−e−tτ ) ( t≥0 ) , with k2 , max the maximal release rate , τ a time constant for the exponential time course of k2 ( t ) , and t = 0 the start of sucrose application . Solving Equation ( 11 ) analytically yields the following solution: ( 13 ) R ( t ) =R0e−k2 , max ( τe−tτ+t ) +k2 , maxτ , with R0 = R ( 0 ) , the initial RRP size at the start of the stimulation . From this follows an exact expression for the fusion rate k2 ( t ) R: ( 14 ) dFdt=k2 ( t ) R=k2 , max ( 1−e−tτ ) R0e−k2 , max ( τe−tτ+t ) +k2 , maxτ . After convolving fusion rates for different values of k2 , max with an average mEPSC , postsynaptic current responses were obtained corresponding to different concentrations of hypertonic sucrose ( Figure 2—figure supplement 1 ) . These current responses display the typical characteristics as experimental responses , with increased peak release rates and shorter time-to-peak are observed for higher concentrations , but obviously do not reproduce the increased standing currents towards the end of depleting stimuli ( 0 . 5M or higher; Figure 3A1 ) , because of the lack of replenishment in this model . In our model described by Equations ( 7 ) and ( 8 ) , the number of release sites is not restricted . When we assume a fixed number of ( instantaneously available ) release sites S , Equation ( 8 ) transforms into ( 15 ) dRdt=k1D ( S−R ) − ( k−1+k2 ) R . Here , the extra factor ( S − R ) captures the idea that priming is hampered when fewer release sites are available for new vesicles to tether to . In this case , the steady-state RRP becomes ( 16 ) R∞=k1DSk1D+k−1+k2 . If , as an approximation , we assume k1D to be constant for the duration of the stimulation , Equations ( 8 ) and ( 15 ) and their respective steady-state RRP expressions Equations ( 9 ) and ( 16 ) are mathematically equivalent under the transformation k1D↔ ( k1DS ) sites and k−1+k2↔ ( k1D+k−1+k2 ) sites . However , priming- and unpriming rate constants have different values in both systems and affect R in a different manner . During hypertonic sucrose stimulation , vesicles are released from the RRP that consists of vesicles that were already primed at the onset of the stimulus R0 and newly primed vesicles Rnew . With R = R0 + Rnew Equation ( 8 ) transforms into ( 17 ) d ( R0+Rnew ) dt=k1D− ( k−1+k2 ) ( R0+Rnew ) , which can be separated in an expression for the depletion of R0 and the replenishment of vesicles into Rnew ( 18 ) dR0dt=− ( k−1+k2 ) R0 , ( 19 ) dRnewdt=k1D− ( k−1+k2 ) Rnew . The postsynaptic current I during the stimulus is given by the sum of the currents IR0 and IRnew , evoked by release from R0 and Rnew , respectively ( 20 ) I=IR0+IRnew=−k2 ( t ) ( R0+Rnew ) , with the minus sign correcting for the fact that we record inward currents but express R in as positive charge ( in nC ) . Interestingly , in this reduced model it follows from Equation ( 8 ) that without a limited number of release sites and assuming k2 ≈ 0 in the absence of sucrose , recovery of the RRP after depletion is given by ( 21 ) R= ( Rend−R∞ ) e−k−1t+R∞ , with Rend the RRP size at the end of the depleting stimulus , R∞ the fully recovered RRP given by Equation ( 9 ) , and 1/k−1 the time constant for recovery . The depleted RRP fraction is defined as the release during a hypertonic stimulus normalized to the steady state RRP size before the stimulation . If we assume that R has an initial steady state value Ri and is at a new steady state value Rf at the end of the stimulus , the depleted RRP fraction can be expressed as ( 22 ) depleted RRP fraction=Ri−RfRi=1−RfRi . Using Equation ( 9 ) , Ri and Rf are defined as ( 23 ) Ri=k1Dk−1+k2 , 0 , and ( 24 ) Rf=k1Dfk−1+k2 , max , When we assume that D is a large depot pool , with little effect on the size of D from replenishment from D to R during a sucrose stimulus ( Df ≈ Di ) , and that the initial fusion ate before stimulation is negligibly small ( k2 , 0 ≈ 0 ) , Equation ( 22 ) transforms into ( 25 ) Depleted RRP fraction=1− ( k−1+k2 , 0 ) ( k−1+k2 , max ) k1Dfk1Di≈1−k−1k−1+k2 , max=k2 , maxk−1+k2 , max . This analytical approximation closely resembles the relation between k2 , max and the depleted RRP fraction obtained with our model simulations using Equations ( 7 ) , ( 8 ) , and ( 10 ) ( Figure 4—figure supplement 1 ) . Fits were performed with an in-house developed analysis program in Matlab ( Source code 1 ) . The software reads Axon binary files ( . abf ) , which can be loaded in batches . When fitting the model to data , Equations ( 8 ) and ( 10 ) are numerically simulated using Matlab's ode45 ordinary differential equation ( ODE ) solver . This one-step solver for non-stiff ODEs makes use of explicit Runge-Kutta methods of order 4 and 5 with a variable time step . Matlab's odeset structure to alter the ODE solver's properties , such as integration error and step size , is set to its default value . R is expressed in nC . The initial condition of the simulation is the steady-state solution of the model assuming k2 = 0 . During the initial fit of a trace , k1D is taken constant and only Equation ( 8 ) is used . Subsequently , one can fit D and k1 separately to capture the decay in the refill phase , for instance during long HS-stimulations , by re-running the fitting procedure with all parameters ( including RRP size and the product k1D ) fixed , except for D and k1 , using both Equations ( 7 ) and ( 8 ) . In this paper , k1D is always obtained from the initial fit . The data time span used for fitting is specified by the user , and is generally taken equal to the duration of the sucrose application , up to the time when the sucrose concentration starts to decay back to baseline . The solution for the R state in this time window resulting from the ODE solver is subsequently interpolated at each measured time point within the fitting time window ( typical sampling frequency 10 kHz ) and the outcome is fed into a cost function , which calculates the sum of squared errors between model prediction and data for each iteration . When fitting multiple sucrose responses of a single cell simultaneously ( e . g . , 0 . 5M and 0 . 25M ) , the sum of squared errors is calculated separately for each concentration and subsequently added up . This cost function is used as input for the optimisation algorithms , all of which are contained in Matlab's Optimization Toolbox . The user has the option to choose between global ( genetic algorithm or simulated annealing ) and local ( Nelder-Mead downhill simplex ) methods . All methods are executed using default options , except for the lower and upper bounds of all parameters as used by the global search methods , which are set to 10−5 and 106 , respectively . The user can control the maximum number of iterations and function evaluations , both of which are by default set to 400 per fitted parameter . Once the global method has reached its stopping criterion at a certain point in parameter space , the local method takes over to search for the optimal set of parameters in the neighbourhood of this point . Quality of the fits was checked by visual comparison of the following features between the fitted curve and the experimental trace: ( 1 ) onset of fit , ( 2 ) peak amplitude and/or time-to-peak , ( 3 ) decay towards steady state phase , and ( 4 ) steady-state phase ( refill ) ( Figure 3—figure supplement 5B ) . When the deviation was too large , traces were fitted again with new initial conditions until no further improvement of the fit was observed . Although the model consists of multiple free parameters , different features of the HS-induced traces are constrained by different parameters in the model ( Figure 2—figure supplement 3 ) and vice versa . The RRP size , and thus the ratio of k1D and k−1 , is constrained by the charge transfer during the peak . In addition , k1D is constrained by the steady state current after the peak , which then also constrains k−1 via the RRP size and Equation ( 9 ) . Note that the RRP itself is not a fit parameter , and that the fit procedure optimizes k1D and k−1 to get the best fit of the experimental trace . Equation ( 9 ) is then used to calculate the RRP post-hoc . tdel is constrained by the delay of the onset of the response . Peak amplitude in combination with steepness of the rise phase constrains τ , and peak amplitude in combination with the decay phase after the peak constrains k2 , max . Simulations show that the fit method can indeed robustly discriminate between the effects of different model parameters on the shape of the sucrose response , that is , changes in one model parameter are reliably detected with the other model parameters being invariant ( Figure 3—figure supplement 5A , C; Figure 3—source data 2 ) . In addition , random examples of experimentally obtained responses to 0 . 3M and 0 . 5M sucrose in the absence and presence of the phorbol ester ( PDBu ) show that this method provides a close fit for almost all traces ( Figure 6—figure supplements 1 , 2 ) . The activation energy as a function of sucrose concentration as shown in Figure 3D was fitted with a mono-exponential function of the form ΔE ( M ) = ae−b·M+c , with M the sucrose concentration in molar , using Matlab's built-in Curve Fitting Tool . Fits of k2 , max as a function of sucrose concentration in Figure 3C were obtained by transformation of the fitted function in Figure 3D , using Equation ( 5 ) . As log-transforming symmetrical error bars in the release rate domain results in asymmetric error bars in the energy domain , we used the largest error of the two for plotting the SEM of fitted activation energy . Data shown in Figure are mean ± SEM . In addition , bootstrap analysis was performed to estimate statistical errors and confidence intervals for the distributions of the mean values of all fitted parameters . We applied the nonparametric bootstrap method ( i . e . , resampling the original data ) using the ‘bootstrap’ function from MATLAB's statistics toolbox with default options . The size of the original data sets used to constitute the bootstrap sample is equal to the number of observations per parameter ( n ) , as given in the figure tables . For each parameter , we bootstrapped 10 , 000 sample means , and subsequently calculated the mean value , the standard deviation ( std ) and the 95% confidence interval ( 95% CI ) of the distributions of these sample means . For the combined effect of PDBu and sucrose on k2 , max we also calculated 95% CI for the absolute change in k2 , max ( Figure 6D ) . Values used for model parameters and fit parameters in the figures and results from bootstrap analysis are given in the supplemental tables provided for each figure . | Information is carried around our nervous system by cells called neurons , which are connected to each other by junctions known as synapses . Within the neurons , there are many small compartments known as synaptic vesicles that are essential to the transfer of information from one neuron to the next . When one neuron is activated , the synaptic vesicles fuse with the membrane surrounding the cell to release molecules called neurotransmitters , which cross the synapse and activate the next neuron . Vesicle fusion is carefully regulated to control the speed and amount of neurotransmitter release , which determines the level of activation of the next neuron . Vesicle fusion requires energy , much of which is provided by a set of proteins found in the synapse . The minimum amount of energy required—called the activation energy—is influenced by many factors , including the shape of the cell's membrane at the synapse . It is thought that altering the activation energy required for fusion may control the activity of synapses , but it is not possible to directly measure this in living cells . To bypass this problem , Schotten , Meijer , Walter et al . established a new method to study vesicle fusion . This method combines a mathematical model with experimental data of the activity of synapses . First , the neurons were placed in a solution containing the sugar sucrose , which triggered vesicle fusion by lowering the activation energy . The increase in vesicle fusion was smaller in neurons that lacked two proteins called complexin I and complexin II—which control vesicle fusion—than in the normal neurons . A molecule called phorbol ester is also able to activate the release of neurotransmitters . When cells were treated with both sucrose and phorbol ester , the speed of vesicle fusion was greater . The experiments show that the effects of sucrose , phorbol ester , and the complexins multiply together to dramatically alter vesicle fusion . Schotten , Meijer , Walter et al . suggest a new model for how the activation energy of vesicle fusion controls the transfer of information across synapses . This might shed new light on how the efficiency of vesicle fusion is altered when neurons are highly active , which is thought to have strong implications for how information is processed in the brain . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"neuroscience"
] | 2015 | Additive effects on the energy barrier for synaptic vesicle fusion cause supralinear effects on the vesicle fusion rate |
RadA ( also known as 'Sms' ) is a highly conserved protein , found in almost all eubacteria and plants , with sequence similarity to the RecA strand exchange protein and a role in homologous recombination . We investigate here the biochemical properties of the E . coli RadA protein and several mutant forms . RadA is a DNA-dependent ATPase , a DNA-binding protein and can stimulate the branch migration phase of RecA-mediated strand transfer reactions . RadA cannot mediate synaptic pairing between homologous DNA molecules but can drive branch migration to extend the region of heteroduplex DNA , even without RecA . Unlike other branch migration factors RecG and RuvAB , RadA stimulates branch migration within the context of the RecA filament , in the direction of RecA-mediated strand exchange . We propose that RadA-mediated branch migration aids recombination by allowing the 3’ invading strand to be incorporated into heteroduplex DNA and to be extended by DNA polymerases .
All organisms have complex mechanisms to accurately replicate and repair their chromosomes to maintain genetic integrity . In E . coli , the RecA protein promotes repair of DNA lesions directly through its role in homologous recombination ( reviewed in [Persky and Lovett , 2008] ) . In addition , it promotes repair indirectly by the recruitment of repair polymerases to damaged replication forks ( Patel et al . , 2010 ) and by activation of the SOS response , a transcriptional response to DNA damage ( reviewed in [Simmons et al . , 2009] ) . Each of these processes depends on the formation of RecA filaments on single-strand DNA ( ssDNA ) . In vitro RecA mediates strand exchange , a key step of recombination , in three distinct phases ( Radding et al . , 1983 ) . The first phase is the formation of the presynaptic filament on ssDNA . RecA filaments form when dimers nucleate on DNA in a slow step ( Bell et al . , 2012 ) ; subsequently the filament is extended in both directions , although at a higher rate at the 3’ end of the RecA:ssDNA filament . When ssDNA is bound in the primary DNA binding site of the RecA filament , it is underwound relative to B-form dsDNA such that the RecA-DNA filament has about 18 bases per turn ( Chen et al . , 2008; Galletto et al . , 2006 ) . ATP binding , but not hydrolysis , is required for active RecA filament formation . The second phase involves the homology search process and strand-pairing in which duplex DNA is bound and sampled for pairing through a secondary DNA binding site ( Mazin and Kowalczykowski , 1996 , 1998 ) . After homologous DNA molecules are paired , the third phase of strand exchange involves branch migration , in which the region of heteroduplex DNA formed between the two strand exchange partners is extended . The heteroduplex is initially bound through RecA primary site interactions , with the displaced strand ( s ) in the secondary site ( Mazin and Kowalczykowski , 1998 ) . ATP hydrolysis is required for complete heteroduplex product formation when the homologous molecules exceed several kilobases in length ( Jain et al . , 1994 ) . Several proteins modulate the RecA filament , regulating the recombination activity of RecA and its potentially mutagenic activity resulting from induction of the SOS response . These accessory proteins include those involved in RecA nucleation onto ssDNA ( PsiB ) , in loading and unloading of RecA onto DNA coated with single-strand DNA binding protein ( SSB ) ( RecFOR ) and in regulating RecA filament stability ( DinI and RecX ) ( reviewed in Cox ( 2007 ) ) . Other proteins such as UvrD ( Petrova et al . , 2015 ) , PcrA ( Fagerburg et al . , 2012; Park et al . , 2010 ) , and RuvAB ( Eggleston et al . , 1997 ) , and DinD dismantle RecA filaments ( Uranga et al . , 2011 ) . In eukaryotes , there are several paralogs of the major recombination protein Rad51 that either modulate Rad51 activity or are specialized strand-exchange proteins themselves ( Adelman and Boulton , 2010; Bernstein et al . , 2013; Gasior et al . , 2001; Qing et al . , 2011; Suwaki et al . , 2011; Taylor et al . , 2015 ) . In bacteria , there is at least one partially characterized RecA paralog , RadA . RadA ( also known as 'Sms' , for 'sensitivity to methyl methanesulfonate' [Song and Sargentini , 1996] was identified as a radiation-sensitive mutant of E . coli [Diver et al . , 1982a] ) and is required for DNA recombination and repair in many diverse bacterial species ( Beam et al . , 2002; Burghout et al . , 2007; Carrasco et al . , 2004; Castellanos and Romero , 2009; Cooper et al . , 2015; Krüger et al . , 1997; Lovett , 2006; Slade et al . , 2009 ) . Thus , RadA is a possible candidate for a RecA accessory protein . Despite its name , RadA of eubacteria is not orthologous to RadA of archaea , the latter being a true strand-exchange protein , functionally and structurally similar to bacterial RecA and eukaryotic Rad51 ( Seitz et al . , 1998; Wu et al . , 2004; Yang et al . , 2001 ) . In E . coli , RadA affects recombination measured by certain in vivo assays , often in a manner partially redundant to other functions that mediate late steps of homologous recombination . Loss of radA , by itself , reduces recovery of genetic rearrangements at tandem-repeated sequences , which are promoted by defects in the replication fork helicase , DnaB ( Lovett , 2006 ) . In addition , loss of RadA reduces homologous recombination when in combination with loss of RuvAB or RecG ( Beam et al . , 2002 ) , as measured by conjugation with Hfr donors . RuvAB and RecG are DNA motor proteins that branch-migrate recombination intermediates such as Holliday junctions during the late stages of homologous recombination ( reviewed in ( Persky and Lovett , 2008 ) . For sensitivity to genotoxins including azidothymidine ( AZT ) , ciprofloxacin ( CPX ) and UV irradiation , mutations in radA are especially strongly synergistic with those in recG ( Beam et al . , 2002; Cooper et al . , 2015 ) . This genotoxin-sensitivity in radA recG strains appears to be due , in part , to the accumulation of recombination intermediates , since it can be suppressed by early blocks to recombination ( by mutations in recF or recA ) or by overexpression of RuvAB ( Cooper et al . , 2015 ) . These genetic observations implicate RadA in late steps of recombination , potentially involving branch migration of recombination intermediate DNA structures such as Holliday junctions . RadA is a 460 amino acid protein that has three well-conserved domains found in other proteins , as well as a 5-amino acid motif highly conserved among radA orthologs . The N-terminal 30 amino acids form a putative zinc-finger domain with a C4 motif , CXXC-Xn-CXXC . In bacteria , proteins with this domain include the DNA repair proteins UvrA and RecR , and the ATP-dependent serine protease ClpX . The E . coli radA100 mutation , a C28Y mutation in the putative Zn finger motif , negates radA function in vivo and is partially dominant ( Cooper et al . , 2015; Diver et al . , 1982b ) . The second RadA domain ( aa 59–184 ) is homologous to the ATPase region of RecA and contains both Walker A and Walker B boxes and regions homologous to its L1 and L2 loops involved in primary site DNA binding . A RadA-K108R mutation at the Walker A sequence is a dominant-negative RadA allele in E . coli ( Cooper et al . , 2015 ) . The C-terminal 150 amino acids comprise a predicted S5 domain 2-type fold , ( EMBL-EBI Interpro subgroup IPR014721 , http://www . ebi . ac . uk/interpro/entry/IPR014721 ) , present in ribosomal proteins S5 and S9 , EF-G , Lon , RNase P , MutL , and several DNA topoisomerases . Deletion of this domain negates RadA functions in vivo ( Cooper et al . , 2015 ) . In BLAST alignments , this region is most closely related to the ATP-dependent serine protease Lon ( Chung and Goldberg , 1981 ) . Mutation of serine 372 of RadA , comparable in alignments to the active site serine of Lon , did not affect RadA genetic function and this serine is not conserved among RadAs; this and the lack of other conserved residues of the Lon protease catalytic triad indicate that RadA is unlikely to possess serine protease activity . Between the RecA and S5 domain 2 domains , there is a conserved motif specific to RadA proteins , KNRFG , a motif also found in the phage 29 structure-specific nuclease ( Giri et al . , 2009 ) . The K258A mutation in this motif negates RadA function and is partially dominant in vivo ( Cooper et al . , 2015 ) . To explore RadA function in E . coli , we purified wild type RadA as well as several site-directed mutants altered in conserved motifs of the protein . We found that the wild-type RadA protein preferentially binds single-strand DNA in the presence of ADP , exhibits ATPase activity stimulated by DNA , and increases the rate of RecA-mediated recombination in vitro by stimulation of branch migration . Branch migration can be mediated by RadA even in the absence of RecA and is highly directional in nature , with preferential extension of the heteroduplex in the 5’ to 3’ direction , relative to the initiating single-strand; this is codirectional with that of RecA-mediated strand exchange . Mutations in the Walker A , KNRFG and zinc finger motifs abolish RadA’s branch migration activity in RecA-coupled reactions and lead to the accumulation of strand exchange intermediate species . The ability of RadA to catalyze branch migration in the context of the RecA filament and its codirectionality with strand exchange distinguish it from other branch migration functions in E . coli , RecG and RuvAB ( Whitby et al . , 1993 ) . RadA’s ability to branch migrate recombination intermediates readily explains radA mutant phenotypes in vivo .
To elucidate RadA structure and function , we purified native wild-type RadA and several RadA domain mutants and then evaluated their biochemical activities , particularly those possessed by the RecA protein . Wild-type RadA was estimated to be more than 98% pure ( Figure 1—figure supplement 1 ) . Using an electrophoretic mobility shift assay ( Figure 1A ) , we examined the binding of purified RadA to poly ( dT ) 30 in the presence of various nucleotide cofactors . RadA bound this oligonucleotide only in the presence of ADP . No binding was detected in the absence of nucleotide , with ATP or dATP or in the presence of poorly-hydrolyzable ATP analogs , ATPγS or AMP-PNP ( Figure 1A ) . These observations suggest that ADP promotes the most stably DNA-bound RadA . This behavior contrasts to that of RecA , which requires ATP for binding to form the active , extended DNA filament and which dissociates in the presence of ADP . Poorly hydrolyzable nucleotide analogs such as ATPγS produce the most stable RecA binding ( McEntee et al . , 1981 ) . 10 . 7554/eLife . 10807 . 003Figure 1 . RadA Binding to DNA . ( A ) Nucleotide dependence of RadA binding to poly d ( T ) 30 . Reactions ( 10 µl ) contained 100 fmol ( molecule ) of radio-labeled poly d ( T ) 30 , 3 . 3 pmol RadA and 1 mM nucleotide . After incubation at 37 °C for 20 min , binding was assessed using EMSA . ( B ) DNA substrate specificity of RadA Binding . Reactions containing 1 mM ADP , 3 . 3 pmol RadA , 100 fmol ( molecule ) of radio-labeled poly d ( T ) 30 and unlabeled competitor DNA ( circles-poly d ( T ) 30 , triangles- poly d ( C ) 30 , diamonds-poly d ( G ) 30 , squares poly d ( A ) 33 were incubated at 37° for 20 min . The extent of binding was determined using scanned autoradiographs of EMSA gels processed with Image J-64 . DOI: http://dx . doi . org/10 . 7554/eLife . 10807 . 00310 . 7554/eLife . 10807 . 004Figure 1—figure supplement 1 . Purification of RadA . 1 . SDS PAGE Gel Showing Purification of RadA . Lane 1 ) MW Markers; Lane 2 ) Crude Lysate; Lane 3 ) Flowthrough; Lane 4 ) Wash; Lane 5 ) eXact peak fraction; Lane 6 ) MonoQ HP peak fraction . DOI: http://dx . doi . org/10 . 7554/eLife . 10807 . 00410 . 7554/eLife . 10807 . 005Figure 1—figure supplement 2 . RadA binding to poly ( dT ) 30 . Electrophoretic mobility shift assay for RadA binding to poly ( dT ) 30 , with RadA concentrations indicated . DOI: http://dx . doi . org/10 . 7554/eLife . 10807 . 00510 . 7554/eLife . 10807 . 006Figure 1—figure supplement 3 . RadA binding curve to poly ( dT ) 30 . The autoradiographs from Figure 1—figure supplement 2 and a second gel were scanned and binding was quantitated using Image J 64 . GraphPad Prism software was used to fit the data to a binding curve . The data from two replicates are shown . The apparent Kd derived from this curve is 110 nM ± 14 nM . DOI: http://dx . doi . org/10 . 7554/eLife . 10807 . 00610 . 7554/eLife . 10807 . 007Figure 1—figure supplement 4 . Binding of RadA to substrate E2 . Electrophoretic mobility shift assay for RadA binding . Substrate E2 has approximately 30 nucleotides on either side of a poly ( dT ) 30 core . Poly ( dA ) 30 was annealed to E2 before the binding reaction was initiated for the samples indicated . The * indicates the strand labeled with 32P . DOI: http://dx . doi . org/10 . 7554/eLife . 10807 . 007 Competition experiments ( Figure 1B ) with unlabeled poly ( dA ) , poly ( dC ) , poly ( dG ) and poly ( dT ) showed that only poly ( dT ) competes for RadA binding to labeled poly ( dT ) , indicating a binding preference for by poly ( dT ) . RecA protein also shows a preference for poly ( dT ) ( Bugreeva et al . , 2005; McEntee et al . , 1981 ) , which may be because poly ( dT ) assumes a more flexible structure ( Mills et al . , 1999 ) . The apparent KD of RadA binding to poly ( dT ) 30 is about 110 nM , with a Hill coefficient of 1 . 5 ( Figure 1—figure supplement 2 , Figure 1—figure supplement 3 ) , indicative of cooperative binding . RadA was observed to bind poly ( dT ) 30 when flanked on both 5’ and 3’ ends by 30 nucleotides of natural DNA sequence ( substrate 'E2' , Figure 1—figure supplement 4 ) ; this binding was inhibited if poly ( dA ) 30 was allowed to anneal to the substrate , showing that RadA binds more poorly to duplex DNA . Using an NADH-coupled assay ( Table 1 ) , we measured the ATPase activity of RadA , in the presence or absence of various DNA cofactors , circular ssDNA ( φX174 virion ) and dsDNA ( φX174 RF DNA ) . Like RecA , RadA’s ATPase activity is strongly stimulated by ssDNA . However , RadA’s ATPase is also substantially stimulated by dsDNA whereas dsDNA stimulates RecA's ATPase only after a lag period ( Kowalczykowski et al . , 1987 ) . RadA’s ATPase activity in the presence of ssDNA was measured to have a kcat of 29 . 4 min-1 , comparable to that of RecA ( Weinstock et al . , 1981 ) . Addition of single-strand DNA-binding ( SSB ) protein after incubation of RadA or RecA with ssDNA was observed to repress the ATPase activity of RadA , whereas it slightly stimulated the ATPase of RecA ( Figure 2 ) . This latter observation suggests that , although RecA exhibits stable binding to ssDNA when challenged by SSB , less stable binding by RadA allows for SSB competition and inhibition of its ATPase activity . RadA ATPase activity measured with a variety of different DNA structures formed by oligonucleotides , including forks , splays , and other branched structures , showed no significant difference from that with ssDNA ( Figure 2—figure supplement 1 ) . Although these DNA molecules are only very weakly bound by RadA in gel shift experiments ( data not shown ) , they are sufficient to stimulate RadA’s ATPase activity , indicating some transient or unstable association . 10 . 7554/eLife . 10807 . 008Table 1 . DNA dependence of RadA ATP hydrolysis . DOI: http://dx . doi . org/10 . 7554/eLife . 10807 . 008DNA substrateApparent kcat ( ATP/RadA/min ) None2 . 9 +/- 0 . 2Circular single-strand29 . 4 +/- 0 . 4Supercoiled double-strand9 . 1 +/- 0 . 4Linear double-strand4 . 2 +/- 0 . 2Nicked double-strand15 . 2 +/- 0 . 5Reactions contained 1 mM ATP and 20 . 3 µM ( nucleotide ) φX174 DNA . The values shown are the average of two experiments , except for the circular single-strand value . It is the average of five experiments . Standard deviations are reported . 10 . 7554/eLife . 10807 . 009Figure 2 . ATP hydrolysis in reactions including RecA , SSB and RadA . ATP hydrolysis was measured in reactions containing DNA and protein concentrations similar to those in recombination reactions and included 21 µM ( nucleotide ) single-strand circular DNA and 6 . 7 µM RecA , 1 . 9 µM SSB , and 630 nM RadA . RecA and/or RadA were pre-incubated with the single-strand DNA for 8 min at 37 °C before the reactions were initiated with ATP +/- SSB . Rate measurement started 5 min after the addition of ATP . Reactions included: SSB ( closed purple circles ) , RadA ( open red squares ) , RadA +SSB ( closed dark red squares ) , RecA alone ( blue inverted triangles ) , RecA+SSB ( closed dark blue inverted triangles ) , RadA + RecA ( open green diamonds ) , RadA+RecA+SSB ( closed dark green diamonds ) . Error bars represent the 95% confidence interval of the linear fit of the data calculated using Prism Graph Pad . DOI: http://dx . doi . org/10 . 7554/eLife . 10807 . 00910 . 7554/eLife . 10807 . 010Figure 2—figure supplement 1 . ATPase activity on model oligonucleotide substrates . NADH coupled ATP hydrolysis assays were performed as described in the procedures . Reactions contained 0 . 3 μM ( molecule ) model forks and 0 . 63 μM RadA . Average rates from three experiments are shown . Error bars represent standard deviations . The sequences for the oligonucleotides used to make the forks are in Table 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 10807 . 010 We purified several forms of RadA , mutated in its characteristic motifs: C28Y ( Zn finger ) , K108R ( Walker A box ) , K258A ( KNRFG RadA motif ) and S372A ( putative Lon protease active site ) ( Figure 3A , Figure 3—figure supplement 1 ) . These were analyzed for poly ( dT ) 30 binding in the presence of ADP and ssDNA-dependent ATPase activity ( using the more sensitive PEI TLC method ) ( Figure 3B , C ) . As expected , the K108R mutant in the Walker A box abolished ATPase activity; a defect in ATPase was exhibited by the K258A mutant as well . Both of these mutants were defective for DNA binding in the presence of ADP . RadA mutant C28Y retained the ATPase activity but was defective in DNA binding . RadA mutant S372A retained ATPase and DNA-binding activities . 10 . 7554/eLife . 10807 . 011Figure 3 . Properties of RadA Domain Mutants . ( A ) Schematic of RadA Domains and Locations of Domain Mutants . ( B ) Binding of RadA Mutants to poly d ( T ) 30 . Reactions contained 100 fmol ( molecule ) of poly d ( T ) 30 , 3 . 3 pmol ( mutant ) RadA and 1 mM ADP . After incubation at 37 °C for 20 min , binding was assessed using EMSA . Binding relative to wild-type RadA is shown for two independent experiments . ( C ) ATP hydrolysis by RadA domain mutants . ATP hydrolysis was assessed using PEI TLC to visualize release of inorganic phosphate ( Pi ) as described in the procedures . Reactions contained either no or 10 . 5 µM ( nucleotide ) single-strand circular M13 DNA and 250 nM RadA or RadA mutant protein . Graph shows the mean ATP hydrolysis activity of RadA mutants from three independent experiments relative to the activity of wild-type RadA . Error bars represent the standard deviation of the mean . DOI: http://dx . doi . org/10 . 7554/eLife . 10807 . 01110 . 7554/eLife . 10807 . 012Figure 3—figure supplement 1 . SDS-PAGE of RadA mutants . SDS-PAGE gel showing purified RadA mutant proteins . DOI: http://dx . doi . org/10 . 7554/eLife . 10807 . 012 We examined standard RecA-mediated strand-exchange reactions between 5386 nucleotide circular φX174 ssDNA and linear duplex DNA in the presence of ATP and an ATP-regeneration system ( Figure 4A ) . Presynaptic filaments are formed by the incubation of RecA with ssDNA , linear dsDNA is then added and the reaction is initiated by the addition of ATP and SSB . Reactions are monitored by agarose gel electrophoresis . In this regimen , RecA catalyzes the formation of branched DNA molecules within 5 min , which are converted to the relaxed circular dsDNA final product by 18 min . Addition of RadA , at a 1:17 stoichiometry relative to RecA , accelerated final product formation , such that the reaction was complete by 5 min , with no detectable accumulation of branched intermediates ( Figure 4B ) . Although RadA did not change the efficiency of the RecA strand-exchange reaction , RadA enhanced the branch migration phase of the reaction to yield the final product more quickly . This branch migration was directional in nature and drove the reaction forward to the nicked circular product rather than back to the linear substrate . 10 . 7554/eLife . 10807 . 013Figure 4 . Three-strand Recombination Reactions in the Presence of RadA . ( A ) Diagram of thethree-strand recombination reaction . Single-strand circular φX-174 DNA ( SCS ) was mixed with double-strand φX174 DNA linearized with PstI ( DLS ) in the presence of RecA , SSB , ATP and an ATP regenerating system . When RadA was included in the reactions , it was added to achieve a RadA:RecA ratio of 1:17 . Initially , branched Intermediates ( INT ) form between the singe-strand circular DNA and its complementary sequence . After continued incubation , nicked circular product ( NP ) and single-strand linear product ( SLP ) are formed . Note: The SLP is not usually visible . The standard order of addition for this reaction is: 1 ) Incubation at 37 °C for 8 min with buffer , ATP regenerating system , φX174 single-strand DNA , and RecA ( and RadA when included ) . 2 ) Addition of double-strand linear φX174 and continued incubation for 5 min at 37° . 3 ) Addition of pre-mixed ATP and SSB to initiate the reaction . Incubation then continued for the times indicated . 4 ) Deproteinization of the reaction and separation of the products from substrates on a 0 . 8% agarose gel run in TAE . ( B ) Effect of RadA on Three-strand Recombination Reactions . Three-strand recombination reactions were performed as described above with either RecA and RadA or RecA alone in the first incubation step . In the third set of reactions , RadA was added to reactions containing RecA ( split from the RecA alone reaction ) five minutes after addition of SSB and ATP . ( C ) Effect of excess ADP on Three-strand Recombination Reactions . Recombination reactions were performed as described except for the following modifications . No regenerating system was included , but the ATP concentration was increased to 5 mM . ADP was added at the concentrations indicated . Finally , incubation at 37 °C was extended to 60 min . DOI: http://dx . doi . org/10 . 7554/eLife . 10807 . 01310 . 7554/eLife . 10807 . 014Figure 4—figure supplement 1 . Effect of order of addition of reaction components on 3-strand recombination reactions in the presence or absence of RadA . Components in the blue box were pre-incubated with buffer and the regenerating system for 8 min at 37 °C . Then , reaction components in the green box were added and pre-incubation was continued for 5 min . Recombination was initiated by the addition of components in the red box . The reactions were then incubated for the times indicated above the gel . DOI: http://dx . doi . org/10 . 7554/eLife . 10807 . 01410 . 7554/eLife . 10807 . 015Figure 4—figure supplement 2 . Titration of RecA in 3-strand recombination reactions in the presence or absence of RadA . Intermediates ( INT ) ; Nicked Product ( NP ) ; Duplex Linear Substrate ( DLS ) ; and Single-strand Circular Substrate ( SCS ) . DOI: http://dx . doi . org/10 . 7554/eLife . 10807 . 01510 . 7554/eLife . 10807 . 016Figure 4—figure supplement 3 . 3-strand recombination reactions with saturating RecA concentrations and RadA concentrations as indicated . Intermediates ( INT ) ; Nicked Product ( NP ) ; Duplex Linear Substrate ( DLS ) ; and Single-strand Circular Substrate ( SCS ) . DOI: http://dx . doi . org/10 . 7554/eLife . 10807 . 01610 . 7554/eLife . 10807 . 017Figure 4—figure supplement 4 . Effect of RadA alone on 3-strand recombination reactions . Reactions were performed as described in the procedures except 4 μM RecA or Rad A was used . DOI: http://dx . doi . org/10 . 7554/eLife . 10807 . 017 The strand exchange reaction was also performed by initiating the reaction with RecA alone and adding RadA to the ongoing reaction after 5 min , when all the duplex linear DNA had been converted to branched intermediates ( Figure 4B ) . In these reactions , final product is visible as early as 2 . 5 min after RadA addition , with the reaction complete after an additional 5 min . Therefore , RadA need not be present in the reaction during RecA presynaptic filament formation and can stimulate branch migration from RecA-promoted strand-exchange intermediates . Order-of-addition experiments ( Figure 4—figure supplement 1 ) measuring RecA-mediated strand exchange , with and without RadA , showed that RadA strongly stimulated branch migration when added before SSB , but not when added after . In addition , RadA did not detectably stimulate RecA-strand exchange in the absence of SSB . RadA stimulated RecA-mediated strand exchange reactions when RecA concentrations were reduced to suboptimal levels but did not when RecA became limiting ( Figure 4—figure supplement 2 ) . When RecA was held at saturating concentration ( 6 . 7 µM ) and RadA was titrated , the stimulation of branch migration was detectable at RadA:RecA stoichiometries of as little as 1:223 and reached the maximal detectable stimulation at 1:13 ( Figure 4—figure supplement 3 ) . In the RadA-stimulated reactions , the time of appearance of final product varied with RadA concentration . In the absence of an ATP-regeneration system , the addition of ADP to the 3-strand reaction inhibits strand exchange , primarily by destabilizing the RecA presynaptic filament ( Kahn and Radding , 1984; Lee and Cox , 1990a; 1990b; Piechura et al . , 2015; Wu et al . , 1982 ) . As an indication of how RadA might affect the RecA filament , we assayed ADP-inhibition of the standard RecA 3-strand reaction ( lacking an ATP-regenerating system ) , at various concentrations relative to a fixed amount of ATP , 5 mM ( Figure 4C ) . In 60 min reactions with RecA alone , ADP completely inhibited strand-exchange at 4 mM . In reactions containing both RecA and RadA , ADP inhibition of strand exchange was almost complete at 1 mM and complete at 2 mM . Furthermore , even without ADP addition , RadA inhibited RecA 3-strand transfer reactions that lack an ATP-regeneration system . We note that under these reaction conditions , which include SSB , RadA’s ATPase is inhibited ( Figure 2 ) , so this result is not simply due to a higher rate of ATP depletion by the mere addition of RadA . Under the conditions used for RecA presynaptic filament formation , addition of RadA at 1:17 RadA:RecA caused a slight reduction in the measured ATPase ( Figure 2 ) , which appears to be primarily due to RecA . This may indicate some removal and replacement of RecA with RadA on ssDNA , or alternatively , that RadA , at substoichiometric amounts , can inhibit RecA’s ATPase while bound to ssDNA . RadA did not promote RecA strand transfer or joint molecule formation when SSB was omitted from the reaction ( Figure 4—figure supplement 1 ) indicating that RadA does not assist nucleation or filament extension under these conditions . Under no conditions examined did we find that RadA by itself could promote strand-pairing and exchange ( for example , see Figure 4—figure supplement 4 ) . To determine which properties of RecA are required for RadA-stimulation of strand exchange , we next examined strand-exchange catalyzed by RecA K72R , a mutant that can bind but not hydrolyze ATP ( Rehrauer and Kowalczykowski , 1993; Shan et al . , 1996 ) . These reactions substitute dATP for ATP , which promotes higher affinity of RecA for ssDNA , and lower Mg2+ concentrations . As reported previously , the RecA mutant readily catalyzes strand-pairing to form branched intermediates but is inefficient at branch migration to form the final relaxed circular dsDNA product . RadA did indeed accelerate branch migration in RecA K72R-mediated reactions , with final nicked circular product ( 'NP' ) visible at 30 min ( Figure 5A ) . Therefore , RadA-acceleration of branch migration in RecA-promoted reactions does not require the ATPase of RecA . 10 . 7554/eLife . 10807 . 018Figure 5 . Mutational Analysis of 3-strand Recombination Reactions . ( A ) Three-strand recombination reactions with RecA K72R and RadA . Recombination reactions with the mutant RecA K72R were performed as described except dATP replaced ATP and the Mg ( OAc ) 2 concentration was decreased to 3 mM . ( B ) Three-strand recombination reactions with RadA domain mutants . The reactions were performed as described except with the indicated RadA mutant protein replacing wild-type RadA . DOI: http://dx . doi . org/10 . 7554/eLife . 10807 . 018 In the standard RecA 3-strand reaction , we examined the ability of RadA mutant proteins to accelerate branch migration , compared to reactions in parallel with wild-type RadA or lacking RadA altogether ( Figure 5B ) . Interestingly , the RadA K258A mutant appeared to arrest branch migration , with the accumulation of slowly-migrating intermediate species . Final product formation was reduced , more so than reactions that lack RadA . RadA mutants C28Y and K108R also slowed the reactions and reduced the efficiency of strand exchange . In contrast , RadA S372A appeared fully wild-type in its ability to accelerate strand-exchange . These experiments show that RadA’s ATPase activity , negated in the K108R mutant , is required to stimulate branch migration . RadA K258A , a dominant-negative mutant in the 'RadA motif' , not only did not accelerate branch migration but also interfered with RecA’s ability to branch migrate , suggesting it binds either to intermediate structures and/or RecA to block the reaction . RadA’s putative zinc finger , affected by the C28Y mutant , is also required to stimulate branch migration . To test whether RadA could promote branch migration in the absence of RecA , we performed standard RecA strand-exchange reactions for 12 min and purified the DNA from such reactions after proteinase K/SDS treatment to remove RecA . This DNA , enriched in branched strand-exchange intermediates , was then incubated with RadA alone in the presence of ATP . RadA caused loss of branched DNA and enhanced formation of relaxed circular dsDNA product , relative to the same substrate incubated without RadA ( Figure 6A , quantitated in Figure 6B ) . Interestingly , branch migration catalyzed by RadA under these conditions was directional in nature , with accumulation primarily of full strand exchange products ( nicked circular dsDNA product , 'NP' ) rather than linear dsDNA ( 'DLS' ) ( Figure 6A , B , D ) . In the presence of SSB , however , the directionality of RadA-mediated branch migration was lost , with accumulation of both linear and circular products ( Figure 6A , C , D ) . The ability of RadA to catalyze branch migration in this assay required ATP ( Figure 6E ) and was not seen in reactions containing no nucleotide , ADP or ATPγS . 10 . 7554/eLife . 10807 . 019Figure 6 . Recombination Intermediate Branch Migration . ( A ) Branch Migration of intermediates Mediated by RadA . Three-strand recombination reactions were stopped after 12 min and deproteinized and purified as described in the procedures . Branch migration assays contained DNA intermediates ( 100 ng ) , 1 µM RadA , 3 mM ATP , and 2 . 4 µM SSB when indicated . After incubation for the times incubated , reactions were stopped and products were resolved on an 0 . 8% TAE agarose gel . The No protein sample includes DNA intermediate fractions and ATP and was incubated for 30 min at 37 °C without RadA or SSB . ( B ) Quantification of DNA Species in the Branch Migration Assay Formed by RadA . Amounts of each DNA species was determined from scanned digital photographs using ImageJ64 ( Nicked Product ( NP ) -squares , Duplex Linear Substrate ( DLS ) -triangles , Single-strand Circular Substrates ( SCS ) -inverted triangles , and Intermediate Substrates ( INT ) -circles ) . ( C ) Quantification of DNA Species in the Branch Migration Assay Formed by RadA and SSB . * Amounts of each DNA species was determined from scanned digital photographs using ImageJ64 . ( Nicked Product ( NP ) -squares , Duplex Linear Substrate ( DLS ) -triangles , Single-strand Circular Substrates ( SCS ) -inverted triangles , and Intermediate Substrates ( INT ) -circles ) . ( D ) Quantification of the Nicked Product ( NP ) and Duplex Linear Substrate ( DLS ) Formed in Branch Migration Assays . Graph shows the mean and standard deviation of the relative amounts of NP and DLS formed in three independent branch migration experiments . Two different RadA preparations and three different DNA Intermediate preparations were used in these experiments . ( E ) Nucleotide Dependence of the Branch Migration Assay . Reactions were performed as above except 1mM of the nucleotide indicated replaced 3mM ATP . Incubation was for 30 min at 37 °C . ( F ) Model Depicting RadA Directionality . In the absence of SSB , RadA ( illustrated by wedge shape ) preferentially migrates DNA , displacing a 5’ ssDNA flap . In the presence of SSB , the directional bias of RadA branch migration is largely eliminated . * No correction for the difference in binding affinity of ethidium bromide for single-strand and double-strand DNA was made . Thus , the absolute amount of the DNA species containing single-strand DNA may be underestimated . DOI: http://dx . doi . org/10 . 7554/eLife . 10807 . 019 The prior strand-exchange reactions involve three DNA strands , with one substrate and one product entirely single-stranded . RecA can also catalyze strand-exchange between two duplex molecules , provided that strand exchange is initiated at a short ssDNA gap in the substrate . These reactions produce full 4-strand Holliday junction intermediates , as opposed to the 3-strand junctions in the 3-strand reactions above ( Figure 7A ) . To determine if RadA could accelerate branch migration between 4 DNA strands involving true Holliday junctions , we performed RecA strand exchange reactions between linear dsDNA molecules and circular dsDNA with a 1346 nt ssDNA gap . All other conditions were identical to the 3-strand reactions above and the 3-strand reactions were performed in parallel to the 4-strand reactions ( Figure 7B ) . Although RecA by itself efficiently promoted joint molecule formation in the 4-strand reaction , formation of the final branch-migrated product ( nicked dsDNA circle , 'NP' ) was inefficient relative to the 3-strand reaction after 30 min , and most DNA was found in various branched intermediate forms . In the RecA RadA-coupled reaction , the final product was visible at the first time point , 10 min , and accumulation of intermediates was not observed . Therefore , RadA can stimulate branch migration of 4-strand Holliday junctions , as well as 3-strand junctions . 10 . 7554/eLife . 10807 . 020Figure 7 . Four-strand Recombination Reactions in the Presence of RadA . ( A ) Diagram of the Four-strand Recombination Reaction . Gapped circular substrate ( GS ) prepared as described in the procedures was mixed with double-strand φX174 DNA linearized with PstI ( DLS ) in the presence of RecA , SSB , RadA , ATP and an ATP regenerating system . Complex , largely duplex DNA intermediates are formed first . The final products are nicked circular double-DNA ( NP ) and Duplex Linear DNA with Single-strand Tails ( DLP ) . Note: The tailed linear product species is not well-resolved from the duplex linear substrate ( DLS ) . ( B ) Comparison of 3-strand and 4-strand Recombination Mediated by RecA in the Presence and Absence of RadA . Recombination reactions between either single-strand circular φX174 DNA ( SCS ) and double-strand φX174 DNA linearized with PstI ( DLS ) -3-strand reactions or double-strand circular φX174 with a 1 . 3 kB single-strand gap ( GS ) and double-strand φX174 DNA linearized with PstI ( DLS ) -4-strand reactions were performed as described . At the times indicated , reactions were stopped and de-proteinated . Products were resolved using an 1 . 0% agarose gel in TAE buffer . DOI: http://dx . doi . org/10 . 7554/eLife . 10807 . 020
RadA is a ubiquitous RecA-paralog protein found in eubacteria and plants . Genetic studies have implicated RadA in homologous recombination , particularly in the late steps of recombination intermediate processing . This work presented here provides a biochemical rationale for this role , showing that purified RadA protein can mediate branch migration of recombination intermediates , in the context of a RecA filament and in a direction 5’ to 3’ with respect to the initiating single-strand . In theory , branch migration can promote homologous recombination in several ways ( Figure 8A ) . If strand exchange is initiated at a site removed from a 3’ end , branch migration can serve to engage the 3’ strand into the heteroduplex region , providing a paired 3’ end that can be extended by DNA polymerases . RadA’s directionality is consistent with this role . Furthermore , RadA's inhibition by SSB may prevent the reverse reaction , the dissolution of this heteroduplex . Branch migration also can extend the heteroduplex region formed between donor and recipient DNA strands , which may aid its stability . Therefore , we might expect RadA to aid the process known as 'break-induced replication' ( Anand et al . , 2013 ) , during which a resected linear DNA fragment invades a homologous duplex region , and establishes a replication fork . Indeed , RadA is strongly required for exchange events believed to be associated with breakage of the replication fork in vivo ( Lovett , 2006 ) . 10 . 7554/eLife . 10807 . 021Figure 8 . How branch migration assists homologous recombination . ( A ) Heteroduplex extension . In reactions between linear resected DNA and an intact chromosome , initial strand pairing and invasion may occur at a distance from the 3’ end . Branch migration of the D-loop ( in direction of the arrow ) allows the heteroduplex region to extend fully to the 3’ end , allowing it to be engaged by DNA polymerases . Branch migration also allows the D-loop to be extended , lengthening and stabilizing the region of heteroduplex and forming a 4-strand Holliday juntion . ( B ) Synthesis-dependent strand annealing ( SDSA ) . After resection of a broken chromosome and strand invasion into a sister molecule , branch migration is required to dissolve the intermediate , allowing broken strands to anneal to one another and the break to be healed . Reactions contained 1 mM ATP and 20 . 3 µM ( nucleotide ) φX174 DNA . The values shown are the average of two experiments , except for the circular single-strand value . It is the average of five experiments . Standard deviations are reported . DOI: http://dx . doi . org/10 . 7554/eLife . 10807 . 021 Branch migration also can dissociate recombination intermediates and is integral to a recombination reaction known as 'synthesis-dependent strand-annealing' ( SDSA ) , a process that can heal double-strand breaks ( DSBs ) without crossing-over ( Figure 7B ) . This mechanism underlies a number of repair events associated with DSBs , including transposon-mediated breaks in Drosophila ( Nassif et al . , 1994 ) , mating-type switching in yeast ( Haber et al . , 2004 ) and radiation damage-repair in the bacterium Deinococcus radiodurans ( Zahradka et al . , 2006 ) . Initial homologous strand-exchange allows DNA synthesis across the region of the break; subsequent branch migration dissolves the intermediate to allow the broken strand to anneal to itself and to be joined . In Deinococcus radiodurans , radA is required for the SDSA reactions that aid extreme radiation resistance in this organism ( Slade et al . , 2009 ) . In this case , RadA protein appears to promote the initial joints that prime DNA synthesis but could , in theory , also participate in the subsequent branch migration that resolves such joints to allow annealing . Alternatively , this dissolution might be catalyzed by RecG , whose polarity seems suited for this role ( Whitby et al . , 1993 ) . Catalysis of branch migration can also allow bypass of barriers such as DNA lesions or regions of non-homology that are sufficient to block spontaneous thermal branch migration . E . coli possesses three branch migration systems that participate in recombination , RadA , RecG and RuvAB . Genetic analysis indicates that these systems are both somewhat specialized and somewhat redundant ( see discussion [Cooper et al . , 2015] ) . How does branch migration catalyzed by RadA differ from that promoted by RecG and RuvAB ? One difference is that RadA can function in the context of the RecA synaptic filament , with branches migrated in the direction of RecA-promoted strand transfer . In contrast , when purified RuvAB or RecG are added to ongoing RecA-strand transfer reactions , they decrease the recovery of full strand-exchange products , by accelerating the reverse reaction back to substrate forms ( Whitby et al . , 1993 ) . On RecA-free intermediates , RecG and RuvAB migrate branches preferentially to substrate and product forms , respectively . There are two mechanisms by which proteins mediate ATP-driven branch migration , the first exemplified by the E . coli RuvAB complex . RuvA forms a tetramer , which specifically binds Holliday junctions; RuvB acts as two hexameric complexes , flanking RuvA and encircling duplex DNA ( Parsons et al . , 1995; Yamada et al . , 2002 ) . RuvB acts as the motor , pumping DNA through the complex and thereby moving the position of the junction . RuvAB has classical helicase activity ( Tsaneva et al . , 1993 ) , unwinding DNA strands , and , through the RuvA complex , special affinity for branched structures ( Parsons and West , 1993 ) . The magnitude of RuvAB ATPase activity depends on the DNA structures to which it is bound ( Abd Wahab et al . , 2013 ) . The RuvAB complex interacts with a nuclease component , RuvC , coupling branch migration to junction cleavage ( West , 1997 ) . In E . coli branch migration might be limited to providing the preferred sequence ( A/T TT G/C ) for RuvC cleavage ( Shah et al . , 1994; Shida et al . , 1995 ) . RecG , the other branch-migration protein in E . coli , is a DNA translocase with special affinity for branched structures; it is also believed to branch migrate DNA via a motor mechanism ( Whitby and Lloyd , 1998; Whitby et al . , 1993 ) . On the other hand , RecA protein catalyzes branch migration by a distinctly different mechanism involving strand-exchange between DNA in the primary and secondary DNA-binding-sites on the RecA filament . Site II-bound DNA has been modeled with the RecA filament as a helix with same average pitch as Site-I-bound DNA ( the latter visible in the crystal structure , [Chen et al . , 2008] ) ; this modeled Site II-bound DNA , however , has a larger radius and an even more highly extended DNA structure than Site I-bound DNA ( reviewed in [Prentiss et al . , 2015] ) . ATP hydrolysis is required for RecA strand transfer over extended distances ( Jain et al . , 1994 ) ; how ATP hydrolysis promotes branch migration via the RecA mechanism is not well understood . Unlike RuvAB , RecA has no special affinity for branched DNA structures nor can it act as a DNA helicase/translocase . The mechanism by which RadA branch migrates DNA is not known , although its RecA-like sequence character , lack of structure-specific binding or helicase activity might suggest a RecA-like mechanism . Further analysis should be revealing . Although we have no direct evidence for this , the RecA-like structure of RadA suggests that it might be recruited to a RecA filamens , interacting at its natural interface . Moreover , our ADP inhibition experiments raise the possibility that RadA may destabilize the RecA filament . In vivo , RecA foci become more numerous and persistent in radA mutants of E . coli ( Massoni et al . , 2012 ) , consistent with a role for RadA in RecA postsynaptic filament destabilization . This property may serve to provide a handoff of recombination intermediates from RecA to RadA , facilitating the completion of recombination . Strand-exchange paralog proteins are universally found in archaea , eubacteria and eukaryotes . Humans have five such Rad51 proteins , in addition to true strand exchange proteins Rad51 and Dmc1 ( reviewed in [Gasior et al . , 2001] ) . These proteins are required for homologous recombination , albeit to a lesser extent , than their true strand exchange-protein counterparts . The few that have been studied biochemically appear to affect the presynaptic phase of strand exchange by enhancing formation or stability of the Rad51 filament . Rad55/Rad57 of yeast interact with Rad51 ( Johnson and Symington , 1995 ) and act as Rad51-mediator proteins to allow Rad51 to overcome inhibition by single-strand DNA-binding protein , RPA , in formation of the Rad51 presynaptic filament ( Sung , 1997 ) . In the archaea , Sulfolobus tokodaii StRad55 protein appears to play a similar role ( Sheng et al . , 2008 ) . In addition , yeast Rad55/Rad57 have an additional role in stabilizing the Rad51 filament against dissociation by the Srs2 helicase ( Liu et al . , 2011 ) . The effect of bacterial RadA on late stages of recombination , evident both in vivo and in vitro , presents a new paradigm for strand-exchange paralog proteins that may be shared in other organisms . RFS-1 , the sole Rad51 paralog of C . elegans has properties consistent with a late recombination role and a RFS-1 peptide can disrupt Rad51 filaments in vitro ( Adelman and Boulton , 2010 ) , although recent evidence supports a presynaptic role in remodeling the Rad51 filament to a more flexible form ( Taylor et al . , 2015 ) . The human Rad51C-XRCC3 paralog complex may also have a late recombination role: depletion of Rad51C from cell extracts reduces branch migration capacity and copurifies with HJ cleavage activity ( Liu et al . , 2004; 2006 ) . Although RecA can promote the homology search process , pair DNA and promote branch migration , RadA may be a specialized form , selected for its ability to catalyze faster branch migration and incompetent for homology-search and pairing . Because of its role in synapsis , RecA binding needs to be highly specific for ssDNA , lest it bind indiscriminately to the undamaged chromosome . RadA cannot pair DNA and has weak ability to bind ssDNA , in comparison to RecA . Its capacity to bind DNA in the presence of ADP is an intriguing property . In a study of a RecA mutant ( P67G E68A , near the Walker A motif ) , strand-exchange between lengthy homologies ( but not homologies <2 kb ) was highly stimulated by ADP and completely inhibited by ATP or an ATP-regenerating system . This finding suggests that the ADP-form of the RecA filament is required , in some way , for the branch migration phase of strand-exchange , as revealed by this particular mutant . RadA binding behavior may naturally reflect this propensity and the stable binding of RadA in the presence of ADP may explain its superior ability to promote branch migration , relative to that by RecA . RadA has stronger dsDNA-stimulated ATPase activity relative to RecA , which may also assist branch migration . Because the role of ATP hydrolysis in RecA strand-exchange is still unclear , further study of RadA-mediated branch migration may provide valuable insights into this mechanism . Our study shows that RadA’s ATPase , 'RadA motif' ( KNRFG ) and Zn-finger motif are essential to the biochemical function of the protein , consistent with our prior genetic results ( Cooper et al . , 2015 ) . We hypothesize , based on the position of KNFRG element in the RadA sequence , that this motif is required to assemble the ATPase site ( comparable to K248 K250 region of RecA [Chen et al . , 2008] ) at an interface . The Zn-finger appears to assist ssDNA binding but it may also facilitate some protein ( SSB , RecA ? ) or DNA interaction ( branched molecules ? ) , since it is required for RadA’s stimulation of branch migration in RecA-coupled reactions . Our study does not address the ability of RadA to form a filament or other oligomeric structure with itself or with RecA , an area for further investigation . Threading of RadA onto the RecA presynaptic crystal structure ( our unpublished results ) suggests that RadA possesses subunit interfaces similar to that of RecA that assemble the ATPase site . In a bacterial one-hybrid assay , RadA was shown to interact with itself ( Marino-Ramirez et al . , 2004 ) , indicating that it forms a multimeric complex . Our early experiments with His6-tagged RadA protein , a less active protein than the more native protein characterized here , exhibited multiple bound species in gel-shift experiments with poly ( dT ) ( data not shown ) , consistent with oligomer formation . RadA’s ability to reduce ATPase activity of RecA in the presence of ssDNA and SSB , its enhancement of ADP-inhibition of RecA-mediated strand exchange and the ability of RadA K258A ( and to a lesser extent C28Y and K108R ) to inhibit RecA-mediated branch migration are consistent with the notion that RadA joins and destabilizes the RecA filament . Biochemical confirmation of this hypothesis is ongoing .
Biochemicals were purchased from USB or Sigma unless noted . Wild-type RecA was a kind gift from Shelley Lusetti ( New Mexico State University ) or purchased from Epicentre Biotechnologies ( Madison , WI ) . D72R Mutant RecA was generously provided by Michael Cox ( University of Wisconsin-Madison ) . Single-strand DNA-binding Protein ( SSB ) was from Promega ( Madison , WI ) and T4 polynucleotide kinase and restriction enzymes were from New England Biolabs ( Ipswich , MA ) . Regenerating system enzymes were from Sigma-Aldrich ( St . Louis , MO ) . φX-174 and M13 RF and virion DNAs were purchased from NEB , pBS SK- was from Agilent Technologies ( Santa Clara , CA ) and pPAL7 was from Bio-Rad Laboratories ( Hercules , CA ) . Oligonucleotides were purchased from Sigma-Aldrich . The wild-type RadA gene was amplified by PCR and cloned into the high copy vector pBS SK- . Site-directed mutants were made from this construct by PCR as described ( Cooper et al . , 2015 ) . Subsequently for protein purification , wild type and mutant RadA DNA was amplified from the pBS SK- constructs using the eXact primers ( Table 2 ) and then cloned into the vector pPAL7 . All constructs were confirmed by sequencing . 10 . 7554/eLife . 10807 . 022Table 2 . Oligonucleotides used in this study . DOI: http://dx . doi . org/10 . 7554/eLife . 10807 . 022Oligonucleotide nameSequenceradAeXactFGGAAGCTTTGACTTCTGTGGCAAAAGCTCCAAAACGradAeXactRTTTGCGGCCGCTTATAAGTCGTCGAACACGCpolyd ( N ) 33TTAGCGGCCGCATAGTCAAGATGACAATGTTCTSubstrate E2CGGTCAACGTGGGCATACAACGTGGCACTG ( T ) 30ATGTCCTAGCAAAGCGTATGTGATCACTGGJxn1CCGCTACCAGTGATCACCAATGGATTGCTAGGACATCTTTGCCCACCTGCAGGTTCACCCJxn2TGGGTGAACCTGCAGGTGGGCAAAGATGTCCTAGCAATCCATTGTCTATGACGTCAAGCTJxn3AGCTTGACGTCATAJxn4GATCACTGGTAGCGGJxn5TGCCGATATTGACAAGACGGCAAAGATGTCCTAGCAATCCATTGGTGATCACTGGTAGCGG Wild-type or mutant eXact-tag constructs were transformed into BL21 Codon-Plus or BL21 AI ( Agilent Technologies ) deleted for endA::Km . Typically , the transformation mix was grown for 1 hr at 37 °C in SOC media ( 2% tryptone , 0 . 5% yeast extract , 20 mM glucose , 10 mM NaCl , 10 mM MgCl2 , 2 . 5 mM KCl ) , diluted into 20 ml SOC supplemented with ampicillin ( 'Ap' , 100 μg/ml ) and then grown standing overnight . This culture was used as the inoculum for a 1 l LB ( 2% tryptone , 1% yeast extract , 0 . 5% NaCl ) culture with Ap . Cultures were grown until the A590 reached approximately 0 . 8 at which time IPTG ( Gold Biotechnology , St . Louis , MO ) was added to 1 mM and arabinose was added to 0 . 2% ( for BL21 AI strains ) . Growth was continued for 3–4 hr at 30 °C . For production of RadAC28Y and RadAK108A mutant protein , growth conditions were altered so that a 500 ml culture was grown from the initial 20 ml inoculum in LB supplemented with 0 . 2% glucose and 100 μg/ml ampicillin until the culture reached an A590 of 1 . 0 . Cells were then diluted into LB with 0 . 4% arabinose , 1 mM IPTG and fresh ampicillin . Growth was continued 2 hr at 30 °C . After growth of all strains , cells were collected by centrifugation , and the resulting pellet was frozen and stored at -20 °C . RadA was purified with slight modifications from the Biorad eXact protocol . Cleavage of the N-terminal eXact tag produces a RadA protein with the addition of a 2 N-terminal amino acids , threonine serine , ( necessary to facilitate efficient cleavage ) . Typically , pellets from 200 ml of wild type RadA culture were resuspended in 20 ml eXact buffer ( 100 mM sodium phosphate , pH 7 . 2 , 10% glycerol , 10 mM beta-mercaptoethanol , 300 mM sodium acetate . ( The final pH was adjusted to 7 . 2 if necessary ) . Cells were lysed by adding lysozyme ( in eXact buffer ) to 200 μg/ml and incubating for 45 min on ice . Lysis was completed by incubation at 37 °C for 2 min followed by homogenization using a Dounce Homogenizer on ice ( 5 passes with B , followed by 5 passes with A ) . The crude lysate was clarified by centrifugation at 17000 x g for 30 min at 4 °C . One third of the cleared lysate was applied to a 1 ml eXact pre-packed column equilibrated in eXact buffer at room temperature . The column was then washed with 15–20 ml eXact buffer . To cleave the eXact Tag from RadA , 2 ml cleavage buffer ( eXact buffer + 100 mM NaF ) was applied to the column , and the column was capped and incubated at room temperature for 45–60 min . RadA was then eluted from the column with eXact buffer + 100 mM NaF . Fractions containing RadA were pooled , diluted in small batches in Q buffer ( 20 mM Tris-HCl , pH 8 . 0 , 0 . 5 mM EDTA , 15% ( v/v ) ethylene glycol , 10 mM beta-mercaptoethanol ) without salt to give a conductivity of approximately 125 mS-1 and applied to a 1 ml Q HP column ( GE Healthcare ) equilibrated with125 mM NaCl Q buffer . After washing with 15 ml 125 mM NaCl Q Buffer , RadA was eluted with a linear gradient from 125 mM NaCl in Q buffer to 750 mM NaCl in Q buffer with 20% glycerol replacing the ethylene glycol . RadA eluted at a conductivity approximately 275 mS-1 . To retain activity , fractions containing RadA were immediately flash-frozen in small aliquots and stored at -80 °C . Thawed aliquots were used within 24 hr . Purification from 200 ml of cells yielded approximately 0 . 75–1 mg of highly purified RadA Protein ( Figure 1—figure supplement 1 ) . For some experiments , purified RadA was concentrated using Q HP Sepharose beads ( GE Healthcare , Chicago , IL ) . RadA protein concentration was spectrophotometrically determined using an extinction coefficient of 22 , 460 M-1 cm-1 ( Gasteiger et al . , 2005 ) or by the Bradford method ( Bradford , 1976 ) , which gave equivalent results . ATP hydrolysis activity of RadA mutants was tested either by measuring release of inorganic 32P from ATP ( PerkinElmer , Waltham , MA ) using thin layer chromatography ( TLC ) with polyethylenimine plates ( PEI , Macherey-Nagel , Düren , Germany ) as the solid phase and 0 . 5 M LiCl/ 4 . 3% formic acid as the mobile phase ( Kornberg et al . , 1978 ) . Reactions were incubated at 37 °C for the times indicated and contained 10 mM Bis-Tris-Propane-HCl , pH 7 . 0 , 10 mM MgCl2 , 2 mM DTT , 1mM ATP . ATP hydrolysis by wild-type RadA protein was measured in reactions that included an ATP regenerating system and were coupled with NADH oxidation . Oxidation of NADH was monitored spectroscopically at 380 nm ( Extinction coefficient = 12100 M-1cm-1 ) using a Synergy H1 Microplate Reader and Gen5 Data Collection and Analysis Software ( Biotek , Winooski , VT ) . Reactions included 1 µg of φX174 single-strand circular DNA , 10 mM Bis-Tris-Propane-HCl , pH 7 . 0 , 10 mM MgCl2 , 2 mM DTT , 1mM ATP , 2 mM DTT , 3 . 5 mM phosphoenol pyruvate , 10 u/ml pyruvate kinase , 2 mM NADH , and 10 u/ml lactate dehydrogenase . For all reactions , ATP concentrations were determined at 260 nm using an extinction coefficient of 15400 M-1cm-1 Oligonucleotides ( Table 2 ) were 5’ end-labeled with 32P-ATP using T4 polynucleotide kinase and manufacturer’s conditions ( New England Biolabs ) . Excess 32P-ATP was removed from the reaction using Sephadex G-50 columns ( Roche , Basel , Switzerland ) . Double-strand substrates were made by heating two complementary oligonucleotides to 95 °C and then cooling to room temperature slowly . Standard DNA-binding reactions contained RadA as indicated , 100 fmol DNA , 50 mM Tris-HCl buffer , pH 7 . 5 , 10 mM MgCl2 , 0 . 1 mM EDTA , 75 mM NaCl , 5 mM dithiothreitol , 100 µg/ml bovine serum albumin and 1 mM nucleotide . Reactions were incubated at 37 °C for 30 min and then resolved on a 6% Tris-borate EDTA polyacrylamide gel ( pre-run for 1 hr ) at 100 V for 45 min at room temperature . Gels were then dried and binding was analyzed using ImageJ-64 software with scanned autoradiographs ( HiBlot CL film-Denville Scientific , Holliston , MA ) . To form branched substrates , equimolar concentrations of oligonucleotides ( Table 2 ) were mixed and heated to 100° for 5 min followed by slow cooling to room temperature in 10 mM Tris Acetate , pH 7 . 4 , 10 mM Mg Acetate , and 50 mM K Acetate . The extent of branched molecules formation was assessed either by electrophoresis using 3% agarose gels in TAE followed by staining with ethidium bromide or by using oligo 2 radio-labeled with polynucleotide kinase and 32P-ATP to form branched molecules ( in parallel reactions ) followed by electrophoresis on 6% acrylamide gels in TBE and autoradiography . Contaminating structures were present at less than 5% of the total , except for the 3-strand fork when contaminating structures were between 5 and 10% . Annealed substrates were used without further purification . Fork 1 was made by annealing Jxn1 and Jxn 2 oligonucleotides . Fork 2 was made from Jxn1 , Jxn 2 , and Jxn3 oligonucleotides . Fork 3 was made by annealing Jxn1 , Jxn2 , and Jxn4 oligonucleotides . Fork 4 was the annealed product of Jxn1 , Jxn2 , Jxn 3 , and Jxn 4 oligonucleotides . The 3-stranded fork was constructed from Jxn1 , Jxn2 , and Jxn 5 . oligonucleotides . Figure 2—figure supplement 1 shows a diagram of each branched substrate structure . φX174 double strand circular DNA was cleaved with BsaAI and the 4 . 1 kb fragment was purified from 0 . 7% low-melt agarose ( USB , Cleveland , OH ) using Gene Jet gel purification columns ( Thermo-Fisher Scientific , Waltham , MA ) . Gapped molecules were made using a large-scale recombination reaction using conditions above ( without RadA ) with the BsaAI fragment replacing the full length linear fragment and incubation extended to 1 . 5–2 . 0 hr . The reaction mix was then extracted with phenol two times . After back-extracting the organic phase with 1 volume of 10 mM Tris-EDTA , all aqueous phases were combined and extracted with chloroform:isoamyl alcohol ( 24:1 ) . Sodium acetate , pH 5 . 2 was added to 0 . 3 M and DNA was precipitated using 2 . 5 volumes of ethanol . DNA pellets were then washed with 70% ethanol and resuspended in 10 mM Tris-EDTA . Finally , gapped DNA was gel-purified as described above . Large-scale standard recombination reactions were stopped with after 12 min by addition of EDTA to 35 mM , SDS to 0 . 65% , and Proteinase K to 800 µg/ml and applied to a 3 . 5 ml Sepharose 4B-CL equilibrated in 20 mM Tris-acetate pH 7 . 4 , 3 mM ammonium glutamate , 1 mM DTT , 2% glycerol , and 11 mM magnesium acetate . Fractions ( 150–200 µl ) containing DNA were identified by staining with Picogreen . DNA fractions containing recombination intermediates were incubated at 37 °C in 20 mM Tris-acetate pH 7 . 4 , 3 mM ammonium glutamate , 1 mM DTT , 2% glycerol , 11 mM magnesium acetate , and approximately 30 mM NaCl from the RadA protein with RadA ( 460 nM ) , 3 mM ATP , and , when included , SSB ( 2 . 1 µM ) . Reactions were stopped and analyzed as outlined above . | Damage to the DNA of a cell can cause serious harm , and so cells have several ways in which they can repair DNA . Most of these processes rely on the fact that each of the two strands that make up a DNA molecule can be used as a template to build the other strand . However , this is not possible if both strands of the DNA break in the same place . This form of damage can be repaired in a process called homologous recombination , which uses an identical copy of the broken DNA molecule to repair the broken strands . As a result , this process can only occur during cell division shortly after a cell has duplicated its DNA . One important step of homologous recombination is called strand exchange . This involves one of the broken strands swapping places with part of the equivalent strand in the intact DNA molecule . To do so , the strands of the intact DNA molecule separate in the region that will be used for the repair , and the broken strand can then use the other non-broken DNA strand as a template to replace any missing sections of DNA . The region of the intact DNA molecule where the strands need to separate often grows during this process: this is known as branch migration . In bacteria , a protein called RecA plays a fundamental role in controlling strand exchange , but there are other , similar proteins whose roles in homologous recombination are less well known . Cooper and Lovett have now purified one of these proteins , called RadA , from the Escherichia coli species of bacteriato study how it affects homologous recombination . This revealed that RadA can bind to single-stranded DNA and stimulate branch migration to increase the rate of homologous recombination . Further investigation revealed that RadA allows branch migration to occur even when RecA is missing , but that RadA is unable to begin strand exchange if RecA is not present . The process of branch migration stabilizes the DNA molecules during homologous recombination and may also allow the repaired DNA strand to engage the machinery that copies DNA . Cooper and Lovett also used genetic techniques to alter the structure of specific regions of RadA and found out which parts of the protein affect the ability of RadA to stimulate branch migration . Future challenges are to find out what effect RadA has on the structure of RecA and how RadA promotes branch migration . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"chromosomes",
"and",
"gene",
"expression",
"microbiology",
"and",
"infectious",
"disease"
] | 2016 | Recombinational branch migration by the RadA/Sms paralog of RecA in Escherichia coli |
Achieving confidence in the causality of a disease locus is a complex task that often requires supporting data from both statistical genetics and clinical genomics . Here we describe a combined approach to identify and characterize a genetic disorder that leverages distantly related patients in a health system and population-scale mapping . We utilize genomic data to uncover components of distant pedigrees , in the absence of recorded pedigree information , in the multi-ethnic BioMe biobank in New York City . By linking to medical records , we discover a locus associated with both elevated genetic relatedness and extreme short stature . We link the gene , COL27A1 , with a little-known genetic disease , previously thought to be rare and recessive . We demonstrate that disease manifests in both heterozygotes and homozygotes , indicating a common collagen disorder impacting up to 2% of individuals of Puerto Rican ancestry , leading to a better understanding of the continuum of complex and Mendelian disease .
During the past two decades major advances in deciphering the genetic basis of human disease have resulted in thousands of disorders that are now understood at a genetic level ( Hamosh et al . , 2000; McCarthy et al . , 2013 ) . This progress has led to the integration of genomic sequencing in clinical care , especially for the diagnosis of rare genetic disease ( Bamshad et al . , 2011; Manolio et al . , 2013 ) , and clinical sequencing is increasingly offered to patients with known or suspected genetic disorders . In the past few years , large national and international efforts ( Manolio et al . , 2015; Philippakis et al . , 2015; Chong et al . , 2015a ) have emerged to enable patients and health systems to share knowledge of rare genetic disorders and improve genetic testing , resulting in improved healthcare management and outcomes for patients . In parallel , many large regional and national biobank efforts ( Collins and Varmus , 2015; Ashley , 2015; Collins , 2012 ) are underway to enable the broad integration of genomics in health systems for genetic identification of disease ( Dewey et al . , 2016 ) . Such efforts have recently revealed clinically actionable variants ( Dewey et al . , 2016; Feldman , 2016 ) and genetic disorders segregating at higher frequencies in general patient populations than previously suspected . The increased promulgation of genomics in health systems represents an opportunity to improve diagnostic sensitivity for more precise therapeutic intervention and better health outcomes ( Ashley , 2016 ) . Despite this progress , most genetic diseases are still under-diagnosed ( Abul-Husn et al . , 2016 ) or misdiagnosed ( Yang et al . , 2014; Taylor et al . , 2015; Centers for Mendelian Genomics et al . , 2015b ) . A number of barriers exist for wholesale genetic testing and diagnoses , including incomplete standardized guidelines for interpreting genetic evidence of disease ( Amendola et al . , 2016 ) , variable penetrance or expressivity of phenotype ( Katsanis , 2016 ) , and that the causal variant may be missed or mis-assigned during testing ( Manrai et al . , 2016 ) . The latter is a particularly pernicious problem in non-European populations due to systematic biases in large genomic and clinical databases ( Popejoy and Fullerton , 2016; Petrovski and Goldstein , 2016 ) . These challenges have led several research groups to attempt to genetically identify disease by examining patient health patterns using data from the Electronic Health Record ( EHR ) ( Gottesman et al . , 2013; Green et al . , 2016 ) . EHRs contain comprehensive information on medical care throughout a patient’s life , including medications , medical billing codes , physician notes and generated reports ( i . e . pathologic , genetic and radiologic reports ) . EHRs have been used to clinically characterize well-known genetic disorders , but have been of limited success for the vast cadre of less-characterized or unknown disorders ( Blair et al . , 2013 ) . The gold standard of genetic disorder diagnosis involves testing both patient and family members to confirm Mendelian segregation of the suspected underlying pathogenic variant ( ClinGen et al . , 2015; ClinGen Resource et al . , 2015; Gahl et al . , 2016 ) . However , as genomic data becomes more ubiquitous in health systems , it can be used to detect genetic relationships in the absence of known family and pedigree information . Specifically , components of pedigrees can be uncovered within the general population; particularly those that have experienced recent founder effects . Pairs of individuals who are related share genetic homology in the form of long genomic haplotypes . These haplotypes are considered to be identical-by-descent ( IBD ) if they are inherited from a common ancestor without any intervening recombination . The chance of any two people sharing a tract of their genome IBD decays exponentially , with a ~50% reduction in the chance of sharing per generation . However , when IBD sharing does occur , the length of an IBD segment can remain long even between distantly related individuals . In practice , long tracts of IBD ( >3 cM ) can be accurately detected using genetic data between individuals with a common ancestor from the past 4–50 generations ( Browning and Browning , 2012 ) . Detection of IBD haplotypes can allow for the identification of distantly related patients with a genetic disorder driven by a locus inherited from a founding ancestor who brought the disease mutation into a population ( Houwen et al . , 1994; Kenny et al . , 2009; Henden et al . , 2016; Meta-Analysis of Glucose and Insulin-related traits Consortium ( MAGIC ) et al . , 2010; Traherne et al . , 2016; Shaw et al . , 2015; Ko et al . , 2014; Lalli et al . , 2014 ) . This is the principle underlying population-scale disease mapping approaches that combine IBD sharing and statistical association to discover novel disease loci , so called IBD-mapping . By detecting genetic relatedness , as inferred by IBD sharing , we hypothesized that we may be able to detect hereditary forms of disease in an EHR-linked biobank . With over 38000 participants , the BioMe biobank , at the Icahn School of Medicine at Mount Sinai , New York City ( NYC ) , is one of the most diverse cohorts ascertained at a single urban medical center under a uniform study protocol . Participants are largely from the local Upper East Side , Harlem and Bronx communities , and represent broad ancestral , ethnic , cultural , and socioeconomic diversity . We initially focused on adult height , which is easily measurable , stable over the adult life course , and one of the most abundantly recorded clinical parameters in EHRs . Height is known to be highly heritable and polygenic ( Visscher et al . , 2010; Lango Allen et al . , 2010 ) , however , extremes of short stature can be caused by rare variants in single genes with large effect sizes ( Durand and Rappold , 2013a ) . Although , many genetic syndromes are known to cause short stature , most of the time no definitive etiology underlying short stature is found in patients . Here we used loci associated with elevated genetic relatedness as measured by IBD to map a locus underlying extreme short stature in the BioMe biobank , and linked it to a known , but little characterized , collagen disorder previously thought to be rare . By interrogating a large global diversity panel , we demonstrated that this variant is actually common in Puerto Rican populations . Furthermore , we leveraged the EHR to show significant musculoskeletal disease in both heterozygous and homozygous patients , indicating the disease is not simply a recessive disorder as had previously been thought . Finally , we showed how this work can generate broad insights for sustainable adoption and large-scale dissemination of genomic medicine .
The BioMe biobank comprises a highly diverse cohort , with over 65% of participants self-reporting as Black/African-American or Hispanic/Latino/a , and over 35% born outside mainland US , representing more than 110 countries of origin . First we estimated patterns of direct relatedness in a subset of BioMe participants genotyped on the Illumina OmniExpress array ( N = 11212 ) by detecting pairwise identity-by-state using RELATEAdmix ( Moltke and Albrechtsen , 2014 ) , a method that accounts for admixture in populations ( Figure 1—figure supplement 1 ) . We observed that 701 individuals had primary ( parent-child , sibling ) or secondary ( avuncular , grandparental ) relationship with another participant in BioMe , and we removed these individuals from all downstream analysis . Next we devised a strategy to divide the diverse BioMe biobank into population groups for downstream analysis . We combined genotype data for BioMe participants ( N = 10511 ) with 26 global populations from the 1000 Genomes Project ( 1KGP; N = 2504 ) ( The 1000 Genomes Project Consortium , 2015 ) and two additional panels of Native American ( N = 43 ) ( Mao et al . , 2007 ) and Ashkenazi Jewish populations ( N = 100 ) ( see Methods ) . Using a common set of 174468 SNPS we performed principal component analysis ( Price et al . , 2006 ) ( PCA; Figure 1—figure supplement 2 ) . Based on both self-reporting and patterns of genetic diversity observed in BioMe participants , we stratified individuals into four broad population groups . The first group self-reported as European American , but were also genetically identified as Ashkenazi Jewish ( AJ; N = 808 ) as they clustered distinctly with an AJ reference panel and separately from other European ancestry groups in PCA space ( Figure 1—figure supplement 3 ) . The other three groups we defined using self-reported race/ethnicity categories , African-American ( AA; N = 3080 ) , Hispanic/Latino/a ( H/L; N = 5102 ) and European-Americans with no AJ genetic ancestry ( Non-AJ EA; N = 1270 ) ( Figure 1—figure supplement 3 ) . An additional 251 individuals who reported ‘Mixed’ ( N = 89 ) or ‘Other’ ( N = 162 ) ethnicity were excluded from further analysis . To evaluate signatures of distant relatedness BioMe biobank participants , we estimated sharing of genomic tracts IBD >3 cM between every pair of individuals using the GERMLINE software ( Gusev et al . , 2009 ) . The minimum length of 3 cM was chosen based on reports of elevated type I error in call rates of smaller lengths ( Gusev et al . , 2012; Chiang et al . , 2016 ) . It is known that population-level rates of distant relatedness are observed to be particularly elevated after population bottlenecks ( i . e . in founder populations ) ( Browning and Thompson , 2012 ) . We summed the length of all IBD-tracts shared between a given pair of individuals if they shared more than one tract and examined the distribution of pairwise sharing at a population level . We observed elevated levels of distant relatedness in both the AJ ( median summed length of IBD sharing within population = 44 . 7 cM; 95% C . I . = 44 . 66–44 . 82 cM ) and HL ( 16 . 2 cM; 16 . 18–16 . 22 cM ) populations , compared to AA ( 3 . 77 cM; 3 . 76–3 . 77 cM ) or non-AJ EA ( 4 . 5 cM; 4 . 45–4 . 55 cM ) populations ( Figure 1A ) . This is congruent with previous reports of founder effects in both AJ populations ( Need et al . , 2009 ) and in some H/L populations ( Moreno-Estrada et al . , 2013 ) . Hispanic or Latino/a is a broad ethnic label encompassing myriad populations with origins in Northern , Southern or Central America , century-long roots in New York City , and genetic ancestry from Africa , Europe and the Americas . To explore the signature of a founder effect in the BioMe H/L population , we leveraged self-reported and genetic information about sub-continental ancestry . By self-reporting , the H/L participants in BioMe were born in New York City ( NYC ) ( 40% ) , Puerto Rico ( 24% ) , Dominican Republic ( 19% ) , Central/South America ( 12% ) , Mexico ( 2% ) or other Caribbean Island ( 2% ) ( Figure 1B ) . We examined IBD tract length distributions within H/L sub-continental populations and observed that the founder effect was predominantly driven in the Puerto Rican-born group ( Figure 1C ) . We assembled a cohort of Puerto Ricans including BioMe participants who were either born in Puerto Rico or , were born in NYC and had 2 parents or 3–4 grand parents who were born in Puerto Rico ( N = 1245 ) . Approximately 5086 NYC-born H/L individuals did not have recorded parental or grandparental country-of-origin , therefore we also devised a selection strategy using PCA analysis . We identified BioMe H/L participants on the cline between the African and European reference panels in PCA space coincident with Puerto Rican-born individuals . We excluded those on the same cline with ancestry from the Dominican Republic or another Caribbean Island ( Figure 1—figure supplement 4 ) , and counted the remainder ( N = 1571 ) in the Puerto Rican group . In total , we estimated 2816 H/L in the BioMe discovery cohort were of Puerto Rican ancestry , and focused the downstream analysis on this group as the largest founder population in BioMe . Next we tested the hypothesis that rare , recessive disease variants may have arisen to appreciable frequency in the Puerto Rican founder population . We linked genomic data to clinical data in the Electronic Health Record ( EHR ) of the Mount Sinai Health System . We focused on height , a stable and ubiquitous health measure . Clinically , rare instances of growth failure or ‘short stature’ may be caused by a large heterogeneous group of genetic disorders ( i . e . skeletal dysplasias ) ( Durand and Rappold , 2013b ) . We first extracted measures of height for the Puerto Rican adult population of BioMe ( mean age = 55 . 3 , standard deviation ( s . d . ) =16 . 1 ) . After making exclusions based on age ( >=18 years old for women , >=22 years old for men , and <80 years old in both sexes , 194 individuals in total ) , mean height measurements ( mean height = 5’ 8 . 2’ , s . d . = 3 . 2’ for men; mean height = 5’ 2 . 8’ s . d . = 2 . 8’ for women ) were consistent with those reported for Puerto Rican populations in a recent global study on height ( NCD Risk Factor Collaboration ( NCD-RisC ) , 2016 ) . We noted that 56 Puerto Ricans met the clinical definition of short stature ( Cohen et al . , 2008 ) ( range of short stature 5’1’−4’0’ in men , and 4’8’−3’8’ in women ) defined as 2 standard deviations below the population-specific mean for men and women separately ( Figure 2—figure supplement 1 ) . To test for recently arisen , recessive variants underlying clinical short stature in Puerto Ricans , we implemented a previously published pipeline for ‘IBD mapping’ ( Kenny et al . , 2009; Vacic et al . , 2014 ) ( Figure 2—figure supplement 2 ) . We first clustered participants into ‘cliques’ of 3 or more individuals whom , at a given genomic region , shared overlapping homologous IBD tracts of at least 0 . 5 cM in length in either a heterozygous or homozygous state . Membership in a clique indicates the sharing of a recent common ancestor at that locus , from which the homologous IBD tract was jointly inherited . Clustering of IBD into cliques in the Puerto Rican population ( N = 2816 ) yielded 1434421 IBD-cliques after quality control filters ( see Methods ) . The site frequency spectrum of IBD-cliques ( Figure 2—figure supplement 3 ) demonstrates an expected exponential distribution of clique sizes ( of 3–77 haplotypes ) , representing a class of rare IBD haplotypic alleles ( allelic frequency 0 . 0005–0 . 0137 ) . To test whether any cliques of IBD haplotypes were significantly associated with height we performed genome-wide association of height as a continuous trait under a recessive model using PLINKv1 . 9 ( Purcell et al . , 2007; Chang et al . , 2015 ) , including the first five PCA eigenvectors as covariates ( see Materials and methods ) . We restricted analysis to 480 out of 1434421 cliques that contained at least 3 individuals who were homozygous for the shared haplotype . Adjusting for 480 tests ( Bonferroni adjusted threshold p<1×10−4 ) one IBD-clique achieved a genome-wide significant signal at the locus 9q32 ( IBD-clique frequency = 0 . 012; β = −3 . 78; p<2 . 57×10−11 ) ( Figure 2A ) , spanning a large mapping interval chr9:112 MB-120MB . The clique contains 59 individuals , 56 of whom are heterozygous and 3 are homozygous for the associated IBD haplotype . The three individuals driving the recessive signal , two women and one man , were less than 2 . 5 s . d . shorter ( height reduction range 6’−10’ ) than the population mean for height in the Puerto Rican cohort ( Figure 2B ) . The IBD-haplotypes driving the signal spanned a genic region with several candidate loci , and the minimum shared boundary overlapped a single gene , COL27A1 , which encodes for Collagen Type XXVII , Alpha 1 ( Figure 2C ) . We performed whole genome sequencing ( WGS ) of the three homozygous individuals , and an additional short-statured individual that we observed to possess a homozygous IBD haplotype that was both directly upstream of and highly correlated with the top IBD-clique . Individuals were sequenced to a depth of 4-18X coverage ( Supplementary file 1 ) . Examination of variants that were observed in at least 6 copies between the four individuals ( to account for sequencing error or missing calls ) revealed a single candidate coding allele , a missense mutation in Collagen Type XXVII , Alpha 1 ( COL27A1 , g . 9:116958257 . C>G , NM_032888 . 1 , p . G697R , rs140950220 ) ( Supplemental file 2 ) . In silico analysis suggest that this glycine residue is highly conserved , and that a molecular alteration to arginine at this position is predicted to be damaging ( SIFT score = 0 . 0; PhyloP score = 2 . 673; GERP NR score = 5 . 67 ) . These findings are consistent with a recent report implicating the same COL27A1 variant as causal for the rare orthopedic condition Steel syndrome in a Puerto Rican family ( Gonzaga-Jauregui et al . , 2015 ) . First described in 1993 , the main clinical features of Steel syndrome include short stature , bilateral hip and radial head dislocations , carpal coalition ( fusion of the carpal bones ) , scoliosis , pes cavus ( high arches ) , and dysmorphic features ( Steel et al . , 1993 ) . To confirm the link between the IBD haplotype and the putative causal variant , we calculated the concordance between the IBD haplotype and carrier status of the COL27A1 . pG697R variant by genotyping all of the homozygotes and carriers of the top IBD-clique in the recessive model ( N = 59 ) , along with a panel of age- and sex-matched controls ( N = 59 ) . This demonstrated 100% concordance between the COL27A1 . pG697R variant and the significant IBD-haplotype in homozygotes ( Supplementary file 3 ) . We note that two Puerto Rican participants in the phase 3 1KGP reference panel were carriers of the COL27A1 . pG697R variant , raising the possibility that we may have been able to detect this association using more a traditional SNP association approach . Therefore , we performed genome-wide association in the same Puerto Rican cohort ( N = 2622 ) by first imputing the 1KGP panel and re-running the recessive test as described above ( n = 10007795 imputed and genotyped SNPs with an INFO score of >0 . 3 and at least two observations of homozygotes ) . The recessive model appeared to be well calibrated ( λ = 1 . 02 ) , however , we observed no genome-wide significant signal ( Figure 2—figure supplement 4 ) . Association with the COL27A1 . pG697R variant was the 11775th most significant association ( MAF = 0 . 014: β = −3 . 0; p<0 . 001 ) . Upon examination of the correlation between the imputed COL27A1 . pG697R and the true carrier status of homozygotes , we noted a concordance of only 66 . 67% , indicating that the IBD haplotype was a better tag of the true COL27A1 . pG697R homozygous state compared to 1KGP imputation in the Puerto Rican cohort ( Supplemental file 3 ) . The association between COL27A1 . pG697R and clinical short stature was replicated using a cohort of 1775 individuals of self-reported Puerto Rican ancestry from BioMe that were not included in the discovery analysis , that were genotyped on the Illumina Infinium Multi-Ethnic Genotype Array ( MEGA ) as part of the Polygeneic Architecture using Genomics and Epidemiology ( PAGE ) Study . The COL27A1 . pG697R ( rs140950220 ) variant was directly genotyped on MEGA , and an association of the variant under a recessive model resulted in a strong signal of association ( allele frequency = 0 . 017; β = −3 . 5; s . e . = 0 . 70; p<4 . 87×10−07 ) . The replication analysis revealed 51 additional BioMe carriers and two individuals that were homozygous for the variant . Both carrier and affected status was confirmed via independent genotyping and Sanger sequencing ( see Materials and methods ) . The two homozygous participants were both short statured ( 2 . 4 and 3 . 6 s . d from the sex specific population mean ) . To determine whether there was any clinical evidence to validate the link between the COL27A1 . pG697R variant and Steel syndrome , a clinical expert manually reviewed the electronic health records ( EHR ) , including clinical diagnoses , surgical procedures , and radiology reports of the five participants ( 3 women , 2 men , age range 34–74 years ) homozygous for the COL27A1 . pG697R variant that we had identified through the discovery and replication efforts . Of note , there was no evidence that any of the five patients had a clinical diagnosis of Steel syndrome . In all five individuals , however , we found EHR-documented evidence of several previously described Steel syndrome characteristics , including developmental dysplasia of the hip ( or congenital hip dysplasia ) , carpal coalition , scoliosis , and cervical spine anomalies ( Table 1 ) ( Steel et al . , 1993; Flynn et al . , 2010 ) . The incidence of cervical spine anomalies , including cord compression and spine surgeries , was higher than previously reported ( four out of five patients ) . There was also evidence of other significant musculoskeletal complications , including lumbar and thoracic spine anomalies in three patients , knee replacements in two patients ( both under age 50 ) , and joint degeneration or arthritis in four patients . Together , these data help further our understanding of Steel syndrome-associated characteristics and potential complications that can occur later in life . To understand the biological mechanism underlying Steel syndrome , we investigated the functional role of the COL27A1 gene . COL27A1 is a fibrillar collagen , which are a class of collagens that contribute to the structural integrity of the extracellular matrix ( Pace et al . , 2003 ) . Enrichment of COL27A1 RNA expression in vertebrae , as well as long bones , eyes , and lungs has previously been observed in embryonic mice ( Pace et al . , 2003 ) . A mouse deletion of 87 amino acids of the COL27A1 homolog exhibited severe chondroplasia consistent with clinical features observed in homozygotes ( Plumb et al . , 2011 ) , a similar musculoskeletal phenotype was observed in knockdown of the col27a1a and col27a1b genes in zebrafish ( Christiansen et al . , 2009 ) . Type alpha-1 collagen genes , of which COL27A1 is a member , contain a conserved Gly-Xaa-Yaa repeat in their triple helical domain ( Persikov et al . , 2004 ) . Therefore , we hypothesized that the COL27A1 . pG697R variant may similarly disrupt stability of the COL27A1 triple helix . To test this hypothesis we modeled the effect of a glycine-to-arginine substitution in the structure of a prototypical collagen peptide ( Bella et al . , 1994 ) . We observed that the glycine residues occupied the center of the crowded triple helix , and that substitution for a bulkier arginine would likely destabilize helix formation through steric hindrance ( Figure 2—figure supplement 5 ) . These data provide support for a functional model of the pathogenicity of COL27A1 . G697R through destabilization of the triple helix , which may occur within developing spinal chords , long bones , and other tissues , resulting in the observed clinical features in homozygotes . We note that many other collagen disorders , including Ehlers-Danlos syndrome ( McGrory et al . , 1996; Tromp et al . , 1995; Anderson et al . , 1997 ) , Alport syndrome ( Knebelmann et al . , 1992; Zhou et al . , 1992 ) and Osteogenesis Imperfecta ( Starman et al . , 1989; Shapiro et al . , 1992 ) , are driven by molecular alterations of a glycine in the triple helix of the underlying collagen genes . However , all of these disorders are inherited under an autosomal dominant mode , in contrast to Steel syndrome , which has only been reported as a recessive disease . This analysis raises the question of whether some and/or milder clinical features of Steel syndrome may be present in carriers . To test for clinical features of Steel syndrome in COL27A1 . pG697R carriers , we performed two analyses using EHR data . The first was a test for associated medical billing codes ( ICD9s ) with COL27A1 . pG697R carrier status , or Phenome-Wide Association Study ( PheWAS ) ( Denny et al . , 2010; Denny et al . , 2013 ) . PheWAS analysis is often performed using a general linear model ( GLM ) , however standard implementations often do not account for scenarios where there is a large imbalance between per-test number of cases and controls , rare variants/ICD9s or the presence of elevated distant relatedness . Therefore , in addition to the GLM , we also ran three other score based tests; ( i ) that use saddlepoint approximation ( SPATest ) ( Dey et al . , 2017 ) to account for case:control imbalance; ( ii ) a linear mixed model ( GCTA ) ( Yang et al . , 2011 ) to account for distant relatedness; and ( iii ) a test that incorporates a bias-reduction for small numbers of observations ( Firth test ) ( Wang , 2014 ) . Each test was run using ICD9 codes in all individuals of Puerto Rican ancestry ( N = 106 COL27A1 . pG697R carriers and N = 4480 non-carriers ) . The five homozygotes were excluded . The ICD9 code was set as the outcome variable and COL27A1 . pG697R as the primary predictor variable , including age , sex and the first five PCAs as covariates in all tests . To avoid spurious associations , we restricted the analysis to diagnosis codes with at least 3 observations ( n = 367 ICD9 codes ) amongst carriers . Results of the GLM test are shown in Figure 3 and Table 2 . Of the five significantly associated ICD9 codes ( False Discovery Rate ( FDR ) < 0 . 05 ) , three involved the musculoskeletal system 730 . 08 ( pGLM < 7 . 1 × 10−6; odds ratio ( OR ) = 34 . 5; 95% Confidence Interval ( CI ) = 7 . 4–162 ) , 721 . 0 ( pGLM < 6 . 6 × 10−5; OR = 5 . 4; CI = 2 . 4–12 . 3 ) , and 716 . 98 ( pGLM < 4 . 4 × 10−4; OR = 5 . 8; CI = 2 . 2–15 . 3 ) . ICD9 730 . 08 encodes for ‘acute osteomyelitis , other specified sites’ . Manual review of chart records for these patients revealed that this code referred to vertebral osteomyelitis in the three carriers with the ICD9 code . ICD9 721 . 0 encodes for cervical spondylosis without myelopathy . Cervical spondylosis refers to degenerative changes of the cervical spine , which can eventually progress to encroach on the cervical canal , causing myelopathy ( spinal cord injury ) . A third diagnosis code , 716 . 98 , encodes for ‘arthropathy , unspecified , or involving other specified sites’ . Manual review of chart records for these patients revealed that this code referred to knee arthropathy in all four patients . Finally , two other ICD9 codes were significantly associated with the COL27A1 . pG697R variant; 622 . 10 ( pGLM < 1 × 10−4; OR = 5 . 4; CI = 2 . 3–12 . 6 ) , which encodes for cervical dysplasia , and 789 . 1 ( pGLM < 2 . 1 × 10−4; OR = 11 . 6; CI = 3 . 2–42 . 2 ) , which encodes for hepatomegaly . Presently , it is unclear whether these two are related to a COL27A1 . pG697R carrier phenotype , or are spurious associations . We observed over inflation in the distribution of the PheWAS test statistic , measured by lambda ( λ ) , for all four score based models ( λGLM=1 . 59; λSPATest=1 . 20; λGCTA=1 . 36; λFirth=2 . 09 ) , indicating that no single model fully accounts for the confounding effects of distant relatedness , case:control imbalance and rare variants/ICD9s ( Figure 3—figure supplement 1 ) . The code linked to vertebral osteomyelitis ( 730 . 08 ) was the top signal in all tests ( pSPATest < 1 . 4 × 10−4; pGCTA < 7 . 9 × 10−10; pFirth < 1 . 5 × 10−9 ) , but only remains significant after genomic control adjustment in one of the tests ( pGCTA_adjusted < 4 . 6 x 10−5 ) . Neither codes linked to cervical spondylosis ( 721 . 0; pSPATest < 3 . 0 × 10−3 ( rank=3rd ) ; pGCTA < 3 . 3 × 10−3 ( 8th ) ) or knee arthropathy ( 716 . 98; pSPATest < 0 . 022 ( 21st ) ; pGCTA < 3 . 5 × 10−3 ( 9th ) ; pFirth < 0 . 001 ( 35th ) ) ) were significant after genomic control correction . Therefore , while PheWAS analysis provided preliminary support of Steel syndrome-associated clinical features in carriers , best practices for PheWAS models for rare variants/ICD9 codes , and in the presence of population structure , remains an open problem for the genomics community . It is also possible that some relevant clinical features of Steel syndrome might be poorly captured by or absent from medical billing codes . To evaluate the preliminary evidence from the PheWAS analysis , we performed a second analysis of EHR data that focused on a comprehensive manual chart review to examine for evidence of Steel syndrome characteristics in the COL27A1 . pG697R carriers in the same manner as performed for homozygotes . We limited the analysis to carriers below the age of 55 ( N = 34; mean age 41 . 8 years ) to reduce confounding from age-related related symptoms of spine and joint pain . We also selected 31 age and sex matched Puerto Rican non-carriers for comparison ( mean age 40 . 6 years ) . Utilizing the same criteria used to characterize Steel syndrome cases , we found no evidence of clinical short stature or hip dislocation in carriers , but did observe a trend of elevated rates of major joint and spine degradation ( Table 1 ) . In general , 38% ( 13/34 ) of carriers showed evidence of spine degeneration varied from severe ( multiple level cord compression and neurological symptoms necessitating corrective surgery ) to moderate ( lower back pain with no neurological symptoms managed with physical therapy and/or pain medication ) compared to 13% ( 4/31 ) of non-carriers ( Fishers exact test p<0 . 03 ) . Specifically , we found an increased risk of cervical stenosis in 15% ( 5/34 ) of carriers compared to 0% ( 0/31 ) of controls ( p<0 . 05 ) . Although not reaching statistically significance , we show a trend of 2-fold higher rates of scoliosis ( 24%; p<0 . 35 ) , arthritis ( 38%; p<0 . 1 ) , and lumbar spine degradation ( 29%; p<0 . 25 ) in carriers compared to non-carriers and previous published reports in similar age groups ( Kebaish et al . , 2011; Reginster , 2002 ) . Together these data suggest an appreciable burden of joint and spine degradation in COL27A1 . pG697R carriers ( Table 1 ) . Having genetically identified and clinically characterized a previously little-known disease variant , we next investigated which populations were at risk for harboring the allele . We assessed the carrier frequencies of the COL27A1 . pG697R variant in global panel of 51745 individuals from Africa ( N = 376 ) , the Americas ( N = 45685 ) , Asia ( N = 5311 ) , Europe ( N = 209 ) , the Middle East ( N = 163 ) and Oceania ( N = 28 ) genotyped on MEGA in the PAGE Study . This included; 13050 in the Multi-Ethnic Cohort ( MEC ) ( Kolonel et al . , 2000 ) Study; 12327 in the Hispanic Community Health Study/Study of Latinos ( HCHS/SOL ) ( Lavange et al . , 2010 ) ; 12852 in the Women’s Health Initiative ( WHI ) Study ( The Women's Health Initiative Study Group , 1998 ) ; 13044 additional BioMe biobank participants ( including the 1775 Puerto Ricans on MEGA described above ) ; and a Global Reference Panel from Stanford University including the Human Genome Diversity Panel ( Cann , 2002 ) ( N = 986 , see Materials and methods ) . Combined , the PAGE and BioMe dataset represented 57316 individuals from 112 global populations ( Supplementary file 4 ) . The COL27A1 . pG697R C allele was present in 183 copies ( 173 heterozygous carriers and 5 homozygous cases ) . We estimate the carrier rate of COL27A1 . G697R to be 1:51 in Puerto Rican-born individuals ( minor allele frequency ( MAF ) = 1 . 1% ) ; 1:9 in individuals born on the island of St . Thomas ( MAF = 11% ) ; 1:346 in Hispanic/Latino/as in the US ( MAF = 0 . 29% ) and 1:72 in BioMe Hispanic/Latino/a populations from New York City ( MAF = 0 . 7% ) ; and 1:746 in individuals born in the Dominican Republic ( MAF = 0 . 067% ) ( Figure 4A ) . We note that only 9 people were assayed from St . Thomas , so the high carrier frequency estimate could be biased by small sample size . Finally , the variant is present in only 4 copies in the 60 , 706 exomes in the ExAC database ( Exome Aggregation Consortium et al . , 2016 ) , likely due to differences in the populations comprising both datasets . To predict what other populations might be at risk for Steel syndrome , we explored the locus-specific demographic history in carriers of the COL27A1 . pG697R risk haplotype . First , by visual inspection , we were able to discriminate a single haplotype of 107 . 5 kb in length that contained 55 SNPs , which uniquely tagged the COL27A1 . pG697R variant ( R2 = 1 ) . This haplotype was present only in individuals born in Puerto Rico ( N = 25 ) , the Dominican Republic , ( N = 2 ) , Columbia ( N = 1 ) , New York City ( N = 40 ) and St . Thomas ( N = 1 ) . Genotyping determined that only the haplotype carriers from Puerto Rico , New York City and St . Thomas also carried the COL27A1 . pG697R variant ( N = 56 in total ) . Second , we inferred continental ancestry along the genomes of the three Puerto Rican homozygotes in the discovery cohort . Local ancestry inference is the task of assigning continental ancestry to genomic segments in an individual with recent ancestors from multiple continents . For the Puerto Rican homozygotes , we estimated local haplotypic similarity with a reference panel of African , European and Native American genomes using RFMix ( Maples et al . , 2013 ) . Examination of local ancestry on the background of the IBD haplotype in all three homozygous individuals revealed all to be homozygous for Native American ancestry , suggesting the COL27A1 . pG697R arose on a Native American haplotype ( Figure 4B ) . To test whether the disease variant arose via genetic drift or selection , we used the IBDNe software ( Browning and Browning , 2015 ) to estimate the historical effective population size ( Ne ) of the Puerto Rican discovery cohort ( N = 2816 ) ( Figure 4C ) . The IBDNe software calculates the effective population size of a given population over past generations by modeling the distribution of IBD tract lengths present in the contemporary population . The analysis suggested evidence of a strong bottleneck in Puerto Ricans approximately 9–14 generations ago , with the smallest effective population sized dating approximately 12 generations ago ( estimated Ne = 2580 , 95% C . I 2320–2910 ) . This is consistent with the timing of European immigration and slave trading on the Island , resulting in admixture and population bottlenecking , followed by demographic growth post-contact ( Moreno-Estrada et al . , 2013; Gravel et al . , 2013 ) . Finally , to see if there was evidence that the locus had undergone a recent selective sweep we calculated the integrated haplotype score ( iHS ) ( Voight et al . , 2006; Szpiech and Hernandez , 2014 ) across chromosome 9 in phased genotype data for BioMe Puerto Rican samples , but did not observe evidence of selection at the locus ( Figure 4—figure supplement 1 ) . Together , this evidence suggests that the COL27A1 . pG697R variant arose in the ancestral Native American populations that peopled the Caribbean , which underwent a strong bottleneck during the period of colonization , which may help explain the prevalence of this disease in amongst contemporary Puerto Rican populations .
Here we describe a new approach to utilize genomic data in health systems for identifying and characterizing genetic disorders , the cornerstone of which is the ability to identify related individuals in the absence of recorded pedigree or genealogy information . By linking medical records of distantly related patients , identified by shared tracts of genetic homology identical-by-descent ( IBD ) , we discovered a recessive haplotype on 9q32 conferring extreme short stature . Whole genome sequencing revealed that a mutation ( Gly697Arg ) in the COL27A1 gene had been previously implicated as the genetic variant underlying Steel syndrome ( Gonzaga-Jauregui et al . , 2015; Steel et al . , 1993; Flynn et al . , 2010 ) . Population screening indicated that the disease variant is more common than previously thought in people with Puerto Rican ancestry , and in some other Caribbean populations , and very rare or absent elsewhere in the world . Extensive analysis of clinical records confirms almost all features of the recessive disorder in cases , and reveals potential complications that can occur later in life . An agnostic survey of the medical records of carriers , supplemented by manual chart review , indicates evidence of joint and spine degradation in heterozygotes . Biochemical modeling suggests that COL27A1 . G697R disrupts a conserved triple helix domain of the alpha-1 collagen in a mechanism similar to dominant forms of other collagen disorders ( Persikov et al . , 2004 ) . Taken together , this study indicates that a single mutation in the COL27A1 gene underlies a common collagen disorder impacting up to 2% of people of Puerto Rican ancestry . This is consistent with our finding , supported by previous work ( Moreno-Estrada et al . , 2013 ) , demonstrating a founder effect in Puerto Rican populations . Despite segregating at an estimated carrier rate of 1:51 , the COL27A1 . pG697R variant was first described very recently ( Gonzaga-Jauregui et al . , 2015 ) . This suggests that there may be other highly penetrant disease variants segregating at appreciable frequencies in Puerto Rican populations ( Anikster et al . , 2001; Cornier et al . , 2008; Daniels et al . , 2013; Al-Zaidy et al . , 2015; Arnold et al . , 2013; Lee et al . , 2014 ) , and other understudied founder populations , the discovery of which could lead to new disease variants and biology . Indeed , although COL27A1 was first implicated as the Steel syndrome disease locus in an extended family from Puerto Rico recently ( Chang et al . , 2015 ) , other variants in COL27A1 have since been linked to Steel syndrome in Indian ( Kotabagi et al . , 2017 ) and Emerati ( Gariballa et al . , 2017 ) families revealing additional clinical features of the disease such as hearing loss . In our own health system , approximately 190 , 000 patients of Puerto Rican descent are treated annually ( Humes et al . , 2011 ) . We estimate that up to 80 may have the severe homozygous form of the disorder and that the milder heterozygous form could be found in up to 1200 patients . A search of progress notes , discharge summaries , and operative reports of over 4 million patients in the Mount Sinai data warehouse discovered mentions of the text term ‘Steel Syndrome’ in 42 patient records . However , all of these patients were on dialysis for end stage renal disease , indicating that this mention was a misspelling of vascular Steal Syndrome , which is common in dialysis patients . This suggests that Steel syndrome might be largely undiagnosed . Attempts are currently being made to re-contact BioMe participants with suspected Steel syndrome , and a genetic test is now available at Mount Sinai ( website: http://sema4genomics . com/products/test-catalog/ ) . This study highlights the benefits of incorporating statistical and population genetics approaches in medical genetic research . First , we demonstrated that leveraging distant relationships via IBD mapping was better powered for discovery of the COL27A1 variant compared to a more typical GWAS approach ( i . e . genotype , imputation , and SNP association ) . As sample sizes increase in health systems and biobanks , the odds of a new individual being a direct or distant relative of someone already in the database increases exponentially ( Henn et al . , 2012 ) , enabling the detection of shared haplotypes harboring rarer causal variants and better-powered IBD mapping studies . Second , we inferred that COL27A1 . pG697R variant arose on a Native American haplotype , and we estimate that the allele may have segregated at a carrier frequency of 25–30% in pre-Columbian Taíno populations and/or been driven to its current frequency by a bottleneck that occurred during the early days of colonization in Puerto Rico . Therefore this study not only helps estimate population attributable risk of COL27A1 . pG697R in Puerto Rican populations , but also to predict other populations potentially at risk , including other Caribbean and Taíno populations . Targeted population screening of COL27A1 . pG697R could potentially provide personalized health management , surveillance for associated complications , guidelines for intervention ( particularly in newborns [Flynn et al . , 2010] ) , and improved reproductive choices . This work also highlights some of the current challenges in the emerging field of genomic medicine . We demonstrated that evidence from EHRs could be readily extracted and retrospectively used to characterize clinical features of a musculoskeletal genetic disorder . However , features of many other genetic disorders may not be detectable via routine clinical exam , lab tests and radiologics , and may not be amenable to such an approach . Furthermore , statistical methods for population-scale disease variant discovery , which were predominantly developed for cohorts collected for genetic research , may not be optimally calibrated for discovery in patient populations encountered in health systems . Finally , many genetic disorders are very rare , or have more complex genetic underpinnings , which would reduce power for detection using the strategy we have described . However , recent efforts , such as the Precision Medicine Initiative , that focus on the broad adoption of genomics in medicine , combined with international efforts to catalog rare genetic diseases , are primed to increase the rate of incidental genetic diagnosis of disease . In summary , this work demonstrates the utility of biobanks for exploring full medical phenomes , and highlights the importance of documenting a wider spectrum of genetic disorders , in large and diverse populations of humans . In particular , this method provides a bridge between classical medical genetic methods and those employed in population-level GWAS . Here we note that the COL27A1 variant is very rare in current large-scale genomic databases used for clinical research . Thus traditional association strategies and ascertainment bias focused on populations of European descent would have failed to identify and characterize this disorder and its public health burden . As ours and other recent studies have demonstrated , EHR-embedded research will be increasingly important for disentangling the pathology of rare genetic disorders , and understanding the continuum of complex and Mendelian disease . As studies grow in size , and healthcare systems learn to leverage the wealth of information captured in the EHR , there is a need to provide relevant medical information to any patient entering the clinic anywhere in the world . Methods like that described here allow for precision medicine with a truly global outlook .
Study participants were recruited from the BioMe Biobank Program of The Charles Bronfman Institute for Personalized Medicine at Mount Sinai Medical Center from 2007 onward . The BioMe Biobank Program ( Institutional Review Board 07–0529 ) operates under a Mount Sinai Institutional Review Board-approved research protocol . All study participants provided written informed consent . Of the approximately 38000 participants currently enrolled in BioMe , N = 10511 unrelated are genotyped on the Illumina Infinium OmniExpress ( OMNI ) array . 5102 of these participants self-report as ‘Hispanic or Latino/a’ , 3080 as ‘African American or African’ , 2078 as ‘White or Caucasian’ , 89 as ‘Mixed’ and 162 as ‘Other’ . Country of origin information is available for all but N = 5 participants , with N = 6553 reporting being born in the United States and the remaining N = 3953 report being born outside of the US . Parental and grandparental country of origin information is only available for a small subset of individuals genotyped on the OMNI array ( N = 43 ) . An additional N = 10471 participants were genotyped on the Illumina Infinium Multi Ethnic Genotyping array ( MEGA ) v1 . 0 . Of these , approximately 4704 self-reported ‘Hispanic or Latino/a’ ethnicity , 3143 self-reported as ‘African American or African’ , 22 self-reported as ‘White/Caucasian’ , 708 self-reported as ‘Mixed’ and 1894 self-reported as ‘Other’ . Country of birth information was available for all but a small number of participants ( N = 228 ) , with 5190 reporting being born in the United States , and the remaining 5053 self-reporting being born elsewhere . Parental and grandparental country of origin information is available for 4323 individuals genotyped on the MEGA array . Genotyping of 12749 BioMe participants was performed on the Illumina Infinium OmniExpress plus HumanExome array . Calling was performed using the GenomeStudio software . A total of 1093 individuals were removed prior to zCall due to plate failure ( N = 672 ) , unambiguous discordance between genetic and EHR recorded sex ( N = 693 ) , a call rate of <98% ( N = 834 ) , or deviances in levels of heterozygosity ( N = 773 in total ) . This was defined as having either an inbreeding coefficient outside the range −0 . 1 and 0 . 3 for common alleles ( MAF >1% ) , or between 0 . 4 and 0 . 9 for rare alleles ( MAF <1% ) . Additional quality control of 11656 individuals was performed using PLINK1 . 7 ( Purcell et al . , 2007 ) ( RRID:SCR_001757 ) . An individual with a call rate of <99% was also excluded ( N = 1 ) , along with intentional genetic duplicates ( PiHat >0 . 8 , N = 444 ) . Site-level quality control consisted of the removal of SNPs with a call rate of <95% ( n = 42217 ) , and the removal of sites that were significantly out of Hardy-Weinberg equilibrium ( p<1×10−5; n = 39660 ) when calculated for self-reported EA , AA and H/L separately . Palindromic sites and those that deviated considerably from the 1KGP allele frequencies ( <40% versus >60% ) were also removed to ensure uniform stranding across datasets . After QC steps , 11212 participants and 866864 SNPs remained for downstream analysis . Pairwise IBS-based relationship estimates were derived for BioMe participants ( N = 11212 ) using the RELATEAdmix software ( Moltke and Albrechtsen , 2014 ) , which accounts for inflation of IBS statistic due to admixture linkage disequillibrium in admixed populations . To include allele frequency information and global ancestry proportions from ancestral populations relevant for each admixed population in the analysis . These were estimated using ADMIXTURE ( Alexander et al . , 2009 ) ( RRID:SCR_001263 ) H/L samples were merged with the Utah Residents ( CEPH ) with Northern and Western European Ancestry ( CEU; N = 100 ) , Yoruba in Ibadan , Nigeria ( YRI; N = 100 ) and Native American ( NA; N = 43; including Nauha , Ayamaran , Mayan and Quechan individuals ) and used as input for the ADMIXTURE software , which was run unsupervised at k = 3 . ADMIXTURE analysis confirmed that NA reference panel comprised >99% proportion Native American genetic ancestry ( Figure 4—figure supplement 2 ) . European-American ( EA ) individuals were merged with the CEU and a panel of self-reported Ashkenazi Jewish individuals genotyped on OMNI from BioMe ( AJ; N = 100 ) and run unsupervised at k = 2 . AA samples were merged with the CEU and YRI reference panels and run unsupervised at k = 2 . After intersecting with reference panels 99296 SNPs were used as the input for RELATEAdmix . Principal Component Analysis ( PCA ) was performed using the SMARTPCAv10210 software from the EIGENSOFTv5 . 0 . 1 ( RRID:SCR_004965 ) ( Price et al . , 2006 ) in 10511 unrelated BioMe participants . Regions containing the Human Leukocyte Antigen ( chr6: 27000000–35000000 ( NCBI37/hg19 ) ) , Lactase gene ( chr2:135000000–137000000 ( NCBI37/hg19 ) ) and a common inversion ( chr8:6000000–16000000 ( NCBI37/hg19 ) ) , all of which are regions known to confound PCA analysis were removed from the genotype data prior to analysis . Data were merged with a reference panel of 2504 individuals from Phase 3 of the 1KGP ( RRID:SCR_006828 ) ( The 1000 Genomes Project Consortium , 2015 ) that was constructed by extracting OmniExpress sites from whole-genome sequence data . Following this , a further two other relevant reference panels were added: the NA ( N = 43 ) , and AJ ( N = 100 ) panels described above . A total of 174468 SNPs remained after intersecting the data with these reference panels . Phased genotype data were filtered to MAF >0 . 01 and converted to PLINK format using the FCGENE software ( Roshyara and Scholz , 2014 ) ( we avoided using PLINK software for the conversion process in order to retain the phase information ) . Recombination maps from HapMap II ( Build GRCh37/hg19 ) were intersected with the genotyped sites ( n = 490510 SNPs ) . GERMLINE ( RRID:SCR_001720 ) ( Gusev et al . , 2009 ) was used to infer tracts of identity by descent >3 cM across all pairs of BioMe individuals ( N = 11212 ) using the following flags: ‘-min_m 3 -err_hom 0 -err_het 2 -bits 25 –haploid’ . IBD haplotypes that fell within or overlapped with centromeres , telomeres and regions of low complexity were removed from the GERMLINE output using an in-house Ruby script . Additional quality control measures consisted of the exclusion of regions of the genome where the depth of IBD-sharing ( that is , the number of pairwise IBD-haplotypes that contain a given locus of the genome ) exceeded 4 standard deviations from the genome-wide mean ( Figure 2—figure supplement 6 ) . IBD clustering to identify ‘cliques' of three or more IBD haplotypes shared between multiple individuals was then performed using the efficient connect-component-based clustering version of the Dash Associated Shared Haplotypes algorithm ( DASH ) ( Gusev et al . , 2011 ) , using the default parameters . As a further quality control measure IBD-sharing ‘cliques’ inferred by DASH that exhibited excessive sharing ( which we defined as clique membership that exceeded 4 s . d . above the genome-wide mean ) were removed ( Figure 2—figure supplement 7 ) . Data was outputted from DASH in PLINK tped format , and alleles were encoded as; homozygote member in a clique as ‘2’ , heterozygote member as ‘1’ and everyone else not a member in the clique encoded as ‘0’ . We calculated the length of any pairwise IBD tract ( or sum of the lengths if a pair of individuals shared more than one tract IBD ) for each IBD sharing pair within each population to obtain an estimate of the mean and variance of pairwise sharing per population . To compare the tract length distribution between populations ( of size N ) , we first binned pairwise IBD tracts by length bin in 0 . 01 cM increments . We then summed the number of pairwise IBD tracts falling into each length bin ( x ) , and divided this number by the number of possible pairwise IBD sharing for each population: N* ( N-1 ) /2 . A self-reported measurement of height in feet and inches was recorded for each participant at enrollment into the BioMe program . Raw height data were stratified on the basis of sex for all individuals who were inferred to be of Puerto Rican ancestry ( N = 2816 ) . Height data was then log transformed and converted to age-adjusted Z-scores . Participants were excluded on the basis of age reported at the point of enrollment , with a minimum cut-off of 18 years old for females ( N = 0 ) and 22 for males ( N = 0 ) , and a maximum cut off of 79 years old for both sexes ( N = 194 ) leaving a total of n = 2622 PR . Association of IBD clique membership with height as a continuous trait was performed under a recessive model using PLINKv1 . 9 ( Chang et al . , 2015 ) using the ‘--linear recessive’ flag . Age and sex adjusted Z-scores for height were used as the outcome variable . IBD clique membership was used as the primary predictor variable and the first five PCA eigenvectors were used as covariates . The model was run across a total of 2622 PR ancestry individuals and a total of 480 IBD-cliques where at least 3 individuals were homozygous for the IBD haplotype . Genotype data for all of the BioMe individuals ascertained on the Illumina OMNI Express array ( N = 11212 ) were phased together using SHAPEIT2 ( Delaneau et al . , 2011; O'Connell et al . , 2014 ) . Imputation was subsequently performed in 5 MB chunks using IMPUTE2 ( RRID:SCR_013055 ) ( Howie et al . , 2009 ) via the flags ‘-Ne 20000 -buffer 250 -filt_rules_l 'ALL<0 . 0002' 'ALL>0 . 9998'’ with a reference panel derived from phase 3 data from the 1KGP . A total of 46538253 SNPs were imputed from 828109 directly genotyped SNPs . We ran a recessive GWAS on the same 2622 inferred Puerto Rican ancestry individuals used in our recessive IBD-mapping model . The association was run over hard-called data using the PLINKv1 . 9 software using the ‘--linear recessive’ flag . Age and sex adjusted Z-scores for height were used as the phenotypic outcome and the first five PC eigenvectors were used as covariates . Analysis was restricted to SNPs with >= 2 observations of individuals homozygous for the minor allele ( as the only 2 of the 3 homozygotes had been imputed correctly ) , and SNPs with an INFO score of >= 0 . 3 ( n = 10007795 SNPs in total ) . Genomic libraries were prepared from DNA obtained for the four IBD homozygous individuals . DNA was sheared to 300 bp on a Covaris E220 , libraries were made using the NEBNext Ultra DNA Library Prep kit for Illumina . The libraries were submitted for Whole Genome Sequencing ( WGS ) at the Mount Sinai Genomic Core using the Illumina HiSeq 2500 system , performed by the Genomics Core Facility of the Icahn Institute for Genomics and Multiscale Biology , Icahn School of Medicine at Mount Sinai . Reads were aligned to the NCBI37/hg19 reference genome and variants were called using the sequence analysis pipeline by Linderman et al . ( 2014 ) Variant calls and coverage at every site at the genomic interval spanned by the candidate IBD haplotype ( chr9:112000000–118000000 bp ( NCBI37/hg19 ) ) were obtained using the ‘-out_mode EMIT_ALL_SITES’ flag in GATKv3 . 2–72 ( RRID:SCR_001876 ) . For summary statistics of whole genome sequencing ( WGS ) see Supplement file 1 . WGS variant calls were annotated with allele frequency information and in silico prediction scores for SIFT , PhyloP , GERP generated using snpEffv3 . 0 ( RRID:SCR_005191 ) as part of the sequence analysis pipeline published by Linderman et al . ( 2014 ) . We identified all genomic variants that were present in at least 6 copies across the four IBD-homozygotes and that lay within the shared boundary of the IBD haplotype . Using this criteria , only one rare , coding variant was found to be shared between all four homozygotes , namely a point mutation in the gene COL27A1 ( g . 9:116958257 . C>G , NM_032888 . 1 , p . G697R , rs140950220 ) which was present in 7 copies ( with 3 individuals being homozygous , and the fourth being heterozygous ) . The rs140950220 G/C allele status was validated by Sanger sequencing of exon 7 in the COL27A1 gene in all four individuals . We also validated COL27A1 . pG697R status in individuals carrying the significant IBD-clique at 9q32 using the Fluidigm SNPType assay adhering to the standard protocol . All individuals carrying at least one copy of the top IBD-haplotype ( N = 59 ) were genotyped for the rs140950220 variant in addition to a panel of age and sex matched Puerto Rican ancestry controls ( N = 59 ) . We estimated the frequency of the COL27A1 . pG697R ( dbSNP = rs140950220 ) variant in the Population Architecture using Genomics and Epidemiology ( PAGE ) study . The PAGE study comprises a diverse global reference panel from five studies . African-American and Hispanic/Latina women from the Women’s Health Initiative ( WHI ) , a multi-center cohort study investigating post-menopausal women’s health in the US and recruited women at 40 centers across the US . Self-identified Hispanic/Latino/as from four sites in San Diego , CA , Chicago , IL , Bronx , NY , and Miami , FL as part of the Hispanic Community Health Study/Study of Latinos ( HCHS/SOL ) . African American , Japanese American , and Native Hawaiian participants from the Multiethnic Cohort ( MEC ) prospective cohort study recruiting men and women from Hawaii and California . The Global Reference Panel ( GRP ) created by Stanford University contributed samples including; a population sample of Andean individuals primarily of Quechuan/Aymaran ancestry from Puno , Peru; a population sample of Easter Island ( Rapa Nui ) , Chile; individuals of indigenous origin from Oaxaca , Mexico , Honduras , Colombia , the Nama and Khomani KhoeSan populations of the Northern Cape , South Africa; the Human Genome Diversity Panel in collaboration with the Centre Etude Polymorphism Humain ( CEPH ) in Paris; and the Maasai in Kinyawa , Kenya ( MKK ) dataset from the International Hapmap Project hosted at Coriell . Finally , the BioMe biobank in the Mount Sinai health system , New York City , contributed African-American , Hispanic/Latino/a , and participants who reported as mixed or other ancestry to the PAGE study , ~50% of whom were born outside New York City and for whom country-of-birth information was available . In all , participants in the PAGE Study represent a global reference panel of 112 populations ranging from 4 to 17773 individuals in size ( Supplement file 4 ) . Samples in the PAGE study were genotyped on the Illumina Multi-Ethnic Genotyping Array ( MEGA ) , which included direct genotyping of the rs140950220 variant . A total of 53338 PAGE and GRP samples were genotyped on the MEGA array at the Johns Hopkins Center for Inherited Disease Research ( CIDR ) , with 52878 samples successfully passing CIDR’s QC process . Genotyping data that passed initial quality control at CIDR were released to the Quality Assurance/Quality Control ( QA/QC ) analysis team at the University of Washington Genetics Coordinating Center ( UWGCC ) . The UWGCC further cleaned the data according to previously described methods ( Laurie et al . , 2010 ) and returned genotypes for 51520 subjects . A total of 1705969 SNPs were genotyped on the MEGA . The COL27A1 . pG697R variant passed the following filters; ( 1 ) CIDR technical filters , ( 2 ) SNPs with missing call rate >= 2% , ( 3 ) SNPs with more than 6 discordant calls in 988 study duplicates , ( 4 ) SNPs with greater than 1 Mendelian errors in 282 trios and 1439 duos , ( 5 ) SNPs with a Hardy-Weinberg p<10−10 , ( 6 ) positional duplicates . We downloaded X-ray crystal coordinates ( 1CAG from Bella et al . , 1994 . ; www . pdb . org ) on January 21 , 2017 . Visualization and modeling of the missense variant were performed in PyMol ( www . pymol . org; RRID:SCR_000305 ) . To test for clinical symptoms of Steel syndrome in COL27A1 . pG967R carriers , we performed a Phenome-Wide Association Study ( PheWas ) with EHR-derived ICD9 billing codes as the phenotypic outcome . In the association model , for each individual ICD9 codes were encoded as ‘1’ if the ICD9 was present in their EHR , and ‘0’ if the ICD9 code was absent . Carrier status for COL27A1 . pG697R was used as the primary predictor variable , with heterozygous individuals encoded as ‘1’ , non-carriers encoded a ‘0’ and homozygotes excluded from the analysis . We restricted the analysis to carriers of COL27A1 . pG697R ( n = 106 ) and non-carriers ( n = 4480 ) who either reported being born in Puerto Rico or who were US-born , self-identified as H/L and overlapped with Puerto Rican born individuals in principal component analysis . Age , sex and the first 5 principal components were included as covariates in our model . The regression was performed using four methods; Generalized Linear Models ( GLM ) using the glm ( ) function in Rv3 . 2 . 1; a score test based on the saddlepoint approximation ( SPATest ) using the SPAtest ( ) function in Rv3 . 2 . 1; a score test using a base adjustment for rare variants ( Firth test ) using the logistf ( ) function in Rv3 . 2 . 1; and a linear mixed model using the GCTAv1 . 24 . 2 software with a genetic relationship matrix constructed from 281666 SNPs shared between the OMNI and MEGA arrays ( MAF >= 1% ) . To adjust for multiple tests , raw p-values were adjusted for false discovery rate using the p . adjust ( ) function in R , and only those below an FDR adjusted p-value of 0 . 05 were reported as significant . Information from inpatient , outpatient , emergency and private practice settings housed in the Mount Sinai health system since 2004 was reviewed by two clinical experts independently . This data includes laboratory reports , radiological data , pathology results , operative and inpatient/outpatient progress notes , discharge summaries , pharmacy , and nurses reports . The clinical experts examined for clinical features similar to those reported for Steel syndrome cases in Flynn et al . ( 2010 ) , including developmental dysplasia of the hip ( or congenital hip dysplasia ) , carpal coalition , scoliosis , and joint and spine anomalies . Both clinical experts reviews patient records independently and compared notes to resolve discrepancies . They reviewed the records of the 34 youngest COL27A1 . pG697R carriers ( mean age 42 ) , and compared their findings to 31 randomly selected age and sex matched Puerto Rican non-carriers , and also to published reports of population prevalences of key clinical features for similar age groups where available . Due to the process of recombination , individuals from populations that have undergone recent admixture can exhibit a mosaic of genetic ancestry along their genome . Their genetic ancestry at a given genomic segment ( referred to as local ancestry ) , can be inferred from genotype data with the use of non-admixed reference panels of known continental ancestry . We calculated local ancestry in the three homozygous Puerto Rican individuals genotyped on OMNI by first extracting the intersecting sites of the Affymetrix 6 . 0 array ( n = 593729 SNPs in total ) and merging them with 3 ancestral reference panels . These reference panels consisted of the CEU and YRI samples from the 1KGP in addition to the Native American reference panel described previously that were used as a proxy for European , African and Native American ancestral source populations , respectively . RFMix ( Maples et al . , 2013 ) was used to infer local ancestry . To investigate evidence of a founder effect in Puerto Ricans we ran the IBDNe software ( Browning and Browning , 2015 ) in 2816 Puerto Ricans from the discovery effort using the cleaned set of pairwise IBD-haplotypes inferred using GERMLINE . IBDNe was run using the default parameters , including an assumed generation time of 25 years . All scripts used to generate main and supplementary figures for this manuscript are available for download on GitHub ( https://github . com/gillian-belbin/ibdmapping_ehr_figscripts [Belbin , 2017]; a copy is archived at https://github . com/elifesciences-publications/ibdmapping_ehr_figscripts ) . The data frames of analysis results that were used to generate the main and supplementary figures are located here: http://research . mssm . edu/kennylab/data_source_file . tar . gz BioMe OmniExpress data: https://www . ncbi . nlm . nih . gov/projects/gap/cgi-bin/study . cgi ? study_id=phs000888 . v1 . p1 BioMe MEGA data: https://www . ncbi . nlm . nih . gov/projects/gap/cgi-bin/study . cgi ? study_id=phs000925 PAGE MEGA data: https://www . ncbi . nlm . nih . gov/projects/gap/cgi-bin/study . cgi ? study_id=phs000356 Location of Native American panels: ftp://ftp . 1000genomes . ebi . ac . uk/vol1/ftp/technical/working/20130711_native_american_admix_train/ Software used in the analysis: SMARTPCA: https://github . com/DReichLab/EIG ADMIXTURE: https://www . genetics . ucla . edu/software/admixture/download . html RelateAdmix: http://www . popgen . dk/software/index . php/RelateAdmix#Download RFMix: https://sites . google . com/site/rfmixlocalancestryinference/ GERMLINE: http://www . cs . columbia . edu/~gusev/germline/ DASH: http://www1 . cs . columbia . edu/~gusev/dash/ PyMol: www . pymol . org IBDNe: http://faculty . washington . edu/browning/ibdne . html | Diseases often run in families . These disease are frequently linked to changes in DNA that are passed down through generations . Close family members may share these disease-causing mutations; so may distant relatives who inherited the same mutation from a common ancestor long ago . Geneticists use a method called linkage mapping to trace a disease found in multiple members of a family over generations to genetic changes in a shared ancestor . This allows scientists to pinpoint the exact place in the genome the disease-causing mutation occurred . Using computer algorithms , scientists can apply the same technique to identify mutations that distant relatives inherited from a common ancestor . Belbin et al . used this computational technique to identify a mutation that may cause unusually short stature or bone and joint problems in up to 2% of people of Puerto Rican descent . In the experiments , the genomes of about 32 , 000 New Yorkers who have volunteered to participate in the BioMe Biobank and their health records were used to search for genetic changes linked to extremely short stature . The search revealed that people who inherited two copies of this mutation from their parents were likely to be extremely short or to have bone and joint problems . People who inherited one copy had an increased likelihood of joint or bone problems . This mutation affects a gene responsible for making a form of protein called collagen that is important for bone growth . The analysis suggests the mutation first arose in a Native American ancestor living in Puerto Rico around the time that European colonization began . The mutation had previously been linked to a disorder called Steel syndrome that was thought to be rare . Belbin et al . showed this condition is actually fairly common in people whose ancestors recently came from Puerto Rico , but may often go undiagnosed by their physicians . The experiments emphasize the importance of including diverse populations in genetic studies , as studies of people of predominantly European descent would likely have missed the link between this disease and mutation . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"medicine",
"genetics",
"and",
"genomics"
] | 2017 | Genetic identification of a common collagen disease in Puerto Ricans via identity-by-descent mapping in a health system |
Zebrafish display a prominent pattern of alternating dark and light stripes generated by the precise positioning of pigment cells in the skin . This arrangement is the result of coordinated cell movements , cell shape changes , and the organisation of pigment cells during metamorphosis . Iridophores play a crucial part in this process by switching between the dense form of the light stripes and the loose form of the dark stripes . Adult schachbrett ( sbr ) mutants exhibit delayed changes in iridophore shape and organisation caused by truncations in Tight Junction Protein 1a ( ZO-1a ) . In sbr mutants , the dark stripes are interrupted by dense iridophores invading as coherent sheets . Immuno-labelling and chimeric analyses indicate that Tjp1a is expressed in dense iridophores but down-regulated in the loose form . Tjp1a is a novel regulator of cell shape changes during colour pattern formation and the first cytoplasmic protein implicated in this process .
One of the most fascinating features of vertebrates is their display of remarkable colour patterns in skin , fur , or plumage , frequently varying strikingly between closely related species . Teleost fish exhibit a particularly high diversity of patterns formed by several types of pigment cells distributed in a multilayered arrangement in the hypodermis ( Singh and Nüsslein-Volhard , 2015 ) . Adult zebrafish display a conspicuous pattern of alternating dark and light stripes; remarkably different from a relatively simple larval pattern , which is generated directly from neural crest cells migrating during embryogenesis ( Kelsh et al . , 1996 ) . The adult pattern is formed from neural crest-derived progenitors during metamorphosis ( 3–6 weeks of development ) . Metamorphic iridophores ( silvery cells containing reflective guanine platelets ) and melanophores ( dark cells containing the black pigment melanin ) arise from neural crest-derived stem cells associated with the peripheral nervous system , whereas metamorphic xanthophores ( yellow–orange cells containing pteridine based pigments ) originate from proliferating larval xanthophores ( Budi et al . , 2011; Dooley et al . , 2013; Mahalwar et al . , 2014; McMenamin et al . , 2014; Singh et al . , 2014 ) . Several adult viable zebrafish mutants displaying abnormal adult pigment patterns have been described ( Haffter et al . , 1996; Kelsh et al . , 1996; Lister et al . , 1999 ) . One class of genes primarily affects the formation of one of the three cell types . For example nacre/mitfa mutants lack melanophores , pfeffer/csf1ra/fms mutants lack xanthophores , and in shady/ltk iridophores are compromised ( Lister et al . , 1999; Parichy et al . , 2000; Lopes et al . , 2008 ) . Genetic analyses and regeneration studies revealed that interactions between all three cell types are necessary for proper stripe formation in the trunk of the fish ( Maderspacher and Nüsslein-Volhard , 2003; Nakamasu et al . , 2009; Frohnhöfer et al . , 2013; Patterson and Parichy , 2013 ) . Long-term in vivo imaging has shown that stripe formation involves intricate cell shape and density changes of metamorphic pigment cells ( Mahalwar et al . , 2014; Singh et al . , 2014 ) . Iridophores take a lead in stripe formation: they appear along the horizontal myoseptum , proliferate and spread as a dense sheet in the skin to form the first light stripe . At the margins of this first light stripe , the dense iridophores undergo a transition into a loose form and spread over the dark stripe region . Past the presumptive dark stripe , they change into the dense form again and aggregate into sheets forming new light stripes ( Singh et al . , 2014 ) . The first two dark stripes form dorsally and ventrally of the first light stripe by melanoblasts migrating along spinal nerves into the skin in the presumptive dark stripe region . They initially appear as stellate cells with the pigment located in the centre of the cells but later expand into the stationary rounded form ( Dooley et al . , 2013; Singh et al . , 2014 ) . Metamorphic xanthophores originate from larval xanthophores , they compact over the dense iridophores of the light stripe and change into a pale stellate shape above the loose iridophores and melanophores of the dark stripe ( Mahalwar et al . , 2014 ) . A different type of iridophores—L-iridophores—underlie the melanophores of the dark stripe . L-iridophores appear only after the first two dark stripes are formed and do not participate in laying out the pattern ( Frohnhöfer et al . , 2013; Hirata et al . , 2003 , 2005 ) . Interestingly , iridophore-deficient mutants are not affected in the stripe pattern of the fins , suggesting differences in the mechanisms involved in patterning of the trunk and fins ( Frohnhöfer et al . , 2013 ) . Mutants in which all three chromatophore types develop , but stripe formation is impaired , are of particular interest , as they can provide insights in the molecular mechanisms of cell–cell interactions underlying stripe formation . Several mutants have been described in which dark stripes are broken into spots . leopard/Cx 41 . 9 , luchs/Cx39 . 4 encode components of gap junctions involved in cell–cell communications ( Maderspacher and Nüsslein-Volhard , 2003; Watanabe et al . , 2006; Irion et al . , 2014 ) . In the absence of leo or luc , iridophores fail to change to the loose form and suppress melanophores . leo and luc presumably form heteromeric gap junctions among and between melanophores and xanthophores , instructing iridophores to change shape in a spatially controlled manner ( Irion et al . , 2014 ) . In this study , we present the mutant schachbrett ( sbr ) ( German for checkerboard ) that exhibits interruptions in dark stripes by light stripe regions . sbr encodes Tight Junction Protein 1a ( Tjp1a/ZO-1 ) . Immunostaining revealed that Tjp1a is expressed in dense iridophores but neither in loose iridophores nor any other pigment cell type . Analysis of double mutants and chimeras shows that sbr is cell-autonomously required in iridophores . During metamorphosis , dense iridophores invade the dark stripe regions and temporarily suppress the expansion of melanophores , suggesting that Tjp1a is required to regulate the transition of dense iridophores into the loose shape and their organisation .
Adult sbr fish display an unchanged arrangement and approximately normal width of stripes , however , the dark stripes are interrupted and undulating ( Figure 1A ) . The allele sbrtnh009b was isolated during a screen for ENU-induced recessive , homozygous viable mutants affecting adult pattern formation . The mutation was mapped to the region 29 . 6–32 . 5 Mb of chromosome 7 ( Ensembl Zebrafish release 72 ) ( Figure 1B ) . Using a candidate approach , we sequenced tjp1a cDNA of sbrtnh009b and detected a nonsense mutation leading to Y1143Stop change in the C-terminal part of the protein ( Figure 1C ) . To confirm the suggestion that this mutation is causative for the sbr phenotype , we performed a screen for additional alleles . ENU-mutagenized Tü males were crossed to sbrtnh009b females; the progeny was raised to the adulthood and screened for the sbr phenotype . Four new alleles not complementing the original allele were isolated . We identified novel stop codons in positions of the tjp1a gene corresponding to the N-terminal part of the protein in all four new alleles . The phenotype is variable , and no qualitative differences between the alleles could be recognized . Individual fish of the sbrtnh009b allele with the C-terminal truncation may show a weaker phenotype not seen in the other alleles , therefore , we cannot exclude that it may have residual function . In subsequent crosses , we never observed a segregation of the sbr phenotype and the tjp1a mutant alleles . These results show that the loss of Tjp1a function causes the sbr phenotype . 10 . 7554/eLife . 06545 . 003Figure 1 . schachbrett encodes Tjp1a . ( A ) All alleles of sbr exhibit interrupted , undulating dark stripes of normal arrangement and width when compared to wild type , but no other obvious defects . Scale bar: 5 mm . ( B ) Scheme of meiotic mapping of sbr . Marked are z-markers and contigs on which SNPs were found with their genomic and genetic ( where applicable ) coordinates . The numbers of recombinants among all fish tested are given in red and blue . The right-most bar shows genes on the ends of the final mapped region . The dotted region is not to scale and contains multiple genes . ( C ) DNA sequence traces for four alleles of sbr . Red rectangles mark the mutated residues . Red asterisks stand for stop codons . ( D ) Scheme of Tjp1a protein . Purple rounded squares indicate regions corresponding to polypeptides used for antibody generation . Red diamonds show the positions of stop codons in the mutants . DOI: http://dx . doi . org/10 . 7554/eLife . 06545 . 003 The larval pigment pattern is unaffected in sbr mutants ( Figure 2A , 6 . 5 mm ) . Repeated photography of individual fish revealed that mutants can be distinguished from wild-type siblings at stage 7 . 5 mm SL ( Standard Length [Parichy et al . , 2009] ) ( about 4 weeks post fertilisation ) shortly after the first metamorphic melanophores appear ( Figure 2A ) . At this and following stages , melanophores in the mutants appear as small dots when compared to melanophores of wild type , giving the metamorphic fish a pale appearance ( Figures 2A , 9 . 0–10 . 2 mm ) . Later ( 11 mm SL ) , the melanophores acquire a shape similar to wild-type cells ( Figure 2A , 11 . 6 mm ) . The melanophore numbers in mutant and wild-type fish do not differ significantly until 10 mm SL ( Figure 2B ) , when the pale phenotype is already established . In older mutant fish , there is a slight decrease in the average number of melanophores , likely due to the interruptions of the stripe areas ( Figure 2C , wt , sbr ) . To assess the impact of melanophore number on stripe integrity , we compared sbr to sparse ( spa ) mutants , which have decreased numbers of melanophores ( Johnson et al . , 1995 ) . spa mutants have only about a third as many melanophores as wild-type fish ( Figure 2C , plot ) ; however , these cells form uninterrupted stripes ( Figure 2C , spa ) . Double mutants sbr;spa display a combination of both phenotypes ( Figure 2C , sbr;spa ) . This indicates that the pale appearance of the mutant metamorphic fish is caused by an abnormal size , shape , or pigment arrangement rather than a reduced number of melanophores . 10 . 7554/eLife . 06545 . 004Figure 2 . Abnormal behaviour of sbr mutant melanophores . ( A ) Pigment pattern during metamorphosis in the mid-trunk of individual wild type and sbr mutant fish . Arrowheads: forming interruptions . White arrowheads: disappearing melanophores ( N = 6 ) . Scale bar: 1 mm . ( B ) Average number of melanophores per segment in the first two dark stripes in wild type and mutant fish plotted against standard length . Red circles—individual wild type fish; blue squares—individual sbr fish . Inset shows the area where melanophores were counted . Distributions of melanophore numbers in mutants and wild type fish do not differ significantly until the 10 mm stage as shown by Kolmogorov–Smirnov statistics . At 10–14 mm stages the distributions are different with p-values < 0 . 05 . ( C ) Close-ups of mid-trunk regions of adult wild type , sbr , spa and spa;sbr and melanophore numbers in a dark stripe dorsal to the first light stripe of adult fish . Red lines—standard deviation . Scale bar: 2 mm . DOI: http://dx . doi . org/10 . 7554/eLife . 06545 . 00410 . 7554/eLife . 06545 . 005Figure 2—figure supplement 1 . Width of the first light stripe in sbr and wild type fish . The width was measured at the anal fin area at five points for every individual . The obtained data measurements were divided by fish body height . These ratios were plotted against SL of metamorphic fish . Orange circles represent individual measurements for wild type . Red circles show the average of five measurements for each wild type fish . Light blue squares—individual measurements for sbr . Dark blue squares—average of five measurements of sbr . DOI: http://dx . doi . org/10 . 7554/eLife . 06545 . 005 In early metamorphic mutant fish , but not in adults , the width of the first light stripe , composed of dense iridophores covered by compact yellow xanthophores , is increased compared to wild type ( Figure 2A , 11 . 6 mm; Figure 2—figure supplement 1 ) . After 10 mm SL , dense S-iridophores and xanthophores can be observed in the dark stripe region in sbr mutants ( Figure 2A , arrowheads ) and melanophores disappear from these areas ( Figure 2A , white arrowheads ) , ultimately leading to the interruptions . To investigate iridophore behaviour , we performed repeated imaging of wild-type and sbr individuals over a period of 2 weeks . To allow a more detailed visualisation of the cell shapes , we imaged fish carrying the Tg ( TDL358:GFP ) transgene ( labelling iridophores and glia with cytosolic GFP [Levesque et al . , 2013] ) alone ( Figure 3 ) or together with a second transgene , Tg ( sox10:mRFP ) ( Figure 4 ) , which labels neural crest derivatives with membrane-bound mRFP . In both , wild type and mutants , iridophores appeared in segmental clusters during early metamorphosis ( about 7 mm SL ) , they increased in number and merged to form the first light stripe ( Figure 3A , Figure 4A ) . In wild type , iridophores proceeded to define the edge of the light stripe , there they delaminated and formed loose iridophores , which spread dorsally and ventrally over the dark stripe regions ( Figure 3A , 8 . 9 mm; Figure 4; Singh et al . , 2014 ) . Dense iridophores occasionally spread too far from the horizontal myoseptum ( Figure 3B , wt ) , but later formed sharp light stripe borders . However , in the mutants the dense iridophores did not delaminate but continued to spread over the metamorphic melanophores as a coherent sheet ( Figure 3A , sbr 8 . 9 mm; Figure 4 , sbr , 8 . 3 mm ) . At later stages , eventually some of them switched to the loose form ( arrowheads in Figure 3B; Figure 4A ) and occasionally seemed to disappear from the dark stripe regions at a time point , which coincided with expansion of melanophores ( 10 . 5 mm SL , Figure 3—figure supplement 1 ) . When this retreat did not happen , the iridophores persisted in interruptions of the dark stripes ( Figure 2A , 11 . 6 mm ) . The failure to precisely form the boundary between light and dark stripes might be a cause for another anomaly observed in sbr mutants: L-iridophores , which are restricted to dark stripe areas in wild type , were observed in light stripes of adult sbr mutants ( Figure 3—figure supplement 2 ) . 10 . 7554/eLife . 06545 . 006Figure 3 . Behaviour of sbr mutant iridophores during metamorphosis . ( A ) Repeated imaging of Tg ( TDL358:GFP ) wild type and mutant metamorphic individual ( N = 5 each , one shown ) . Scale bar: 300 µm . ( B ) Same individuals with another magnification . Empty patches in the light stripe of wild type fish are caused by variegation of the transgene expression . Arrowheads: loose iridophores . Scale bar: 300 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 06545 . 00610 . 7554/eLife . 06545 . 007Figure 3—figure supplement 1 . Invading sbr iridophores occasionally retreat . Iridophores ( marked with the yellow outline ) are retreating from the area between two groups of melanophores . Scale bar: 150 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 06545 . 00710 . 7554/eLife . 06545 . 008Figure 3—figure supplement 2 . L-iridophores in wt and sbr . In wild type dense S-iridophores and L-iridophores are separated , but in sbr L-iridophores can be observed in S-iridophore area of light stripes . Pigment assembles in the centres of melanophores due to prolonged light exposure prior to fixation . DOI: http://dx . doi . org/10 . 7554/eLife . 06545 . 00810 . 7554/eLife . 06545 . 009Figure 4 . Behaviour of sbr mutant iridophores during establishment of the first dark stripes . ( A ) Tg ( TDL358:GFP ) ; Tg ( sox10:mRFP ) wild type and sbr metamorphic fish ( N = 4 each , one shown ) . Arrowheads point to delaminating loose iridophores . Arrow shows dense iridophores failing to delaminate . Scale bar: 150 µm . ( B ) Close-ups of Tg ( TDL358:GFP ) ; Tg ( sox10:mRFP ) wild type and sbr metamorphic fish 8 . 3 SL . Note difference in iridophore shapes in wild-type . Scale bar: 50 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 06545 . 009 Analysing fish carrying the transgene Tg ( kita:GalTA4:UAS:mCherry ) , which labels melanophores ( Anelli et al . , 2009 ) , we observed that in sbr mutants individual melanophores moved away from invading dense iridophores , while maintaining a migratory stellate shape , or they disappeared after being trapped ( Figure 5 , Figure 5—figure supplement 1 ) . This is in agreement with the observed reduction in the number of melanophores in sbr during later stages of development ( Figure 2B ) . 10 . 7554/eLife . 06545 . 010Figure 5 . Two closely positioned melanophores in sbr ( arrowheads ) , are migrating away from the iridophores in posterior and anterior directions . Scale bar: 100 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 06545 . 01010 . 7554/eLife . 06545 . 011Figure 5—figure supplement 1 . Melanophores trapped in the mass of iridophores are disappearing in sbr . Number of melanophores in the marked light stripe area is shown in the upper right corner . Scale bar: 200 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 06545 . 011 To investigate in which cell type sbr function is required , we analysed sbr in combination with mutants lacking one of the three pigment cell types . Both shady mutants , lacking iridophores ( shd , Figure 6C , Figure 6—figure supplement 1A ) and shd;sbr double mutants ( Figure 6D , Figure 6—figure supplement 1B ) , display the shd phenotype with no detectable differences , suggesting that sbr function is only required in iridophores . In contrast , the phenotypes of double mutants with nacre ( nac , no melanophores , Figure 6E ) or pfeffer ( pfe , no xanthophores , Figure 6G ) differ from the single mutants . Both pfe and nac alone exhibit expanded areas of dense iridophores . In combination with sbr , both double mutants show a further expansion of these dense iridophore regions ( Figure 6F , H ) , covering most of the body . This phenotypic enhancement suggests that the cell type affected in sbr is still present in nac and pfe mutants , again pointing to iridophores . To confirm these findings , we created chimeric animals by blastomere transplantations . Experiments with sbr donors and nac or pfe recipients revealed that sbr melanophores and xanthophores can participate in normal pattern formation ( Figure 6I , J ) . When we used shd;sbr double mutants as recipients ( Figure 6D ) and nac;pfe ( Figure 6K ) as donors , which can provide only iridophores , we observed regional restoration of the striped pattern in the chimeric fish ( Figure 6L ) . This indicates that sbr is required cell autonomously in iridophores and confirms that mutant sbr melanophores and xanthophores can contribute to the normal pattern when confronted with wild-type iridophores . 10 . 7554/eLife . 06545 . 012Figure 6 . tjp1a is required in iridophores , but not melanophores or xanthophores . ( A ) Wild type fish . ( B ) sbr fish . ( C ) shady ( shd ) mutant , which lacks iridophores . ( D ) shd;sbr mutant is indistinguishable from shd . ( E ) nacre ( nac ) mutant , which lacks melanophores . ( F ) nac;sbr double mutant exhibiting expanded dense iridophore areas in comparison to nac alone . ( G ) pfeffer ( pfe ) mutant , which has no xanthophores . ( H ) pfe;sbr double mutant exhibiting expanded dense iridophore areas in comparison to pfe alone . ( I ) Chimeras , obtained from transplantation of sbr blastomeres into nac recipient blastulas , show clonal rescue . ( J ) Chimeras obtained from transplantation of sbr blastomeres into pfe recipient blastulas , show clonal rescue . ( K ) nac;pfe fish have only one type of pigment cells—iridophores . ( L ) Chimeras obtained from transplantation of nac;pfe blastomeres into shd;sbr recipient blastulas , show clonal rescue . Scale bars: 5 mm . DOI: http://dx . doi . org/10 . 7554/eLife . 06545 . 01210 . 7554/eLife . 06545 . 013Figure 6—figure supplement 1 . Phenotypes of shd and shd;sbr mutants . Shown are ( A ) shd and ( B ) shd;sbr . Scale bars 5 mm . DOI: http://dx . doi . org/10 . 7554/eLife . 06545 . 013 The fins of sbr mutants are striped , although we detect branching and supernumerary stripes to various extents in the caudal fins of some sbr mutant fish but not in their anal fins suggesting that there is no systematic defect in fin patterning . This is in agreement with the finding that iridophores are not required for striping the fins ( Hirata et al . , 2005; Frohnhöfer et al . , 2013; Krauss et al . , 2013 ) . We raised two polyclonal antibodies in rabbits specific to zebrafish Tjp1a ( Figure 1D ) . α-Tjp1aN was designed to recognize both , truncated sbrtnh009b and wild-type Tjp1a protein , whereas α-Tjp1aC would only bind to the wild-type protein . Both antibodies allow the detection of Tjp1a in epithelial cells of larval and adult zebrafish skin ( Figure 7—figure supplements 1 , 2 ) . This staining is absent in mutants with stop codons in the N-terminal part of tjp1a but present in sbrtnh009b mutants stained with α-Tjp1aN ( Figure 7—figure supplement 1 ) . We also detected expression of Tjp1a in blood vessels during larval and adult stages ( Figure 7—figure supplement 2 ) , corroborating earlier reports on the expression of Tjps in zebrafish and mice ( Anderson and Itallie , 1995; Blum et al . , 2008 ) . Immunostaining of skin in metamorphic fish carrying the Tg ( TDL358:GFP ) transgene ( Figure 7A ) shows that Tjp1a is expressed in dense iridophores of the light stripe . Intriguingly , delaminated loose iridophores still express GFP , but no Tjp1a is detectable ( Figure 7B ) . This indicates that Tjp1a is down-regulated during delamination of loose iridophores from the dense sheet in the light stripe . In adult skin preparations , the signal can be observed in dense iridophores of the light stripes but not in xanthophores , melanophores , L- , or loose iridophores ( Figure 7C ) . Together with our observation that the loss of tjp1a function in sbr mutants compromises the transition of iridophores from dense to loose state , these results suggest that Tjp1a is a component of the molecular switch that regulates iridophore shape changes during their dispersal . 10 . 7554/eLife . 06545 . 014Figure 7 . Tjp1a is expressed in dense iridophores . ( A ) Double antibody staining of metamorphic Tg ( TDL358:GFP ) fish with α-Tjp1aC and α-GFP antibodies . Note: not all iridophores are expressing GFP due to transgenic line variegation . Scale bar: 100 µm . ( B ) Loose iridophores migrating over the dark stripe in 8 . 3 mm metamorphic Tg ( TDL358:GFP ) fish express GFP , but not Tjp1a , although the epithelial staining is still visible . Scale bar: 30 µm . ( C ) α-Tjp1aC staining in skin of adult wild type fish . The protein is detected in the sheet of dense S-iridophores of the light stripe , but not in L-iridophores ( black arrowheads ) , loose iridophores ( white arrowheads ) , melanophores or xanthophores . Scale bar: 100 µm . ( D ) Double antibody staining with α-Tjp1aC and α-GFP of skin of adult chimera , obtained by transplanting sbr;Tg ( H2A:GFP ) blastomeres into wild type blastula . Either GFP or Tjp1a was detected in cells , never both . Some sbr cells express no GFP due to variegation of the transgene expression . Scale bar: 30 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 06545 . 01410 . 7554/eLife . 06545 . 015Figure 7—figure supplement 1 . Tjp1a stainings in wild type and sbr . ( A ) α-Tjp1aC staining in adult wild type fish skin sample shows signal colocalizing with E-cadherin , expressed in epithelial cells . In adult sbrtnh009b skin samples the Tjp1a staining is not observed , but E-cadherin is detected . Scale bar: 20 µm . ( B ) α-Tjp1aN antibody stains skin epithelium of both wild type and sbrtnh009b adult fish but not sbrtwl4 . Scale bar: 20 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 06545 . 01510 . 7554/eLife . 06545 . 016Figure 7—figure supplement 2 . Characterization of the Tjp1a expression domain . ( A ) Signal in epithelium of wild type 5 dpf larva stained with α-Tjp1aC . Scale bar: 20 µm . ( B ) Signal in two layers of adult wild type epithelium ( about 7 µm apart ) stained with α-Tjp1aC st . Scale bar: 20 µm . ( C ) Double staining of whole-mount Tg ( kdrl:GFP ) 5 dpf larvae with α-Tjp1aC and α-GFP demonstrates the expression of Tjp1a in blood vessels ( arrowheads ) and intestinal epithelium ( arrows ) . Scale bar: 20 µm . ( D ) α-Tjp1aC staining shows expression of Tjp1a in vasculature of adult Tg ( kdrl:GFP ) animal . Scale bar: 50 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 06545 . 01610 . 7554/eLife . 06545 . 017Figure 7—figure supplement 3 . Correlation between clonal rescue of sbr phenotype and Tjp1a expression . ( A ) shd;sbr with nac;pfe clones show rescue of the wild type phenotype . Scale bar: 5 mm . ( B ) Immunostaining of shd;sbr with nac;pfe clone with α-Tjp1aC antibody demonstrates the presence of the protein in dense iridophores but not in the epithelium . Scale bar: 500 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 06545 . 017 Additionally , we analysed chimeras obtained by transplanting blastomeres from sbrtwl4embryos , where transplanted cells were labelled with expression of the ubiquitous Tg ( H2A:GFP ) transgene , into blastula stage wild-type embryos . Double stainings with α-Tjp1aN and α-GFP antibodies show that the donor-derived sbr dense iridophores integrate with the wild-type recipient iridophores but do not express Tjp1a ( Figure 7D ) . This suggests that the sbr phenotype is not caused by over-proliferation of iridophores , since they do not produce large clusters . We stained skin of nac;pfe/shd;sbr chimeras with α-Tjp1aC and detected Tjp1a in donor-derived iridophores but not in the epithelium , suggesting that loss of Tjp1a function in the epithelium does not affect pattern formation ( Figure 7—figure supplement 3 ) . To investigate the genetic interactions between tjp1a and potential partners , cx39 . 4 and cx41 . 8 , we evaluated the phenotypes of double mutants with luct32241 and leot1 ( Figure 8 ) . luc mutant fish display meandering and broken stripes , whereas in leot1 the stripes are broken into spots . In the double mutants with sbr , we observe considerably stronger patterning defects than in the single mutants . In the case of sbr;luc , an almost complete loss of melanophore clustering is observed; the upper part of the body is covered with a layer of dense iridophores . In the case of sbr;leo , the melanophore spots are even smaller and the dense iridophore-free areas around them are narrower . These results suggest that connexins and tjp1a do not act in a linear pathway affecting pigmentation . To investigate whether zebrafish Tjp1a can interact directly with connexins , we performed yeast two-hybrid assays ( Figure 8—figure supplement 1 ) . We observed interactions between Cx41 . 8 and all three PDZ domains of Tjp1a and between Cx39 . 4 and PDZ-2 and 3 in this assay . 10 . 7554/eLife . 06545 . 018Figure 8 . Genetic interactions between luc , leo and sbr . ( A ) luchst32241 ( luc ) mutant affects Cx39 . 4 and results in meandering and broken stripes . ( B ) luchst32241;sbrtwl4 mutant exhibits complete loss of stripes and expansion of dense iridophore area . ( C ) leopardt1 ( leo , cx41 . 8 ) stripes are broken into spots . ( D ) leot1;sbrtnh009b double mutant displays decrease in the size of the spots . Scale bars: 5 mm . DOI: http://dx . doi . org/10 . 7554/eLife . 06545 . 01810 . 7554/eLife . 06545 . 019Figure 8—figure supplement 1 . Interaction of PDZ domains of Tjp1a with connexins . Shown are the results of a yeast two-hybrid experiment . Left column: growth on SD plates lacking Trp and Leu indicates the presence of both plasmids ( bait and prey ) ; right column: growth and blue colour on SD plates lacking Trp , Leu and His , supplemented with x-α-Gal , indicates interaction of the two proteins in the yeast cell . All three PDZ domains of Tjp1a strongly interact with Cx41 . 8 ( leo ) , PDZ-2 and PDZ-3 strongly interact with Cx43 as well . Only PDZ-3 shows strong interaction with Cx39 . 4 ( luchs ) . DOI: http://dx . doi . org/10 . 7554/eLife . 06545 . 019
One surprising finding of this study is that Tjp1a-deficient zebrafish are viable unlike embryonic lethal Tjp1−/− mice ( Katsuno et al . , 2008 ) . There are three tjp genes ( 1–3 ) in mammals and five in zebrafish ( 1a–b , 2a–b , 3 ) , due to the whole genome duplication in teleosts ( Amores et al . , 1998 ) . The lack of Tjp1a function in epithelial cells in sbr mutants might be compensated for by other Tjps , for example , Tjp1b , which does not exist in mammals . This is supported by the observation that morpholino-mediated knockdown of tjp1b in sbr mutants , but not wild-type embryos , results in impaired blood flow and death at 5 dpf ( Videos 1–3 ) . This suggests that Tjp1b and Tjp1a have redundant functions at least in the vasculature epithelial cells . This notion is supported by experiments with mammalian cell cultures showing that absence of ZO-1 leads to increased recruitment of ZO-2 to cell membranes , which is suggested to compensate for the absence of ZO-1 ( Umeda et al . , 2004 ) . 10 . 7554/eLife . 06545 . 020Video 1 . 50 hpf wild type embryo . Note normal blood flow . DOI: http://dx . doi . org/10 . 7554/eLife . 06545 . 02010 . 7554/eLife . 06545 . 021Video 2 . 50 hpf wild type embryo injected with morpholino against Tjp1b . Note normal blood flow ( N = 53/53 ) . The result shows non-toxicity of morpholino . No defects are observed in the injected fish ( observed until adulthood ) . DOI: http://dx . doi . org/10 . 7554/eLife . 06545 . 02110 . 7554/eLife . 06545 . 022Video 3 . 50 hpf sbr embryo injected with morpholino against Tjp1b . No blood flow is observed , possibly due to disrupted angiogenesis ( N = 30/38 ) . None of the 30 individuals without blood flow survived past 5 dpf . DOI: http://dx . doi . org/10 . 7554/eLife . 06545 . 022 Our data show that in sbr dense iridophores fail to switch to the loose form in dark stripe regions . In wild type , dense iridophores normally stay restricted to developing light stripes , but occasionally spread into the prospective dark stripe areas . This irregularity is usually corrected and sharp stripe boundaries are formed . However , in sbr , the invasion of dense iridophores occurs along the whole length of stripes . Not all dense iridophores persist in dark stripe regions in sbr mutants . In summary , we hypothesize that the loss of Tjp1a impairs the ability of iridophores to recognise the ( as yet unknown ) cues defining the dark stripe areas or their ability to react to them efficiently . So far , only a rather small number of molecules have been identified , which are involved in the various interactions between chromatophores . Tjp1a is the first for which a molecular distribution and cell type specific expression has been shown . Several zebrafish mutants including leopard ( Haffter et al . , 1996; Watanabe et al . , 2006 ) , luchs ( Irion et al . , 2014 ) , and seurat ( Eom et al . , 2012 ) exhibit a spotted pattern formed by ingressions of iridophores into the dark stripe area . luc and leo encode Connexin41 . 8 ( Cx41 . 8 ) and Connexin39 . 4 ( Cx39 . 4 ) , respectively , which are , in contrast to sbr , required in melanophores and xanthophores ( Maderspacher and Nüsslein-Volhard , 2003; Irion et al . , 2014 ) . Irion et al . suggest that Cx39 . 4 and Cx41 . 8 form heteromeric gap junctions , promoting interactions of melanophores and xanthophores that result in the appropriate patterning of iridophores . In the absence of xanthophores or melanophores , dense iridophore regions are expanded ( Frohnhöfer et al . , 2013 ) , suggesting that Tjp1a in iridophores may be involved in cell communication with xanthophores and/or melanophores . However , the downstream cytoplasmic partners of the transmembrane proteins shown to be involved in patterning in melanophores and xanthophores are unknown as well as transmembrane molecules in iridophores that are responsible for the interactions . Another mutant displaying a spotted pattern is seurat , encoding the transmembrane protein immunoglobulin superfamily member 11 ( Igsf11 ) ( Eom et al . , 2012 ) . Interestingly , Cx41 . 8 and Igsf11 are possible interacting partners of Tjps since they have putative PDZ-binding motifs on their extreme C-termini ( Hung and Sheng , 2002; Suzu et al . , 2002 ) . The multiple protein–protein interacting domains in Tjps allow for many interacting partners and facilitate formation of large complexes in proximity of cell membranes that are associated with tight , adherens , and gap junctions . These provide a link between transmembrane proteins and the cytoskeleton and were shown to participate in regulation of many cellular processes such as junction assembly , cell proliferation , and differentiation ( Balda and Matter , 2000; Bauer et al . , 2010; Xu et al . , 2012; González-Mariscal et al . , 2014 ) . Our results show that sbr enhances the phenotypes of both luc and leo mutants . This suggests that Tjp1a and connexins do not act in a linear pathway to regulate pattern formation , but most likely work through different mechanisms . One possible explanation is that Tjp1a is required for spatially and temporally controlled reaction of iridophores in response to melanophores ( directly or through xanthophores ) . Absence or truncation of Tjp1a results in a delayed switch to the loose form , which in turn forces melanophores to reorganize according to the presence of dense iridophores in normally iridophore-free regions . In luchs and leopard , the melanophore and xanthophore autonomous mutations also affect patterning of iridophores , likely due to the failure to properly guide iridophores ( Irion et al . , 2014 ) . The combined effect of failure of melanophores and xanthophores to provide cues to iridophores , and the delayed reaction of iridophores might be responsible for the enhanced phenotypes in the double mutants . Interestingly , Tjp1a and Cx39 . 4 and Cx41 . 8 can interact in a yeast 2-hybrid assay . As in the fish they are required in different pigment cell types , this may point to the existence of other , as yet unknown , connexins , similar to the Cx41 . 8 and Cx39 . 4 , that are expressed in iridophores and interact with Tjp1a . ZO-1 was shown to regulate gap junction assembly , localization , and regulate plaque size in mammalian cell cultures ( Hunter et al . , 2005; Laing et al . , 2005; Rhett et al . , 2011 ) . Defective Tjp1a in sbr might affect proper interaction of iridophore connexins with their counterparts in melanophores and xanthophores , compromising cell–cell communication and recognition . Iridophore-specific connexins or other molecules , responsible for communication between pigment cells , might also transmit signals via Tjp1a , controlling iridophore migration or shape change in a spatiotemporally appropriate manner . Immunostainings show that Tjp1a is expressed in dense iridophores , but not in loose iridophores . Intriguingly , the absence of Tjp1a does not obviously affect the morphology of dense iridophores , which display normal shape and organisation . In vitro studies of the past decade have demonstrated a function of ZO-1 in organisation of confluent cell layers . Counterintuitively , ZO-1 −/− Eph4 cells polarize and form tight junctions morphologically indistinguishable from those of ZO-1 +/+ cells , but the formation is delayed . These cells do not exhibit abnormal growth or motility in scratch assays ( Umeda et al . , 2004 ) . However , knockdown of endogenous ZO-1 in COS-7 cells hampers delamination and migration of cells to fill the wound area in scratch assays ( Huo et al . , 2011 ) . These data suggest that epithelial cells of different origin may react differently to the absence of ZO-1 . Our Tjp1aN antibody shows that the non-functional truncated protein is at least partially retained and normally localized in sbrtnh009n mutants , suggesting that the missing domains ( ZU-5 and possibly parts of afadin- and actin-binding regions [Bauer et al . , 2010] ) are crucial for the function of Tjp1a in iridophores . It was shown that absence of the ZU5 domain of ZO-1causes defective delamination and migration of COS-7 cells ( Huo et al . , 2011 ) . Furthermore , mis-expression of truncated ZO-1 in the presence of the wild-type protein in CE culture leads to the expression of mesenchymal markers and to an epithelial–mesenchymal transition ( EMT ) ( Ryeom et al . , 2000 ) . Taken together with our findings , these data suggest a role of ZO-1 in regulating and fine-tuning of cell shape and state . We show that transitions in cell shape and organisation are crucial for the arrangement of pigment cells in stripes and identify Tjp1a as a regulator of this process . It appears that the presence of Tjp1a allows iridophores to change into the loose form at the appropriate positions . This suggests that Tjp1a is required for interaction of iridophores with other pigment cells ( for example through controlling assembly of gap junctions ) and/or appropriate reaction of iridophores to perceived cues ( through control of delamination and cell shape ) . The spatial and temporal regulation of iridophore shape transitions by Tjp1a might underlie the generation of a variety of patterns observed in teleosts . Moreover , the viability of sbr mutants presents exciting opportunities for studying the behaviour of Tjp1 deficient cells in vivo .
Fish were bred and maintained as described ( Nüsslein-Volhard and Dahm , 2002 ) . Fish of the following genotypes were used: Tü , WIK , TE wild-type strains ( Tübingen zebrafish stock centre ) , luchst32241 ( Irion et al . , 2014 ) , leot1 ( Watanabe et al . , 2006 ) , nacrew2 ( Lister et al . , 1999 ) , pfeffertm236b ( Odenthal et al . , 1996 ) , shadyj9s1 ( Lopes et al . , 2008 ) , sparseb134 ( Kelsh et al . , 1996 ) , Tg ( TDL358:GFP ) ( Levesque et al . , 2013 ) , Tg ( kdrl:GFP ) ( Jin et al . , 2005 ) , Tg ( kita:GalTA4:UAS:mCherry ) ( Anelli et al . , 2009 ) , Tg ( sox10:mRFP ) ( M Levesque; CN-V laboratory ) , Tg ( H2A:GFP ) ( A Mongera; CN-V laboratory ) . Fish were staged according to the normal table of zebrafish development ( Parichy et al . , 2009 ) . The original allele sbrtnh009b was identified in a screen for mutants induced with N-ethyl-N-nitrosourea ( N5809 , Sigma-Aldrich , St . Louis , Missouri ) in Tü wild-type background . Mutagenesis was carried out as described previously ( Rohner et al . , 2011 ) . Subsequently , fish were crossed to TE and later maintained in homozygosity by regular outcrossing . Four new alleles were isolated by crossing mutagenized Tü males to sbrtnh009b females and screening the adult progeny for the sbr phenotype . sbrtnh009b/WIK fish were incrossed and used for meoitic mapping as described previously ( Nüsslein-Volhard and Dahm , 2002 ) . The mutation was mapped to the region between microsatellite markers z4706 ( 36 . 7 cM ) and z52932 ( 41 . 4 cM ) on chromosome 7 . Further , the interval was narrowed to the region 29 . 6–32 . 5 Mb of chromosome 7 , between contigs CR356242 and BX3235912 ( Ensembl Zebrafish release 72 ) . The following primers were used: CR356242_F GTAGTATATGGATATGGATG CR356242_R CCACCGCTGCATACCCTGC BX3235912_F CTTGCACAGGGAATGTGT BX3235912_R CTGCAGTGTTCTCACGCT To check for presence of lesions in tjp1a , RNA was extracted from blastema of adult wild-type and sbr fish using TRIzol reagent ( 15596 , Thermo Fisher Scientific , Waltham , Massachusetts ) . cDNA was obtained using Omniscript RT kit ( 205111 , Qiagen , Netherlands ) . Four overlapping regions of the coding region of tjp1a ( ENSDART00000148347 ) were amplified using Taq polymerase S ( M3001 . 0250 , Genaxxon , Germany ) and the following primers: tjp1a_1F 5′-GACTGCGGGATTTCAGTTGT-3′ tjp1a_1R 5′-CACTATTCGCCGGTACACATC-3′ tjp1a_2F 5′-GCAGAAGAAGAAAGATGTGTAC-3′ tjp1a_2R 5′-ATGTGAACCGTCCGCCTTG-3′ tjp1a_3F 5′-CAACCATCATCTCTTCACAGCCACT-3′ tjp1a_3R 5′-GATTTTCTCCACTGACTCTGCTCTGG-3′ tjp1a_4F 5′-CTGGATCAAGAGAAGACCTTTAGAACTC-3′ tjp1a_4R 5′-TCCCTGCAGTCTCAGAGGTT-3′ . PCR products were cloned into pGEM-T Easy ( A360 , Promega , Fitchburg , Wisconsin ) and sequenced using Big Dye Terminator v3 . 1 kit ( 4337455 , Thermo Fisher Scientific ) . Two parts of the tjp1a cDNA corresponding to 992–1143 a . a . ( α-Tjp1aN ) and 1293–1397a . a . ( α-Tjp1aC ) of Tjp1a ( ENSDART00000148347 ) were cloned both into pET28-nusA ( Novagen ) and pOPT-GST-Kan ( gift from U Irion and O Perisic ) plasmids to produce 6xHis-nusA and GST-tagged fusions . The following primers were used to amplify these regions: tjp1aN_F 5′-CATATGTACAAGAAGGATATCTACCGACCC-3′ tjp1aN_R 5′-GGATCCTTAGGAAGGCCTTTGGG-3′ tjp1aC_F 5′-CATATGAAACCCTCCACACAGCTGACAC-3′ tjp1aC_R 5′-GGATCCTTAGCTGGACGTGGCAG-3′ . Obtained constructs were used to transform BL21-CodonPlus DE3-RIPL ( 230280 , Agilent Technologies , Santa Clara , California ) cells . The cells were grown in 1 ml of 2xTY medium containing 20 mM glucose and 15 µg/ml kanamycin for 3 hr on 37°C , 220 rpm . This culture was used to inoculate 50 ml of the same medium and was grown overnight on 20°C , 220 rpm . His-tagged polypeptides were purified using HiTrap IMAC FF 1 ml ( 17-0921 , GE Healthcare , UK ) charged with Ni2+ and 250 mM imidazole in the elution buffer . GST-tagged polypeptides were purified using GSTrap FF 1 ml ( 17-5130 , GE Healthcare ) . In all cases , the samples of eluted proteins were loaded on NuPage Novex 4–12% Bis-Tris gel ( NP0322BOX , Thermo Fisher Scientific ) and stained with Coomassie Brilliant Blue G-250 to assess the purity . The polypeptides were dialyzed in PBS using Slide-A-Lyzer Dialysis Cassettes 10K MWCO ( 66383 , Thermo Fisher Scientific ) . The protein concentrations were assessed using Bradford method . His-tagged polypeptides were used to immunize rabbits with Freund's complete adjuvant ( F5881 , Sigma-Aldrich ) as immunopotentiator . GST-tagged polypeptides were bound to HiTrap NHS-activated HP columns ( 17-0716 , GE Healthcare ) and used to purify the corresponding antibodies from rabbit serum , using PBS as binding buffer and 100 mM glycine pH 2 . 3 as elution buffer . The purified antibodies were neutralized with Tris-HCl pH 9 . 5 and mixed 1:1 with glycerol . Antibody staining was performed as described previously ( Singh et al . , 2014 ) omitting methanol hydration/rehydration and HCl steps . Antibodies used were mouse α-E-cadherin ( 610181 , BD Biosciences , Franklin Lakes , New Jersey ) , mouse α-GFP ( 11814460001 , Roche , Germany ) , goat α-rabbit coupled with Cy3 ( 111-165-003 , Dianova , Germany ) , goat α-mouse AlexaFluor 488 ( A21131 , Molecular Probes , Eugene , Oregon ) . All antibodies were used in 1:400 dilution , except α-Tjp1aN and α-Tjp1aC , which were used in 1:100 dilution . Chimeras were produced as described ( Nüsslein-Volhard and Dahm , 2002 ) using mid-blastula stage ( 1000 cell stage ) embryos , transplanting 30–60 cells . We used Zeiss LSM 780 NLO confocal microscope and Canon 5D Mk II camera to obtain images . Fiji ( Schindelin et al . , 2012 ) , Adobe Photoshop , and Adobe Illustrator CS6 were used for image processing and analysis . Maximum intensity projection was made for fluorescent channels of confocal scans . For bright-field images , we used ‘stack focuser’ plugin or a single slice on an appropriate depth . For adult fish photos , multiple RAW camera images were taken in different focal planes and auto-align and auto-blend functions of Photoshop were used . Repeated imaging of metamorphic fish and anaesthesia were performed as described previously ( Singh et al . , 2014 ) . Melanophores in metamorphic fish were counted in five segments in the middle 70% of myotome starting with the one above the first ray of the anal fin and proceeding posteriorly . A Kolmogorov–Smirnov test was conducted in SciPy ( Jones et al . , 2001 ) to compare the distributions of melanophore counts in mutant vs wild-type fish . An initial comparison was conducted on fish of 4–6 mm SL . Sample sizes were then increased to include the melanophore counts of fish of 6–7 mm SL , and each subsequent data set was formed in a similar fashion by 1 mm increment . The null hypothesis of the samples being drawn from the same distribution was rejected with a p-value of 0 . 011 when a data set composed of 4–10 mm SL fish was used . For measuring the first light stripe width , the light stripe was defined as an area taken by dense iridophores . The width of the stripe was measured along five lines , perpendicular to the lateral line and drawn from the bases of each second fin ray in the anal fin starting with the first . The body height was measured along the first line . The knockdown was performed as described before ( Nüsslein-Volhard and Dahm , 2002 ) using 3 ng of tjp1b-MO ( CGAGTATGTGATCAGTCTTACTGCA ) , obtained from Gene Tools , LLC , Philomath , Oregon . The PDZ domains of ZO-1 were amplified by RT-PCR from wild-type zebrafish RNA with the following primer pairs: T878: 5′-CATATGGTGACTCTTCACAGGGCACC-3′ T879: 5′-GGATCCTTCCGCTTCCTGCGGATAG-3′ T880: 5′-CATATGGTCACACTCGTCAAGTCCCGC-3′ T881: 5′-GGATCCTTCATCTCTCTGCACCACCAT-3′ T882: 5′-CATATGAAGTTTAAGAAAGGGGAAAGTG-3′ T883: 5′-GGATCCTTTCTTCTTCTGCGCAAGGATGG-3′ and cloned in the vector pGBKT7 ( Takara , Japan ) via NdeI and BamHI . Similarly , the C-termini of Cx39 . 4 , Cx41 . 8 , and Cx43 were amplified by RT-PCR with the following primer pairs: T886: 5′-CATATGCTTCAGTTGGTGATAAC-3′ T887: 5′-GGATCCTCAAACATAATGTCTCGGTTTG-3′ T884: 5′-CATATGGCATGGAAGCAGTTGAGG-3′ T885: 5′-GGATCCTATACCGCAAGGTCGTCCGG-3′ T888: 5′-CATATGCTCTTCAAACGAATCAAGGACC-3′ T889: 5′-GGATCCTAGACGTCCAGGTCATCAGG-3′ and cloned into the vector pGADT7 ( Clontech ) via NdeI and BamHI . The plasmids were transformed into the yeast strain Y2HGold ( Clontech ) by standard procedures , and we screened for positive interactions using X-α-Gal and His as markers . | The striking horizontal striped pattern of the zebrafish makes it a decorative addition to many home aquariums . The stripes are a result of three different pigment cells interacting with each other , and first begin to emerge when the animal is two to three weeks old . At that time , iridescent cells called iridophores begin to multiply and spread in the skin . In the light-coloured stripes , the iridophores are compact and ‘dense’; in the dark stripes the cells change into a ‘loose’ shape and organisation . Black-pigmented cells fill in the dark stripes , and a third cell type with a yellow hue condenses over the light stripes . How the three types of cell work together to make the striped pattern is not fully understood . Fadeev et al . examined a zebrafish variant with a genetic mutation that disrupts the function of a protein called Tight Junction Protein 1a ( or Tjp1a ) —a fish variant of a mammalian protein called ZO-1 . This protein helps cells to interact with each other . The mutant fish appear spotted rather than striped , because light regions containing sheets of the dense iridophores interrupt the dark stripes . Experiments using fluorescent markers showed that Tjp1a is produced in much lower amounts in the loose iridophores in the dark stripes than in the dense iridophores of the light stripes . This led Fadeev et al . to suggest that the transition from the dense to the loose shape is dependent on the presence of Tjp1a in the cell . Tjp1a is likely to regulate how colour patterns form by controlling how iridophores interact with other types of pigment cell . The Tjp1a mutant fish provides the first glimpse into the machinery inside cells that underlies colour pattern formation , and will help to identify other components and cues responsible for cell interactions . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"developmental",
"biology",
"cell",
"biology"
] | 2015 | Tight Junction Protein 1a regulates pigment cell organisation during zebrafish colour patterning |
A common strategy by which bacterial pathogens reside in humans is by shifting from a virulent lifestyle , ( systemic infection ) , to a dormant carrier state . Two major serovars of Salmonella enterica , Typhi and Typhimurium , have evolved a two-component regulatory system to exist inside Salmonella-containing vacuoles in the macrophage , as well as to persist as asymptomatic biofilms in the gallbladder . Here we present evidence that SsrB , a transcriptional regulator encoded on the SPI-2 pathogenicity-island , determines the switch between these two lifestyles by controlling ancestral and horizontally-acquired genes . In the acidic macrophage vacuole , the kinase SsrA phosphorylates SsrB , and SsrB~P relieves silencing of virulence genes and activates their transcription . In the absence of SsrA , unphosphorylated SsrB directs transcription of factors required for biofilm formation specifically by activating csgD ( agfD ) , the master biofilm regulator by disrupting the silenced , H-NS-bound promoter . Anti-silencing mechanisms thus control the switch between opposing lifestyles .
Salmonella enterica serovar Typhimurium is a rod-shaped enteric bacterium which easily infects diverse hosts such as humans , cattle , poultry and reptiles through contaminated food or water , causing gastroenteritis . A human-restricted serovar of Salmonella enterica , serovar Typhi , causes typhoid fever and continues to be a dangerous pathogen throughout the world . Salmonella lives as a facultative pathogen in various natural and artificial environments as independent planktonic cells , cooperative swarms ( Harshey and Matsuyama , 1994 ) or as multi-cellular communities called biofilms ( see Steenackers et al . , 2012 for a review ) . Upon successful invasion of host cells , Salmonella is phagocytosed by macrophages , where it resides in a modified vacuole in a self-nourishing niche called a Salmonella-Containing Vacuole ( SCV ) . This intracellular lifestyle eventually adversely affects the host . Salmonella also resides as multi-cellular communities on intestinal epithelial cells ( Boddicker et al . , 2002 ) , gallstones ( Prouty et al . , 2002 ) and tumors ( Crull et al . , 2011 ) . It is believed that biofilms in the gall bladder are important for maintaining the carrier state , allowing Salmonella to persist ( Crawford et al . , 2010 ) . Each of these lifestyles of Salmonella are regulated by two-component regulatory systems ( TCRS ) . TCRSs are comprised of a membrane-bound sensor histidine kinase and a cytoplasmic response regulator . The virulence genes of Salmonella are encoded on horizontally acquired AT-rich segments of the genome called Salmonella Pathogenecity Islands ( SPIs ) , and are also tightly regulated by TCRSs . For example , the SsrA/B TCRS is essential for the activation of the SPI-2 regulon genes encoding a type-three secretory needle and effectors that are involved in formation of the SCV ( Cirillo et al . , 1998 ) . Interestingly , the SsrA/B system itself is regulated by upstream two-component systems such as EnvZ/OmpR and PhoP/Q , which regulate gene expression in response to changes in osmolality , pH and the presence of anti-microbial peptides ( Fields et al . , 1989; Miller et al . , 1989; Lee et al . , 2000; Feng et al . , 2003 ) . The ssrA and ssrB genes are present on the SPI-2 pathogenecity island adjacent to each other and are regulated by a set of divergent promoters ( Feng et al . , 2003; Ochman et al . , 1996 ) . Under acidic pH and low osmolality , the ssrA and ssrB genes are transcriptionally activated by the binding of OmpR~P and PhoP~P to their promoters ( Feng et al . , 2003; Bijlsma and Groisman , 2005; Walthers and Kenney unpublished ) whose levels are in turn regulated by the respective sensor kinases , EnvZ and PhoQ . SsrA is a tripartite membrane-bound histidine sensor kinase that undergoes a series of intra-molecular phosphorylation reactions before it transfers the phosphoryl group to the N-terminal aspartate residue of the response regulator , SsrB . SsrB belongs to the NarL/FixJ family of transcriptional regulators that require phosphorylation-dependent dimerization to bind DNA . The X-ray crystal structure of NarL revealed that the C-terminal DNA binding domain was occluded by the N-terminus ( Baikalov et al . , 1996 ) , and phosphorylation was predicted to relieve this inhibition . Full-length SsrB is unstable in solution , but an isolated C-terminal domain of SsrB , SsrBc , is capable of binding to the regulatory regions of nine genes belonging to the SPI-2 regulon , including ssrA and ssrB ( Feng et al . , 2004; Walthers et al . , 2007 ) and activating transcription . A role for SsrB~P was identified by its dual function as a direct transcriptional activator and as an anti-silencer of H-NS-mediated repression ( Walthers et al . , 2007 ) . The Histone like Nucleoid Structuring protein H-NS is involved in silencing many of the SPI-2 regulon genes in accordance with its role in binding to xenogenic AT-rich sequences and repressing their expression ( Walthers et al . , 2007; Navarre et al . , 2006 ) . H-NS binding to DNA leads to the formation of a stiff nucleoprotein filament which is essential in gene silencing ( Lim et al . , 2012; Liu et al . , 2010; Amit et al . , 2003; Winardhi et al . , 2015 ) . Moreover , relief of repression occurs due to the binding of SsrBc to this rigid H-NS-DNA complex ( Walthers et al . , 2011 ) . Salmonella reservoirs in host and non-host environments produce a three-dimensional extracellular matrix which consists of curli fimbriae , cellulose , proteins and extracellular DNA , to encase clusters of bacteria and form a mature biofilm . CsgD ( AgfD ) is the master regulator of biofilm formation ( Gerstel et al . , 2003 ) ; it is a LuxR family transcriptional activator that activates the expression of curli fimbriae encoded by csgDEFG/csgBAC operons ( Collinson et al . , 1996; Romling et al . , 1998 ) . CsgD also activates expression of adrA , increasing intracellular c-di-GMP levels , and activating the cellulose biosynthetic operon bcsABZC ( Zogaj et al . , 2001 ) . Two other biofilm matrix components are also positively regulated by CsgD: BapA and the O-antigen capsule ( Latasa et al . , 2005; Gibson et al . , 2006 ) . Transcriptional profiling of biofilms formed by S . Typhimurium SL1344 showed that many SPI-2 genes were down-regulated , yet SsrA was required for biofilms ( Hamilton et al . , 2009 ) . This apparent paradox drove us to explore the underlying mechanism of biofilm formation . The role of SsrA/B in this process was of particular interest , since our previous comparison of SsrA and SsrB levels at neutral and acidic pH had shown that the expression of ssrA and ssrB was uncoupled ( Feng et al . , 2004 ) . We examined the ability of the wild type S . Typhimurium strain 14028s to form biofilms in the absence of ssrA and ssrB and found it to be dependent only on the expression of ssrB . We further showed that H-NS was a negative regulator of csgD . Surprisingly , the SsrB response regulator positively regulated the formation of biofilms by activating csgD expression in the absence of any phospho-donors . Moreover , AFM imaging revealed that unphosphorylated SsrB was able to bind to the csgD regulatory region and binding was sufficient to relieve H-NS-mediated repression and favor formation of S . Typhimurium biofilms . As a result of these studies , we propose that SsrB , a pathogenicity island-2-encoded response regulator , sits at a pivotal position in governing Salmonella lifestyle fate: to either exist inside the host ( in the SCV ) as a promoter of virulence; or as a surface-attached multicellular biofilm , maintaining the carrier state . This switch is achieved merely by the ability of unphosphorylated SsrB to function as an anti-repressor of H-NS and the additional role of SsrB~P in activating SPI-2 transcription ( Walthers et al . , 2011 ) .
The SsrA/B TCRS is activated by environmental stimuli such as pH and osmolality , stimulating the expression of virulence genes essential for intra-cellular growth and survival of Salmonella ( Feng et al . , 2003; 2004 ) . Thus , it was surprising that SsrA/B and SPI-2 were implicated in the multi-cellular lifestyle of Salmonella ( Hamilton et al . , 2009 ) . Furthermore , the SsrA kinase was required , but SPI-2 genes were down-regulated . At SPI-2 , SsrB~P de-represses H-NS and also activates transcription ( Walthers et al . , 2007; 2011 ) . We therefore wanted to explore this seeming paradox during biofilm regulation . In order to quantify the defect in biofilm-forming ability resulting from the absence of ssrB , the wild type strain 14028s , an ssrA null strain and an ssrB null strain were grown in 96-well polystyrene microtiter plates for 2 days . The surface-attached cells in 14028s , ssrA and ssrB strains were stained by crystal-violet and quantified . As shown in the figure , biofilm levels in the ssrB mutant strain decreased by around 60% compared to the wild type or an ssrA null strain ( Figure 1A ) , and it could be complemented by over-expressing SsrBc in trans . Thus , we establish a new role for SsrB , but not the SsrA kinase in positively regulating biofilm formation . The ssrB strain was not compromised for planktonic growth , as measured by total viable counts of the wild type , ssrA and ssrB cultures ( Figure 1—figure supplement 1 and Supplementary methods ) . 10 . 7554/eLife . 10747 . 003Figure 1 . Loss of ssrB but not ssrA decreases Salmonella Typhimuirum biofilms . ( A ) The defect in formation of biofilms in the ssrB null was complemented by the overexpression of SsrBc from plasmid pKF104 in trans as measured by crystal violet staining . ( B ) The typical rdar morphotype of the wild type strain was lost in the ssrB strain as shown on congo red plates . ( C ) A two day old macrocolony of the ssrB strain is not fluorescent under UV light with Fluorescent Brightener 28 . ( D ) The wild type strain forms thick solid-surface biofilms , while the ssrB strain remains poor for biofilms as monitored for six days by SYTO-9 staining of flow cell biofilms; scale bar = 1 mm . ( E ) SEM images showing extensive mesh-like network of wild type biofilms and sparse extracellular matrix of the ssrB biofilms; scale bar = 1 µm . ( F ) The amount of biofilms formed by the wild type strain ( solid black bars ) increases after 24 hr but the ssrB null ( hatched black bars ) remains defective up to 84 hr . n = 2 , Mean ± SD , p < 0 . 0001 between wild type and ssrB strains from 36 hr till 84 hr . ( G ) The amount of cholesterol-attached biofilms formed by the ssrB strain were significantly less than that produced by the wild type . n = 3 , Mean ± SD , p < 0 . 0001 . Source data file: Figure 1—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 10747 . 00310 . 7554/eLife . 10747 . 004Figure 1—source data 1 . Source data for crystal violet staining in Figure 1A , F and G . DOI: http://dx . doi . org/10 . 7554/eLife . 10747 . 00410 . 7554/eLife . 10747 . 005Figure 1—figure supplement 1 . The ssrB mutant is not defective in growth compared to the wild type strain . Number of colonies formed by the wild type , ssrA , ssrB and D56A strains were the same order of magnitude across all the time points tested . Source data file: Figure 1—figure supplement 1—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 10747 . 00510 . 7554/eLife . 10747 . 006Figure 1—figure supplement 1—source data 1 . Growth curves of wild type , ssrA , ssrB and D56A strains . DOI: http://dx . doi . org/10 . 7554/eLife . 10747 . 00610 . 7554/eLife . 10747 . 007Figure 1—figure supplement 2 . The planktonic sub-population of the ssrB strain was higher by two orders of magnitude compared to the wild type , ssrA and D56A strains at 2 days . n = 2 , Mean ± SD , p < 0 . 05 . Source data file: Figure 1—figure supplement 2—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 10747 . 00710 . 7554/eLife . 10747 . 008Figure 1—figure supplement 2—source data 1 . Number of cells in the planktonic sub-population of each strain . DOI: http://dx . doi . org/10 . 7554/eLife . 10747 . 00810 . 7554/eLife . 10747 . 009Figure 1—figure supplement 3 . Total wet weight of the adherent sub-population was decreased by at least 50% in the ssrB strain compared to the wild type , ssrA and D56A strains at 2 days . n = 2 , Mean ± SD , p < 0 . 05 for the ssrB strain versus ssrA/D56A strains and p = 0 . 08 for the ssrB strain versus wild type . Source data file: Figure 1—figure supplement 3—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 10747 . 00910 . 7554/eLife . 10747 . 010Figure 1—figure supplement 3—source data 1 . Total wet weight of the adherent sub-population of each strain . DOI: http://dx . doi . org/10 . 7554/eLife . 10747 . 010 The ability to form biofilms was analysed for three strains , wild type 14028s , an ssrA null strain and an ssrB null strain , by observing appearance of the rough dry and red ( rdar ) morphotype on Congo Red plates . After seven days , each of the 14028s and ssrA mutant plates showed the appearance of rough , dry and red macrocolonies . However , the ssrB null strain exhibited a smooth , wet and pale brown morphotype which showed a poor ability to exhibit a multicellular phenotype ( Figure 1B ) . The absence of Congo Red staining in the ssrB strain also indicated that there were greatly reduced levels of extracellular curli fibers in the macrocolony . These fibers form one of the key components of the extra-cellular matrix of a Salmonella biofilm . To examine the levels of cellulose , the other main component of the extra-cellular matrix , the three strains were grown on Calcofluor plates . The macrocolonies formed by 14028s and ssrA appeared white under UV , as cellulose binds the Calcofluor fluorescent dye ( Figure 1C ) . The dull and non-white macrocolony produced by the ssrB null strain was starkly different , indicating again a poor extracellular matrix . We visualized the nature of biofilms formed by the wild type and ssrB strains by Confocal Fluorescence Imaging of solid-surface biofilms formed under a continuous flow system in a micro-fluidic chamber . Wild type and the ssrB null strain were grown in minimal M9 medium and inoculated in their respective flow cells . After allowing the cells to attach for 1h , flow commenced and the growing biofilms were monitored daily for six days . Each day , the flow cell was removed , and the glass-attached cells were stained with SYTO-9 green , a fluorescent dye that stains nucleic acids . The cells were then imaged using fluorescence microscopy ( Figure 1D ) . The figure clearly shows that for the wild type strain , the attached biomass not only increased progressively in size , but also in its complexity of design . This correlated with the appearance of typical mushroom-like constituent microcolonies by around 2 days . Such type of mature biofilms were never observed in the ssrB null strain and even after 6 days it failed to form biofilms , as evident by fewer and thinly attached clusters of SYTO-9 green-labeled cells . To compare the differences in the structural properties of biofilms formed by the ssrB strain with the wild type and the ssrA mutant , cells were grown on APTES-coated cover slips in 16-well polystyrene microtiter plates for two days . The surface-attached communities were fixed , dehydrated and dried for Scanning Electron Microscopy ( SEM ) . As expected , the wild type and ssrA mutant showed the presence of tightly knit communities of cells containing an extensive network of curli fibres surrounding large groups of cells ( Figure 1E ) . In contrast , the ssrB null strain formed sparsely spaced groups of few cells with a large reduction in the surrounding matrix and poor network properties . It was possible that the ssrB strain was just slower in biofilm formation and needed more time to exhibit its biofilm capability or that it caused a growth defect . We ruled out this possibility by measuring growth curves ( Figure 1—figure supplement 1 ) and by monitoring the wild type and ssrB strains by crystal-violet staining every 12 hr for around 4 days . This determined that the ssrB mutant was not deficient for growth and remained incapable of forming biofilms up to 84 h , while the wild type formed large biofilms by 2 days ( Figure 1F ) . We next investigated the ability of Salmonella to form biofilms , on cholesterol-coated eppendorf tubes . The cholesterol-attached biofilms were estimated by crystal violet staining for the strains 14028s , ssrA null and ssrB null . There was a drastic reduction in the amount of biofilms formed by the ssrB strain on cholesterol-coated eppendorf tubes ( Figure 1G ) . If cholesterol biofilms are indeed an indicator of an ability to form biofilms on gallstones ( Crawford et al . , 2008 ) , then SsrB may be crucial for establishment of the carrier state in Salmonella . Taken together , our results modify the initial hypothesis that SsrA was involved in the formation of biofilms . SsrB , but not its cognate kinase SsrA , is involved in the switch from planktonic growth to a multi-cellular lifestyle , in addition to its role in regulating genes required for pathogenesis . Phospho-relay between the sensor kinase , SsrA , and the response regulator , SsrB , was crucial for activation of SPI-2 virulence genes ( Feng et al . , 2003 ) . SsrA is a tripartite sensor kinase that is presumably autophosphorylated by ATP at His405 , followed by intra-molecular phosphorylation reactions at Asp739 and His867 , based on homology to other tripartite kinases . His867 would then participate in the transfer reaction of the phosphoryl group to Asp56 in the N-terminus of SsrB . To confirm that SsrA-dependent phosphorylation was not required for SsrB-mediated regulation of biofilms ( Figure 1A–G ) , we examined biofilm formation of Salmonella strains possessing H405Q , D739A and H867Q mutations in ssrA ( Walthers and Kenney unpublished ) and compared them to the wild type , ssrA and ssrB null mutants . Figure 3A indicates that the extent of biofilms formed by the three kinase point mutants was comparable to that of the wild type and the ssrA null mutant . A complete loss of ssrB was the only genetic change which adversely affected biofilm formation . In the absence of SsrA kinase activity , small inorganic phosphate donors such as acetyl phosphate can phosphorylate SsrB , albeit with a lower efficiency ( Feng et al . , 2004 ) . We therefore constructed strains that were deficient in both SsrA kinase activity as well as in the production of acetyl phosphate and assayed their biofilm capabilities . Strains H405Q ack pta , D739A ack pta and H867Q ack pta , were all capable of biofilm formation ( Figure 2A ) , demonstrating that none of the known phosphate donors for SsrB were required for regulating biofilms . In contrast , SPI-2 activation required ssrA , as measurement of β-galactosidase activity of a sifA-lacZ transcriptional fusion confirmed that activation of sifA required SsrA/B phosphorylation ( Figure 2C ) . 10 . 7554/eLife . 10747 . 011Figure 2 . Phosphorylation of SsrB is not required for biofilm formation . Amount of biofilms formed as measured by crystal violet staining for ( A ) Strains ssrAH1 , ssrAD , ssrAH2 , ssrAH1AcP , ssrADAcP , ssrAH2AcP and ( B ) D56A SsrB shows similar levels to that of the wild type , and higher than the ssrB mutant . Source data file: Figure 2—source data 1 . ( C ) Beta-galactosidase activity of a sifA-lacZ chromosomal fusion was significantly lower in the ssrB null and the D56A SsrB mutant compared to the wild type . n = 3 , Mean ± SD , p < 0 . 0001 . Source data file: Figure 2—source data 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 10747 . 01110 . 7554/eLife . 10747 . 012Figure 2—source data 1 . Source data for crystal violet staining in Figure 2A and B . DOI: http://dx . doi . org/10 . 7554/eLife . 10747 . 01210 . 7554/eLife . 10747 . 013Figure 2—source data 2 . Source data for the measurement of beta-galactosidase activity in Figure 2C . DOI: http://dx . doi . org/10 . 7554/eLife . 10747 . 013 SsrB is in the NarL subfamily of response regulators . NarL requires phosphorylation to relieve inhibition of C-terminal DNA binding by the N-terminus ( Baikalov et al . , 1996 ) . Therefore , we substituted the conserved Asp56 residue with Ala ( D56A ) , and determined its ability to form biofilms ( Figure 2B ) . The D56A strain was as competent as the wild type , ssrA null and ssrA kinase mutants in biofilm formation . The ability of unphosphorylated SsrB to regulate biofilm formation was unique , as it was unable to activate sifA-lacZ levels . The beta-galactosidase activity was significantly lower in the ssrB null and D56A strains ( 8 . 7% and 14 . 9% , respectively ) compared to the wild type ( Figure 2C ) . The D56A mutant was also able to form rdar macrocolonies on Congo Red plates . In addition , SEM imaging of D56A biofilms revealed that they maintained their intricate architecture and wild type structure ( data not shown ) . When cultures of the wild type , ssrA , ssrB and D56A strains were separated as a non-adherent ( or free-swimming ) sub-population and an adherent sub-population ( or multicellular aggregates ) ( MacKenzie et al . , 2015 ) , we observed that the ssrB strain had a greater number of cells in the non-adherent fraction and fewer in the adherent fraction as compared to the other three strains . This confirms that the absence of SsrB , and not SsrB~P , led to a defect in the ability of Salmonella to form surface-attached communities ( Figure 1—figure supplement 2 and 3 ) . Taken together , these results indicate that SsrB can activate biofilm formation in the absence of any phosphorylation signals . This is the first evidence for a role of unphosphorylated SsrB in gene regulation . These results also strongly advocate for a larger contribution of SsrB in Salmonella pathogenesis owing to its dual regulation of intracellular ( i . e . , the SCV ) as well as mutli-cellular lifestyles ( the carrier state ) . SsrB is encoded on SPI-2 and regulates SPI-2 genes involved in assembly and function of the type three secretion system encoded by SPI-2 , as well as effectors that are encoded on SPI-2 and outside of SPI-2 ( Feng et al . , 2004; Walthers et al . , 2007 ) . To determine whether the SsrB target of biofilm regulation was dependent on the SPI-2 injectisome or any of its secreted SPI-2 effectors , we examined ssaC and ssaJ null strains for their ability to form biofilms ( Figure 3A ) . SsaC is an outer ring component of the injectisome and SsaJ forms the inner ring of the SPI-2 needles . Both of these strains formed biofilms to an extent similar to the wild type and hence , we ruled out any involvement of the SPI-2 secretory apparatus or its related secreted proteins in the SsrB-dependent regulation of Salmonella biofilms . 10 . 7554/eLife . 10747 . 014Figure 3 . SsrB regulates biofilms by a CsgD-dependent mechanism . ( A ) The SPI-2 needle , ssaC and ssaJ mutant strains were not affected in biofilm formation . ( B ) Over-expression of csgD from a plasmid ( pBR328::csgD ) in trans rescued biofilm formation in the ssrB mutant , as measured by crystal violet staining , n = 3 . Source data file: Figure 3—source data 1 . An estimate of csgD expression by ( C ) Real-time qRT-PCR showed a significant decrease in csgD transcription in the ssrB null , but not in the D56A SsrB and ssrA mutants . rrsA transcript levels were used as control; n = 2 , Mean ± SD , p < 0 . 0001 . Source data file: Figure 3—source data 2 and ( D ) Immunoblot analysis showing the absence of CsgD in the ssrB null strain in two day old biofilms , using GroEL as a loading control . DOI: http://dx . doi . org/10 . 7554/eLife . 10747 . 01410 . 7554/eLife . 10747 . 015Figure 3—source data 1 . Source data for crystal violet staining in Figure 3A and B . DOI: http://dx . doi . org/10 . 7554/eLife . 10747 . 01510 . 7554/eLife . 10747 . 016Figure 3—source data 2 . Source data for Real-time qRT-PCR in Figure 3C . DOI: http://dx . doi . org/10 . 7554/eLife . 10747 . 016 Biofilm formation in Escherichia coli and Salmonella Typhimurium is governed by the master regulator , CsgD , that acts as a transcriptional activator of genes involved in curli biogenesis and cellulose synthesis . Some of the environmental conditions that favor the formation of biofilms in Salmonella such as low salt and acidic pH also up-regulate levels of SsrB in the cell ( Feng et al . , 2003 ) . An obvious null hypothesis was therefore to test whether SsrB ( in its unphosphorylated state ) , activated the expression of csgD . We first over-expressed csgD from a plasmid in the ssrB null mutant and determined whether it could rescue the defect in biofilm formation . The presence of csgD in trans restored the biofilm capability to wild type levels as measured by crystal-violet staining of 2 day old biofilms ( Figure 3B ) . We next examined the levels of csgD transcripts in 2 day macrocolonies formed by the wild type , ssrA null , ssrB null and D56A strains by quantitative real time RT-PCR ( qRT-PCR ) . As shown in Figure 3C , there was around a 60-fold decrease in csgD transcripts when ssrB was deleted ( Figure 3C ) . We also corroborated our previous findings ( Figure 1A and 2B ) , as csgD transcripts in the ssrA null and D56A strains were maintained at levels similar to the wild type . Furthermore , we probed the whole-cell lysates obtained from such macrocolonies with a monoclonal anti-CsgD antibody in order to measure the CsgD protein levels by western blotting . CsgD levels were undetected in the ssrB null compared to the wild type , ssrA null or D56A strains ( Figure 3D ) . Hence , unphoshorylated SsrB was able to up-regulate the expression of csgD and positively influence the shift to a sessile lifestyle . Both in E . coli and S . Typhimurium , the expression of csgD is sensitive to various environmental stimuli such as starvation , temperature , pH and osmolality due to the action of upstream global regulators at the intergenic region of the csgDEFG and csgBAC operons . We noted that the SsrA/B system also responded to similar environmental conditions such as acidic pH and low osmolality . Furthermore , H-NS , a known repressor of SPI-2 genes ( Walthers et al . , 2007 ) , was also known to regulate the expression of csgD in E . coli and S . Typhimurium ( Ogasawara et al . , 2010; Gerstel et al . , 2003 ) . In regulating SPI-2 genes , SsrBc and H-NS work in an opposing fashion ( Walthers et al . , 2011 ) . SsrBc antagonises the repressive action of H-NS at regulatory regions upstream of the promoters , while also acting as a direct transcriptional activator . We therefore examined whether or not this paradigm for SsrB-mediated transcriptional activation was applicable to csgD regulation . In order to test this , we completely deleted hns from the ssrB null strain and determined if its deletion rescued biofilm formation . Crystal-violet staining of static biofilms formed by the wild type , ssrB and ssrB hns strains was performed; the amount of biofilms formed by ssrB hns was the same as the wild type . In contrast , the ssrB null strain remained at around 35% of these levels ( Figure 4A ) . Macrocolonies formed by the ssrB hns strain also displayed a rugose morphology , reminiscent of the wild type strain after 2 days ( Figure 4B ) . In addition , SEM images of 2 day-old biofilms formed by the ssrB hns strain showed a typical ‘biofilm’ architecture , i . e . , densely packed communites of cells surrounded by biofilm matrix ( Figure 4C ) . This result indicated that H-NS was functioning to repress csgD ( see Discussion ) . Thus , we strengthened our prediction of a role for SsrB in the activation of csgD expression by acting as an anti-H-NS molecule . 10 . 7554/eLife . 10747 . 017Figure 4 . hns deletion rescues biofilm formation in the ssrB mutant as shown by Crystal violet staining . ( A ) The amount of biofilms formed is higher in the wild type , hns and ssrB hns strains compared to the ssrB null , n = 3 , Mean ± SD , p < 0 . 0001 . Source data file: Figure 4—source data 1 . Macrocolony phenotype ( B ) ssrB hns forms a highly rugose and dry macrocolony , while the ssrB macrocolony was smooth and mucoidy . SEM imaging ( C ) ssrB hns biofilms were covered by a thick extra-cellular matrix; scale bar = 2 µm . ( D ) qRT-PCR: csgD levels were restored in the ssrB hns strain and were higher than the wild type ( p < 0 . 03 ) and the ssrB mutant ( p < 0 . 003 ) against rrsA transcripts as a control . Note that the normalized csgD levels in the ssrB null were 0 . 0009 , too low for the scale . n = 2 , Mean ± SD . Source data file: Source data file: Figure 4—source data 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 10747 . 01710 . 7554/eLife . 10747 . 018Figure 4—source data 1 . Source data for crystal violet staining in Figure 4A . DOI: http://dx . doi . org/10 . 7554/eLife . 10747 . 01810 . 7554/eLife . 10747 . 019Figure 4—source data 2 . Source data for Real-time qRT-PCR in Figure 4D . DOI: http://dx . doi . org/10 . 7554/eLife . 10747 . 019 We next examined the levels of csgD transcripts in mature biofilms formed by the wild type , ssrB null , hns null and ssrB hns null strains by qRT-PCR ( Figure 4D ) . Normalized levels of csgD transcripts were highly up-regulated in the ssrB hns double mutant compared to the ssrB null strain , indicating that the loss of H-NS repression rescued the defect in csgD expression in the absence of SsrB . Expression of csgD was slightly higher than the wild type levels in the hns single mutant and the ssrB hns double mutant , which also correlated with their biofilm capabilities ( Figure 4A ) . Since the levels of csgD transcripts were equal in the hns and ssrB hns null strains , it seemed likely that SsrB influenced csgD expression by silencing H-NS-mediated repression and not by direct activation of transcription , which requires phosphorylation for a productive interaction with RNA polymerase ( Walthers et al . , 2011 ) . Taken together , these results identified a regulatory role for unphosphorylated SsrB by orchestrating anti-silencing at the H-NS-repressed csgD locus . This work clearly shows that , as observed in E . coli , H-NS represses expression of csgD in Salmonella ( see Discussion ) . SsrB is a NarL family member and X-ray crystallography suggested that phosphorylation was required for DNA binding ( Baikalov et al . , 1996 ) . Thus , it was of interest to determine how SsrB relieved H-NS silencing at csgD . We used atomic force microscopy ( AFM ) to visualize the csgDEFG-csgBAC intergenic region and examined the effect of addition of full-length SsrB , D56A SsrB and SsrBc ( DNA binding domain alone ) . Surprisingly , we observed binding of SsrB , D56A SsrB and SsrBc to distinct regions of the csgD regulatory region ( Figure 5B and Figure 5—figure supplement 1A and B ) . Binding occured at low protein concentrations , suggesting high affinity interactions and showed that unphosphorylated SsrB was capable of binding csgD . An SsrB mutant , K179A , that was incapable of binding DNA ( Carroll et al . , 2009 ) did not bind csgD even at 300 nM , indicting that binding was specific ( Figure 5E and F ) . Closer examination revealed a sharp curvature at the regions where SsrB was bound to csgD ( Figure 5B , arrows ) . This result indicates that , like its NarL homologue ( Maris et al . , 2002 ) , SsrB bends DNA upon binding , as predicted ( Carroll et al . , 2009 ) . We estimate that on average , binding of SsrB to the csgDEFG-csgBAC intergenic sequence occurred with a bending angle of around 82° ( Figure 5D and Figure 5—figure supplement 2 for naked DNA ) . The observation that SsrB bending was more severe than NarL ( 82° compared to the 42° of NarLc at its cognate palindrome ) ( Maris et al . , 2002 ) , likely contributes to its ability to promote anti-silencing . 10 . 7554/eLife . 10747 . 020Figure 5 . SsrB binds upstream of csgD . ( A ) AFM images of the 755 bp csgD regulatory region ( csgD755 ) . ( B ) At 120 nM SsrB , distinct areas of SsrB binding were visualized as sharp bends ( yellow arrows ) . ( C ) At 300 nM SsrB , areas of condensation ( pink arrows ) were observed . ( D ) Binding of SsrB bends the DNA by an average angle of 82º ( for the naked DNA angle , refer to Figure 5—figure supplement 2 and for analysis refer to Supplementary method ) , Scale bar = 200 nm as in ( A ) . ( E ) and ( F ) The SsrB mutant , K179A SsrB , which is defective in DNA binding , was unable to bind csgD755 both at 30 nM or 300 nM; Scale bar = 200 nm . DOI: http://dx . doi . org/10 . 7554/eLife . 10747 . 02010 . 7554/eLife . 10747 . 021Figure 5—figure supplement 1 . AFM images of the 755 bp csgD regulatory region ( csgD755 ) ( A ) with 120 nM D56ASsrB and ( B ) SsrBc . Distinct areas of binding were visualized as sharp bends ( yellow arrows ) and as areas of condensation ( pink arrows ) at 300 nM D56A SsrB ( C ) and SsrBc ( D ) ; Scale bar = 200 nm as in Figure 5A . DOI: http://dx . doi . org/10 . 7554/eLife . 10747 . 02110 . 7554/eLife . 10747 . 022Figure 5—figure supplement 2 . Bending angle of the naked csgD755 fragment . DOI: http://dx . doi . org/10 . 7554/eLife . 10747 . 02210 . 7554/eLife . 10747 . 023Figure 5—figure supplement 3 . Electrophoretic mobility shift assay with the 122 bp csgD regulatory region , csgD122 , showing a DNA-protein complex in the presence of SsrB . The K179A SsrB mutant did not bind to csgD122 . Addition of competitor unlabelled csgD122 fragment decreased the SsrB-DNA complex as apparent by an increase in free , labelled csgD122 . DOI: http://dx . doi . org/10 . 7554/eLife . 10747 . 023 When the SsrB concentration was increased to 300 nM , we detected large-scale condensation of the DNA-protein complexes by AFM ( Figure 5C ) . Condensation at the csgD regulatory region was similar when incubated with 300 nM D56A SsrB or SsrBc , indicating that protein binding led to DNA structural changes , irrespective of whether SsrB was phosphorylated ( Figure 5—figure supplement 1C and D ) . Biochemical analysis of the protein-DNA complexes by an electrophoretic mobility shift assay using a shorter fagment of the csgD regulatory region ( Figure 6—figure supplement 4 ) , also indicated the presence of an SsrB-DNA complex . This complex was dissociated when an unlablelled csgD fragment was added as a competitor . In contrast , K179A SsrB failed to form a complex with DNA ( Figure 5—figure supplement 3 ) . The DNA bending ability of SsrB is important for SsrB relief of H-NS-mediated transcriptional silencing at csgD ( see below and [Winardhi et al . , 2015] ) . The DNA binding behavior of SsrB was recapitulated in solution AFM imaging experiments , indicating that binding and condensation were not an artifact of drying the samples ( Figure 6—figure supplement 1C ) . Our previous work established that H-NS silenced genes by forming a rigid filament on DNA ( Liu et al . , 2010; Walthers et al . , 2011; Lim et al . , 2012 ) . We purified H-NS , incubated it with the csgD regulatory region , and immobilised it on a glass coverslip . Subsequent AFM imaging indicated the presence of a straight and rigid nucleoprotein filament ( Figure 6A ( i ) ) , which was distinct from the random conformation adopted by naked DNA ( Figure 5A ) . Thus , we reaffirmed our earlier observations that H-NS repressed expression of csgD by filament formation , leading to transcriptional silencing ( Figure 4D ) . When we pre-formed stiffened filaments by addition of H-NS and then added SsrB , evidence of SsrB condensation was immediately apparent ( Figure 6A ( ii ) ) . For example , areas of SsrB binding to DNA led to the formation of condensed nucleoprotein complexes and abolished the prior structural rigidity introduced by H-NS binding ( see arrows , Figure 6A ( ii ) ) . Interestingly , Figure 6A ( ii ) also indicates that H-NS continued to form straight nucleoprotein complexes in regions devoid of SsrB ( yellow line ) , i . e . H-NS was still bound to some regions of DNA when SsrB was also bound . Similar binding behavior with D56A SsrB and SsrBc at the H-NS-bound csgD regulatory region was also observed ( Figure 6—figure supplement 2A and B ) , and similar binding patterns were observed in solution AFM ( Figure 6—figure supplement 1B–D ) . This combined nucleoprotein complex was also detected as a supershift by electrophoretic mobility shift assay using a shorter fragment of the csgD regulatory region and specific concenterations of H-NS , SsrB and anti-SsrBc serum ( Figure 6—figure supplement 3 ) . This csgD regulatory element harbored the H-NS binding region ( Gerstel et al . , 2003 ) as well as an SsrB binding motif ( Feng et al . , 2004; refer Figure 6—figure supplement 4 ) . Thus , unphosphorylated SsrB can bind to the csgD regulatory region when it has been coated with the repressor H-NS . Anti-silencing results from SsrB-induced local topological changes in the DNA , in part as a result of its bending ability ( Figure 6B ) . This likely provides enough free DNA to enable access to the promoter for RNA polymerase to activate transcription . 10 . 7554/eLife . 10747 . 024Figure 6 . SsrB condenses H-NS bound csgD DNA . ( A ) ( i ) AFM imaging in the presence of 600 nM H-NS shows a straight and rigid filament on csgD755 . ( ii ) Addition of 600 nM SsrB to the H-NS bound csgD DNA resulted in areas of condensation ( pink arrows; an ‘SsrB signature’ ) along with a few areas where the straight H-NS bound conformation persisted ( yellow line; an ‘H-NS signature’ ) ; Scale bar = 200 nm as in Figure 5A . ( B ) A model for the mechanism of anti-silencing by SsrB at csgD wherein SsrB likely displaces H-NS from the ends of a stiffened nucleoprotein filament and relieves the blockade on the promoter for RNA polymerase to activate transcription . For details refer to ( Winardhi et al . , 2015 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 10747 . 02410 . 7554/eLife . 10747 . 025Figure 6—figure supplement 1 . Liquid AFM imaging of ( A ) the 755 bp csgD regulatory region . ( B ) In the presence of 600 nM H-NS , a straight and rigid filament was observed ( yellow line ) . ( C ) In the presence of 300 nM SsrB , areas of condensation were evident ( pink arrow ) . ( D ) Addition of 600 nM SsrB to the H-NS bound csgD DNA resulted in areas of condensation ( pink arrows; an ‘SsrB signature’ ) along with a few areas where the straight H-NS bound conformation persisted ( yellow line; an ‘H-NS signature’ ) ; Scale bar = 200 nm . DOI: http://dx . doi . org/10 . 7554/eLife . 10747 . 02510 . 7554/eLife . 10747 . 026Figure 6—figure supplement 2 . SsrB D56A and SsrBc condense H-NS-bound csgD DNA . AFM imaging in the presence of H-NS shows areas of condensation upon addition of ( A ) 600 nM D56ASsrB and ( B ) 600 nM SsrBc; Scale bar = 200 nm as in Figure 5A . DOI: http://dx . doi . org/10 . 7554/eLife . 10747 . 02610 . 7554/eLife . 10747 . 027Figure 6—figure supplement 3 . SsrB and H-NS form a complex on csgD . Electrophoretic mobility shift assay with the 122 bp csgD regulatory region , csgD122 ( left to right ) ; in the presence of SsrB and H-NS , the DNA-protein complex ( * ) is super-shifted in the presence of anti-SsrBc serum ( ** ) . A DNA-protein complex is also observed when SsrB and H-NS were present alone . Note the absence of any complex in a control reaction with csgD122 and anti-SsrBc , while anti-SsrBc recognizes the SsrB-csgD122 complex . DOI: http://dx . doi . org/10 . 7554/eLife . 10747 . 02710 . 7554/eLife . 10747 . 028Figure 6—figure supplement 4 . The sequence of the 755 bp csgD regulatory region indicating the H-NS binding region according to Gerstel et al . ( 2003 ) ; and the SsrB binding motif as found by Feng et al . ( 2004 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 10747 . 028
When the SsrA kinase is present and activated by acid stress , SsrB is phosphorylated and SsrB~P de-represses H-NS and activates transcription at SPI-2 and SPI-2 co-regulated genes , including: sifA ( Walthers et al . , 2011 ) , ssaB , ssaM , sseA and ssaG ( Walthers et al . , 2007 ) . In the absence of the SsrA kinase , SsrB is not phosphorylated , but it can counter H-NS silencing at csgD ( Figure 4A–D and Figure 6A ) . SsrB binding and bending at the csgD promoter causes a sufficient change in the DNA secondary structure ( Figure 5B , C ) that likely enables access for RNA polymerase , stimulating csgD transcription . It is interesting to note that SsrB is located on the SPI-2 pathogenicity island , and thus was acquired as Salmonella enterica diverged from Salmonella bongori . However , the capability to form biofilms is an ancestral trait , as phylogeny studies have shown that most of the natural or clinical isolates of Salmonella belonging to all the five sub-groups form rdar colonies ( White and Surette , 2006 ) . The SsrB response regulator can control two distinct lifestyle choices: the ability to assemble a type three secretory system and survive in the macrophage vacuole or the ability to form biofilms on gallstones in the gall bladder to establish the carrier state . What then controls the presence or activation of the kinase SsrA ? Our early experiments indicated that SsrA and SsrB were uncoupled from one another ( i . e . , SsrB was present in the absence of SsrA ) and ssrA transcription was completely dependent on OmpR ( Feng et al . , 2004 ) . The EnvZ/OmpR system is stimulated by a decrease in cytoplasmic pH when Salmonella enters the macrophage vacuole ( Chakraborty et al . , 2015 ) . This may also be the stimulus for activating SsrA , since the Salmonella cytoplasm acidifies to pH 5 . 6 during infection and the cytoplasmic domain of EnvZ ( EnvZc ) was sufficient for signal transduction ( Wang et al . , 2012; Chakraborty et al . , 2015 ) . Previous reports also identified a role for PhoP in ssrA translation ( Bijlsma and Groisman , 2005 ) , which would further add to fluctuating SsrA levels . The present work describes a novel role for the unphosphorylated response regulator SsrB in de-repressing H-NS ( Figure 6B ) . We show that under biofilm-inducing conditions , unphosphorylated SsrB is sufficient to activate the expression of csgD . There are only a few such examples of unphosphorylated response regulators playing a role in transcription such as DegU ( Dahl et al . , 1992 ) in Bacillus subtilis and RcsB ( Latasa et al . , 2012 ) in S . Typhimurium . In recent years , it has become apparent that H-NS silences pathogenicity island genes in Salmonella ( Lucchini et al . , 2006; Navarre et al . , 2006; Walthers et al . , 2007; 2011 ) . Understanding how H-NS silences genes and how this silencing is relieved is an active area of research ( Will et al . , 2015; Winardhi et al . , 2015 ) . Because the anti-silencing style of gene regulation is indirect and does not rely on specific DNA interactions , searching for SsrB binding sites has not been informative in uncovering this type of regulation ( Tomljenovic-Berube et al . , 2010; Worley et al . , 2000; Shea et al . , 1996 ) . Even a recent report in which the proteomes of wild type , hilA null ( a transcriptional regulator of SPI-1 genes ) and ssrB null were analyzed by SILAC and compared with an existing CHIP dataset failed to identify csgD as an SsrB-regulated locus ( Brown et al . , 2014 ) , as sequence gazing alone does not help in identifying mechanisms of transcriptional regulation . SsrB is well suited to this style of regulation , because it does not recognize a well-defined binding site ( Feng et al . , 2004; Walthers et al . , 2007; Tomljenovic-Berube et al . , 2010 ) , it has a high non-specific binding component ( Carroll et al . , 2009 ) and it bends DNA upon binding ( Carroll et al . , 2009; Figure 6B , this work ) . Furthermore , previous microarray studies disrupted both ssrA and ssrB , which would not uncover a distinct role for SsrB in gene regulation under non SPI-2-inducing conditions in the absence of the SsrA kinase . It is worth mentioning here that in our AFM images , it was apparent that H-NS was still bound to some regions of the csgD promoter when SsrB condensed the DNA ( Figure 6A ( ii ) ) . Thus , H-NS does not have to be completely stripped off the DNA for de-repression to occur , a finding that was also evident in our previous studies ( Liu et al . , 2010 ) and others ( Will et al . , 2014 ) . SsrB binds and bends DNA , resulting in highly curved DNA conformations . This DNA binding property of SsrB is distinct from H-NS , which forms rigid nucleoprotein filaments and thus straight DNA conformations ( Figure 6A ( i ) ) . Bent DNA is therefore an energetically unfavorable substrate for H-NS binding , and a likely mechanism of SsrB-mediated anti-silencing of H-NS repressed genes . SsrB-dependent displacement of H-NS is more energetically favored to occur predominantly at the ends of H-NS-bound filaments , which requires disruption of fewer H-NS protein-protein interactions ( Winardhi et al . , 2015 and Figure 6B ) . In an equal mixture of H-NS and SsrB ( Figure 6A ( ii ) ) , we do not see evidence of sharply bent filaments . This is expected because H-NS dissociation is likely restricted to the filament ends . Such events occur due to the cooperative nature of H-NS binding that results in a chain of linked H-NS proteins . Hence , H-NS displacement by SsrB likely occurs progressively from the filament end . This behavior has been observed in our single-molecule stretching experiments with H-NS filaments in the presence of SsrB . This ability of H-NS to re-orient on the DNA without being released would also promote its re-binding and silencing when SsrB or other anti-silencers are released ( Figure 6B ) . Response regulators are grouped into subfamilies on the basis of the structures of their DNA binding domains . SsrB is in the NarL/FixJ subfamily , which possess a helix-turn-helix ( HTH ) motif in the C-terminus ( Baikalov et al . , 1996 ) . NarL was the first full-length structure of a response regulator and it showed that the N-terminal phosphorylation domain physically blocked the recognition helix in the HTH motif ( Maris et al . , 2002 ) . Thus , phosphorylation is required to relieve the inhibition of the N-terminus . In the results presented herein , it is apparent that SsrB has adapted to relieving H-NS-silencing and that phosphorylation is not required for this behavior , nor is it required for DNA binding ( Figure 5B ) . In summary , we showed that the response regulator SsrB is required for biofilm formation because it can de-repress H-NS at the csgD promoter ( Figure 6B ) . This leads to the production of CsgD , the master regulator of biofilms . It is noteworthy that a laterally acquired gene product , SsrB , has evolved the job of regulating the levels of csgD , a transcriptional regulator encoded by the core genome . For this activity , phosphorylation of SsrB was not required , which is rare amongst response regulators . Furthermore , we identify H-NS as a repressor of csgD in Salmonella , instead of an activator ( Gerstel et al . , 2003 ) . This unifies the regulation of CsgD by H-NS in E . coli ( Ogasawara et al . , 2010 ) and Salmonella . This work places SsrB at a unique decision point in the choice between lifestyles by Salmonella and makes it crucial for the entire gamut of pathogenesis , i . e . , biofilms and virulence .
The bacterial strains and plasmids used in this study are listed in Table 1 . Salmonella enterica serovar Typhimurium strains were grown in LB medium with shaking at 37°C in the presence of 100 μg/ml ampicillin , 12 . 5 μg/ml Tetracycline ( Tet ) or 50 μg/ml Kanamycin ( Km ) when necessary . For observing the rdar morphotype , plates of LB medium ( without salt ) containing 1% Tryptone and 0 . 5% Yeast Extract supplemented with congo red ( 40 μg/ml ) ( Sigma-Aldrich , Singapore ) were prepared and kept at 30°C after inoculation . To detect cellulose production , LB without salt medium plates were supplemented with Fluorescent Brightener 28 ( 200 μg/ml ) ( Sigma-Aldrich ) , stored under darkness and observed under UV light after incubating at 30°C . For the SsrBc complementation test using the plasmid pKF104 , 0 . 2% arabinose was added to the medium . 10 . 7554/eLife . 10747 . 029Table 1 . List of Bacterial strains and plasmids . DOI: http://dx . doi . org/10 . 7554/eLife . 10747 . 029StrainDescription/NomenclatureReferencewild typeSalmonella enterica serovar Typhimurium strain 14028sLab strain collectionssrAssrA::TetRA derivative of 14028sThis workssrBDW85Don Walthers ( originally from Stephen Libby ) ssaCssaC::TetRA derivative of 14028sChakraborty et al . ( 2015 ) ssaJssaJ::TetRA derivative of 14028sHideaki Mizusaki unpublishedD56AD56A SsrB derivative of 14028sThis workssrAH1DW748Don Walthers unpublishedssrADDW749Don Walthers unpublishedssrAH2DW750This workssrAH1 sifA-LacZMade by transducing sifA-lacZ from DW636 to DW748This workssrAD sifA-lacZMade by transducing sifA-lacZ from DW636 to DW749This workssrAH2 sifA-lacZMade by transducing sifA-lacZ from DW636 to DW750Don Walthers; Lab strain collectionDW636sifA-lacZ at attP site in 14028sDon Walthers; Lab strain collectionDW637ssrB::Km derivative of DW636This work/Don Walthers ( lab strain collection ) ssrAH1 AcPackA-pta::Km ( from DW142 ) transduced in DW748This work/Don Walthers ( lab strain collection ) ssrAD AcPackA-pta::Km ( from DW142 ) transduced in DW749This work/Don Walthers ( lab strain collection ) ssrAH2 AcPackA-pta::Km ( from DW142 ) transduced in DW750This work/Don Walthers ( lab strain collection ) hnshns::TetRA derivative of DW636This work/Don Walthers ( lab strain collection ) hns ssrBhns::TetRA derivative of DW637This work/Don Walthers ( lab strain collection ) Plasmid pKF46D56A His-SsrB pMpM-A5Ω constructFeng et al . ( 2004 ) Plasmid pKF43His-SsrB pMpM-A5Ω constructFeng et al . ( 2004 ) Transformant DW160DH5α harboring His-HNS ( S . typhimurium ) in pMpM-A5ΩWalthers et al . ( 2011 ) Plasmid pKF104His-SsrBc pMpM-A5Ω constructFeng et al . ( 2004 ) pBR328::csgDcsgD constructProf Iñigo Lasa’s groupPlasmid pRC24K179A His-SsrB pMpM-A5Ω constructCarroll et al . ( 2009 ) All DNA manipulation procedures were carried out according to ( Sambrook , 1989 ) using reagents procured from Qiagen , Invitrogen or Fermentas , Singapore . Transformation in S . Typhimurium strains was performed by standard electroporation protocols ( Sambrook , 1989 ) . Polymerase chain reaction ( PCR ) was carried out using oligonucleotides as listed in the Table 2 following standard protocols ( Sambrook , 1989 ) . 10 . 7554/eLife . 10747 . 030Table 2 . List of oligonucleotides . DOI: http://dx . doi . org/10 . 7554/eLife . 10747 . 030Purpose/nameSequence ( 5’-3’ ) Digf ( forward 755bp csgD regulatory region ) tgatgaaactccactttttttaDigr ( reverse 755bp csgD regulatory region ) tgctgtcaccctggacctggtcssrA knockout ( forward ) atgaatttgctcaatctcaagaatacgctgcaaacatctt ttaagacccactttcacattssrA knockout ( reverse ) agccgatacggcattttcaatatcagccagcaagaggtcc ctaagcacttgtctcctgcsg1 ( forward csgD internal ) ggaagatatctcggccggttgccsg2 ( reverse csgD internal ) tcagcctagggataatcgtcagrrsA1 ( forward rrsA internal ) gcaccggctaactccgtgccrrsA2 ( reverse rrsA internal ) gcagttcccaggttgagcccgPSsrBF ( forward for pKF46 ) atgaaagaatataagatcttatPSsrBTR ( hybrid reverse for pKF46 ) ttaatactctaattaacctcattcttcgggcacagttaagtctaagcacttgtctcctgTSsrBF ( forward TetRA-ssrB ) acttaactgtgcccgaagaatgaggttaatagagtattaattaagacccactttcacattTSsrBR ( reverse TetRA- after ssrB stop ) catcaaaatatgaccaatgcttaataccatcggacgcccctggctaagcacttgtctcctgDigb ( Forward for EMSA ) Biotin- tgatgaaactccactttttttaCsgDigRS ( Reverse for EMSA ) aatatttttctctttctggatahns knockout ( forward ) gctcaacaaaccaccccaatataagtttggattactacattaagacccactttcacatthns knockout ( reverse ) atcccgccagcggcgggattttaagcatccaggaagtaaactaagcacttgtctcctg The S . Typhimurium strain harboring a D56A mutation in SsrB was generated by the homologous recombination technique as decribed in ( Karlinsey , 2007 ) . Plasmid pKF46 ( Feng et al . , 2004 ) was used to construct the necessary linear DNA fragment by a two-step overlapping PCR ( Sambrook , 1989 ) . Homologous recombination was also used to construct the hns and ssrA deletion mutants in the respective strain backgrounds . The sifA-lacZ marker was transduced into the lambda attP site of strains DW748 ( ssrAH1 ) , DW749 ( ssrAD ) and DW750 ( ssrAH2 ) using standard P22 transduction protocols ( Davis , 1980 ) . Biofilms were grown in three-channel glass-bottomed flow cells ( individual cell dimensions 1 × 4 × 40 mm3 ) with an individual-channel flow rate of 8 ml/h of 0 . 5X M9 glucose minimal medium . Each channel was inoculated with M9 minimal medium cultures adjusted to an OD600 of 0 . 04 . Flow was started after allowing initial attachment for 1 hr . After specific time points , each channel of the flow cell was stained with 200 μl SYTO-9 green ( Invitrogen , Singapore ) as per the manufacturer’s protocol . Subsequently , for each flow cell channel image acquisition was performed using a LSM 780 Carl Zeiss confocal fluorescence microscope at 480/500 nm ( 20X magnification ) . Five image stacks were acquired starting from the center of the channel to a distance of 5–10 mm from the inlet , approximately 5 mm apart . This experiment was repeated twice in duplicates for the wild type and ssrB mutant strains . All images were processed using the Image J software . Static biofilms were grown by inoculating the bacterial strains on APTES-coated coverslips in 24-well Nunclon polystyrene plates in 500 μl LB without salt medium and kept shaking at 30°C , 100 rpm . After two days , the growth medium was removed; wells were washed with PBS and incubated with 200 μl of the fixative solution ( 4% para-formaldehyde+0 . 2% glutaraldehyde in filter-sterilized PBS ) for 1 hr at room temperature . After washing twice with sterile water , 35% , 50% , 75% , 90% , 95% and 100% ethanol were added sequentially and kept for 10 min at room temperature . Dehydrated samples were stored in absolute ethanol and kept at 4°C until subjected to critical point drying . Images were obtained using a Carl Zeiss Merlin field emission scanning electron microscope by detecting secondary electrons ( SE2 ) under low electron beam acceleration voltage ( 1 kV ) and low probe current ( 76 pA ) . To estimate the amount of biofilms typically a single colony was inoculated in LB broth medium and incubated at 37°C/250 rpm . 2 μl of this culture was added to 198 μl LB without salt medium in a 96-well polystyrene plate and kept at 30°C with gentle shaking of 100 rpm . After two days , the growth medium was removed and each well was washed twice with 200 μl of Phosphate-buffered Saline ( PBS ) . The attached bacterial communities or biofilms in each well were then stained with 200 μl crystal-violet solution ( 0 . 1% ) for a minimum of 1 min . This was followed by washing twice with PBS and addition of 200 μl absolute ethanol . Appropriate dilutions were measured for absorbance at 580 nm using a Tecan Infinite M200 plate reader . Each experiment was performed at least thrice in triplicates . For the time course experiment , the above procedure was performed twice at 12 , 24 , 36 , 47 , 60 , 72 and 84 hr respectively in triplicates . To study the capability of the Salmonella strains to form biofilms on cholesterol-attached surfaces , a tube biofilm assay was performed as described in ( Crawford et al . , 2008 ) , but without the use of bile salts in the growth medium . LB without salt medium was used for growth at all steps . All experiments were performed at least thrice in triplicates for 7 days on a nutator shaker at room temperature . Biofilms formed at the end of seven days were estimated using the above described crystal violet staining protocol with appropriate controls . Each of the bacterial strains were grown overnight in LB , washed twice and resuspended in 50 μl of PBS buffer . These were then transferred into 1 ml of fresh MgM medium , pH 5 . 7 , as in ( Feng et al . , 2003 ) and kept shaking at 37°C for around 6–7 h , until growth reached OD600 between 0 . 6–0 . 8 . At this stage , 50 μl of the culture was removed in a 96-well microtiter plate and 145 μl of lysis buffer was added ( 0 . 01% SDS , 50 mM Beta Mercaptoethanol in Z-buffer ) as performed previously ( Feng et al . , 2003 ) . The β-galactosidase activity was represented in Miller Units and calculated as 1000 x [ ( OD420-1 . 75 x OD550 ) ]/ t ( min ) x volume ( ml ) x OD600 ) . Measurements were made in a Tecan Infinite M200 plate reader and repeated thrice in triplicates . Total RNA was isolated from two-day old macrocolonies using the Qiagen RNeasy kit ( Qiagen , Singapore ) as per the manufacturer’s protocol with some modifications . Briefly , the macrocolonies were selected and re-suspended in 1 ml Qiagen RNAprotect reagent followed by immediate total RNA isolation or storage at -80°C . Appropriate volume of lysis buffer , 15 mg/ml lysozyme in TE buffer , was used . After dissolving the RNA in 20 μl RNase-free water , samples were assessed by gel electrophoresis and quantified using the Nanodrop system ( THERMO Scientific , Singapore ) . All RNA samples were treated with TURBO DNase as per the manufacturer’s protocol ( Life Technologies , Singapore ) . PCR analysis using rrsA specific primers was performed to ensure the absence of genomic DNA in the RNA preparations . Reverse transcription reaction was carried out using the iScript reverse transcription supermix for RT-qPCR ( BIO-RAD , Singapore ) according to the recommended protocol . This was followed by amplifying 5–10 ng cDNA by real-time qPCR using SsoFast EvaGreen Supermix ( BIO-RAD ) and csgD-specific primers . Similar reactions were set up using rrsA-specific primers for their use as normalization controls . All experiments were performed in triplicate with two independent RNA preparations . Relative transcript abundance was determined using the △CT method using a reference gene as per the manufacturer’s guide . Two-day old macrocolonies were resuspended in 400 μl Laemmli buffer ( Sambrook , 1989 ) and total protein was separated by 10% SDS-PAGE . This was followed by electro-transfer to a PVDF membrane as described before ( Feng et al . , 2004 ) . The membrane was incubated with anti-CsgD ( 1:10 , 000 ) or anti-GroEL ( 1:5000 ) antibodies in PBS buffer containing 0 . 05% Tween-20 and 3% BSA . Anti-rabbit secondary antibody ( Santa Cruz Biotechnology Inc . , Dallas , TX ) was used for detection as described previously ( Feng et al . , 2004 ) . The E . coli BL21 ( DE3 ) strain was used as a host for the overproduction of proteins His-SsrBc , His-SsrB , D56A His-SsrB , K179A His-SsrB and His-H-NS . The respective plasmids harboring the constructs have been listed in Supplementary Table 2 . Detailed procedures for the purification of His-SsrBc , His-SsrB , D56A His-SsrB and K179A His-SsrB have been described before ( Feng et al . , 2004; Walthers et al . , 2007; Carroll et al . , 2009 ) . The DW160 plasmid harboring His-tagged H-NS from Salmonella Typhimurium was used to overexpress and purify His-H-NS following the procedure described in ( Walthers et al . , 2011 ) . Glutaraldehyde-modified mica surface was prepared as decribed previously in ( Liu et al . , 2010 ) and references therein . A 755 bp sequence upstream to the +1 start site ( Gerstel et al . , 2003 ) of csgD was amplified and gel purified . This fragment also harbored a SsrB-specific site ( Feng et al . , 2004; Walthers et al . , 2007 ) , TTATAAT sequence ( Figure 6—figure supplement 4 ) . A typical 50 μl reaction contained 10 ng of this DNA ( 755 bp of the csgD regulatory region ) mixed with an appropriate amount of SsrB , SsrBc or D56A SsrB and incubated for 15 min at room temperature . This mixture was then deposited on glutaraldehyde-modified mica for 15 min . Images were acquired on a Bruker Dimension FastScan AFM system using the tapping mode with a silicon nitride cantilever ( FastScan C , Bruker ) . Raw AFM images were processed using Gwyddion software ( http://gwyddion . net/ ) . The bending angle was analysed using a home-written Matlab code as described in the Supplementary methods . GraphPad Prism 6 version 6 . 00 , GraphPad Software , La Jolla California , www . graphpad . com was used to make all the graphs and do statistical analysis by Student’s t-test wherever required . | Salmonella bacteria can infect a range of hosts , including humans and poultry , and cause sickness and diseases such as typhoid fever . Disease-causing Salmonella evolved from harmless bacteria in part by acquiring new genes from other organisms through a process called horizontal gene transfer . However , some strains of disease-causing Salmonella can also survive inside hosts as communities called biofilms without causing any illness to their hosts , who act as carriers of the disease and are able to pass their infection on to others . So how do Salmonella bacteria ‘decide’ between these two lifestyles ? Previous studies have uncovered a regulatory system that controls the decision in Salmonella , which is made up of two proteins called SsrA and SsrB . To trigger the disease-causing lifestyle , SsrA is activated and adds a phosphate group onto SsrB . This in turn causes SsrB to bind to and switch on disease-associated genes in the bacterium . However , it was less clear how the biofilm lifestyle was triggered . Desai et al . now reveal that the phosphate-free form of SsrB – which was considered to be the inactive form of this protein – plays an important role in the formation of biofilms . Experiments involving an approach called atomic force microscopy showed that the unmodified SsrB acts to stop a major gene that controls biofilm formation from being switched off by a so-called repressor protein . Salmonella acquired SsrB through horizontal gene transfer , and these findings show how this protein now acts as a molecular switch between disease-causing and biofilm-based lifestyles . SsrB protein is also involved in the decision to switch between these states , but how it does so remains a question for future work . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"microbiology",
"and",
"infectious",
"disease"
] | 2016 | The horizontally-acquired response regulator SsrB drives a Salmonella lifestyle switch by relieving biofilm silencing |
Coupling of synaptic vesicle fusion and retrieval constitutes a core mechanism ensuring maintenance of presynaptic function . Recent studies using fast-freeze electron microscopy and capacitance measurements reported an ultrafast mode of endocytosis operating at physiological temperatures . Here , using rat hippocampal neurons , we optically monitored single synaptic vesicle endocytosis with high time resolution using the vesicular glutamate transporter , synaptophysin and the V0a1 subunit of the vacuolar ATPase as probes . In this setting , we could distinguish three components of retrieval operating at ultrafast ( ~150–250 ms , ~20% of events ) , fast ( ~5–12 s , ~40% of events ) and ultraslow speeds ( >20 s , ~40% of events ) . While increasing Ca2+ slowed the fast events , increasing temperature accelerated their time course . In contrast , the kinetics of ultrafast events were only mildly affected by these manipulations . These results suggest that synaptic vesicle proteins can be retrieved with ultrafast kinetics , although a majority of evoked fusion events are coupled to slower retrieval mechanisms .
Synapses form the basic computational units in the brain as they transfer and process information with exquisite temporal and spatial precision . In presynaptic terminals , this is achieved through fast neurotransmitter release via fusion of synaptic vesicles within hundreds of microseconds upon the arrival of a presynaptic action potential ( Südhof , 2013 ) . In order to maintain synaptic performance during activity , synaptic vesicle proteins and lipids need to be retrieved and recycled . In contrast to the rapidity of the synaptic vesicle fusion process , the time course of endocytosis and its temporal relationship to presynaptic activity are widely debated . Early studies using electron microscopy ( EM ) and fluorescence imaging with FM dyes postulated retrieval and recycling of synaptic vesicles to occur within 30—90 s ( Miller and Heuser , 1984; Ryan et al . , 1996 ) . However , subsequent experiments proposed a faster time course — spanning only about 2 s — for synaptic vesicle endocytosis based on capacitance measurements ( von Gersdorff and Matthews , 1994 ) , a time course also corroborated by FM dye-based studies ( Klingauf et al . , 1998; Kavalali et al . , 1999 ) . In recent years , experiments using fast-freeze EM ( Watanabe et al . , 2013 ) or improved capacitance measurements ( Delvendahl et al . , 2016 ) produced estimates within 100—500 ms for synaptic vesicle retrieval in central synapses . This so-called ultrafast synaptic endocytosis appeared to predominantly occur near physiological temperatures ( ~36°C ) ( Watanabe and Boucrot , 2017 ) . In contrast , experiments using synaptic vesicle proteins labeled with pHluorin ( a pH-sensitive derivative of GFP ) to optically monitor single synaptic vesicle endocytosis gave rise to variable and often contradicting results ( Balaji and Ryan , 2007; Gandhi and Stevens , 2003; Granseth et al . , 2006; Leitz and Kavalali , 2011; Zhu et al . , 2009 ) with the extra complication that these studies were typically conducted at room temperature . In the present study , we aimed to monitor retrieval of single synaptic vesicles with improved time resolution at near physiological temperatures to resolve the discrepancies between fast EM and capacitance measurements and optical approaches . Besides enabling direct visualization of single vesicle retrieval kinetics , visualizing pHluorin-tagged synaptic vesicle protein dynamics provides the unique advantage of examining synaptic vesicle retrieval in a molecularly specific manner ( Kavalali and Jorgensen , 2014 ) . Under these conditions , using fluorescently tagged vesicular glutamate transporter , synaptophysin and V0a1 subunit of the vacuolar ATPase as probes , we detected three components of synaptic vesicle endocytosis: ultrafast ( ~150–250 ms ) , fast ( ~5–12 s ) and ultraslow ( >20 s ) . This indicates that multiple endocytic pathways with different kinetics are operating at presynaptic terminals . Interestingly , while Ca2+ and temperature were robust regulators of fast events , the kinetics of ultrafast events were relatively impervious to these manipulations . Overall , these results provide a bridge between dynamic optical measurements of synaptic vesicle protein retrieval and EM or capacitance based methods that monitor membrane trafficking .
The vesicular glutamate transporter 1 ( vGluT1 ) bound to the pH sensitive derivative of GFP , pHluorin has been widely used to track synaptic vesicle trafficking based on its low plasma membrane expression and high signal-to-noise ratio ( Voglmaier et al . , 2006; Balaji and Ryan , 2007; Leitz and Kavalali , 2011 ) . The pHluorin fluorescence is quenched in the acidic environment of the synaptic vesicle lumen and peaks after fusion with the plasma membrane ( Figure 1A ) . Under strong stimulation ( e . g . 40 Hz 5 s in Figure 1A–C ) , recapture of synaptic proteins from the plasma membrane results in a decay in fluorescence , reflecting synaptic vesicle retrieval and subsequent vesicle re-acidification ( Figure 1A ) . At room temperature ( ~24°C ) , increasing extracellular Ca2+ concentration from 2 mM to 8 mM leads to an increase in the number of fused vesicles ( seen as an increase in peak amplitude ) and a slight increase in the decay time , suggesting saturation of the endocytic machinery under strong , repetitive stimulation ( Figure 1A–C ) . Earlier studies have demonstrated that at physiological temperatures ( 34–36°C ) the time course of bulk synaptic vesicle endocytosis seen after strong stimulation shows accelerated kinetics ( Fernández-Alfonso and Ryan , 2004; Renden and von Gersdorff , 2007; Soykan et al . , 2017 ) possibly via activation of ultrafast endocytosis mechanisms ( Delvendahl et al . , 2016; Watanabe et al . , 2013 ) . In agreement with this premise , in our experiments , increasing the temperature from ~24 to 34°C led to a ~ 2 . 5 fold decrease in the fluorescence decay time constants after 40 Hz stimulation ( Figure 1C ) . Moreover , at 34°C we observed faster fluorescence rise times at stimulation onset compared to room temperature ( Figure 1D ) , consistent with an increase in the synchronicity of release seen in previous electrophysiological studies ( Pyott and Rosenmund , 2002 ) . In the next set of experiments , we focused on single synaptic vesicle fusion and retrieval events induced during sparse low frequency stimulation ( 0 . 05 Hz ) . At the single vesicle level , fluorescence dwell times detected after fusion are representative of the time the pHluorin-tagged protein resides at the presynaptic plasma membrane before being retrieved . For action potential ( AP ) evoked release the kinetics of this process is negatively regulated by Ca2+ concentration in a synaptotagmin-dependent manner ( Leitz and Kavalali , 2011; Li et al . , 2017 ) . To visualize potential ultrafast retrieval of single synaptic vesicles , we monitored single fusion events at near physiological temperatures . Moreover , we increased the time resolution of our measurements , by a combination of faster image acquisition settings and time-domain de-noising of fluorescence signals based on methods originally developed to de-noise electrophysiological single channel recordings ( Chung and Kennedy , 1991 ) . This non-linear algorithm consists of a series of backward and forward predictors , similar to moving averages , used to calculate the probability of occurrence of fast changes in the signal , i . e . sharp changes in amplitude above the noise , in a Bayesian framework ( Figure 2A ) . This type of noise reduction preserves the amplitude of the measured signal without major distortion of the edges compared to other classical methods based on the fast Fourier transform ( see Figure 2B and Figure 2—figure supplement 1 ) . As the filtering method we used assumes Gaussian distribution of the noise ( Chung and Kennedy , 1991 ) , we first validated that our experimental noise shows a Gaussian distribution ( Figure 2—figure supplement 1A–B ) . These new settings enabled us to probe the fastest rate at which synaptic vesicle protein retrieval takes place at physiological temperatures . Noise reduction of the fluorescence traces from each region of interest ( ROI ) allowed us to detect small amplitude fusion events and reliably estimate release probability using failure analysis ( Figure 2C ) . Moreover , dwell times of the fluorescent probe at the plasma membrane prior to retrieval could be determined with higher precision , enabling a clear discrimination between dwell times and the subsequent decay in fluorescence due to reacidification . Dwell time was defined as the total time after fusion that maintains maximum ( peak ) amplitude until the first derivative starts to be negative ( due to re-acidification and the consequent decay of the signal ) . To estimate the accuracy of our event detection algorithm , we generated negative imaging controls with no stimulation in the presence of TTX or in 0 mM extracellular Ca2+ following EGTA-AM treatment to buffer intracellular Ca2+ . In these control experiments , we detected events with a probability of ~0 . 03% or less ( likely reflecting false positives along with spontaneous fusion events ) , well within acceptable boundaries ( ~5 fold less than typical event detection rate under normal conditions ) ( Figure 2D–E ) . To assess the precision of the dwell time calculation , artificial traces with different release probabilities and dwell time distributions were simulated and similarly de-noised and analyzed . This analysis corroborated the limited contribution of false positives to the dwell time durations obtained ( Figure 2—figure supplement 1C–E ) . In conclusion , all these experimental and simulated controls enabled us to tune the de-noising parameters to analyze the duration of dwell times in the experimental traces ( Chung and Kennedy , 1991 ) . Example vGluT1-pHluorin fluorescence traces of single synaptic vesicle fusion events after de-noising are depicted in Figure 3A . As shown in Figure 3B and C , mean release probability at 2 mM extracellular Ca2+ is ~0 . 15 , in agreement with previous reports ( Leitz and Kavalali , 2011; Murthy et al . , 1997 ) . Increasing Ca2+ concentration to 8 mM leads to a marked increase in release probability ( Figure 3B–C ) , although switching from room temperature ( ~24°C ) to 34°C , did not result in a major change in release probability , consistent with previous electrophysiological measurements ( Kushmerick et al . , 2006; Pyott and Rosenmund , 2002 ) . Initially , we detected that 40 ± 2% of all fusion events did not return to baseline during the course of imaging ( >20 s , and see Figure 3H ) suggesting an ultraslow pathway for endocytosis ( Gandhi and Stevens , 2003; Zhu et al . , 2009 ) . The remaining events ( ~60% ) showed dwell times of variable lengths which displayed a non-normal distribution ( Figure 3D–G ) with a peak at ~250 ms . Further analysis of the histograms revealed that they are best represented by a double exponential decay function , pointing to the co-existence of two endocytic mechanisms with different mean rates . The fitting of dwell time distributions for the experimental groups revealed an ultrafast endocytic component with an average time course of about 150–250 ms , and a fast process that proceeds with an order of magnitude slower time course of around 5–12 s ( arrows in Figure 3D–G ) . Increasing extracellular Ca2+ concentration from 2 to 8 mM led to only a 10–15% slowdown of ultrafast endocytosis at both temperatures ( Figure 3D–G ) . The fast component of endocytosis , on the other hand , showed a steeper temperature dependence . Increasing extracellular calcium concentration from 2 to 8 mM led to a ~ 60% slowdown of fast endocytosis at 24°C , while increasing calcium at 34°C led to only a ~ 10% slowdown ( Figure 3D–G ) . This result indicates that increasing temperature triggers an overall acceleration of vGluT1-pHluorin retrieval mainly through regulation of the fast component of endocytosis , and this form of regulation is more evident at higher Ca2+ concentrations ( compare insets in Figure 3E and G ) . Kolmogorov-Smirnov pair-wise comparison of cumulative distributions , as well as Kruskal-Wallis non-parametric analysis of variances , further corroborated that Ca2+ concentration and temperature had a significant effect on the shift of the distributions ( Figure 3D–G , legend ) . While the mean speed of endocytosis for both fast and ultrafast pathways was regulated by temperature and Ca2+ , the relative contribution of each mode to total protein retrieval did not change significantly ( Figure 3H ) . Overall under all conditions around 40% of fusion events were not followed by endocytosis in the imaging period ( >20 s ) , 13–18% undergo ultrafast endocytosis and 38–48% of vGluT1-pHluorins were retrieved through a slow pathway ( Figure 3H ) . The mean amplitude of fusion events was similar for all types of retrieval ( data not shown ) . In addition , to test whether the two components of retrieval we identified were a consequence of denoising , we averaged non-denoised traces in an unbiased manner and found two phases of fluorescence decay with ultrafast ( ~250 ms ) and fast kinetics ( ~3–7 s ) ( Figure 4A ) . Although , this analysis does not distinguish dwell time and re-acidification kinetics , it supports our conclusion that two kinetically distinct processes retrieve vesicle proteins with ultrafast and fast speeds . Interestingly , in neurons where the Ca2+ sensor synaptotagmin-1 ( syt1 ) was knocked-down ( KD ) the fast component is greatly reduced and vGluT1-pHluorin is retrieved almost exclusively by ultrafast endocytosis ( 70–80% of all fusion events , Figure 4B–C ) , suggesting that ultrafast endocytosis is syt1-independent ( see Li et al . , 2017 ) . Moreover , ultraslow retrieval ( >20 s ) is also abolished in syt1 KD neurons ( Figure 4C ) . This was not the case for synaptotagmin-7 ( syt7 ) KD neurons , where averaged single vesicle traces showed a double exponential decay ( τ1 = 0 . 31 ± 0 . 05 s , τ2 = 8 . 5 ± 4 . 8 s ) similarly to control synapses . Taken together , these results indicate that the speed of fast single vesicle retrieval is markedly accelerated by temperature and slowed down by extracellular Ca2+ concentration , possibly through syt1 , whereas the kinetics of ultrafast events are less sensitive to these factors . Approximately 60% of total fusion events displayed dwell times and subsequent decay . Among these , only in 50% the fluorescence returned back to baseline completely , consistent with full quantal retrieval of vGluT1-pHluorins ( fractional retrieval of 1 . 0 ± 0 . 2 , see Figure 4D–E ) . Approximately , 35% of the fusion events that were followed by a measurable dwell time , showed partial retrieval , indicating that some vGluT1-pHluorin molecules might diffuse away from the site of fusion and are temporarily unavailable for endocytosis . The remaining ~15% of events showed excess retrieval , suggesting that more vGluT1-pHluorin molecules were retrieved compared to the ones that fused ( Figure 4D–E ) , similar to what was described in other systems ( e . g . Chung and Kavalali , 2009; Van Hook and Thoreson , 2012; Zhu et al . , 2009 ) . Plotting the fraction of retrieval for each fusion event as a function of the duration of its dwell time revealed that the dispersion of retrieval values increased with longer dwell times ( Figure 4F ) . While the dispersion was ±20% for ultrafast events ( dwell times of 0 . 0 to 0 . 5 s ) , for fast events dispersion increased from approximately ±40% to around ±80–90% as the values of dwell times increased ( see legend of Figure 4F ) . This increasing variability in the retrieved fraction occurred similarly for all the temperatures and extracellular Ca2+ concentrations tested . This result implies that as endocytosis gets slower , the probability of having partial retrieval of synaptic vesicle proteins or of having retrieval of double amount of protein compared to the amount that fused increases . As mentioned before , for neurons lacking syt1 the majority of fusion events were followed by ultrafast endocytosis , for these events vGluT1-pHluorin is endocytosed very efficiently , with ~80% of vGluT1-pHluorin fusion events showing quantal retrieval ( 1 . 0 ± 0 . 2 , Figure 4G–H ) . Our results suggest that ultrafast endocytosis is Ca2+ , syt1 and temperature independent , and it rapidly retrieves approximately the same amount of protein that fused . Distinct synaptic vesicle proteins are typically coupled to diverse endocytic mechanisms ( Chanaday and Kavalali , 2017; Li et al . , 2017; Pan et al . , 2015; Voglmaier and Edwards , 2007 ) Therefore , to evaluate the general validity of our results , we tested the trafficking of synaptophysin-1 fused to the red pH sensitive protein pHTomato ( Syp1-pHTomato , Figure 5A ) ( Li and Tsien , 2012 ) under the same conditions . At 40 Hz stimulation , the increase of temperature from 24 to 34°C caused a ~ 2 . 5 fold rise in the bulk endocytic rate of Syp1-pHTomato ( Figure 5B ) coupled with a faster signal rise time ( Figure 5C ) , consistent with our observations using vGluT1-pHluorin . At the single vesicle level , as before , we did not detect a change in release probability in response to increase in temperature ( ~0 . 15 at both tested temperatures ) ( Figure 5D ) . Sample traces of single synaptic vesicle fusion events detected with Syp1-pHTomato are shown in Figure 5E . Histograms of detectable dwell times ( 50 ± 2% of events ) were best fitted with a double exponential , revealing the presence of two parallel endocytic processes for synaptophysin ( Figure 5F–I ) . The mean time courses of these two processes were about 175–300 ms and 5–11 s , comparable to our previous results with vGluT1-pHluorin ( arrows in Figure 5F–I ) . While the increase in temperature did not produce a significant effect in the overall endocytic rate at 2 mM Ca2+ ( Figure 5F–G , and inset in 5G ) , it caused an almost two-fold acceleration of retrieval at 8 mM extracellular Ca2+ ( Figure 5H–I , and inset in 5I ) . As observed for vGluT1-pHluorin , even though both ultrafast and fast processes are regulated by Ca2+ and temperature , the effect of these manipulations are more striking for the fast component and at 8 mM Ca2+ . Statistical analysis confirmed this finding ( Figure 5F–I , legend ) . In contrast , the charts in Figure 5J show that there is only a modest effect of temperature and Ca2+ in the proportion of the different modes of endocytosis after single AP stimulation . Our results so far revealed the co-existence of two rapid endocytic processes at hippocampal synapses ( ultrafast and fast ) , whose speed is differentially regulated by Ca2+ and temperature . VGluT1 and Syp1 proteins are present in high copy numbers on synaptic vesicles ( ~9 and~31 per vesicle , respectively ) ( Takamori et al . , 2006 ) . As discussed above , it is possible that they are not completely retrieved after fusion ( Gimber et al . , 2015 ) ; and also see Figure 4D–H ) . In our experiments , about 40% of all vGluT1-pHluorin events and 50% of Syp1-pHTomato events do not show protein retrieval during the recording period , probably reflecting fully collapsed vesicles after fusion ( ultraslow retrieval events ) . Taken together , these observations suggest that three modes of vesicle recycling take place; a mode where synaptic vesicle proteins remain at the surface membrane for extended time period albeit in a clustered fashion , an ultrafast mode of retrieval that occurs in the timescale of 150–300 ms , and a fast mode of endocytosis happening in the order of several seconds . To evaluate this premise further , we designed a probe with V0a1 subunit of the vacuolar ATPase ( v-ATPase ) responsible for vesicle acidification with pHluorin attached to one of its intraluminal loops ( V0a1-pHluorin , Figure 6 ) . V-ATPase and its subunits show low levels of expression in synapses and they have a low copy number per synaptic vesicle ( ~1 ) ( Takamori et al . , 2006 ) . Co-localization experiments revealed that V0a1-pHluorin is properly trafficked to synapses , with about 10–30% of the probe co-localizing with the presynaptic marker Synapsin-1 , also showing a similar subcellular distribution to the endogenous V0a1 protein ( Figure 6A–B ) and similar to what was described previously ( Bagh et al . , 2017 ) . V0a1 over-expression or V0a1-pHluorin expression did not produce significant alterations in synaptic transmission , as indicated by similar to control single AP evoked response amplitudes and charge transfer , as well as similar paired-pulse ratios ( Figure 6—figure supplement 1A–D ) . V0a1-pHluorin expression also had no effect on spontaneous , miniature excitatory postsynaptic current ( mEPSC ) amplitudes and frequency , and only caused a 2-fold increase in miniature inhibitory postsynaptic currents ( mIPSC ) frequency , with no changes in amplitude ( Figure 6C–E ) . Application of a depolarizing high potassium solution accentuated synaptic localization and demonstrated activity-dependent trafficking of V0a1-pHluorin ( Figure 6F ) . We used acid quenching ( pH ~5 ) to suppress surface fluorescence followed by NH4Cl perfusion to alkalinize intracellular compartments and determined the ratio of intracellular and surface V0a1-pHluorin ( Figure 6G–H ) . The subcellular distribution was highly variable among synapses , with ~40% of the probe present in internal organelles and ~60% on the plasma membrane on average , resembling what was described for endogenous V0a1 ( Morel et al . , 2003 ) . Due to the low copy number per synapse , 40 Hz stimulation only led to a small increase in fluorescence , which could be clearly visualized after treatment with folimycin to prevent re-acidification ( Figure 6—figure supplement 1E ) . Nevertheless , we could detect single vesicle fusion events in response to single AP stimulation albeit with a low probability where ~ 50% of the ROIs did not respond to stimulation . Application of v-ATPase inhibitor folimycin ( 200 μM ) converted 80–90% of these events to a non-decaying time course with a monophasic distribution of amplitudes supporting their basis in quantal fusion of single vesicles ( Leitz and Kavalali , 2011; Li et al . , 2017 ) ( Figure 6—figure supplement 1F–G ) . The distribution of V0a1-pHluorin dwell times at 34°C could be fitted with a double exponential decay , again reflecting ultrafast and fast time constants of ~250–300 ms and 3–10 s , respectively ( Figure 6J–K ) . As observed with the previous indicators , the time scale of the fast component of endocytosis was more sensitive to changes in extracellular Ca2+ concentration , showing more than two-fold decrease in speed compared to a ~ 30% slow down of the ultrafast component . The proportion of each mode of endocytosis for V0a1-pHluorin retrieval is shown in Figure 6L . Our measurements suggest that the ultrafast endocytic mechanism occurs at an average speed of ~150–300 ms , consistent with the recent flash-and-freeze electron microscopy-based estimates of ultrafast endocytosis in hippocampal synapses ( Watanabe et al . , 2013 ) . Although , these estimates are close to the Nyquist limit of detection based on our imaging speed of 10 Hz , our analysis of simulated single event traces ( see Figure 2—figure supplement 1C–E ) suggests that aliasing at this rate is not a major factor altering our measurements . Nevertheless , to address this potential concern , we repeated the experiments using vGluT1-pHluorin at an imaging speed of 40 Hz ( Figure 7 ) , taking advantage of the considerable improvement in signal-to-noise ratio and sensitivity added by signal de-noising ( see negative controls and artificial traces test in Figure 7—figure supplement 1B–I ) . Example single synaptic vesicle fusion events are shown in Figure 7A . The distribution of dwell times calculated at 40 Hz imaging speed closely resembled those obtained at 10 Hz , with a peak at 200 ms and a double exponential behavior ( Figure 7B–E ) . The mean endocytic speed for the ultrafast component of the distribution is ~150–250 ms , 3–5 times higher than the Nyquist limit . As observed before , at both 24°C and 34°C , endocytosis is slowed down by higher Ca2+ concentration , although this effect is more pronounced at room temperature and particularly for the fast endocytic component ( at 24°C there is a ~50 and ~60% increase in decay time at 8 mM Ca2+ for the ultrafast and fast components , respectively , while at 34°C the change is ~20 and ~40% , respectively; see arrows in Figure 7B–E ) . Statistical analysis of the distributions corroborated this observation , demonstrating that the overall effect of Ca2+ rise in the whole event distribution is only significant at 24°C while the effect of temperature rise is significant exclusively at 8 mM extracellular Ca2+ concentration ( insets in Figure 7D and E , and legend ) . As observed previously , temperature and Ca2+ did not have a major impact in the proportion of each mode of endocytosis ( Figure 7F ) .
In this study , using a combination of fluorescent reporters based on the vesicular glutamate transporter , synaptophysin or the V0a1 subunit of the vacuolar ATPase , we were able to identify three kinetically different pathways that retrieve individual synaptic vesicles after single AP driven fusion . As the retrieval patterns of these three probes largely overlapped , we could draw conclusions not only on the kinetic pathways that govern the retrieval of individual proteins but also on retrieval of associated synaptic vesicles . Visualization of vesicle retrieval was aided by rapid image acquisition settings and a time-domain signal de-noising algorithm adapted from earlier work focusing on estimation of ion channel kinetics parameters from single channel recordings ( Chung and Kennedy , 1991 ) . Furthermore , in these experiments monitoring endocytosis of single synaptic vesicles following fusion helped us to uncouple regulation of synaptic vesicle fusion by temperature and Ca2+ from the putative effects of these factors on retrieval . Since synaptic vesicle fusion occurs in a binary fashion , once a vesicle is fused its subsequent trajectory can be kinetically isolated from its fusion probability . Therefore , the nature of single vesicle fusion eliminates the requirement to normalize the extent of endocytosis to exocytosis and thus enables isolation of potential direct effects on endocytosis . One of the pathways we visualized retrieved synaptic vesicle proteins with a time scale on the order of 200 milliseconds . This time frame is consistent with ultrafast endocytosis as the time course is an order of magnitude faster than earlier estimates of fast endocytosis ( Klingauf et al . , 1998; von Gersdorff and Matthews , 1994 ) and in line with the time scales detected in recent EM-based or capacitance-based methods ( Delvendahl et al . , 2016; Watanabe et al . , 2013 ) . In addition to the ultrafast retrieval of fluorescent probes , we also detected , as in earlier optical experiments , a ‘fast’ mode of endocytosis operating within seconds as well as an ultraslow form of trafficking where we could not reliably determine a decrease in signal after fusion for the duration of our optical recordings ( Gandhi and Stevens , 2003; Zhu et al . , 2009 ) . Importantly , the parallel operation of the three pathways persisted in response to changes in temperature or Ca2+ levels as we did not observe major shifts in the relative segregation and composition of the dwell time distributions . However , the intrinsic kinetics of the fast pathway were modulated by temperature and Ca2+ , whereas the intrinsic kinetics of the ultrafast pathway were relatively unperturbed by these manipulations . As indicated above , our estimates for the ultrafast endocytosis are in line with previous measurements using EM based methods or capacitance measurements in multiple systems . For instance , experiments using ‘flash-and-freeze’ electron microscopy ( Watanabe et al . , 2013 ) revealed an ultrafast pathway of endocytosis that takes place at physiological temperature , but is not present or measurable at room temperature . Through this mechanism large invaginations appeared as soon as 50 ms after optogenetic stimulation and fully endocytosed membranes , endosome-like closed structures , peaked at 100–300 ms ( Watanabe et al . , 2013 ) . Subsequent capacitance measurements in hippocampal and cerebellar neurons corroborated this temperature-dependent ultrafast kinetics , revealing a mean time constant of ~470 ms for single AP evoked endocytic events only at 36°C ( Delvendahl et al . , 2016 ) . However , previous capacitance measurements in pituitary terminals as well as at the calyx of Held , which were performed at room temperature , reported the presence of fast flickering fusion pores in neurons with a mean open duration of ~300 ms ( Klyachko and Jackson , 2002; He et al . , 2006 ) . These fast flickering pores constituted around 20% of all exocytic events ( He et al . , 2006 ) . In another report , the time constant of endocytosis after vesicle fusion was estimated to be in the order of 100 ms ( Sun et al . , 2002 ) . These earlier findings would imply that ultrafast retrieval could occur at room temperature . In agreement with this premise , our experiments show that endocytosis of distinct synaptic vesicle proteins after single AP stimulation occur through similar kinetically distinguishable mechanisms . The ultrafast endocytosis process we measured has a mean duration of 150–250 ms , consistent with the time course of ultrafast endocytosis reported previously . Importantly , the fastest speed at which synaptic vesicle proteins can be fully retrieved after exocytosis seems to be ~150 ms , since the probability distribution of dwell times shows a dramatic reduction for shorter times . In our measurements , the mean speed of ultrafast endocytosis was only mildly affected either by changes in temperature or Ca2+ concentration , suggesting that the underlying molecular mechanism is insensitive to these factors and probably already operating at the limiting speed . The vast majority of ultrafast endocytotic events in our experiments showed quantal retrieval — that is the same number of proteins that fused were retrieved — indicating that synaptic vesicle molecular identity can be conserved through ultrafast endocytosis . One possible mechanism underlying this observation could be the re-closure of the fusion pore ( compatible with kiss-and-run ) where proteins remain on the synaptic vesicle membrane and are unable to diffuse away to the plasma membrane ( Alabi and Tsien , 2013 ) . Additionally , the high efficiency in protein retrieval during ultrafast endocytosis could also be explained by a mechanism where synaptic vesicle components , lipids and proteins , remain clustered in the plasma membrane after fusion as was proposed previously ( Bennett et al . , 1992; Willig et al . , 2006; Opazo and Rizzoli , 2010 ) . Subsequently , these clusters could be rapidly retrieved following a mechanism similar to the one found using rapid freeze EM methods ( Watanabe et al . , 2013 ) . Our results showed that at the level of single vesicle retrieval , Ca2+ appears to predominantly target fast events that occur in the order of seconds but leave the time course of ultrafast events relatively unperturbed . Recent work from our group showed that synaptotagmins , besides their role in coupling fusion to Ca2+ signals , can also regulate the endocytic time course of single vesicle fusion events in a Ca2+-dependent manner ( Leitz and Kavalali , 2016; Li et al . , 2017 ) . Particularly , in the absence of the main Ca2+-sensor for synchronous release , syt1 , ~80% of endocytosis occurs only through a rapid retrieval mechanism – less than 1 s – with no detectable delays ( also see Li et al . , 2017 ) . Moreover , in the absence of syt1 we detected very limited excess or partial retrieval of synaptic vesicle proteins . This finding suggests that the syt1-dependent delay in vesicle retrieval kinetics also to some extent disrupts the fidelity of vesicle protein retrieval , possibly facilitating the generation of vesicles with diverse protein compositions during retrieval ( Crawford and Kavalali , 2015; Raingo et al . , 2012 ) . Furthermore , the kinetics of single vesicle protein retrieval are not altered by syt7 knock down and knocking down syt7 on top of syt1 KD does not further accelerates endocytosis ( Li et al . , 2017 ) . Taken together , these results raise the possibility that ultrafast endocytosis is relatively syt1 and Ca2+-independent , while fast endocytosis might be regulated by these factors , supporting the notion that different molecular pathways govern the two processes . Previous reports of ultrafast endocytosis showed that it does not involve a clathrin-mediated pathway ( Delvendahl et al . , 2016; Watanabe et al . , 2013 ) , but it is dependent on actin . In this regard , a recent study implicated formins — actin remodeling proteins that serve several cell biological functions — in synaptic vesicle endocytosis in hippocampal synapses as well as at the calyx of Held terminals ( Soykan et al . , 2017 ) . Moreover , GTP and dynamin independent endocytosis have been proposed to coexist in neurons with dynamin-dependent pathways ( Van Hook and Thoreson , 2012; Xu et al . , 2008 ) . Taken together with our results , this indicates that a novel , still molecularly uncharacterized mechanism may operate independently of the classical endocytic machinery and synchronize synaptic vesicle retrieval with fusion in the timescale of milliseconds . As mentioned before , a tempting mechanism that fits this description is kiss-and-run , although further experimental support is needed to validate this hypothesis ( Alabi and Tsien , 2013; Chanaday and Kavalali , 2017 ) . Around 40% of the fusion events detected in our experiments were not endocytosed during the imaging period ( 20 s ) , indicating that even for single synaptic vesicle exocytosis slow modes of retrieval contribute to synaptic vesicle recycling . Elegant studies showed that synaptic vesicle proteins undergo localized diffusion after exocytosis , followed by re-clustering and endocytosis ( Gimber et al . , 2015 ) . The re-clustering and re-capturing of synaptic vesicle proteins is an extremely slow process , requiring more than 60 s , and it was mediated by the clathrin adaptors CALM and AP-180 ( Gimber et al . , 2015 ) . The ultraslow retrieval presented here could be compatible with this mechanism . The speed of endocytosis is a limiting step in the maintenance of synaptic transmission under repetitive stimulation ( Kavalali , 2006 ) . Synaptic vesicle endocytosis can rapidly replenish vesicle pools and prevent vesicle depletion , while removing the potential hindrance of subsequent fusion events by previously fused vesicles ( Fernández-Alfonso and Ryan , 2004; Hua et al . , 2013; Sara et al . , 2002 ) . Accordingly , it is plausible to envision the physiological necessity of an ultrafast endocytic mechanism clearing future sites of fusion in preparation for subsequent rounds of neurotransmitter release . We believe the single vesicle imaging approach we present here will facilitate the identification of heretofore poorly understood mechanisms underlying the ultrafast vesicle retrieval process .
Postnatal day 2–4 Sprague-Dawley rats were used for the experiments . Both hippocampi were dissected in sterile conditions and posteriorly dissociated using 10 mg/ml trypsin and 0 . 5 mg/ml DNAase for 10 at 37°C . After careful trituration using a P1000 pipette , cells were resuspended to a concentration of 1 pups per 16 coverslips and plated onto 12 mm coverslip coated with 1:25 MEM:Matrigel solution . Basic growth medium consisted of MEM medium ( no phenol red ) , 5 g/l D-glucose , 0 . 2 g/l NaHCO3 , 100 mg/l transferrin , 5% of heat inactivated fetal bovine serum , 0 . 5 mM L-glutamine , 2% B-27 supplement , and 2–4 μM cytosine arabinoside . Cultures were kept in humidified incubators at 37°C and gassed with 95% air and 5% CO2 . The super-ecliptic pHluorin was inserted between Gly-677 and Thr-678 of V0a1 from Mus musculus ( Atp6v0a1 gene – UniProtKB database number Q9Z1G4 ) . The construct was subcloned into pFU-GW lentiviral vector from Addgene . Lentiviruses were produced in HEK293T cells ( catalog number CRL-1573; ATCC , Manassas , VA , US ) by contransfection of pFUGW transfer vectors and three packaging plasmids ( pCMV-VSV-G , pMDLg/pRRE , pRSV-Rev ) using Fugene six transfection reagent ( catalog number E2692; Promega , Madison , WI , US ) . The supernatants of the cultures were collected 72 hr after the transfection and clarified by centrifugation ( 2000 rpm 15 min ) , and subsequently used for infection of DIV four hippocampal neurons . All experiments were performed on 16–20 DIV cultures when synapses were mature and lentiviral expression of constructs of interest was optimal ( Mozhayeva et al . , 2002; Deák et al . , 2006 ) . All experiments were performed following protocols approved by the UT Southwestern Institutional Animal Care and Use Committee . Western blots were performed as described in Nosyreva and Kavalali ( 2010 ) . Primary antibodies against GDI and V0a1 were used in 1:1000 and 1:500 dilution , respectively . Immunoreactive bands were visualized by enhanced chemiluminescence ( ECL ) , captured on autoradiography film and analyzed using GelAnalyzer2010 software ( http://www . gelanalyzer . com ) . V0a1 protein levels were normalized to GDI loading control . 16–18 DIV neuron cultures were fixed for 10 min in PBS containing 4% para-formaldehyde ( PFA ) and processed for as previously described ( Ramirez et al . , 2008 ) . Primary antibody against GFP was used to detect pHluorin-tagged proteins ( 1:200 ) , antibody against V0a1 subunit of the V-ATPase was used to detect total V0a1 in control and V0a1-pHluorin expressing neurons ( 1:250 ) , and anti-Synapsin1 antibody was used to detect presynaptic boutons ( 1:1000; control experiments were performed using Synaptobrevin two antibody to corroborate the results ) . Alexa-conjugated secondary antibodies ( 1:1000 ) were used to label the cells and then coverslips were mounted and imaged using an LSM 510 META confocal microscope ( Carl Zeiss , Oberkochen , Germany ) with a 63X ( NA1 . 4 ) objective . Cultured pyramidal neurons between 14 to 18 DIV were used for whole cell recordings at a clamped voltage of −70 mV by means of Axopatch 200B and Clampex 8 . 0 software ( Molecular Devices , San Jose , CA , US ) , filtering at 2 kHz and sampling at 5 kHz . The cells were visualized using a Nikon DIAPHOT 200 microscope ( Nikon , Minato , Tokyo , Japan ) . The internal pipette solution contained 115 mM CsMeSO3 , 10 mM CsCl , 5 m M NaCl , 10 mM HEPES , 0 . 6 mM EGTA , 20 mM tetraethylammonium chloride , 4 mM Mg-ATP , 0 . 3 mM Na2GTP and 10 mM QX-314 ( lidocaine N-ethyl bromide ) . The final solution was adjusted to pH 7 . 3 and 304 mOsM . Final resistance of the electrode tips was ~3–6 MΩ . For all experiments , the extracellular solution was a modified Tyrode’s solution containing 150 mM NaCl , 4 mM KCl , 10 mM glucose , 10 mM HEPES , 2 mM MgCl2 and 2 mM CaCl2 , adjusted to pH 7 . 4 and 310 mOsM . To isolate inhibitory postsynaptic currents , agonists of ionotropic glutamate receptors were added: 10 μM 6-cyano-7-nitroquinoxaline-2 , 3-dione ( CNQX ) and 50 μM aminophosphonopentanoic acid ( AP-5 ) . To isolate excitatory currents ( AMPA-mediated ) 50 μM AP-5 and 50 μM picrotoxin ( PTX , ionotropic GABA receptor inhibitor ) were added to the bath solution . To elicit evoked responses , electrical stimulation was delivered through parallel platinum electrodes with a constant current unit ( WPI A385; World Precision Instruments , Sarasota , FL , US ) set at 35 mA . Spontaneous activities ( mIPSCs and mEPSCs ) were recorded with the addition of 1 μM TTX . Miniature events were identified with a 5 pA detection threshold and analyzed with MiniAnalysis ( Synaptosoft , Fort Lee , NJ , US ) . Cultured hippocampal neurons at 16–20 DIV transfected with either vGluT1-pHluorin , Syp1-pHTomato or V0a1-pHluorin were used for the imaging experiments . The modified Tyrode's buffer from above containing 2 or 8 mM Ca2+ was used with 10 μM CNQX and 50 μM AP-5 to prevent recurrent network activity . For experiments performed at ~34°C solutions were heated using a bipolar temperature controller ( CL-1000 ) attached to a multi-line solution heater ( SHM-828; Warner Instruments , Hamden , CT ) . The objective was heated with an objective collar connected to a single channel temperature controller ( H401-T-SINGLE-BL; Okolab , Shanghai , China ) , and the stage and microscope were isolated from the room with a protective case to minimize temperature variances . Fluorescence was recorded using a Nikon Eclipse TE2000-U microscope with a 100X Plan Fluor objective ( Nikon , Minato , Tokyo , Japan ) attached to an Andor iXon + back illuminated EMCCD camera ( Model no . DU-897E-CSO-#BV; Andor Technology , Belfast , UK ) . For illumination , we used a Lambda-DG4 illumination system ( Sutter Instruments , Novato , CA , US ) with a FITC or TRITC filter . Images were acquired at 10 or 40 Hz with binning of 4 by four to optimize the signal-to-noise ratio . Neurons were stimulated using parallel bipolar electrodes ( FHC , Bowdoin , ME , US ) delivering 35 mA pulses at 20 s intervals , followed by a rest period prior to the delivery of 200 APs at 40 Hz . Boutons were visualized by the addition of Tyrode’s solution with 50 mM NH4Cl at the end of each experiment . Circular regions of interests ( ROIs ) of 2 . 27 μm diameter were automatically drawn around local fluorescence maximums using a custom-made macro for Fiji ( Schindelin et al . , 2012 ) and the fluorescent traces obtained were exported to Matlab ( Mathworks , Natick , MA , US ) for analysis . To calculate surface versus intracellular distribution of V0a1-pHluorin , we perfused a modified Tyrode’s solution at pH = 5 . 5 ( buffered by MES instead of HEPES ) to quench surface pHluorin and recorded for 30 s . After imaging putative boutons for another 30 s , a modified Tyrode’s solution containing 50 mM NH4Cl was perfused for 30 s , in order to alkalinize all compartments . The difference of the mean fluorescence during acid buffer perfusion and normal Tyrode’s solution correspond to the surface pool of pHluorin , while the difference in mean fluorescence between NH4Cl perfusion and normal Tyrode’s solution correspond to the internal pool of pHluorin . Fluorescence intensity traces were analyzed using a custom made Matlab script ( Chanaday , 2018; https://github . com/nchanaday/Single-vesicle-fusion-events; copy archived at https://github . com/elifesciences-publications/Single-vesicle-fusion-events ) , based on our previous analysis with some modifications ( Leitz and Kavalali , 2011 ) . Photobleaching was corrected with a single exponential decay and background was subtracted linearly , both photobleaching and background values were calculated based on fluorescence measurements of background in each imaging experiment and also for each ROI . De-noising was performed using a time-domain forward-backward non-linear filter developed originally by Chung and Kennedy ( 1991 ) and implemented for Matlab by Nigel Reuel ( Reuel et al . , 2012 ) . Minor changes were made to the code in order to improve de-noising of our data , based on the original paper by Chung and Kennedy ( 1991 ) . To find single vesicle fusion events , successful events were defined as those whose fluorescence amplitude was greater than three times the standard deviation of the baseline ( average of ~2 s prior to the event ) . To avoid multivesicular fusion events , the upper limit was set at the mean value of single event amplitude plus half the difference between the mean single vesicle amplitude and the mean of the next amplitude peak in the distribution ( corresponding to two quantums ) . This value was calculated comparing amplitude distributions for experiments performed in different extracellular Ca2+ concentration with or without TTX and with or without folymicin . Also , the event time has to be coincident with the time of stimulation . Dwell times were calculated as the time between the initial fluorescence step and the start of fluorescence decay defined as a switch to negative values of the first derivative . For 40 Hz stimulation , amplitude measurement , single exponential decay fitting and rise slope linear fitting were also performed in an automatized way using Matlab ( Mathworks , Natick , MA , US ) . The authors are open to share the Matlab script developed by our lab for the analysis of single synaptic vesicle fusion and endocytosis events used in the present work , upon request to ETK or NLC . Histograms of single vesicle dwell time distributions were fitted using Matlab and OriginPro 8 . 1 ( OriginLab , Northampton , MA , US ) , to corroborate the results . The comparison of reduced R-square and F ( from F test ) values between single exponential decay and double exponential decay fittings revealed a better fit ( higher R-square and F values ) for the double exponential model . R-square and residuals are informed in the figure legends . N for each group and experiment are informed in the figure legends . The Kolmogorov–Smirnov ( K-S ) test was used to determine differences in cumulative probability histograms when comparing two groups , for three or more groups histograms were compared using Kruskal-Wallis analysis of medians and Dunn’s multiple comparison post-test . Averaged fluorescence traces are shown as mean ± SD . Bar graphs always inform mean values ± SEM , except for non-parametric data where they express median ±confidence interval ( 5–95% ) . | Nerve cells or neurons exchange information at junctions called synapses . To send a message to its neighbor , a neuron must release molecules called neurotransmitters into the synapse . These then bind to receptor proteins on the neighboring cell . But neurons do not release neurotransmitter molecules one at a time . Instead they release them in packages called vesicles . Each vesicle contains about 1 , 000 molecules , which it releases by fusing with the cell membrane . The entire process takes less than one thousandth of a second . Synaptic vesicles are complex structures made up of many different proteins and lipids . To help ensure that neurons do not run out of vesicles , cells retrieve and recycle these components via a process called endocytosis . A number of studies have attempted to measure how long this retrieval process takes . But the studies – which used a variety of different techniques – yielded results ranging from a few hundredths of a second to more than a minute . Chanaday and Kavalali have now resolved this discrepancy by using fluorescence microscopy to study the retrieval process in rat brain cells . By attaching a fluorescent tag to specific molecules within the vesicle membrane , Chanaday and Kavalali were able to track individual vesicles . The results revealed that neurons retrieve vesicles from synapses via three different pathways . At temperatures like those in the rodent or human body , an ‘ultraslow’ pathway takes more than 20 seconds to retrieve vesicles . By contrast , a ‘fast’ pathway takes about 5 to 12 seconds . The quickest option , an ‘ultrafast’ pathway , retrieves vesicles in about 150 to 250 milliseconds . Increasing the temperature speeds up the fast pathway but has no effect on the ultrafast pathway . Neurons can thus retrieve vesicles from synapses in about 200 milliseconds , or one fifth of a second . Nevertheless , they retrieve about 80% of their vesicles using the two slower pathways . Identifying the mechanisms responsible for vesicle retrieval will help reveal how synapses work , as well as what can go wrong . Changes in components of synaptic vesicles contribute to several neurological and psychiatric diseases . Developing drugs that target synaptic vesicle recycling could be a promising therapeutic avenue . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"neuroscience"
] | 2018 | Optical detection of three modes of endocytosis at hippocampal synapses |
Cell cortex remodeling during cell division is a result of myofilament-driven contractility of the cortical membrane-bound actin meshwork . Little is known about the interaction between individual myofilaments and membrane-bound actin filaments . Here we reconstituted a minimal actin cortex to directly visualize the action of individual myofilaments on membrane-bound actin filaments using TIRF microscopy . We show that synthetic myofilaments fragment and compact membrane-bound actin while processively moving along actin filaments . We propose a mechanism by which tension builds up between the ends of myofilaments , resulting in compressive stress exerted to single actin filaments , causing their buckling and breakage . Modeling of this mechanism revealed that sufficient force ( ∼20 pN ) can be generated by single myofilaments to buckle and break actin filaments . This mechanism of filament fragmentation and compaction may contribute to actin turnover and cortex reorganization during cytokinesis .
The actin cortex consists of a thin actin meshwork bound to the inner cytosolic face of the plasma membrane by various anchor proteins ( Morone et al . , 2006 ) . It plays a pivotal role in providing mechanical stability to the cell membrane , and in controlling cell shape changes during cell locomotion and cell division ( Wessells et al . , 1971; Bray and White , 1988; Diz-Muñoz et al . , 2010; Sedzinski et al . , 2011 ) . Many of these features rely on proper functioning of the actin motor myosin II ( De Lozanne and Spudich , 1987; Cramer and Mitchison , 1995 ) . Myosin II in the actin cortex functions as assemblies of motor proteins forming anti-parallely arranged bipolar filaments ( myofilaments ) with motor domains on both filament ends ( ( Verkhovsky and Borisy , 1993; Verkhovsky et al . , 1995 ) , Figure 1A ) . Besides being the force generator necessary for cell cortex remodeling and actomyosin ring constriction during cytokinesis , there is evidence that myofilaments also contribute to actin filament turnover during cyotkinesis ( Burgess , 2005; Guha et al . , 2005; Murthy and Wadsworth , 2005 ) . In all these processes , the microscopic mechanism of the interaction between individual myofilaments and membrane-bound actin filaments is not understood . Due to the vast complexity of cellular systems , much effort has been spent to investigate the consequences of actin-myosin interactions from the in vitro perspective ( Backouche et al . , 2006; Smith et al . , 2007; Schaller et al . , 2010; Kohler et al . , 2011; Soares e Silva et al . , 2011; Gordon et al . , 2012; Reymann et al . , 2012 ) However , these studies focused on myosin-induced actin structure formation on a mesoscopic scale , rather than on the interaction between actin and individual myofilaments . In addition , membrane-bound minimal actin systems have only recently begun to be functionally reconstituted ( Vogel and Schwille , 2012 ) . To fill the gap in understanding individual myofilament–actin interactions , we directly visualized the action of myofilaments on membrane-bound actin filaments in a minimal in vitro system and complemented the experimental findings with a theoretical model . 10 . 7554/eLife . 00116 . 003Figure 1 . MAC composition and actin pattern formation by myofilaments . ( A ) Scheme of the MAC . Biotinylated actin filaments are coupled to a supported lipid bilayer ( Egg PC ) containing biotinylated lipids ( DSPE-PEG ( 2000 ) -Biotin ) via Neutravidin . ( B ) TIRFM images of MACs containing Alexa-488-phalloidin labeled actin filaments . The increase of actin filament densities ( left to right ) corresponds to an increase in the amount of DSPE-PEG200-Biotin ( low = 0 . 01 mol% , medium = 0 . 1 mol% , high = 1 mol% ) in the membrane . Scale bars , 10 µm . ( C ) Length distribution of myofilaments . The median length ( Lm ) and the 25th and 75th percentile ( brackets ) are indicated in µm . Inset shows a topographical AFM image of a myofilament . Height , 12 nm; scale bar 200 nm . ( D ) Dual-color TIRFM time-lapse images of a medium actin density MAC with Alexa-488-phalloidin labeled actin filaments ( green ) and myofilaments ( 0 . 3 µM unlabeled myosin II doped with Alexa 647 myosin II ( red ) ) before ( left image ) and during actin pattern formation . Scale bars , 10 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 00116 . 003
To mimic the cell cortex , we developed a ‘minimal actin cortex’ ( MAC ) consisting of actin filaments coupled to a supported lipid bilayer via biotin neutravidin bonds ( Figure 1A ) . We used Alexa-488-phalloidin stabilized as well as non-stabilized ( data not shown ) actin filaments for the MAC composition . By varying the amount of biotinylated lipids present in the lipid bilayer , we have control over the density of the actin layer ( Figure 1B ) . In order to understand the origin of contractility observed in cell cortices , we tested the response of the MAC upon addition of myosin motors . Rabbit muscle myosin II was purified and reassembled forming bipolar synthetic myofilaments with a typical length of 500–600 nm ( Figure 1C ) . Time lapse imaging of MACs with various actin densities using total internal reflection fluorescence microscopy ( TIRFM ) showed a dynamic rearrangement of the actin filaments , and subsequently the formation of actomyosin foci in an ATP-dependent manner immediately after addition of myofilaments ( Figure 1D , Movie 1 ) . Actin pattern formation occurred at ATP concentrations between 0 . 1–1 µM in systems where ATP is enzymatically regenerated ( Table 1 ) in 94% of the experiments ( n = 45 experiments ) . At higher and lower ATP concentrations , actin pattern formation in the MAC is absent . Actin structure formation after ATP depletion has been also reported for experiments using actin and myofilaments in solution ( Smith et al . , 2007 ) . At low ATP concentrations , myofilaments were predicted to function as active temporary crosslinkers , which may drive self-assembly of actin filaments into actin clusters . 10 . 7554/eLife . 00116 . 004Movie 1 . Actin pattern formation by myofilaments . Medium density MAC containing Alexa-488-phalloidin labeled actin filaments ( green ) exhibits dynamic rearrangements of actin filaments after addition of myofilaments ( 0 . 3 µM unlabeled myosin II doped with Alexa 647 myosin II [red] ) . Original image sequence was acquired at 200 ms intervals and contained 1500 frames . The frame number in the video was reduced to 187 frames and is displayed at 15 frames per second ( fps ) . Total time: 5 min . Corresponds to Figure 1D ( compressed avi; 32 . 7 MB ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00116 . 00410 . 7554/eLife . 00116 . 005Table 1 . ATP dependency of contraction and fragmentation . ATP was kept at a constant level during the experiment by enzymatic regeneration ( see ‘Material and methods’ ) . Buffer containing ATP concentrations listed in the table and 0 . 3 µM myofilaments were added to medium and or low-density MACs . Contraction here is defined as visible dynamic rearrangements of actin filaments after myofilament addition . Fragmentation implies visible actin filament breakage events after myofilament additionDOI: http://dx . doi . org/10 . 7554/eLife . 00116 . 005Regenerated ATP concentration ( µM ) ContractionFragmentation100NoNo50NoNo25NoNo12 . 5NoNo10NoNo1YesYes0 . 3YesYes0 . 1YesNo0NoNo To understand the details of myofilament–actin interactions at low ATP concentrations during actin pattern formation , we added myofilaments to low density MACs . If crosslinking of actin by myofilaments were the only prerequisite for pattern formation , we would expect no ( fast ) pattern formation in low density MACs , due to the long distances between the actin filaments in relation to the length of myofilaments . Strikingly , myofilament addition to low density MACs displayed breakage events and compaction of actin filaments , resulting in their shortening over time in all experiments ( n = 21 , Figure 2A–C , Movie 2 ) . After 20 min , the majority of the actin filaments have been shortened to , on average , half of their original length , and most of the fragments coalesced into single foci ( Figure 2A , B , Movie 2 ) . Note that actin filaments remained intact when imaged in the absence of myofilaments ( data not shown ) . Upon myofilament addition , actin filaments frequently showed deformations prior to actin filament breakage ( Figure 2C , yellow arrowheads , Movie 3 ) , indicating that force is exerted by the myofilaments and stress along the actin filament may build up until the actin filament breaks . Similarly , recent evidence implied that exposure of actin/fascin bundles to myofilaments can induce their disassembling and severing by an unknown process ( Backouche et al . , 2006; Haviv et al . , 2008; Thoresen et al . , 2011 ) . 10 . 7554/eLife . 00116 . 006Figure 2 . Actin filament shortening and compaction by myofilaments . ( A ) TIRFM time-lapse images of a low actin density MAC with Alexa-488-phalloidin labeled actin filaments before ( left image ) and after addition of ( non-labeled ) myofilaments ( 0 . 3 µM ) . Scale bars , 10 µm . ( B ) Actin filament length distribution at 0 , 20 and 53 min after myofilament addition . The median length ( Lm ) and the 25th and 75th percentile ( brackets ) are indicated in µm . ( C ) TIRFM time-lapse sequence of an Alexa-488-phalloidin labeled actin filament in the presence of myofilaments ( 0 . 3 µM ) . Yellow arrowheads point at deformation and breakage events . White arrowheads indicate an increase in fluorescence intensity . Scale bar , 5 µm . ( D ) and ( E ) image and the corresponding intensity profile ( blue curve ) of the actin filament . The intensity was measured along the yellow dashed line shown in ( D ) . The line started and ended outside the actin filament to indicate the background level . Asteriks in ( C ) and ( D ) mark the image taken for the intensity profile measurement . DOI: http://dx . doi . org/10 . 7554/eLife . 00116 . 00610 . 7554/eLife . 00116 . 007Movie 2 . Shortening of individual actin filaments by myofilaments . Individual Alexa-488-phalloidin labeled actin filaments in a low density MAC shorten over time in the presence of myofilaments . Original image sequence was acquired at 400 ms intervals and contained 8000 frames . The frame number in the video was reduced to 248 frames and is displayed at 30 fps . Total time: 53 min . Corresponds to Figure 2A . ( compressed avi; 7 . 7 MB ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00116 . 00710 . 7554/eLife . 00116 . 008Movie 3 . Fragmentation of a single actin filament . Example of a single Alexa-488-phalloidin labeled actin filament during its fragmentation and compaction in the presence of myofilaments . Images were acquired at 400 ms intervals . The video contains 1451 frames and is displayed at 100 fps . Total time: 9 . 7 min . Corresponds to Figure 2C . ( compressed avi; 2 . 8 MB ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00116 . 008 Furthermore , an increase in fluorescence intensity along the remaining actin filament is often observed after a fragmentation event ( Figure 2C , white arrowheads , Figure 2D , E ) . The fluorescence intensity proximal to the breakage site is approximately twice as high compared to the rest of the actin filament , suggesting that the fragment is dragged along the remaining actin filament by the myofilaments leading to its compaction ( Figure 2C–E , Movie 3 ) . We propose that fragmentation and compaction contribute to the observed coalescence of actin fragments into single foci during the dynamic rearrangement of the actin filaments ( Figures 1D and 2A , Movies 1 and 2 ) . To determine how myofilaments execute fragmentation and compaction and to test whether these processes demand ( concerted ) actions of a multitude of myofilaments , we reduced the myofilament concentration to the single molecule level . Alexa-647 labeled myofilaments were added to MACs with low actin density and imaged by two-color TIRFM . Upon binding of single myofilaments to individual actin filaments , we observed directed movement of myofilaments along actin filaments and actin fragmentation at low ATP concentrations ( Figure 3A , Movie 4 , Table 1 , Figure 5 ) . 68% of the individually observed myofilaments displayed directed movement and 50% exhibited fragmentation and compaction of an actin filament ( total number of myofilaments = 152; 7 experiments ) . In cases where both fragmentation and compaction occurred , 95% of the observed myofilaments showed directed movement along the actin filaments , while 5% remained stationary at their original binding site ( total number of fragmenting myofilaments = 75; seven experiments ) . 10 . 7554/eLife . 00116 . 009Figure 3 . Single molecule analysis of the myofilament movement and actin fragmentation . ( A ) Dual-color TIRFM time-lapse sequence of a Alexa-647 labeled myofilament ( red ) moving along an Alexa-488-phalloidin labeled actin filament ( green ) . White asterisks mark the position of the myofilament . Yellow arrowheads point to actin filament deformations . White arrowheads indicate an increase in fluorescence intensity . Scale bars , 5 µm . ( B ) x ( grey curve ) and y ( red curve ) positions of the myofilament movement shown in ( A ) as a function of time . Inset depicts the trajectory ( green curve ) . ( C ) Myofilament velocity ( red curve ) calculated from the xy positions in ( B ) and actin filament intensity ( blue [raw data] and black [smoothed] curves ) over time . Red arrowheads denote acceleration events . Black arrows point to fluorescence intensity increases . Red arrowheads in ( A ) – ( C ) mark corresponding time points in ( A ) . ( D ) Proposed model for myofilament driven actin fragmentation and compaction ( details in text ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00116 . 00910 . 7554/eLife . 00116 . 010Movie 4 . Movement and actin fragmentation by a single myofilament . Example of a single Alexa-647 labeled myofilament during its directed movement along an Alexa-488-phalloidin labeled actin filament . During the movement the myofilament breaks and compacts the actin filament . Images were acquired at 200 ms intervals . The video contains 668 frames and is displayed at 60 fps . Total time: 2 . 2 min . Corresponds to Figure 3A . ( compressed avi; 1 . 3 MB ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00116 . 010 In contrast , at high ATP concentrations , processive movement of myofilaments was barely visible , and fragmentation of actin filaments was absent ( supplementary text in ‘Material and methods’ , Figure 5 , Movie 5 ) . Shortly after binding of myofilaments , actin filaments deformed and eventually broke , indicating that the force generated by a single myofilament is sufficient to break an actin filament ( Figure 3A , yellow arrowheads , Movie 4 ) . Subsequently , the fluorescence signal of actin filaments proximal to the breakage site increased ( Figure 3A , white arrowheads , Figure 3C ) . The detected increase in the fluorescence intensity of actin filaments implies that actin fragments are further dragged along the remaining actin filaments by the myofilaments during their movement ( Figure 3A , D , Movie 4 ) . We analyzed our TIRFM data in greater detail by tracking single myofilaments during their directed movement along an actin filament ( Figure 3B , ( Rogers et al . , 2007 ) ) and determining the velocity from the trajectories ( Figure 3C , red curve ) . Simultaneously , the fluorescence intensity in the green ( actin ) channel of the area occupied by the myofilament during its movement was measured , indicating breakage events as the dragged fragment leads to an increase of fluorescence proximal to the breakage site ( Figure 3C , blue and black curve ) . The velocity fluctuates during the movement of the myofilament along the actin filament ( Figure 3C , red curve ) , whereby acceleration events ( Figure 3C , red arrowheads ) are closely followed by an increase in the actin fluorescence , indicating that a breakage event occurred ( Figure 3C , black arrows , blue and black curve ) . We hypothesize that phases of increasing tension between the ends of the myofilament lead to deformation of the actin filament and coincide with a decrease in velocity , while the phases of tension release immediately after actin filament breakage may result in acceleration of the myofilament ( Figure 3C , D ) . In order to explain the buckling and eventual breakage of the actin filament , we propose the following model: The myosin filament aligns parallel to the actin filament and interacts with it via the myosin heads ( ( Sellers and Kachar , 1990 ) , Figure 3D ) . One end of the myofilament , which we will refer to as the ‘leading end’ , is oriented towards the actin plus end ( barbed end ) , as the filaments in a muscle , and can therefore walk along the actin filament while hydrolyzing ATP ( Figure 3D ) . The other end of the myofilament , which we will refer to as the ‘trailing end’ , separated from the leading end by the approximately 160 nm long bare zone ( Al-Khayat et al . , 2010 ) without myosin heads , is oriented in the opposite direction . While the trailing end still interacts with the actin filament , its opposite orientation results in a much slower directed motion towards the actin plus end ( Spudich et al . , 1985; Sellers and Kachar , 1990 ) . Every myosin head independently follows the biochemical cycle consisting of ATP hydrolysis , binding to actin , powerstroke accompanied by phosphate dissociation , ADP dissociation , ATP binding , and finally detachment from actin ( ( Howard , 2001 ) , Figure 8 ) . We assume that the heads on the trailing end still interact with the actin filament and go through the same cycle , but either do not perform steps that would lead to a processive motion ( probability of making a step pst=0 ) , or make steps with a small probability ( pst=0 . 1 ) . When a myosin head makes a step ( step size d = 5 nm ( Howard , 2001 ) ) it tries to move the whole myosin filament towards the actin plus end , but other attached myosin heads are holding it back , thus generating friction ( Figure 3D ) . The myofilament trailing end functions mainly as a source of friction ( an effective ‘brake’ ) , but is also moved towards the plus end by the pulling force exerted by the leading end ( Figure 3D ) . Since mainly the leading end is actively moving and the trailing end is either passively pulled ( pst=0 ) or contributes only weakly to its own motion ( pst=0 . 1 ) , tension builds up within the myofilament , and is transferred onto the actin filament as a compressive force . If sufficiently high , this compressive force can cause buckling , and finally breakage , of the actin filament ( Figure 3D ) . After the breakage , the leading end can move unhindered further towards the plus end , while dragging along the broken-off part of the actin filament attached to the trailing end ( Figure 3D ) . The trailing end can also attach again to the actin and further breakage events on the same filament can follow . In order to estimate the force needed to break an actin filament , we model the filaments as flexible rods with bending rigidity EI = 60 nN μm2 , determined from the persistence length of actin: lp = EI/ ( kT ) = 15 μm ( Yanagida et al . , 1984 ) . The force needed to buckle and break a filament is F = π2 EI/l2 . With the length l of the myofilament bare zone of 160 nm , this gives a force of 23 pN . Can the tension within the myofilament reach up to this force ? We performed simulations of this model , describing the attached myosin head as a spring with a spring constant equal to the myosin head stiffness κ = 1 pN nm−1 ( Kaya and Higuchi , 2010 ) , and equilibrate all forces after every step and every detachment of a myosin head . From the AFM image of the myofilament ( Figure 1C ) and the known myofilament structure ( Woodhead et al . , 2005 ) , we estimated that there are nm = 30 interacting myosin heads per filament . The result is a net movement of the whole myofilament towards the actin plus end , with fluctuating tension force , velocity and number of attached myosin heads , the mean values of which depend on the ATP concentration ( Figures 4A , B and 6 ) . We note that the tension force increases with decreasing ATP concentration , reaching forces necessary for breakage only at low ( lesser than ∼3 µM ) ATP concentrations , in agreement with the experiments ( Figure 4A , see also Table 1 ) . When the actin filament was allowed to bend in simulations at the point where the buckling force of 23 pN was reached ( Figure 4B ) , thus releasing the excess tension , the filament curvature steadily increased at low ATP concentrations , reaching the critical curvature of breaking of 5 . 6 µm−1 ( ( Arai et al . , 1999 ) , Figure 7A ) . At intermediate ATP concentrations , the actin curvature fluctuated , at times exceeding the curvature threshold ( Figures 4C and 7B ) . At higher ATP concentrations , the threshold force was reached for too short periods for the filament to be bent to the breakage point ( Figures 4D and 7C ) . The results do not depend significantly on whether the myosin heads on the trailing end perform steps ( pst=0 . 1 ) or not ( pst=0 ) . The simulations thus support the idea that in the ATP concentration range used in the experiments , the differences in the interactions of the trailing and leading ends of the myofilament with actin can generate compressive forces on the actin filament , and that these forces are sufficiently high to bend and break the actin filament . 10 . 7554/eLife . 00116 . 011Figure 4 . Simulation of the interaction between myofilaments and an actin filament . ( A ) Mean tension force F within the myofilament when bending of actin is not allowed; dependence on ATP concentration for several different numbers of interacting myosin heads nm . The forces when the myosin heads of the trailing end are not performing steps ( pst=0 , points connected by a solid line ) are slightly higher than the forces when the steps occur with the probability pst=0 . 1 ( dashed lines ) . ( B ) Mean tension force when the actin filament is allowed to bend at the threshold force of 23 pN ( points and solid line: pst=0 , dashed line: pst=0 . 1 ) . ( C ) Actin filament curvature fluctuations during 20 s of the simulation at 0 . 0025 mM ATP concentration , showing that the critical curvature of 5 . 6 µm−1 needed for actin filament breakage is often reached , while at higher ATP concentration ( 0 . 01 mM ) , the critical curvature is never reached ( D ) ( pst=0 in ( C ) and ( D ) ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00116 . 01110 . 7554/eLife . 00116 . 012Figure 5 . Trajectories of individual myofilaments moving along actin filaments . Left panel displays the trajectories of myofilaments at low ( 1 µM ) ATP concentration ( see also the corresponding Movie 5 ) . Right panel shows trajectories of myofilaments at high ( 4 mM ) ATP concentration ( see also the corresponding Movie 5 ) . Myofilaments were tracked for one minute and those who stayed less than 900 ms attached to the actin filament were filtered out ( Rogers et al . , 2007 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00116 . 01210 . 7554/eLife . 00116 . 013Movie 5 . Behavior of single myofilaments at low and high ATP concentrations . The video shows the binding and moving behavior of Alexa-647 labeled myofilaments added to medium actin density MACs at low ( 1 µM ) and high ( 4 mM ) ATP concentrations respectively . The white bar in the middle separates the movies of low ATP concentration ( left ) and high ATP concentration ( right ) . Note the comparatively higher number of myofilaments that are bound to actin filaments and exhibit movement at the low ATP concentration ( left ) . By contrast at high ATP concentration ( right ) bound myofilaments are less frequently visible and do not move . Images were acquired at 300 ms intervals . The video contains 200 frames and is displayed at 15 fps . Total time: 1 min . The video was used for the tracking data presented in Figure 5 . ( compressed avi; 3 . 3 MB ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00116 . 01310 . 7554/eLife . 00116 . 014Figure 6 . Dependence of the mean velocity and mean number of attached myosin heads on the ATP concentration obtained from the simulations of myofilament–actin filament interaction . ( A ) Mean velocity ( points and solid line: pst=0 , dashed line: pst=0 . 1 ) . ( B ) Mean fraction of attached myosin heads . DOI: http://dx . doi . org/10 . 7554/eLife . 00116 . 01410 . 7554/eLife . 00116 . 015Figure 7 . Fluctuations of the compression force ( upper row ) , the actin curvature ( middle row ) and the fraction of attached myosin heads ( lower row ) during 20 s of the simulation , for three different ATP concentrations ( pst=0 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00116 . 015
In our in vitro study we provide a potential mechanism how actin turnover in cells may be mediated by myofilament driven actin fragmentation . To bend and break the actin filament , the myofilament has to be attached to the actin for sufficient time , and sufficient force has to be generated , requiring a certain mean number of heads being attached to actin at any given time . Given the small number of interacting myosin heads this translates to a requirement that a large fraction of heads is in the bound state . In our assay this was achieved with low ATP concentrations . It is tempting to speculate that actin breakage by myofilaments in vivo is governed by the ATP level in the cell . Recent evidence exists that ATP levels indeed vary significantly inside living cells ( Imamura et al . , 2009 ) . However , physiological ATP concentrations , derived from methods that provide averaged ATP levels with no high spatial and temporal resolution from cell extracts , are usually found in the millimolar range ( Beis and Newsholme , 1975 ) . Here it is important to mention that skeletal muscle myosin II ( used in this study ) has a lower duty ratio and is therefore less processive than non-muscle myosin in cells ( Harris and Warshaw , 1993; Wang et al . , 2003 ) . By lowering the ATP concentration in our assay , we increased the duty ratio of our myofilaments and thereby made them more processive ( supplementary text in ‘Material and methods’ , Figure 5 , Movie 5 ) , similar to non-muscle myosin , which is in line with previous studies ( Humphrey et al . , 2002; Smith et al . , 2007; Soares e Silva et al . , 2011 ) . Moreover , processivity of myofilaments in animal cells may be also controlled by phosphorylation of myosin light chains through other proteins that modify the kinetic rates of the biochemical cycle and thereby increase the duty ratio of the myosins shifting the observed behavior to a region of high physiological ATP concentrations ( Tan et al . , 1992; DeBiasio et al . , 1996; Matsumura et al . , 2001 ) . Our simulations show that the model of interaction between the actin and myosin filaments described in the text is plausible given the quantitative parameter values known from literature ( rate constants ) . Most importantly , it shows that sufficient force needed to bend and break the actin filament can be generated by a single myofilament , assuming only the difference in the interaction between the leading and trailing myofilament ends ( making a step vs not making a step ) . This result implies that neither actin nor myosin attachment to a support or any other rigid structure is necessary for the observed process . The forces act within a single myofilament . Binding of the actin filaments to the membrane provides confinement of the filaments to a plane , without rigidly tethering the filaments to a support or a scaffold . The motion of actin is only restricted by the effectively higher viscosity of the membrane compared to the buffer solution . The simulation further yields quantitative details of the model , for example , the mean fractions of myosin heads in the six possible states ( Figure 9 ) , and describes the fluctuations of the actin filament curvature ( Figure 4C , D ) . Other parameters that can be derived from the simulations and could be compared with new experiments include the processivity of myosin , as a function of the ATP concentration , and the dependence of the overall behavior on the length of myofilaments . 10 . 7554/eLife . 00116 . 016Figure 8 . The biochemical cycle of the myosin heads with rates k1–k6 assumed in the model and the simulations . The rate k5 is ATP-dependent . DOI: http://dx . doi . org/10 . 7554/eLife . 00116 . 01610 . 7554/eLife . 00116 . 017Figure 9 . The fractions of myosin heads in states 1–6 , and in the actin-bound state ( sum of states 3–6 ) in dependence on the ATP concentration . The values are calculated from the model of the myosin head cycle in Figure 8 . DOI: http://dx . doi . org/10 . 7554/eLife . 00116 . 017 In conclusion , our findings show the distinct functions that myosin motors can execute . In our minimal system , myofilaments fragment and compact membrane-bound actin . We directly show that single myofilaments can interact with actin in such a way that sufficient force can be generated to break the filament , without either the myofilament or the actin being firmly attached to a solid support or scaffold . We suggest that fragmentation and compaction by myofilaments contributes to the observed large-scale pattern formation of actomyosin networks also in other in vitro systems ( Backouche et al . , 2006; Smith et al . , 2007; Soares e Silva et al . , 2011; Gordon et al . , 2012 ) . In vivo breakage of actin bundles has been shown to occur in neuronal growth cones in a myosin II dependent manner , and is thought to be important for recycling actin ( Medeiros et al . , 2006 ) . We propose that the observed fragmentation and compaction of membrane-bound actin filaments by myosins in our in vitro system may serve as a possible general mechanism for actin turnover and actin cell cortex remodeling .
Traditionally , single muscle myosin II motors are described as non-processive motors ( Howard , 2001 ) . By assembling myosin motors into filaments , the myofilaments become processive due to the coupling of a higher number of myosin heads that are in contact with the actin filament . In our minimal system directed movement accompanied by actin filament fragmentation only occurred at low ATP concentrations between 0 . 1–1 µM ( Figure 3 , Table 1 , Movie 3 ) . The lower ATP concentrations are expected to increase the duration of the actin-bound post-working stroke state of the myosin head , thus increasing the duty ratio and processivity ( Howard , 2001 ) . Extremely low ATP concentrations are on the other hand not sufficient for myosin motor activity . To check the effect of ATP concentration on processivity we added Alexa-647 labeled myofilaments to medium density MACs and tracked ( Rogers et al . , 2007 ) the movement of individual myofilaments at low ( 1 µM ) and high ( 4 mM ) ATP concentrations ( Figure 5 , Movie 5 ) . ( Note that fragmentation and compaction of actin filaments only occurred at low ATP concentrations ) . Comparison of the trajectories illustrates that low ATP trajectories are on average longer than high ATP trajectories indicating a higher processivity at low ATP conditions ( Figure 5 , Movie 5 ) . Moreover the number of tracked myofilaments at low ATP concentration with a dwell time greater than 900 ms was more than six times higher than at high ATP concentration suggesting a higher duty ratio at low ATP levels ( n = 681 at low ATP , n = 102 at high ATP; Figure 5 , Movie 5 ) . High ATP concentrations therefore lead to a faster detachment of myofilaments from actin filaments than at low ATP concentrations ( Movie 5 ) . We propose that the increase in processivity due to ATP deprivation is necessary for the processive movement and a prerequisite for fragmentation and compaction of actin filaments . | Actin is a multi-functional protein that is found in almost all eukaryotic cells . When polymerized , it forms robust filaments that participate in a variety of cellular processes . For example , actin filaments are involved in the contraction of muscles , and they are also a major component in the various structures that maintain and control the shape of cells as they move and divide . These structures include the cell cortex , a meshwork of actin filaments that is bound to the inner surface of the plasma membrane by anchor proteins . However , both the cell cortex and the plasma membrane must undergo dramatic changes when a cell divides , and the forces that drive these changes are generated by another protein , myosin II . Myosin II contains three domains: a head domain , also known as the motor domain , that binds to actin; a neck domain; and a tail domain . Like actin , myosin II proteins also form filaments , but these myofilaments have a distinctive structure: the tail domains of two Myosin II proteins join together , with the motor domains being found at both ends of the filament . When activated , the motor domains grab actin filaments and pull against them in a ‘powerstroke’ . However , the details of the interactions between the myofilament motor domains and the actin filaments in the cell cortex , which are bound to the plasma membrane , are not fully understood . Studying these processes in living cells is extremely challenging , so Vogel et al . have built an in vitro model of the cell cortex , and then used single-molecule imaging to watch the interactions between the myofilaments and the actin filaments in this model . They show that the myofilaments move along the actin in the cortex , breaking up the filaments and compressing them in the process . They propose that tension builds up between the ends of the myofilaments , leading to compressive stress being exerted on the actin filaments . Computer simulations confirm that the forces generated are high enough to cause the actin filaments to buckle and break . The in vitro model developed by Vogel et al . should allow researchers to clarify the basic biophysical principles that underpin the structure and function of the cell cortex . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Material",
"and",
"methods"
] | [
"structural",
"biology",
"and",
"molecular",
"biophysics"
] | 2013 | Myosin motors fragment and compact membrane-bound actin filaments |
Self-esteem is shaped by the appraisals we receive from others . Here , we characterize neural and computational mechanisms underlying this form of social influence . We introduce a computational model that captures fluctuations in self-esteem engendered by prediction errors that quantify the difference between expected and received social feedback . Using functional MRI , we show these social prediction errors correlate with activity in ventral striatum/subgenual anterior cingulate cortex , while updates in self-esteem resulting from these errors co-varied with activity in ventromedial prefrontal cortex ( vmPFC ) . We linked computational parameters to psychiatric symptoms using canonical correlation analysis to identify an ‘interpersonal vulnerability’ dimension . Vulnerability modulated the expression of prediction error responses in anterior insula and insula-vmPFC connectivity during self-esteem updates . Our findings indicate that updating of self-evaluative beliefs relies on learning mechanisms akin to those used in learning about others . Enhanced insula-vmPFC connectivity during updating of those beliefs may represent a marker for psychiatric vulnerability .
A positive sense of the self is the bedrock of mental health and well-being ( Orth et al . , 2012; Trzesniewski et al . , 2006 ) . Low self-esteem is a vulnerability factor for a range of psychiatric problems , including eating disorders ( Button et al . , 1996; Vohs et al . , 2001 ) , anxiety disorders ( Sowislo and Orth , 2013 ) and depression ( Orth et al . , 2008; Orth et al . , 2009 ) . Classical theories in psychology view self-esteem as an internalization of actual and imagined appraisals from close others across development ( Cooley , 1902; Leary et al . , 1995; Mead , 1934 ) . Indeed , an enduring sense of self-worth ( referred to as ‘trait self-esteem’ ) reflects an accumulation of past appraisals from others ( Cole et al . , 2001; Felson and Zielinski , 1989; Gruenenfelder-Steiger et al . , 2016; Harter , 1983; Ladd and Troop-Gordon , 2003 ) , while momentary feelings of self-worth ( ‘state self-esteem’ ) are highly responsive to positive and negative social evaluations ( Denissen et al . , 2008; Gerber and Wheeler , 2009; Thomaes et al . , 2010 ) . Despite its importance for mental health , we lack a mechanistic understanding of how self-esteem depends on social evaluation . Here , using a novel social evaluation task , in combination with computational modeling and functional magnetic resonance imaging ( fMRI ) , we characterize computational and neural processes underpinning changes in self-esteem . A candidate neural substrate for integrating social evaluation with self-evaluation is ventromedial prefrontal cortex ( vmPFC ) , given evidence that being evaluated by others ( Dalgleish et al . , 2017; Gunther Moor et al . , 2010; Somerville et al . , 2010 ) , and evaluating the self ( Chavez et al . , 2016; D'Argembeau et al . , 2012; Hughes and Beer , 2013; Kelley et al . , 2002 ) activates subregions of vmPFC . The vmPFC comprises multiple distinct cytoarchitectonic zones including parts of the anterior cingulate cortex ( ACC; e . g . , Brodmann areas 24 and 32 ) and orbitofrontal cortex ( OFC; e . g . , Brodmann areas 11 , 13 , 14 ) ( Haber and Knutson , 2010 ) . These subregions have distinct connectivity profiles ( Neubert et al . , 2015 ) and are thought to show specialization for cognitive functions important in valuation ( Bartra et al . , 2013; FitzGerald et al . , 2009; Lebreton et al . , 2015 ) , social learning ( Apps et al . , 2016; Hill et al . , 2016; Suzuki et al . , 2012 ) and their intersection ( Apps and Ramnani , 2017; Garvert et al . , 2015; Nicolle et al . , 2012 ) . A recent study showed that the perigenual ACC ( pgACC ) tracks a history of one’s own success and failures in a social context , while dorsomedial prefrontal area 9 tracked the performance history of an interaction partner ( Wittmann et al . , 2016 ) . Crucially , activity in pgACC was higher in individuals whose subjective evaluation of their own performance was affected more strongly by their actual performance , making this region a candidate substrate for online updating of self-esteem . Value representations in vmPFC are updated by teaching signals in the form of reward prediction errors that mediate an effect on vmPFC ( Frank and Claus , 2006; Garvert et al . , 2015; Hampton et al . , 2006; Jocham et al . , 2011; Pasupathy and Miller , 2005 ) . By applying a computational modeling approach that has been shown to explain changes in subjective well-being during value-based decision-making ( Rutledge et al . , 2014; 2015 ) , we test whether self-esteem is dynamically updated during social evaluations through the cumulative impact of ‘social approval prediction errors’ ( i . e . , the difference between expected and received social feedback ) . More specifically , using fMRI , we could test whether dynamic self-esteem updates are reflected in vmPFC activity , and whether individual variation in neural encoding of social approval prediction errors ( SPEs ) explains inter-individual differences in self-esteem . Through combining computational parameters with psychiatric symptoms in a single multivariate analysis , we identified a dimension of ‘interpersonal vulnerability’ . Vulnerability , so defined , was associated with enhanced SPEs in anterior insula and greater insula-vmPFC functional connectivity during self-esteem updates , suggesting potential biomarkers for psychiatric vulnerability .
We scanned 40 participants ( mean age = 23 . 4 , SD = 3 . 3 , 14 male ) while they performed a newly designed social evaluation task . In this task participants received approval and disapproval feedback . This feedback was ostensibly derived from 184 strangers who had viewed an online character profile each participant had uploaded to an online database , one week prior to the experiment ( see Supplementary file 1 for details ) . Participants’ expectations about feedback were manipulated by sorting raters into four groups based on their ostensible overall approval rates toward all participants in the experiment . All participants received approval feedback in 85% , 70% , 30% , and 15% of the trials , spanning the first to fourth quartile rater groups respectively ( see Figure 1 ) . Participants were not informed about these exact probabilities , but to better orientate them they learned the rank ordering of the four groups prior to performing the task . On each trial , subjects were provided with the name of a rater and a color cue that indicated the rater’s group ( see Figure 1 ) . They were then asked to predict whether the rater liked or disliked them . After a delay , they received approval ( in the form of a thumbs up symbol ) , or disapproval ( in the form of a thumbs down symbol ) feedback . After every 2–3 choice trials , participants reported their current level of self-esteem by answering how good they felt about themselves at that exact moment using a visual analogue scale ( see Materials and methods for further details on the task ) . First , we established that participants adapted their predictions about being liked based on a rater’s group membership . A repeated-measures ANOVA with group ( 4 levels: 85% , 70% , 30% and 15% profiles liked ) as a within-subjects factor and percentage predictions of being liked as a dependent variable showed a main effect of group , F ( 3 , 117 ) = 209 . 47 , p<0 . 001 , ηp2=0 . 843 . Pairwise comparisons showed that participants predicted they would be liked more by raters from group 1 ( 95% ) than group 2 ( 87% , p<0 . 001 ) , more by raters from group 2 than 3 ( 30% , p<0 . 001 ) and more by raters from group 3 than 4 ( 13% , p<0 . 001; see Figure 2A ) . These choice patterns were reliably predicted by a Rescorla-Wagner reinforcement-learning model ( Rescorla and Wagner , 1972 ) fitted to participants’ choice behavior ( mean r2 = 0 . 40 ± 0 . 27; mean ± SD; see Materials and methods for details on this model ) . The modeling results were consistent with participants using SPEs ( the difference between received feedback and expected social feedback ) , weighted by a learning rate , to update their expectations about approval from raters from each of the four groups ( ESV ) , which in turn guided their predictions about being liked ( see Figure 2 ) . The model includes a bias parameter ( ESV0 ) that captures persistent beliefs about the probability of being liked or disliked . Participants with a larger ESV0 persisted in predicting they would be liked by raters from groups for whom they had a negative ESV ( e . g . , raters from groups 3 or 4 ) . Next , we tested the hypothesis that dynamic changes in self-esteem depend on both the valence of social feedback and expectations about feedback . If SPEs explain changes in self-esteem better than outcome valence alone , trial-by-trial changes in self-esteem should correlate positively with outcome valence and negatively with expectations ( Behrens et al . , 2008 ) . We found that , after regressing out the positive effect of outcome valence ( r = 0 . 18 , p<1×10−9 ) , there remained a significant negative correlation with expectations ( r = −0 . 06; p=0 . 012 ) ( see Figure 2B ) . We formally modeled the cumulative impact of SPEs on moment-to-moment variation in self-esteem using exponential kernel regression models . Parameters were fit to both choice behavior and self-esteem ratings in individual participants . Our winning model ( Equation 1 ) successfully captured dynamic changes in self-esteem at the level of the individual ( r2 = 0 . 32 ± 0 . 24; mean ± SD; see Figure 2 ) . We chose this model as it outperformed a range of alternative plausible models , including a model that accounted for the valence of social feedback , but did not feature expectations about approval ( ‘Outcome valence only’ model 6; see Table 1 ) . ( 1 ) Self-esteem ( t ) =w0+w1∑j=1tγt−jSPEj+ϵ For each trial ( t ) , we entered a term for baseline self-esteem throughout the task ( w0 ) and a term capturing the weight of SPEs ( w1 ) into the equation . Expectations about social approval were estimated using the previously mentioned reinforcement-learning model . Through the inclusion of a forgetting factor ( γ ) the influence of SPEs was allowed to decay exponentially in time , such that recent events had greater impact than earlier events . The Gaussian noise term ε ~N ( 0 , σ ) allowed Equation 1 to serve as a generative model of self-esteem . The average learning rate η ( involved in updating expectations about approval from the raters ) was 0 . 04 ± 0 . 07 ( mean ± SD ) and the average forgetting factor γ ( involved in updating self-esteem ) was 0 . 65 ± 0 . 35 ( mean ± SD; see Table 2 for means and standard deviations of all model parameters ) . This indicates that SPEs induce rapid changes in feelings about the self , but impact learning about the probability of approval from the four groups relatively slowly . Bayesian model comparison showed our favored model outperformed alternative models ( see Table 1 ) , including: ( 1 ) models without a response bias parameter ( ESV0 ) which captured persistent beliefs about the probability of being approved or disapproved , ( 2 ) models without learning ( including a model where participants had correct initial beliefs about approval expectations for the four groups that were not dynamically updated based on feedback ) , ( 3 ) a model with a separate expectations term to test whether expectations have additional effect on self-esteem , above and beyond their effect captured by the SPE term and ( 4 ) a model that only took the valence of feedback into account , but did not feature expectations about approval ( see methods for details on models ) . This comparison provides support for an hypothesis that self-esteem is sensitive to a cumulative impact of recent prediction errors arising out of expectations concerning social approval . Furthermore , these expectations are not stable , but are dynamically updated and depend on persistent beliefs about approval . Finally , changes in self-esteem are better described by a model without a separate effect of expectation , indicating that the effect of expectation on self-esteem – unlike in the case of mood ( Rutledge et al . , 2014; 2015 ) - only operates through prediction errors realized at the moment that feedback is delivered . Results from a control experiment demonstrated that the observed self-esteem changes in the fMRI task are specific to situations where the self is the object of evaluation and are unlikely to be the result of demand characteristics ( see Figure 2—figure supplement 1 ) . To examine relationships between computational self-esteem parameters and symptoms linked to low self-esteem we performed a canonical correlation analysis ( CCA; Hair et al . , 1998 ) . CCA finds the maximal correlation between a linear combination of one set of variables ( in our case self-esteem parameters from our computational model ) and a linear combination of another set ( in our case symptoms linked to low self-esteem and interpersonal sensitivity measured using questionnaires; see Materials and methods for details ) . The CCA yielded one significant canonical dimension ( Wilks’s λ = 0 . 01 , F ( 99 , 152 . 5 ) = 1 . 40 , p=0 . 029 ) , which had a canonical correlation of 0 . 87 between computational parameters and symptoms . We labeled the dimension as ‘interpersonal vulnerability’ based on the constellation of positive and negative associations of the different computational parameters and symptom measures with the identified dimension ( see Table 2 and Figure 3 ) . With respect to symptoms , trait and state self-esteem showed a strong negative association with ‘interpersonal vulnerability’ . Symptoms of depression , social anxiety , and trait and state anxiety showed a positive association with ‘interpersonal vulnerability’ . As for the computational parameters , baseline self-esteem ( w0 ) and average initial approval beliefs showed a negative association with ‘interpersonal vulnerability’ . Weight on SPEs ( w1 ) and the range of initial approval beliefs showed a positive association with ‘interpersonal vulnerability’ . The results highlight that people with lower self-esteem and greater anxiety and depression symptoms have lower expectations about approval and greater self-esteem fluctuations in response to SPEs . We first examined encoding of SPEs in a whole-brain regression analysis with trial-by-trial SPEs ( inferred using our computational model and time-locked to feedback onset ) as a parametric modulator . This analysis revealed SPEs correlated with activity in a cluster in bilateral ventral striatum extending into subgenual anterior cingulate cortex ( sgACC; BA 25 ) ( see Figure 4A; left peak coordinates −8 , 21 , –5; t ( 39 ) = 4 . 50; right peak coordinates 5 , 20 , –8 , t ( 39 ) = 5 . 42; Z = 4 . 65 , k = 1172 , p=0 . 005 , Family-wise Error [FWE] cluster-corrected ) . Next , we tested whether neural activity at feedback reflected self-esteem updates contingent on feedback . We regressed activity at feedback presentation against trial-by-trial updates in self-esteem ( i . e . , inferred using our computational model ) . This analysis revealed a significant cluster in ventromedial prefrontal cortex ( vmPFC ) with a peak in left medial OFC ( Brodmann Area [BA] 14m ) extending into pgACC ( BA 32pl ) ( see Figure 4B; peak coordinates −6 , 33 , –15 , t ( 39 ) = 3 . 83; Z = 3 . 51 , k = 868 , p=0 . 047 , FWE cluster-corrected ) . Our next analyses focused on testing whether individual differences in ‘interpersonal vulnerability’ are reflected in neural representations of SPEs and self-esteem updates . We used our CCA results to obtain subject-specific scores on the ‘interpersonal vulnerability’ dimension . These scores are the result of a new weighting of symptoms that maximally correlated with computational self-esteem parameters . First , we ran a whole-brain regression analysis with trial-by-trial SPEs as parametric modulator and ‘interpersonal vulnerability’ scores as a between-subject regressor to test which brain regions responded more strongly to SPEs in individuals who are more vulnerable relative to those who are less vulnerable . This revealed a positive association between ‘interpersonal vulnerability’ and activity in a left anterior insula cluster extending into inferior frontal gyrus ( see Figure 4C; peak coordinates −44 , 11 , 9; t ( 38 ) = 4 . 70; Z = 4 . 15 , k = 5463 , p<0 . 001 , FWE cluster-corrected ) . Our behavioral finding of a positive association between vulnerability and self-esteem updating in response to SPEs , motivated us to test an hypothesis that vulnerability-related variation in self-esteem updating is mediated through variation in functional connectivity between the vmPFC and the insula region where SPE-related activity was modulated by vulnerability . Consequently , we examined functional connectivity between the insula ( using a 6 mm sphere around the peak of the cluster from the previous analysis as a seed region ) and the rest of the brain during self-esteem updates using a psychophysiological interaction ( PPI ) analysis ( Friston et al . , 1997; O'Reilly et al . , 2012 ) . This connectivity analysis showed that ‘interpersonal vulnerability’ correlated positively with functional coupling between the insula and a cluster in vmPFC with a peak in right medial OFC ( BA 14m ) during self-esteem updates ( see Figure 4D; peak coordinates 11 , 32 , –11; t ( 38 ) = 6 . 27; Z = 5 . 16 , k = 78570 , p<0 . 001 , FWE cluster-corrected ) . Interpersonal vulnerability did not correlate with SPE-related activity in the striatum/sgACC cluster ( Spearman’s ρ = 0 . 238 , p=0 . 140 ) or updating-related activity in the vmPFC cluster ( Spearman’s ρ = −0 . 005 , p=0 . 978 ) . Thus , greater interpersonal vulnerability ( i . e . , more symptoms and amplified self-esteem parameters ) is associated with both increased SPE responses in anterior insula and greater functional connectivity between the insula and the vmPFC during self-esteem updates . Together these results hint at potential mechanisms for vulnerability to psychiatric illness .
Self-esteem is shaped by what other people think of you ( Cooley , 1902; Denissen et al . , 2008; Gerber and Wheeler , 2009; Leary et al . , 1995; Mead , 1934; Thomaes et al . , 2010 ) . Our findings reveal how this form of social influence on self-esteem is implemented in the brain . Our computational modeling results are consistent with people using prediction errors to learn what to expect from others and to update their self-esteem based on the outcome of these expectations . Using fMRI , we show that SPEs correlate with activity in ventral striatum and the sgACC , while self-esteem belief updates are reflected in vmPFC activity . The findings highlight that learning from social evaluative feedback and updating self-evaluative beliefs rely on learning mechanisms as seen in social and non-social reward learning at both an algorithmic ( i . e . , prediction error driven ) and neural level ( i . e . , shared neural substrates in the striatum , ACC , and vmPFC ) . Self-esteem has characteristics akin to a ‘gauge of social acceptance’ as articulated within ‘Sociometer theory’ ( Leary et al . , 1995 ) . Our study confirms predictions made by sociometer theory and places this notion within a quantitative and neurobiologically grounded framework . A notable feature of our data is that it indicates that self-esteem is not a sociometer that merely maintains an on-going tally of social acceptance , but is more akin to a read-out of the extent to which our social standing has undergone change recently . Our data indicate that forming accurate expectations about rejection exerts a buffering effect against expected rejection , fitting observations that self-esteem is more volatile in individuals with aberrant expectations about rejection ( Dandeneau and Baldwin , 2004; Leary et al . , 1995 ) . These findings lead to a new testable hypothesis , namely that self-esteem encompasses a form of learning signal that we use to gauge our social standing in new environments . Our winning computational model entailed a mean forgetting rate such that state self-esteem depended on the most recent six appraisals acquired from raters . In contrast , people’s beliefs about the global social milieu ( related to estimates about how approving people are in the current social environment ) accumulated slowly and were much more resistant to change . Thus , it appears that people use SPEs as an estimate of the local gradient of social approval that informs their experienced self-esteem , as well as for slowly updating beliefs about expectations of approval from the global social milieu . In this way , changes in self-esteem can be thought of as learning about the self , which is distinct but related to learning about others . Learning about social approval not only has algorithmic similarity to learning about non-social stimuli ( Montague et al . , 1996; Schultz , 2013; Sutton and Barto , 1998 ) , but also depends on similar neural circuitry . SPEs correlated with activity in a cluster including the ventral striatum and the sgACC . The striatum encodes prediction errors in learning about primary ( D'Ardenne et al . , 2008; Hart et al . , 2014 ) and secondary rewards ( Caplin et al . , 2010; Pessiglione et al . , 2006 ) , including learning about social acceptance ( Jones et al . , 2011; Jones et al . , 2014 ) . While the ventral striatum has been shown to process both prediction errors about rewards for the self and other people , there is also evidence that sgACC exclusively encodes prediction errors about rewards for other people ( Lockwood et al . , 2016 ) . Consistent with this notion , a recent social learning study showed that ventral striatum processes prediction errors about the accuracy of another person’s advice , while a region in the sgACC/septum encodes prediction errors about the person’s general level of trustworthiness ( Diaconescu et al . , 2017 ) . The social approval prediction errors in our task drive changes in self-esteem in response to social rewards for self and at the same time guide learning about other people’s general level of ‘niceness’ . The presence of a cluster spanning both striatum and sgACC in response to this multiplexed prediction error signal dovetails with these prior findings . A goal for future research is to disentangle the different contributions of ventral striatum and sgACC during learning about the self through interactions with others . Based on prior work showing the influence of secondary reward ( i . e . , monetary ) prediction errors on mood ( Rutledge et al . , 2014; 2015 ) , it is likely that our experimentally induced SPEs also affect mood . Like mood , self-esteem depends on expectations that lead to prediction errors . However , model comparison demonstrated a key difference between an established model of momentary variation in mood and the self-esteem model we introduce here . Mood increases at the moment in time when subjects know they may gain future rewards , even in the absence of feedback ( Rutledge et al . , 2014; 2015 ) . Our results show that self-esteem does not increase when individuals are in an environment with socially accepting others and , unlike for mood , model comparison does not support self-esteem models with a separate expectation term . Given a wealth of studies on neurocomputational mechanisms supporting valuation ( Bartra et al . , 2013; Rangel et al . , 2008 ) , it is surprising how little is known about the computations supporting the most fundamental type of valuation , namely the evaluation of our own worth . Our results show that updates of self-worth are represented in a cluster in vmPFC with a peak in medial OFC ( BA 14m ) , extending into pgACC ( BA 32pl ) . Activity in the pgACC has been shown to increase and decrease in response to recent success and failure of the self , but not of others , in joint decision-making tasks ( Wittmann et al . , 2016 ) . Our results extend the role of the pgACC in self-related processing by showing that this region not only keeps track of how well one is performing in a given task , but that it continuously updates a more general value ascribed to the self when learning how others value us . Neurons in adjacent subregion BA 14m show correlated tuning for reward size and reward probability in monkeys , suggesting that BA 14m neurons encode an integrated value signal ( Strait et al . , 2014 ) . This is consistent with a large-scale meta-analysis of 81 human fMRI studies showing that activity in BA 14m correlates with subjective value ascribed to a range of primary and secondary rewards , both upon receipt of a reward and during choice formation ( Clithero and Rangel , 2014 ) . Adjacent subregions in vmPFC may thus integrate separate strands of information about current value assigned to the self in order to estimate self-value in the future . Our results suggest that social approval may act on self-value representations in a manner similar to the effects of primary and secondary rewards on value representations about external stimuli . As in the latter , updates in self-esteem upon receipt of feedback co-varied with activity in vmPFC akin to updates of value attributed to external stimuli ( Behrens et al . , 2008; FitzGerald et al . , 2009; Rushworth et al . , 2011 ) , strengthening the idea that self-evaluation may be reducible to valuation , but where now the object is the self ( D'Argembeau , 2013 ) . In this light the social modulation of a self-esteem representation in vmPFC is also consistent with findings showing that the vmPFC integrates social with personal preferences to compute a new value of an object based on to the opinions of other people ( Campbell-Meiklejohn et al . , 2017 ) . By combining computational self-esteem parameters with measures of psychiatric symptomatology in a single multivariate analysis , we identified a dimension of ‘interpersonal vulnerability’ . Vulnerability was associated with low self-esteem , internalizing symptoms and self-esteem instability in response to SPEs . This was mirrored at a neural level by augmented SPE processing in anterior insula and a greater positive functional connectivity between the insula and a cluster in vmPFC with a peak in BA 14m . This suggests self-esteem instability may result from a greater malleability of self-value representations in vmPFC driven by prediction error signals arising from the insula . The location of the cluster in the insula shows striking overlap with findings from social anxiety patients during reappraisal of negative self-beliefs ( Goldin et al . , 2009 ) as well as findings in a range of anxiety disorders ( Etkin and Wager , 2007 ) . As such , increased responsivity to SPEs and greater insula-vmPFC coupling during self-esteem updates may represent neurobiological markers of a dimension of ‘interpersonal vulnerability’ that confers increased risk for a number of common mental health problems . A question for future research is whether psychiatric patients , especially those suffering from internalizing disorders like depression or anxiety , fall at the extreme end of the ‘interpersonal vulnerability’ dimension that we identified . Such thinking is at the core of the Research Domain Criteria ( rDOC; Insel et al . , 2010 ) which aims at re-conceptualizing psychiatric nosology by identifying dimensions of biologically plausible trans-diagnostic markers . Our approach allows identification of a new weighting of questionnaire measures with sensitivity to individual differences in neural processes relevant to rapid changes in self-esteem . This questionnaire weighting , which relates to both our new self-esteem computational model and the neural responses to social feedback processing , might index risk of future mental health outcomes . We demonstrate that state self-esteem can be conceptualized as a self-value representation in vmPFC , a representation that is dynamically updated through prediction errors resulting from violations of expectations about evaluative feedback . Inter-individual variation on a symptom dimension that cuts across traditional diagnostic categories mapped closely to indiviual differences in insula responses to social feedback and insula-vmPFC coupling during self-esteem updates . Our framework thus reveals fundamental mechanisms that underlie how we use social information when evaluating ourselves and holds promise as a trans-diagnostic predictor of psychiatric outcomes .
We recruited forty-four participants through participant pools at University College London ( UCL ) . Sample size was based on prior fMRI studies examining inter-individual differences in social feedback processing ( Powers et al . , 2013; Somerville et al . , 2010 ) . Exclusion criteria included a prior history of head injury , neurological or psychiatric disorder , color blindness , or being left-handed . We monitored participants using an eye tracker and we excluded participants who were shown to have fallen asleep during scanning ( n = 4 ) . The target sample comprised 40 participants ( mean age = 23 . 3 , SD = 3 . 2 , 14 male ) who were paid a fee of £8 per hour plus earnings based on an additional experiment after the MRI scans ( Dictator Game; see Figure 2—figure supplement 2 ) . Informed consent was obtained from every participant and experimental procedures were approved by the local research ethics committee . Participants were invited to come to the lab seven days prior to the MRI experiment to create and upload a personal profile ( see Supplementary file 1 ) . As part of the cover story for the experiment , we showed them an online database and explained that we needed several days to receive a sufficient number of evaluations , and as a consequence the scanning session would take place seven days later . During this first session , participants also filled out a battery of questionnaires ( see below ) . The MRI session included a training part during which participants learned , and were tested on , the structure of the fMRI task before going into the scanner . Subsequent to scanning they were given a set of additional experimental tasks ( see Figure 2 – figure supplement 1 and 2 ) . Participants performed a new social evaluation task , which was inspired by existing paradigms ( Eisenberger et al . , 2011; Gunther Moor et al . , 2010; Somerville et al . , 2006 ) . In this task they received feedback indicating ostensible approval or disapproval from 184 strangers ( ‘raters’; 92 males and 92 females ) . Participants were told that raters were sorted into four groups based on their overall approval rates toward all participants in the experiment . On each trial , participants were presented with the name of a rater and a color cue that indicated the rater’s group , assigned according to overall approval rates ( see Figure 1 ) . After a jittered fixation display ( uniformly distributed between 2 and 5 s ) , participants had 3 s to predict whether the rater approved or disapproved of them . Following a button press the unchosen option disappeared immediately . After a 6 s delay approval ( in the form of a thumbs up symbol ) or disapproval ( in the form of a thumbs down symbol ) was revealed . To critically test for the role of positive and negative surprise , we also added 24 trials where feedback was not displayed ( empty grey circle instead of a thumb symbol ) . After every 2–3 trials , participants were probed as to their current self-esteem ( total of 78 self-esteem ratings ) by being presented with the question ‘How good do you feel about yourself at this moment ? ’ for 5 s , after which they had 4 s to move a cursor along a visual analog scale with endpoints ‘very bad’ and ‘very good’ . Self-esteem probes were preceded by a jittered fixation display ( uniformly distributed between 2 and 5 s ) . During instructions , we emphasized a distinction between self-esteem and mood ( the former reflecting feeling good about yourself vs . the latter involving feeling good in general ) , self-esteem and self-concept ( an affective evaluation of the self rather than ‘cold’ semantic knowledge about the self , e . g . , ‘I am a student’ , ‘I am English’ ) , and state vs . trait self-esteem ( right now in the task vs . general feelings of self-worth reflecting the last weeks or months ) . As part of a cover story participants were told that each rater made their ratings independently and that a visual color cue signaled how many profiles raters liked or disliked . For example , a rater who liked 36 out of 40 profiles would be placed in the first quartile group . To better orientate participants , they learned the rank ordering of the four groups prior to performing the task and were tested on the rank order before performing the task . In addition , explicit instructions emphasized that colors were unrelated to how a specific rater had evaluated them . To keep participants engaged they were told that for every trial where they failed to make a prediction before the time limit , 50 pence would be subtracted from a potential amount of money they would play with in a game after the scanning experiment ( Dictator Game; see Figure 2—figure supplement 2 ) . Missed trials were excluded from further analysis ( the median number of missed trials was 1 ) . Unbeknownst to participants , social feedback was generated by the computer . All participants received positive feedback on 80 trials , negative feedback on 80 trials and no feedback on 24 trials . Importantly , the probability of receiving positive feedback was dependent on the rater’s group . Participants received ‘approval’ feedback in 85% , 70% , 30% , and 15% of the trials , spanning the first to fourth quartile rater groups respectively . Participants were not informed about these exact probabilities before the task , but the percentages were consistent with information given to them about the order of the groups . Trial order was randomized such that participants never saw cues of the same color more than twice in a row , while the same color was never displayed more than seven trials in the past . The task was administered in three blocks with feedback that randomly ordered but always according with the ‘approval frequency’ of each group . Blocks lasted approximately 17 min in total . After each block participants received feedback about how many correct predictions they had made to increase engagement in the task . After scanning , participants performed two additional tasks: an ‘Other evaluation’ task ( see Figure 2—figure supplement 1 ) and the Dictator Game ( see Figure 2—figure supplement 2 ) . The ‘other evaluation’ task was identical to the scanning task except that participants were not the object evaluation and that there were fewer trials ( 64 evaluations; by 16 raters of each group ) . In this task , participants were asked to predict whether another participant ( of the same gender and age ) was liked and then observed the feedback this other person received . After every 2–3 trials they reported on their own level of self-esteem ( 27 self-esteem ratings ) . For the Dictator Game , participants were endowed with £5 . They played 12 independent Dictator games of which one was randomly selected for payout . They played a Dictator game with three members of each of the four groups . To be specific , out of every group they played with: one person that approved of them , one person that disapproved of them and one person where no feedback was displayed . At the end of the experiment participants were told that the feedback was computer-generated . We modeled dynamic changes in self-esteem for all ratings preceded by choices ( 74 in total ) using exponential kernel regression models that assume an exponential decay of previous events . The winning model ( Equation 1 ) contains separate terms for baseline self-esteem throughout and social approval prediction errors ( SPE; the difference between received feedback and expected social approval from rater on each trial ( Equation 2 ) . Expectations about approval ( ESV ) were derived from participants’ choice behavior in the task; see Equation 3 below ) . The Gaussian noise term ε ~N ( 0 , σ ) allowed Equation 1 to serve as a generative model of self-esteem . The influence of SPE was assumed to decay exponentially in time such that recent events had greater impact than earlier events ( with forgetting factor: 0 < γ <1 ) . SPEs on a given trial were operationalized as the difference between received social feedback and the ESV: ( 2 ) SPEt=Social feedback−ESVt Where social feedback was 1 for approval , −1 for disapproval and 0 for ‘no feedback’ . ESV on each trial was estimated using a Rescorla-Wagner reinforcement-learning model ( Rescorla and Wagner , 1972 ) . This integrates information over trials by updating the ESV of raters from each of the four groups ( k = 1–4 ) as follows: ( 3 ) ESVkt+1=ESVkt+η SPEt where η is a learning rate capturing the weight that participants give to SPEs in updating expectations of social approval ESV . A softmax function transformed an ESV into an action probability for predicting to be liked or disliked: ( 4 ) πL=11+e− ( ESV+ESV0 ) T Here ESV0 is a response bias and T a decision temperature parameter . A positive bias ESV0 describes the ‘extra credit’ people give themselves , the willingness to predict being liked even in the absence of good evidence a rater will approve . Note that in our learning model the accumulation of ESV itself is unbiased . Participants with larger ESV0 persisted in predicting they would be liked by raters from groups for whom they had a negative ESV ( e . g . , raters from groups 3 or 4 ) . The decision temperature T captures the ‘motivational power of outcomes’ , i . e . , the difference in ESV that will increase the probability of predicting social approval by a fixed amount from the indifference point πL=0 . 5 . Initial ESVs were two free parameters specifying initially expected approval rates for the most positive and the least positive group ( i . e . , ESV1 ( 1 ) and ESV4 ( 1 ) ) . Initial approval expectations for the other two groups were spaced equally in between: ( 5 ) ESV2 ( 1 ) =ESV1 ( 1 ) − ( ESV1 ( 1 ) −ESV4 ( 1 ) ) / 3 ( 6 ) ESV3 ( 1 ) =ESV1 ( 1 ) −2 ( ESV1 ( 1 ) −ESV4 ( 1 ) ) / 3 For each individual participant all free parameters in Equations 1-4 ( w0 , w1 , σ , η , ESV0 , T , and initial ESVs ) were fitted together so as to maximize the summed log-likelihood of self-esteem ratings and approval predictions . This model best explained choice behavior ( i . e . , predictions about approval ) and changes in self-esteem in terms of how well the model described the data and its complexity ( i . e . , number of parameters ) based on Bayesian model comparison . We compared this model against the following alternative models . First , to justify the need of the response bias parameter , we compared model 1 to model 2 ( ‘Learning , but no bias’ ) , which is identical to model 1 , but omits the response bias parameter ESV0 . Second , to critically test whether a model that includes updating of expectations based on prediction errors weighted by a learning rate can better explain changes in self-esteem , we compared model 1 to model 3 ( ‘Correct initial beliefs about approval’ ) . This model assumes that participants do not update their expectations based on social feedback across the experiment , but start the experiment with the actual approval probabilities for each group . A comparison between model 3 and model 1 is critical given that participants were instructed about the rank order of the 4 groups prior to the experiment . Third , to test whether expectations have an additional effect on changes in self-esteem above and beyond their effect captured by the SPE term , we compared the model against model 4 that had a separate expectations term based on existing models of prediction-error driven changes in subjective states ( Rutledge et al . , 2014; 2015 ) ( Equation 7 ) . ( 7 ) Self-esteem ( t ) =w0+w1∑j=1tγt−j ESVj+w2∑j=1tγt-j SPEj+ϵ Finally , to verify that self-esteem not only depends on the valence of social feedback , but on errors arising out of expectations about feedback , we fitted model 6 to the self-esteem data ( ‘Outcome valence only’; Equation 8 ) . ( 8 ) Self-esteem ( t ) =w0+w1∑j=1tγt-jSFj+ ϵ This model is most comparable to previous investigations of effects of social feedback on self-esteem where self-esteem is assumed to increase after positive social feedback ( SF ) and decrease after negative social feedback , but where expectations about approval are not modeled . Consistent with this notion , this model assumes that participants start the experiment without expectations about approval from raters from the four groups and do not update expectations based on feedback ( i . e . expectations of . 5 for each of the four groups ) . To allow for a fair comparison between this model and our winning model , we fit this model to the self-esteem ratings only , because a model without expectations would by definition not provide a good fit for the behavioral choice data ( as prediction choices clearly differed for the four groups ) . Therefore , we compared model 6 against models 5 that we also fitted to the self-esteem ratings only . Like model 6 , model 5 assumed that participants did not update expectations . However , model 5 assumed that participants had initially expected approval rates for the most positive and the least positive group model that were specified by 2 free parameters and is therefore comparable to our winning model 1 . A comparison of model 5 against this model 6 is critical for confirming that self-esteem is sensitive to prediction errors arising out of expectations about approval rather than merely to approval and disapproval per se . To fit the parameters of the different computational models , we used maximum likelihood fitting with flat priors over the parameters . In order to examine whether models improved description of the experimental data we considered the mean squared error over self-esteem ratings . We considered the summed log likelihood of the model prediction over the predictions and the summed log density of the model prediction over the self-esteem ratings . To compare between models , we computed Bayesian Information Criterion ( BIC ) by penalizing the model evidence to account for model complexity as follows: BIC=ln ( n ) k−2ln ( L^ ) , where n is the number of choices + self esteem ratings used to compute the likelihood , k is the number of fitted parameters and L^ is the maximized value of the likelihood function of the model . BIC measures are summed across all participants . Lower BIC values indicate a more parsimonious model fit . We performed CCA ( Hair et al . , 1998 ) to find how computational self-esteem parameters correlate with symptoms linked to low self-esteem . Symptoms were measured using questionnaires assessing self-evaluation , interpersonal sensitivity , as well as symptoms of psychiatric disorders characterized by negative self-beliefs . Self-evaluation measures included trait self-esteem ( Rosenberg , 1965 ) , state self-esteem ( Heatherton and Polivy , 1991 ) , self-perception ( Neemann and Harter , 1986 ) , and narcissism ( Raskin and Terry , 1988 ) . Interpersonal sensitivity measures included the brief fear of negative evaluation scale ( Carleton et al . , 2011; Leary , 1983 ) and the rejection sensitivity questionnaire ( Downey and Feldman , 1996 ) . Symptom measures included state and trait anxiety ( Spielberger et al . , 1970 ) , social anxiety ( Liebowitz , 1987 ) , and depression ( Angold et al . , 1995; Beck et al . , 1996 ) . Prior to running the CCA , aggregate questionnaire scores were z-scored across all participants and parameter estimates that were not normally distributed were log-transformed . For the CCA , we substituted the parameter estimates of initial expectations about the most positive and the least positive groups ( ESV1 ( 1 ) and ESV4 ( 1 ) ) with two summary variables that carry the same information , but are more informative about the psychological processes involved ( i . e . , the average and the range of those two estimates ) . Scans were acquired using a 3T Siemens Trio MRI scanner ( Siemens Healthcare , Erlangen , Germany ) equipped with a standard transmit-receive 32-channel whole-head coil . After obtaining a localizer scan , we collected field maps ( TE = 10 and 12 . 46 ms , TR = 102 ms , matrix size 64 × 64 , with 64 slices , voxel size = 3 mm3 ) for distortion correction . Subsequently , we acquired functional MRI data in three runs ( mean amount of volumes = 1135; range 1105–1168; total number of volumes acquired varied depending on participants’ choice times ) with a blood oxygenation level-dependent ( BOLD ) sensitive T2*-weighted single shot echo-planar imaging ( EPI ) sequence ( repetition time ( TR ) = 2 . 8 s , echo time ( TE ) = 30 ms , slice matrix = 64 × 64 × 40 matrix , slice thickness = 2 mm , slice gap = 1 mm gap , slice tilt of −30° ( T > C ) , field of view ( FOV ) = 192 × 192 mm2; ascending slice acquisition order ) , which was optimized to minimize signal dropout in ventral frontal and temporal cortex ( Deichmann et al . , 2003 ) . The first five volumes from each functional run were discarded to allow for equilibration of T1 saturation effects . After the functional images , we obtained a 3D T1-weighted structural scan for anatomical reference ( TR = 7 . 92 ms; TE = 2 . 48 ms , TI = 910 ms , flip angle α = 16° , 176 = slices , 1 × 1 × 1 mm voxels , FOV = 256 × 240 mm2; Deichmann et al . , 2004 ) . We used a pulse-oximeter and breathing belt to collect physiological data during scanning . Stimuli were presented in MATLAB ( MathWorks , Inc . , Natick , MA ) using Cogent 2000 ( Wellcome Trust Centre for Neuroimaging ) onto a screen in the magnet bore , which participants could see through a mirror attached to the head coil . Participants could respond by using a fiber optic response box . During scanning foam inserts restricted head motion . Preprocessing and analysis of MRI data was implemented using Statistical Parametric Mapping 12 ( SPM12 ) ( Wellcome Trust Centre for Neuroimaging , University College London ) . Functional images were slice-time corrected , corrected using field maps , unwarped and realigned , co-registered with structural MRI , normalized to MNI space ( using the DARTEL toolbox; Ashburner , 2007 ) and smoothed using a 8 mm , full-width at half-maximum isotropic Gaussian kernel . To examine the neural correlates of SPEs and changes in self-esteem , we deployed two event-related general linear models ( GLMs ) that included separate regressors indicating onsets of choice , delay period , feedback , self-esteem question screen , and initial button press for self-esteem rating . All durations in the model were set to 0 s . The GLMs also contained 6 motion parameters regressors and 18 regressors for cardiac and respiratory regressors to correct for motion-induced and physiological noise . The first GLM contained parametric modulators for ‘ESV’ ( derived from computational modeling [Equation 1] and time-locked to choice onset ) , ‘SPE’ ( derived from computational modeling and time-locked to feedback onset ) , and ‘self-esteem rating' ( z-scored and time-locked to question onset ) . The second GLM contained parametric modulators for 'inferred self-esteem rating' ( derived from computational modeling; z-scored and time-locked to choice onset and onset of feedback ) , and actual ‘self-esteem ratings' ( z-scored and time-locked to question onset ) . Subject-specific contrast images were submitted to group level random-effects analyses . Results were corrected for multiple comparisons with Family-wise Error ( FWE ) cluster-correction at p<0 . 05 ( cluster-forming threshold of p<0 . 005 ) . To ensure that the FWE correction provided adequate control of false positives , we additionally performed non-parametric permutation tests ( 10 , 000 Monte-Carlo simulations ) that take into account the smoothness of the data and the normalized voxel size ( Slotnick et al . , 2003 ) . These permutation tests determined that a cluster-extent threshold of k > 120 voxels is needed to control for multiple comparisons ( p<0 . 05 ) at a cluster-forming threshold of p<0 . 005 . All reported clusters exceed this threshold , supporting the validity of the FWE multiple-comparison correction procedure . We used the Marsbar toolbox ( Brett et al . , 2002; http://marsbar . sourceforce . net/ ) to extract subject-level contrast values in clusters of activity derived from our whole-brain analyses . To examine functional connectivity , we performed a psychophysiological interaction ( PPI ) analysis ( Friston et al . , 1997; O'Reilly et al . , 2012 ) . We set up a GLM with regressors capturing the physiological effect , ( i . e . , time series for a 6 mm sphere centered on the peak voxel of the insula cluster [−44 , 11 , 9] derived from the whole-brain regression analysis testing for individual differences in SPE processing related to the interpersonal vulnerability dimension ) , the psychological contrast of interest ( i . e . , trial-by-trial self-esteem updates upon receipt of feedback ) and the psychophysiological interaction term ( i . e . , physiological effect x psychological contrast of interest ) . The GLM also included 6 motion parameters regressors and 18 regressors for cardiac and respiratory regressors to correct for these sources of noise . | Self-esteem – our evaluation of our own worth – is shaped by what other people think of us . It increases when others appreciate and value us , and decreases when we are rejected and start to question our own worth . Maintaining a positive sense of self is crucial for mental health and well-being . People with low self-esteem are more likely to develop psychiatric conditions , such as anxiety disorders , eating disorders and depression . Despite its importance for mental health , it was not known how the brain accumulates social feedback to determine our self-esteem . To address this question , Will et al . developed a computational model that precisely predicts how self-esteem changes from moment to moment as people learn what others think of them . Activity in the brain was measured while young adults received approving or disapproving feedback from peers who had seemingly viewed their online character profile . After every second or third peer judgment , participants reported their current level of self-esteem . Will et al . found that self-esteem depended both on whether other people liked the participants and on whether they were liked or disliked more than expected . Self-esteem decreased the most when participants received negative feedback from someone they expected to receive positive feedback from . The model then identified signals in specific parts of the brain that explain why self-esteem goes up and down according to the feedback received . Moment-to-moment changes in self-esteem correlated with activity in the ventromedial prefrontal cortex , which is a brain region important for valuation . Will et al . combined the model with responses to questionnaires that assessed psychiatric symptoms , and showed that vulnerable individuals had elevated responses in a part of the brain called the anterior insula . In vulnerable individuals , activity in this region of the brain was strongly coupled to activity in the part of the prefrontal cortex that explained changes in self-esteem . A better understanding of the brain mechanisms that mediate a decline or improvement in self-esteem may help to find more effective treatments for a range of mental health problems . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"neuroscience"
] | 2017 | Neural and computational processes underlying dynamic changes in self-esteem |
A key challenge in precise genome editing is the low efficiency of homology-directed repair ( HDR ) . Here we describe a strategy for increasing the efficiency of HDR in cells by using a chromatin donor template instead of a naked DNA donor template . The use of chromatin , which is the natural form of DNA in the nucleus , increases the frequency of HDR-edited clones as well as homozygous editing . In addition , transfection of chromatin results in negligible cytotoxicity . These findings suggest that a chromatin donor template should be useful for a wide range of HDR applications such as the precise insertion or replacement of DNA fragments that contain the coding regions of genes .
The ability to manipulate genomes precisely is revolutionizing the biological sciences ( Doudna , 2020 ) . Of particular utility is the modification or insertion of customized DNA sequences at a specific genomic location by homology-directed repair ( HDR ) ( Jasin and Rothstein , 2013 ) . For genome engineering in cells , HDR typically involves the generation of a specifically targeted DNA double-strand break ( DSB ) in the presence of a homologous DNA donor template that contains the desired sequence to be modified or inserted ( Urnov et al . , 2005; Bedell et al . , 2012; Jinek et al . , 2012; Cong et al . , 2013; Pickar-Oliver and Gersbach , 2019 ) . A key challenge in successful genome editing has been the low efficiency of HDR ( Carroll , 2014; Harrison et al . , 2014 ) . For the generation of specific alterations in a short stretch of DNA ( <50 nt ) , recently developed techniques such as base editing ( Rees and Liu , 2018; Molla and Yang , 2019 ) and prime editing ( Anzalone et al . , 2019 ) have been shown to be highly effective . In addition , for the imprecise insertion of larger DNA fragments , homology-independent approaches can be used ( Auer et al . , 2014; He et al . , 2016; Suzuki et al . , 2016 ) . These powerful methods cannot , however , be used for the precise insertion or replacement of >50 bp DNA fragments , such as those containing the coding regions of genes . For such applications , we considered a different strategy for increasing the efficiency of HDR in cells . Based on our previous observation that homologous strand pairing , an early step in HDR , occurs more efficiently with a chromatin donor template than with a plain ( naked ) DNA donor template in vitro ( Alexiadis and Kadonaga , 2002 ) , we postulated that HDR in cells might similarly be more efficient with a chromatin relative to a naked DNA donor template . In this study , we tested this idea by comparing the efficiency of HDR with chromatin versus naked DNA donor templates in conjunction with DSBs generated by the clustered regularly interspaced short palindromic repeats ( CRISPR ) -Cas9 system . We found that the overall HDR efficiency as well as the frequency of homozygous editing is enhanced by the use of a chromatin donor template relative to a DNA donor template . We thus envision that a chromatin donor template , which resembles the natural form of DNA in the nucleus , could be widely used to increase the success of HDR-mediated applications , particularly those that involve the targeted insertion of DNA fragments such as the coding regions of genes .
To ascertain whether the use of chromatin donor templates affects the efficiency of HDR in cells , we reconstituted three DNA donor templates ( corresponding to the human GAPDH , RAB11A , and ACTB loci ) into chromatin and tested the relative efficiencies of the targeted insertion of the GFP coding sequence with chromatin versus naked DNA versions of these templates ( Figure 1 and Figure 1—figure supplements 1–4 ) . The chromatin was reconstituted by using salt dialysis methodology with plasmid DNA and purified core histones from Drosophila embryos , which contain a broad mixture of covalent modifications that have not been precisely resolved ( Levenstein and Kadonaga , 2002 ) . With standard CRISPR-Cas9 methodology and human MCF10A cells ( non-tumorigenic epithelial cells derived from human mammary glands ) , we observed that the use of a chromatin donor template relative to a naked DNA donor template resulted in a 7 . 4- , 2 . 9- , and 2 . 3-fold increase ( average of three biological replicates ) in the directed insertion of GFP sequences at the GAPDH , RAB11A , and ACTB loci , respectively ( Figures 1B , C and D and Figure 1—figure supplements 3 and 4 ) . Thus , at three different loci ( GAPDH , RAB11A , and ACTB ) in human MCF10A cells , there was a higher efficiency of HDR-mediated GFP insertion with chromatin donor templates than with naked DNA donor templates . For many applications of HDR , it is essential to modify all of the copies of the target gene . Therefore , to test the frequency of occurrence of precise homozygous gene editing in the diploid MCF10A cells , we carried out PCR analyses of the individual GFP-positive clones , and we observed a variable but consistently higher frequency of homozygous HDR insertions with chromatin donor templates than with naked DNA donor templates at all three loci ( GAPDH , RAB11A , and ACTB ) in MCF10A cells ( Figure 2 and Figure 2—figure supplements 1–5 ) . At the GAPDH locus , the use of chromatin relative to naked DNA donor templates resulted in a 2 . 1-fold increase in homozygous editing . At the RAB11A locus , there was a high frequency of homozygous insertions with the naked DNA donor template , and the use of a chromatin donor template only slightly augments ( 1 . 1-fold increase ) the percentage of homozygous clones . Strikingly , at the ACTB locus , homozygous insertions were observed only with a chromatin donor template . These findings thus show that the use of chromatin relative to naked DNA donor templates can increase the efficiency of homozygous editing . We also observed imperfect editing , in which there was at least one improperly edited chromosome , as indicated by either the absence of an edited chromosome or the presence of a PCR product whose size is not consistent with that of an edited or wild-type chromosome . In addition , by performing long-range PCR as in Kosicki et al . , 2018 , we identified two apparently homozygous clones that contained one chromosome with a precisely edited allele and one chromosome with a large deletion at the other allele ( Figure 2—figure supplement 2 ) . Hence , in the generation of homozygous clones , it is important to carry out both standard and long-range PCR analyses . The overall efficiency of achieving homozygous editing in diploid MCF10A cells was 15-fold ( 7 . 4 × 2 . 1 ) at the GAPDH locus , 3 . 2-fold ( 2 . 9 × 1 . 1 ) at the RAB11A locus , and large but not quantifiable at the ACTB locus , at which we saw homozygous editing only with a chromatin donor template . The ACTB locus serves as an example in which the use of a chromatin template relative to a naked DNA template was the difference between a successful and an unsuccessful HDR experiment . To determine whether a chromatin donor template affects the efficiency of HDR in a different cell line , we examined the insertion of GFP sequences at the GAPDH locus in HeLa cells , which are human cervical adenocarcinoma cells that are widely used in biomedical research . HeLa cells are aneuploid and contain four copies of the GAPDH gene , which is located on chromosome 12 . In these experiments , we observed that the use of a chromatin donor template results in a 2 . 3-fold increase ( average of three biological replicates ) in the efficiency of insertion of the GFP sequence in at least one GAPDH locus in HeLa cells ( Figure 3A , B and Figure 3—figure supplement 1 ) . We then examined the formation of homozygous edited clones that are generated upon targeted insertion of the GFP sequence at all four copies of the GAPDH locus in HeLa cells . In this analysis , we found a substantial increase ( 5/18 clones versus 1/21 clones ) in the efficiency of formation of homozygous clones with the use of a chromatin donor template instead of a naked DNA donor template ( Figure 3C , D , E and Figure 3—figure supplement 2 ) . Hence , these results show a strong enhancement of HDR by using a chromatin relative to a naked DNA donor template in HeLa cells . We additionally tested the effect of varying the amount of donor template DNA ( as chromatin or naked DNA ) upon the efficiency of HDR ( Figure 3—figure supplement 3 ) . To this end , we used 0 . 5 , 1 . 0 , and 1 . 5 times the mass of DNA as in a standard experiment with the GAPDH donor template in HeLa cells . At each of the three amounts of donor template , we consistently saw a higher efficiency of generation of GFP-positive cells with chromatin relative to naked DNA . Moreover , there was an increase in the fold-enhancement by chromatin as the amount of donor template was increased . We thus observed that a chromatin donor template functions better than a naked DNA donor template for HDR at different concentrations . Because chromatin has rarely been used in cell transfection experiments , we also investigated the toxicity of chromatin relative to naked DNA in five different human cell lines ( Figure 3—figure supplement 4 ) . These experiments revealed that chromatin is of comparable or lower toxicity to cells relative to naked DNA in transfection experiments . This low toxicity of chromatin to cells could be useful for HDR applications in which there is low cell viability after transfection .
Here we show that the efficiency of HDR-mediated gene editing can be increased by using a chromatin donor template instead of a naked DNA donor template . Why is chromatin more effective as an HDR donor template than naked DNA ? We suggest that chromatin , as the natural form of DNA in the eukaryotic nucleus , is the preferred substrate ( relative to naked DNA ) for the factors that mediate homologous recombination in cells . In previous biochemical studies , we and others found that eukaryotic Rad51 and Rad54 , but not bacterial RecA , can mediate homologous strand pairing , an early step in HDR , with a chromatin donor template ( Alexiadis and Kadonaga , 2002; Jaskelioff et al . , 2003 ) . Moreover , we observed that homologous strand pairing occurs more efficiently with a chromatin donor template than with a naked DNA donor template ( Alexiadis and Kadonaga , 2002 ) . Hence , the new findings on HDR with chromatin donor templates in cells are consistent with the results of the earlier biochemical studies on homologous strand exchange . In general , a wide range of efficiencies of HDR has been observed in different cell types and with different methodologies . A common factor in these HDR experiments has been , however , the use of a non-chromatin donor template . In this work , we sought to focus specifically on directly comparing the relative efficiencies of HDR with chromatin versus naked DNA donor templates . In these experiments , we consistently observed a higher efficiency of HDR with chromatin relative to naked DNA . These effects include the increased efficiency of targeted insertion of GFP sequences in both loci of a diploid chromosome and in all loci of a tetraploid chromosome . These findings therefore suggest that the use of a chromatin donor template instead of a naked DNA donor template would be a broadly useful strategy for the precise insertion or replacement of DNA sequences via HDR with different methods . Moreover , transfection of chromatin donor templates , which can be simply prepared by salt dialysis methodology with purified DNA and core histones , does not affect cell viability . Thus , current methods for HDR can be easily adapted to include chromatin donor templates in place of their naked DNA counterparts . In this regard , it is notable that we reconstituted chromatin by using native core histones from Drosophila embryos . These histones contain an undefined broad mixture of covalent histone modifications ( Levenstein and Kadonaga , 2002 ) . Because the core histones and their modifications are highly conserved throughout eukaryotes , it seems likely that similar results would be obtained with core histones from other sources . It is possible , however , that the magnitude of enhancement of HDR by chromatin could be further increased by variation of the core histone sequences and modifications . In conclusion , although there are excellent techniques for the alteration of short ( <50 bp ) stretches of DNA ( Rees and Liu , 2018; Molla and Yang , 2019; Anzalone et al . , 2019 ) , there remains a need for increasing the efficiency of the specific insertion or replacement of longer DNA segments that may contain sequences such as the coding regions of genes . We anticipate that chromatin donor templates might be particularly useful for such applications . In addition , we expect that many new gene editing techniques will be developed in the future , and that some of these methods will benefit from the use of chromatin donor templates . Furthermore , the low toxicity of chromatin to cells may be useful for many current and future methods . There is considerable potential to the use of the natural form of the donor template in gene editing experiments . It is our hope that these findings will advance the utility of precise genome editing in basic , translational , and clinical research .
CRISPR RNA ( crRNA ) sequences targeting the GAPDH , RAB11A , or ACTB loci were each inserted into the pU6- ( BbsI ) CBh-Cas9-T2A-mCherry vector ( Addgene plasmid # 64324 ) as described ( Ran et al . , 2013 ) . The crRNA sequences that were used are as follows: GAPDH , GAGAGAGACCCTCACTGCTG; RAB11A , GGTAGTCGTACTCGTCGTCG; ACTB , GGTGAGCTGCGAGAATAGCC . The donor template plasmid for the modification of the GAPDH locus was generated as follows . Two homology arm ( HA ) sequences ( ~1 kb each ) were PCR-amplified with Phusion polymerase ( NEB ) and genomic DNA ( gDNA ) from HeLa cells . The oligonucleotides that were used are as follows ( the upper case letters are complementary to GAPDH or T2A-EGFP sequences ) : 5' HA , agagataagcttGGACACGCTCCCCTGACTT , agagatggatccCTCCTTGGAGGCCATGTGGG; 3' HA , tgatagggtaccCCTGCCACACTCAGTCCC , tgataggaattcGCTGGGGTTACAGGCGTGCG . The T2A-EGFP sequence was PCR-amplified from the PX461 plasmid ( Addgene plasmid # 48140 ) with the following oligonucleotides: agagatggatccGAGGGCAGAGGAAGTCTGCT and agagatggtaccTTACTTGTACAGCTCGTCCA . Then , the three DNA fragments were sequentially subcloned into the pBluescript KS vector ( Stratagene ) . The 3' HA sequence was inserted between the KpnI and EcoRI sites; the T2A-EGFP sequence was inserted between the BamHI and the KpnI sites; and the 5' HA sequence was inserted between the HindIII and the BamHI sites . All restriction enzymes were from NEB . The donor template plasmid for the modification of the RAB11A locus was Addgene plasmid # 112012 , and the donor template plasmid for the modification of the ACTB locus was Addgene plasmid # 87425 . Native Drosophila core histones from embryos collected from 0 to 12 hr after egg deposition were purified as described ( Fyodorov and Levenstein , 2002; Khuong et al . , 2017 ) . The donor repair template plasmids were purified with the HiSpeed plasmid kit ( Qiagen ) . The optimal histone:DNA ratio for each donor repair template was determined by carrying out a series of reactions with different histone:DNA ratios and then assessing the quality of chromatin by the micrococcal nuclease digestion assay , as described ( Fyodorov and Levenstein , 2002; Khuong et al . , 2017 ) . Chromatin was reconstituted with purified core histones by using the salt dialysis method ( Stein , 1989; Fei et al . , 2015 ) . In a typical chromatin reconstitution reaction , 50 µg plasmid DNA and 50 µg core histones were combined in TE buffer ( 10 mM Tris-HCl , pH 8 , containing 1 mM EDTA ) containing 1 M NaCl in a total volume of 150 µL . The mixture was dialyzed at room temperature against the following buffers in the indicated order: 2 hr in TE containing 0 . 8 M NaCl; 3 hr in TE containing 0 . 6 M NaCl; 2 . 5 hr in TE containing 50 mM NaCl . The quality of the resulting chromatin was assessed by using the micrococcal nuclease digestion assay , and the chromatin was stored at 4°C until use . HeLa cells were a gift from Dr . Anjana Rao ( La Jolla Institute for Immunology ) . MCF10A cells were a gift from Dr . Jichao Chen ( The University of Texas MD Anderson Cancer Center ) . The MCF10A and HeLa cells were not authenticated . The MCF10A cells and HeLa cells were tested for mycoplasma and found to be negative for mycoplasma contamination . MCF10A cells ( non-tumorigenic mammary epithelial cells ) were maintained in DMEM/F-12 medium ( Gibco ) supplemented with 20 ng/mL EGF , 500 ng/mL hydrocortisone ( Sigma ) , 10 μg/mL insulin ( Sigma ) , 100 ng/mL cholera toxin ( Sigma ) , 100 U/mL penicillin and 100 µ/mL streptomycin ( Gibco ) , and 5% horse serum ( Gibco ) at 37°C and 5% CO2 . HeLa cells ( human cervical carcinoma cells ) , HT1080 cells ( human fibrosarcoma cells ) , SW480 cells ( human colorectal adenocarcinoma cells ) , and 293 T cells ( derived from primary human embryonic kidney cells ) were maintained in DMEM , high glucose medium ( Corning ) supplemented with 10% fetal bovine serum ( Gibco ) and 100 U/mL penicillin and 100 µ/mL streptomycin ( Gibco ) at 37°C and 5% CO2 . In each series of experiments , cell transfections with chromatin or DNA donor templates were performed by following standard protocols under exactly the same conditions . Transfection of HeLa cells was performed with Lipofectamine 3000 ( Invitrogen ) according to the manufacturer's recommendations . Linear polyethylenimine ( PEI 25K; 25 , 000 MW; Polysciences , Inc ) was used for transfection of MCF10A cells at a PEI:DNA mass ratio of 3:1 . The transfections were performed as follows . 5 × 105 cells/well were plated in six well plates the day before transfection . For each CRISPR-Cas9 target locus , cells were co-transfected with equal amounts of the target-specific donor repair template ( as free plasmid DNA or chromatin ) and the Cas9 coding plasmid containing the target-specific single guide RNA sequence . For HeLa cells , DNA ( 1 . 25 µg ) or chromatin ( containing 1 . 25 µg of DNA ) was used in each transfection ( except for the experiment in Figure 3—figure supplement 3 , in which 1 . 25 µg of the Cas9 coding plasmid containing the single guide targeting the GAPDH locus was co-transfected with 0 . 625 µg , 1 . 25 µg , or 1 . 875 µg of donor template DNA as naked DNA or chromatin ) ; for MCF10A cells , DNA ( 1 . 5 µg ) or chromatin ( containing 1 . 5 µg of DNA ) was used in each transfection . At 24 hr post-transfection , cells were detached with 0 . 25% trypsin ( Corning ) . After centrifugation , the cell pellets were resuspended in culture media containing 250 ng/mL DAPI ( Sigma ) . mCherry-positive , DAPI-negative cells were sorted by FACS and collected in six well plates ( HeLa cells; 100 , 000 cells/well ) or 24 well plates ( MCF10A cells; 30 , 000 cells/well ) . Then , the cells were passaged twice before the analysis of the expression of GFP by flow cytometry . GFP-positive single-cells were sorted by FACS into 96 well plates . To determine the percentage of GFP-positive cells , at least 100 , 000 cells of each condition were analyzed by flow cytometry with a BD FACSAria Fusion or a BD FACSAria2 instrument at the Human Embryonic Stem Cell Core Facility ( UCSD ) . The BD FACSDiva Software was used for data acquisition , and data analysis was performed with FlowJo version 10 . 6 . 1 ( BD ) . Genomic DNA samples from wild-type cells as well as from independent GFP-positive clones were isolated with the Quick Extract DNA extraction solution ( Lucigen ) by following the manufacturer's recommendations , and were then subjected to PCR analysis . First , the occurrence of edited alleles was analyzed with primers that flank the 5' and 3' homology arm sequences ( and thus do not contain sequences in the donor template ) at the location in which the GFP DNA was inserted . The specific primers that were used are as follows: GAPDH , F1: TGACAACAGCCTCAAGATCATCAGG , R1: GATGGAGTCTCATACTCTGTTGCCT; RAB11A , F1: TGGGAAGTGGACATCATTGG , R1: GACCCTCCAATATGTTCTGT; ACTB , F1: AATGCTGCACTGTGCGGCGA , R1: ATGGCATGGGGGAGGGCATA . Then , genomic DNA from potentially homozygous GFP-positive clones was analyzed by long-range PCR analysis with LongAmp Hot Start Taq DNA Polymerase ( NEB ) , as described by Kosicki et al . , 2018 . The primers that were used are as follows . GAPDH , F2: CTCCTGCAGTGATTTGTTTCTTCTT , R2: ACTCATTCTCCCAACACACATCAAA; RAB11A , F2: GCTTTATCTTCTTTTTGCTCACCTG , R2: GTGTCCCATATCTGTGCCTTTATTG; ACTB , F2: ATGAATAAAAGCTGGAGCACCCAA , R2: TTGTGCAGCTATACGCAAGATTAAG . The locations of the PCR primers at the GAPDH , RAB11A , and ACTB loci are depicted in Figure 2—figure supplement 1 . To confirm the integrity of the homozygous clones obtained with chromatin donor templates , we determined the DNA sequences of three GAPDH clones and three ACTB clones across the insertion junctions and found that the GFP sequences were precisely inserted into the target sites in all six clones . The two-tailed Welch t-test with alpha = 0 . 05 was performed by using GraphPad Prism version 8 . 4 . 1 ( GraphPad Software ) . | Genome editing is a powerful tool used across a wide range of biomedical research . There are several different techniques used , depending on the type of edit being made , and one known as homology-directed repair – or HDR for short – is a common technique for precisely inserting large sections of DNA , such as those needed to make desired proteins in cells . HDR takes advantage of the cell’s mechanisms for repairing damage to DNA if both strands of the DNA double helix are broken . The mechanism relies on a DNA template to stitch the strands back together . To insert or replace a new DNA sequence , scientists can add a customized piece of DNA of their choosing to the cell so that it might be incorporated into the genome . However , HDR is not very efficient , and the success rate is often less than a few percent . In HDR gene editing , the DNA template is typically added as purified , or ‘naked’ , DNA . However , the natural form of DNA in cells , known as chromatin , is where the DNA helix is wrapped around a cluster of proteins known as histones . Cruz-Becerra and Kadonaga tested the idea that DNA in the form of chromatin might be more effective as a template for HDR than naked DNA . The two approaches were compared to see which was better at inserting a sequence at three different locations in the genome of lab-grown human cells . In these experiments , the chromatin templates were 2 . 3- to 7 . 4-fold more efficient than the naked DNA . Also , the DNA in human cells is organized as pairs of chromosomes , and chromatin was better than naked DNA for editing both copies of the chromosome pairs rather than only one of them . In addition , the chromatin is potentially less toxic to the cells . Cruz-Becerra and Kadonaga hope that this will be useful for increasing the success rate of HDR experiments and potentially other methods of gene editing in the future . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"chromosomes",
"and",
"gene",
"expression",
"tools",
"and",
"resources"
] | 2020 | Enhancement of homology-directed repair with chromatin donor templates in cells |
Southeastern Asia is a recognised hotspot for emerging infectious diseases , many of which have an animal origin . Mammarenavirus infections contribute significantly to the human disease burden in both Africa and the Americas , but little data exists for Asia . To date only two mammarenaviruses , the widely spread lymphocytic choriomeningitis virus and the recently described Wēnzhōu virus have been identified in this region , but the zoonotic impact in Asia remains unknown . Here we report the presence of a novel mammarenavirus and of a genetic variant of the Wēnzhōu virus and provide evidence of mammarenavirus-associated human infection in Asia . The association of these viruses with widely distributed mammals of diverse species , commonly found in human dwellings and in peridomestic habitats , illustrates the potential for widespread zoonotic transmission and adds to the known aetiologies of infectious diseases for this region .
Rodents of several species are known hosts of numerous zoonotic pathogens ( Luis et al . , 2013 ) , and are also amongst the peridomestic ensemble that benefit from how humans are modifying the landscape ( Shochat et al . , 2006 ) . Their increased presence will amplify human-rodent encounter rates and opportunities for zoonotic transmission , likely creating an increased risk for human health ( Young et al . , 2014 ) . A concerted effort to establish which human pathogens are present in the environments heavily utilized by humans would therefore be beneficial for the prevention , control and where possible , treatment of such zoonoses . Mammarenaviruses are predominantly rodent-borne viruses , several of which have been associated with human disease . In Western Africa , Lassa virus ( LASV ) infection causes an estimated 100 , 000 to 300 , 000 infections and approximately 5000 deaths annually ( McCormick et al . , 1987 ) , whilst Junín virus ( JUNV ) causes regular seasonal outbreaks of viral haemorrhagic fever in Argentina ( Gómez et al . , 2011 ) . So far only Wēnzhōu virus and lymphocytic choriomeningitis virus ( LCMV ) have been conclusively demonstrated to be present in Asia ( Li et al . , 2014; Morita et al . , 1996; 1991 ) . Wēnzhōu virus was identified in China in rodents belonging to four Rattus species , Niviventer niviventer and in Asian house shrews ( Suncus murinus ) , but has not yet been associated with disease . However , LCMV is zoonotic , is primarily hosted by the widespread house mouse ( Mus musculus ) , and consequently has the widest known distribution of any mammarenavirus ( Buchmeier et al . , 2007 ) . Human infection with LCMV predominantly results in a relatively mild influenza-like illness ( Bonthius , 2012; Macneil et al . , 2012; Centers for Disease Control and Prevention ( CDC ) , 2012 ) . This propensity of some mammarenaviruses to cause mild disease with symptoms common to those found for many other viral infections , illustrates the potential for these infections to go undiagnosed or misdiagnosed . That human infections caused by arenaviruses other than LCMV have not previously been detected in Southeastern Asia ( Charrel and de Lamballerie , 2003 ) should therefore not be considered as evidence of absence . As each mammarenavirus is primarily adapted to rodents of distinct species or to closely related rodents of diverse species ( Gonzalez et al . , 2007 ) , the apparent absence of these viruses from a region with high rodent biodiversity ( Pagès et al . , 2010 ) is surprising . A large proportion of infectious diseases remain undiagnosed in resource poor countries where people rarely visit healthcare professionals and where clinicians often do not have the training or laboratory support for accurate diagnosis ( Mueller et al . , 2014 ) . Here , in the course of a survey for rodent-borne mammarenaviruses , we identified a genetic variant of the recently identified Wēnzhōu virus in Cambodia - provisionally named Cardamones - in rodents belonging to two widespread species , brown rats ( Rattus norvegicus ) and Pacific rats ( R . exulans ) , which are commonly found in proximity to humans . We also identified a novel mammarenavirus species in Thailand in rice-field associated greater bandicoot rats ( Bandicota indica ) , Savile’s bandicoot rats ( B . savilei ) and an Indomalayan niviventer ( Niviventer fulvescens ) . In addition , we detected Wēnzhōu virus in several Cambodian patients presenting with fever and respiratory symptoms , indicating that this virus may be causally related with human disease .
To assess whether mammarenaviruses are present in Southeastern Asian small mammals , we used a previously described RT-PCR ( Vieth et al . , 2007 ) ( Supplementary file 1A ) to screen 627 homogenised lung samples from small mammals of twenty species ( Supplementary file 1B ) collected from seven sites ( two in Cambodia , three in Thailand and two in Laos ) . Based on the hypothesis of rodent-mammarenavirus interrelation ( Gonzalez et al . , 2007 ) and the presence of only Old World rodents in the region ( Marshall et al . , 1988 ) , primers targeting Old World mammarenaviruses were used . We detected evidence of mammarenavirus RNA in twenty seven animals from two locations: Veal Renh in Cambodia and Loei in Thailand . Seventeen individuals from two peridomestic species , brown rats ( four individuals ) and Pacific rats ( thirteen individuals ) , were identified positive in Veal Renh . In Loei , ten individuals from three agricultural-associated species were found positive , namely six Savile’s bandicoot rats , three greater bandicoot rats and one Indomalayan niviventer ( Supplementary file 1B ) . All but one of the Cambodian positive rodents were collected in settlements , whilst five of the Thai rodents were collected in plantations ( Supplementary file 1C ) . Sanger sequencing of the amplicons indicated the presence of a novel mammarenavirus in the Thai samples . Full genome sequencing was attempted on a single sample from each site: C0649 , originally obtained from a wild-caught Pacific rat from Veal Renh , and R5074 , originally obtained from a wild-caught Savile’s bandicoot rat from Loei . Coding-complete genomes of isolates were acquired by deep sequencing ( Illumina HiSeq 2000 ) of random amplified RNAs from the lungs of Wistar laboratory rats inoculated with either the C0649 or R5074 lung homogenate ( C0649: S segment accession KC669696 , length 3331 nucleotides ( nt ) ; L segment accession KC669690 , length 7181nt . R5074: S segment accession KC669698 , length 3345nt; L segment accession KC669693 , length 7186nt ) . Primers were designed based on this sequence and in conjunction with a previously published protocol ( Bowen et al . , 2000 ) ( Supplementary file 1A ) , were used to obtain the coding-complete genome of two further isolates: C0617 , obtained from a wild-caught brown rat , ( S segment accession KC669694 , length 3344nt; L segment accession KC669691 , length 7171nt ) and R4937 , obtained from a wild-caught greater bandicoot rat ( S segment accession KC669697 , length 3379nt; L segment accession KC669692 , length 7185nt ) . Each segment encoded two open reading frames ( ORFs ) in an ambisense organization with an intergenic region containing a predicted hairpin between the ORFs . Deduced amino acid ( aa ) and nt sequences from the four isolates were compared to those of other representative mammarenaviruses . An aa sequence divergence of >25% for the nucleoprotein ( NP ) was found between the Thai isolates and all other known mammarenavirus species , whilst a 3 . 5–12 . 7% aa sequence divergence was found between the Cambodian isolates and Wēnzhōu virus ( Table 1 ) . PAirwise Sequence Comparison ( PASC ) was performed on both segments for each of the four sequenced viruses . All samples were found most closely related to Wēnzhōu virus , with the Cambodian virus demonstrating 88 . 5–88 . 8% identity for the S segment and 86 . 0–86 . 3% identity for the L segment , whilst the Thai virus showed 70 . 3–70 . 6% identity for the S segment and 62 . 7–63 . 1% identity for the L segment ( Table 2; Table 2—source data 1 ) . The International Committee on Taxonomy of Viruses ( ICTV ) arenavirus species demarcation criteria includes , among others , the association with a main host species or group of sympatric hosts , presence in a defined geographical area , and significant protein amino acid sequence differences , including a difference of at least 12% in the amino acid ( aa ) sequence of the nucleoprotein to other species in the genus ( King et al . , 2012; Radoshitzky et al . , 2015 ) . A recent update from the ICTV Arenaviridae study group has also recommended the use of the PASC tool for the assessment of novel arenaviruses . Cut-off values chosen for classifying arenaviruses belonging to the same species using this tool are >80% and >76% nucleotide sequence identity in the S and L segments respectively ( Radoshitzky et al . , 2015 ) . The Thai virus is the first mammarenavirus to be detected in Bandicota species and , alongside the Cambodian virus is the first to be detected in this geographic region . As this virus also meets nucleoprotein amino acid sequence identities and PASC requirements , we propose that this Thai virus represents a member of a novel species . We suggest to call this novel virus Loei River mammarenavirus ( after a river close to the site where it was detected ) with the abbreviation LORV . Although the Cambodian virus is the first mammarenavirus to be associated with Pacific rats ( R . exulans ) and Cambodia is geographically distant to the Chinese region in which Wēnzhōu virus was originally detected , the sequence homology to Wēnzhōu virus and the association with rodents of a common host species ( R . norvegicus ) , indicates that the Cambodian virus represents a genetic variant of Wēnzhōu mammarenavirus ( Li et al . , 2014 ) . We propose to provisionally name this variant Cardamones ( after the chain of mountains near Veal Renh ) . 10 . 7554/eLife . 13135 . 003Table 1 . Nucleotide and amino acid sequence identities ( % ) between Cambodian and Thai isolates and selected other arenaviruses . DOI: http://dx . doi . org/10 . 7554/eLife . 13135 . 003IsolatesSegment or ORFnt/ aaCambodian isolatesThai isolatesWēnzhōuLassaIppyMopeiaLCMVJunínLunaMorogoroCambodian isolatesL segmentnt98 . 969 . 2-69 . 487 . 5-88 . 655 . 8-56 . 457 . 7-57 . 860 . 856 . 950 . 2-50 . 461 . 1-61 . 460 . 9-61 . 0L ORFnt99 . 367 . 3-67 . 588 . 0-89 . 059 . 7-60 . 660 . 4-60 . 559 . 2-59 . 455 . 650 . 6-50 . 959 . 6-60 . 959 . 6aa99 . 669 . 2-69 . 492 . 2-94 . 855 . 5-56-457 . 7-57 . 856 . 6-57 . 148 . 537 . 9-38 . 055 . 8-55 . 955 . 6Z ORFnt99 . 573 . 4-74 . 5837 . -87 . 966 . 8-67 . 463 . 662 . 557 . 6-58 . 254 . 9-55 . 469 . 0-70 . 164 . 7-65 . 2Aa98 . 879 . 289 . 4-93 . 970 . 1-75 . 370 . 164 . 959 . 440 . 363 . 6-64 . 961 . 0S segmentnt99 . 571 . 7-72 . 187 . 5-89 . 861 . 7-68 . 166 . 6-66 . 866 . 6-67 . 061 . 6-61 . 854 . 6-54 . 867 . 6-68 . 266 . 7-66 . 9NP ORFnt99 . 373 . 1-74 . 486 . 6-90 . 067 . 1-68 . 268 . 3-68 . 467 . 6-67 . 962 . 9-63 . 255 . 1-55 . 669 . 0-70 . 368 . 7aa99 . 882 . 9-84 . 287 . 3-96 . 572 . 2-73 . 874 . 4-74 . 673 . 5-73 . 864 . 051 . 9-52 . 573 . 3-74 . 074 . 0-74 . 6GPCnt99 . 769 . 1-70 . 088 . 6-89 . 767 . 3-68 . 664 . 7-65 . 765 . 3-65 . 761 . 2-61 . 453 . 4-53 . 666 . 3-66 . 649 . 9-65 . 2aa99 . 879 . 5-81 . 195 . 5-96 . 474 . 2-76 . 269 . 5-71 . 371 . 5-72 . 857 . 2-57 . 548 . 2-43 . 072 . 6-74 . 473 . 1-74 . 4Thai virusL segmentnt68-68 . 194 . 666 . 6-67 . 560 . 7-61 . 861 . 3-61 . 561 . 1-61 . 457 . 150 . 2-50 . 560 . 7-61 . 461 . 4-61 . 5L ORFnt67 . 3-67 . 595 . 167 . 9-68 . 859 . 6-59 . 959 . 9-60 . 461 . 6-62 . 155 . 7-56 . 249 . 9-50 . 860 . 4-61 . 659 . 9-60 . 2aa69 . 2-69 . 496 . 769 . 6-70 . 755 . 7-56 . 456 . 556 . 6-57 . 149 . 0-49 . 437 . 1-37 . 555 . 9-56 . 356 . 5-56 . 7Z ORFnt73 . 4-74 . 595 . 469 . 4-75 . 065 . 2-66 . 864 . 1-65 . 869 . 0-69 . 658 . 7-59 . 254 . 3-56 . 065 . 8-72 . 866 . 3-66 . 8aa79 . 298 . 573 . 1-74 . 667 . 5-71 . 468 . 870 . 158 . 442 . 9-44 . 270 . 1-71 . 466 . 2S segmentnt71 . 7-72 . 194 . 471 . 0-72 . 265 . 8-67 . 366 . 1-66 . 866 . 2-6761 . 7-62 . 653 . 9-55 . 065 . 2-68 . 166 . 1-67 . 9NP ORFnt73 . 1-74 . 494 . 672 . 2-74 . 165 . 2-67 . 866 . 6-67 . 366 . 9-67 . 762 . 4-63 . 254 . 1-54 . 466 . 1-68 . 067 . 3-68 . 9aa82 . 9-84 . 298 . 178 . 3-87 . 273 . 3-74 . 672 . 0-72 . 973 . 8-74 . 464 . 2-64 . 649 . 9-51 . 072 . 0-73 . 574-75 . 5GPCnt69 . 1-7094 . 169 . 2-69 . 966 . 0-67 . 665 . 0-66 . 166 . 4-6761 . 5-62 . 653 . 9-56 . 065 . 8-69 . 266 . 1-67 . 5aa79 . 5-81 . 197 . 580 . 1-81 . 173 . 5-75 . 771 . 5-72 . 471 . 0-74 . 259 . 7-61 . 043 . 0-43 . 974 . 2-74 . 873 . 1-75 . 9Note where ( L & S gene ) : Cambodian = KC669690 , KC669691& KC669694 , KC669696; Thai = KC669692 , KC669693 & KC669697 , KC669698; Wenhzhou = KM386661 , KM051421 , KJ909795 , KM051420 and KM386660 , KM051423 , KJ909794 , and KM051422; Lassa = GU481076 & GU481077; Ippy = DQ328877 & DQ328878; Mopeia = DQ328874 & DQ328875; LCMV = AY847350 & AY847351; Junín = D10072 & AY216507; ORF = Open reading frame; NP = Nucleoprotein; GPC = Glycoprotein; nt = Nucleotide; aa = Amino acid; LCMV =: Lymphocytic Choriomeningitis Virus10 . 7554/eLife . 13135 . 004Table 2 . Summary of PASC analysis . DOI: http://dx . doi . org/10 . 7554/eLife . 13135 . 00410 . 7554/eLife . 13135 . 005Table 2—source data 1 . PASC analysis . DOI: http://dx . doi . org/10 . 7554/eLife . 13135 . 005Sample IDCountry of originPASC: S segmentPASC: L segmentSequence identity ( % ) Closest virusSequence identity ( % ) Closest virusC0617Cambodia88 . 80Wēnzhōu virus86 . 27Wēnzhōu virusC0649Cambodia88 . 51Wēnzhōu virus85 . 98Wēnzhōu virusR4937Thailand70 . 63Wēnzhōu virus62 . 71Wēnzhōu virusR5074Thailand70 . 29Wēnzhōu virus63 . 08Wēnzhōu virus In maximum likelihood phylogenetic analyses , the Southeastern Asian viruses , along with the Chinese isolates of Wēnzhōu virus , formed an independent clade within the Old World mammarenaviruses . The Cambodian virus sequences clustered with the Wēnzhōu virus sequences , whilst LORV sequences clustered independently . This Asian mammarenavirus clade formed a sister clade to Ippy virus ( IPPYV ) in both full RNA-dependent RNA polymerase ( L ) and NP gene analysis , and to the African Lassa-related viruses as a whole in glycoprotein precursor ( GPC ) gene analysis ( Figure 1a and Figure 2 ) . 10 . 7554/eLife . 13135 . 006Figure 1 . Maximum likelihood phylogenetic tree of novel arenavirus isolates and other representative arenaviruses for a , the complete ORF of the L gene with sequences from rodents only , b , partial L sequences including sequences from rodents and patients . Cambodian strains detected in rodent ( triangle ) , human ( square ) and Thai strains detected in rodent ( circle ) are in bold . Clade A , B and C are three evolutionary lineages of New World arenaviruses within the Tacaribe complex . A/Rec denotes the recombinant clade including the three Northern American viruses . The virus names are in abbreviation according to Radoshitzky et al . ( 2015 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 13135 . 00610 . 7554/eLife . 13135 . 007Figure 2 . Maximum likelihood phylogenetic tree of novel arenavirus isolates and other representative arenaviruses for a , the complete ORF of GPC gene and b , complete ORF of NP gene . Cambodian strains ( triangle ) and Thai strains ( circle ) detected in rodents are in bold . Clade A , B and C are three evolutionary lineages of New World arenaviruses within the Tacaribe complex . A/Rec denotes the recombinant clade including Northern American viruses . The virus names are in abbreviation according to Radoshitzky et al . ( 2015 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 13135 . 007 To establish if infection and transmission of the Cardamones variant of Wēnzhōu virus could be replicated in rodents of two of the suspect host species , experimental infections were performed . Despite several attempts to obtain a culture isolates in tissue culture , including multiple passages , we were unable to propagate the Cambodian isolates in the cell lines available in our laboratory ( Vero E6 , MDCK , BHK ) , or in primary cell lines from lungs obtained from a Wistar laboratory rat . Experimental infections were therefore conducted using homogenised lung tissue from wild-caught RNA-positive animals to infect both Wistar laboratory rats and wild-caught or first generation Pacific rats captured in the Mondulkiri site . Male and female adult animals were used . Due to logistical limitations , parallel experiments were not conducted for LORV . Of 85 animals inoculated intraperitoneally with the Cardamones variant of Wēnzhōu virus-positive homogenate , 77 ( 64/68 Wistar laboratory rats; 13/17 Pacific rats ) were successfully infected as measured by detection of viral RNA by RT-PCR in organs and/or seroconversion by IFA and/or ELISA ( Supplementary file 1D ) . Organs positive for Wēnzhōu virus ( i . e . brain , heart , lung , liver , spleen , kidney , bladder , peritoneum ) varied between individuals ( data not shown ) , with individuals tested usually demonstrating a generalised infection . Organs from randomly selected adult individuals collected between 3dpi and 22dpi or 28dpi for Wistar laboratory rats and Pacific rats respectively , were subjected to quantitative real-time RT-PCR ( qRT-PCR ) and copy number was determined . Copy numbers increased over time with high copy numbers ( 108–1011 cDNA copy/mg ) detected in several organs from 7dpi in Pacific rats and 11dpi in Wistar laboratory rats . Specifically , high copy numbers were detected in the liver , spleen and lung of most individuals , with the latter suggesting the potential for respiratory transmission . Mammarenavirus RNA was detected in most inoculated Wistar laboratory rats killed between 3-56dpi ( 52/68 individuals ) and Pacific rats killed between 3-37dpi ( 11/17 individuals ) , although no animals were tested beyond these time points . Mammarenaviruses are known to cause both acute and chronic infections in their rodent hosts 17 and the duration of infection observed here is comparative to that seen in experimental infections of the natural hosts of other mammarenaviruses ( Fulhorst et al . , 1999; Walker et al . , 1975 ) . However , as lung homogenates were used instead of titrated virus , the variability of the dose received by each animal may have impacted on the infection course , likely resulting in the variations and inconsistencies observed ( Childs et al . , 1993 ) . In both species , we observed horizontal transmission , but only between pups and their dams ( Supplementary file 1D ) . Vertical transmission could not be confirmed for either species . Mammarenavirus RNA was detected in a single pup from each of the two Pacific rat litters but all other pups from these litters were negative , and antibodies detected in additional pups may have been of maternal origin . LCMV infection of house mice must occur shortly before or after mating in order for dams to transmit the virus to a high proportion of their pups ( Skinner and Knight , 1974 ) . Therefore the absence of vertical transmission observed here could have been due to the late stage of gestation at which most dams were inoculated ( Supplementary file 1D ) , although vertical transmission has not been confirmed in all mammarenaviruses ( Mills et al . , 2007 ) . Clinical signs are rarely seen in mammarenavirus rodent reservoirs , although more subtle signs including reduced weight and body mass index have been observed ( Borremans et al . , 2011 ) . To establish if Wēnzhōu virus infection of rodent hosts results in lesions , histopathological analysis and chromogenic immunohistochemistry were performed on tissues from two Pacific rats inoculated intra-peritoneally with lung tissue homogenates obtained from a wild rat that tested positive for Cardamones variant of Wēnzhōu virus . Electron microscopy clearly identified virus particles with a specific mammarenavirus-type morphology ( Figures 3a and b ) in the lung tissues . Histopathological examination of the lungs revealed only severe diffuse pneumonia ( Figures 3c–h ) . Other organs were not affected except for vascular congestion , probably due to pre mortem conditions . This relatively limited pathology is in congruence with the suspected reservoir status of this species . 10 . 7554/eLife . 13135 . 008Figure 3 . Electron microscope images of the Cardamones variant of Wēnzhōu virus in the rat lung tissues , a ( scale: 200 nm ) & b ( scale: 2 µm ) . Histopathological examination of the lungs revealed only severe diffuse pneumonia , with lesions associated with acute exudative inflammation characterised by foci of consolidation surrounded by extremely rare aerated parenchyma , c , vascular congestion , acute bronchiolitis and diffuse leukocytic infiltration with lymphocytes , polymorphonuclear and macrophages ( suppurative exudate in the lumen and parietal inflammation ) , c and d . Reticulin staining demonstrated the severe destruction of lung parenchyma , e . Chromogenic immunohistochemistry identified inflammatory foci with numerous positive cells , f , primarily inflammatory cells including macrophages , but in the more preserved parenchyma and aerated parenchyma some epithelial ( pneumocytes ) , g , and alveolar macrophages , h , were also clearly stained . DOI: http://dx . doi . org/10 . 7554/eLife . 13135 . 008 The peridomestic nature of the mammalian hosts of Wēnzhōu virus ( Ivanova et al . , 2012 ) , indicates that this virus would have ample opportunities to transmit to humans . We therefore decided to establish if zoonotic transmission does occur and if it is associated with clinical disease by testing human clinical samples from eight different sources ( Table 3 ) . In group 1 , a total of 89/510 patients presenting with dengue-like and influenza-like syndromes but who tested negative for dengue and influenza infection ( 17 . 4% ) tested positive for anti-arenavirus IgG . The mean age of positive patients was 13 years ( 95% confidence interval ( CI ) : 10 . 4–15 . 6; range: 4 months to 70 years ) which was not statistically different ( p=0 . 06 ) from negative patients ( mean age: 10 . 6 years; 95% CI: 9 . 6–11 . 6; range: 7 months to 65 years ) . The highest antibody prevalence was detected in the 6 to 10 year old age group ( 40 . 5% ) . There was no difference in the sex ratio between seropositive and seronegative patients ( 49 . 5% and 43% , respectively; p=0 . 29 ) . The seroprevalence was particularly high in some villages of Kampong Cham province ( Figure 4 ) . Seroconversions were observed in 7 young patients with undocumented fever or influenza-like illness ( Table 4 ) . These patients had a mean age of 8 . 1 year ( 95% CI: 4 . 3–12; range: 3–16 years ) with a sex ratio of 0 . 71 , which were not significantly different to mammarenavirus negative patients with paired sera ( mean age: 9 . 6 years; 95% CI: 8 . 2–10 . 9; range: 7 months to 65 years; p=0 . 67; sex ratio: 0 . 45; p=0 . 17 ) . Intervals between the collection of the acute and the convalescent serum samples for these patients ranged between 2 and 15 days with a mean of 7 . 4 days ( Supplementary file 1E ) . Retrospective clinical data were available for two of these patients . The first case ( S1 ) was a 3 year old boy who presented with fever ( 38°C ) , rhinorrhea and nausea . The second case ( S2 ) was a 9 year old boy presenting with mild fever ( 37 . 5°C ) , headache , cough , rhinorrhoea , and nausea . Muscle pain , joint pain , rash and bleeding were absent in both cases . Only fever ( 37 . 5°C to 38 . 9°C ) was recorded in the five remaining cases ( S3-7 ) . Of note , self-treatment with antipyretics is extremely common in Cambodia and therefore the temperature measured at the time of medical examination is often normal . In addition to clinical patients , 529 samples from healthy individuals from Kampong Cham province who participated in a community-based dengue seroprevalence study ( group 2 ) were also screened for anti-mammarenavirus IgG antibodies and 70 ( 13 . 23% ) were found positive ( Table 3 ) . The mean age of positive individuals was 10 . 71 ( 95% CI: 9 . 76–11 . 67; range: 1 . 1–19 years ) with a sex ratio 0 . 47 . The positivity rate and mean age of positive cases in healthy individuals compared to patients presenting with dengue-like and influenza-like symptoms were not significantly different ( p=0 . 059 and p=0 . 09 , respectively ) . A limited number of human samples ( Total = 372 , Supplementary file 1F ) were tested by both mammarenavirus IFA and IgG ELISA , demonstrating in general consistent results ( coefficient of correlation of 0 . 565 with a p<0 . 001 ) . The comparison those two assays is only indicative as the tests were used for different purposes , due to different properties . IFA was chosen for screening purposes because of the ability of the test to detect antibodies directed against a wide range of arenaviruses while the ELISA was developed with the nucleoprotein of the Cardamones variant of Wēnzhōu virus is order to increase the specificity of the serology . Indeed , the Wēnzhōu virus IgG ELISA was tested against LCMV positive human sera , the only other zoonotic mammarenavirus currently presumed to be present in Southeastern Asia , but no serological cross-reaction was observed . The IgG ELISA seemed to be slightly more sensitive than the mammarenavirus IFA , detecting 11 additional positive samples ( Supplementary file 1F ) . The mammarenavirus IFA identified 2 samples found negative by the IgG ELISA as positive . However these infections may have been caused by more distantly related viruses , such as LCMV , not capable of detection by the more specific IgG ELISA . Despite this , as mammarenaviruses are notoriously serologically cross-reactive by ELISA ( Fukushi et al . , 2012 ) it is unlikely that this assay is perfectly specific to Wēnzhōu virus . Instead it is likely that this test would be capable of detecting other closely related arenaviruses , including novel species . However antisera to other arenaviruses were not available for testing . As for most serological assays , the precise causative agent in these seropositive patients can therefore not be identified conclusively . 10 . 7554/eLife . 13135 . 009Table 3 . Details of patient samples tested . DOI: http://dx . doi . org/10 . 7554/eLife . 13135 . 009GroupClinical signsSample type ( number of samples ) Collection datesCollection locationsMean age in years ( range ) 95% CI ( years ) Sex ratioTest usedNumber ( proportion ) of positive1Dengue-like/ influenza-like illnessAcute and convalescent sera ( n=510 including 98 acute , 214 convalescent and 198 paired sera ) 2005-2010Kampong Cham , various11 . 0 ( 4 months to 70 years ) 10 . 0-12 . 00 . 44IgG ELISA89 ( 17 . 4% ) - Acute sera: 33 ( 33 . 7% ) - Convalescence sera: 25 ( 11 . 7% ) - Paired sera: 31 ( 15 . 7% ) with evidence of seroconversion in 7 pairs ( 3 . 5% ) 2Healthy individuals ( community dengue seroprevalence study ) Sera ( n=529 ) 2009Kampong Cham10 . 0 ( 1 month-20 years ) 9 . 59-10 . 420 . 50IgG ELISA70 ( 13 . 23% ) 3Meningo-encephalitisCerebrospinal fluid ( n=200£ ) 2013-2014Various6 . 5 ( 3 months-15 years ) 5 . 95-7 . 060 . 40Real Time RT-PCR0 ( 0% ) 4Dengue-like febrile illnessSera collected during febrile stage ( n=253* ) 2009 , 2011-2013Various8 . 2 ( 1 to 38 years ) 7 . 6-8 . 80 . 48Semi-nested RT-PCR0 ( 0% ) 5aInfluenza-like illness ( negative for four common respiratory viruses ) Nasopharyngeal swabs ( n=328 ) 2007-2012Various13 . 2 ( 1 month to 83 years ) 12 . 0-14 . 40 . 51Semi-nested RT-PCR4 ( 1 . 2% ) 5bInfluenza-like illness ( positive for four common respiratory viruses ) Nasopharyngeal swabs ( n=392# ) 0 ( 0% ) 6Acute lower respiratory infectionNasopharyngeal swabs ( n=279$ ) 2008-2009Various2 . 7 ( 7 months to 63 years ) 2 . 1-3 . 40 . 41Semi-nested RT-PCR2 ( 0 . 7% ) 7Healthy individuals ( H5N1 contacts ) – negative controlNasopharyngeal swabs ( n=266§ ) 2005-2011Various29 . 3 ( 1 to 79 years ) 27 . 1-31 . 50 . 45Semi-nested RT-PCR0 ( 0% ) 8Healthy anti-rabies vaccination volunteers – negative controlNasopharyngeal swabs ( n=238¥ ) 2013Institute Pasteur Cambodia ( Phnom Penh ) 26 . 7 ( 9 months to 83 years ) 24 . 5-28 . 80 . 56Semi-nested RT-PCR0 ( 0% ) CI = Confidence Interval£ 3 no information on age; * 8 no information on age and sex; # 2 no information on age and sex; $ 1 no information on age; § 5 no information on age and sex; 51 no information on sex; ¥ 5 no information on age and sex10 . 7554/eLife . 13135 . 010Figure 4 . Representation map of the percentage ( % ) of patients who tested positive by anti-arenavirus IgG ELISA and of 6 patients who tested positive by L gene RT-PCR in the villages from Kampong Cham province . Village names: AC=Andoung Chea , AK=Ampil Kraom , AL=Ampil Leu , AS=Andoung Svay , BH=Banteay Thma , BT=Boeng Tras , CH=Chachak , CM=Chong Thnal Muoy , CP=Chong Thnal Pir , KD=Kdei Boeng , KH=Kakaoh , KK=Kouk Kream , KP=Krasang Pul , KR=Krala , MM=Memay , OD=Ou Da , PP=Prey Phdau , RK=Roung Kou , RM=Romeas , SS=Srae Siem , SY=Sya , TA=Tuol Ampil and TV=Tuol Vihear . DOI: http://dx . doi . org/10 . 7554/eLife . 13135 . 01010 . 7554/eLife . 13135 . 011Table 4 . Clinical details of human arenavirus infections . DOI: http://dx . doi . org/10 . 7554/eLife . 13135 . 011Case no . Test usedSexAgeHospitalisedFeverCoughRhinorrheaNausea/ vomitingSevere headacheMuscle painOther symptomsCo-infection1Screening nested RT-PCRFemale3 yearsNoYesYesYesHuman para-influenza virus 12Screening nested RT-PCRMale9 yearsNoYesYesYesYesYes3Screening nested RT-PCRMale31 yearsNoYesYesYesYesYes4Screening nested RT-PCRFemale45 yearsNoYesYesYesYesYesYes5Screening nested RT-PCRFemale8 monthsYesYesYesDyspnea , wheezing , moderate anaemia ( 93g/L ) Unspecified Rhinovirus6Screening nested RT-PCRMale3 monthsYesYesYesDyspnea , wheezing , hypoglycaemia ( 28mmol/L ) Unspecified RhinovirusS1ELISAMale3 yearsNoYesYesYesS2ELISAMale9 yearsNoYesYesYesYesYesS3-7ELISANoYes The remaining six groups were tested by a 'screening' semi-nested RT-PCR derived from the RT-PCR method previously described by Vieth et al . ( Supplementary file 1A ) or by the qRT-PCR developed in this study . All the 200 patients who presented with signs and symptoms of meningo-encephalitis ( group 3 ) and tested negative by PCR and serology for the main etiological agents of central nervous system infections , tested negative for mammarenavirus in the cerebrospinal fluid . No trace of mammarenavirus RNA was detected in any of the 253 sera collected during the acute febrile phase of patients presenting with possible dengue fever or dengue hemorrhagic fever ( group 4 ) , and who tested negative for dengue virus infection . But mammarenavirus RNA was detected in the respiratory specimens of 6 of 999 ( 0 . 6% ) Cambodian patients from groups 5 and 6 , who presented with either mild ( ILI: influenza-like illness ) or more severe ( ALRI: acute lower respiratory infection ) respiratory symptoms respectively ( Table 3 and Table 4 ) . In group 5a , the first case was a 3 year-old female with no known underlying disease , sampled 1 day after the onset of fever associated with cough and rhinorrhea . A co-infection with human parainfluenza virus 1 ( HPIV-1 ) was also detected . Case 2 was a 9 year-old male with no known underlying disease sampled 2 days after onset of fever associated with cough , rhinorrhea , nausea , vomiting and severe headache . Case 3 was a 31 year-old male farmer , who presented 2 days after the onset of fever associated with productive cough , rhinorrhea , nausea and severe headache . Case 4 was a 45 year-old female farmer sampled 1 day after onset of fever associated with productive cough , rhinorrhea , nausea , severe headache and muscle pain . None of these 4 patients reported other signs and symptoms often associated with influenza ( otalgia , sore throat ) or dengue ( rash , retro-orbital pain , joint pain , bleeding ) . Hospitalization was not required in any of these cases . In group 6 the first case ( case 5 ) was that of an 8 month-old female with a recent history of fever ( <4 days ) who presented to hospital with fever ( 38°C ) , productive cough , cardiac frequency at 130bpm , respiratory rate at 46/min , dyspnea and wheezing . The blood cell count performed at admission demonstrated anaemia ( haemoglobin: 93 g/L ) , a normal white cell count and formula ( leucocytes: 12 . 6x109/L , lymphocytes: 5 . 2x109/L , neutrophils: 5 . 7 109/L ) . There was no liver cytolysis ( SGOT: 46 U/L; normal value: 5–40 U/L , SGPT: 33 U/L; normal value: 5–50 U/L ) , the creatinine was normal ( 40 µmol/L ) as was glycaemia ( 5 . 0 mmol/L ) . The chest radiograph which was retrospectively reviewed by an expert pulmonologist was normal and the final clinical diagnosis was of an acute bronchiolitis of probable viral origin . The patient received amoxicillin , gentamicin and bronchodilators for 4 days and was discharged after full recovery . The second hospitalized case ( case 6 ) was a 3 month-old boy , admitted with a history of fever but with a normal body temperature at the time of medical examination , associated with productive cough , cardiac frequency of 132bpm , respiratory rate of 50/min , dyspnea and wheezing . The blood cell count ( haemoglobin: 112 g/L , leucocytes: 11 . 6x109/L , lymphocytes: 7 . 2x109/L , neutrophils: 2 . 4x109/L ) and renal function ( creatinine: 47 µmol/L ) were normal for the patient’s age . A hypoglycaemia of 2 . 8 mmol/L was reported . SGOT and SGPT were at 53 U/L and 44 U/L , respectively . The independent expert pulmonologist who reviewed the chest radiograph retrospectively described patchy infiltrates and confirmed the initial diagnosis of acute bronchiolitis of probable viral origin made by the hospital paediatrician . The patient fully recovered and was discharged after 3 days of treatment with bronchodilators , amoxicillin and gentamicin . In both of the hospitalised cases a rhinovirus was also detected in the nasopharyngeal swab samples , but it should be noted that rhinovirus infection is not always associated with symptomatic disease ( Buecher et al . , 2010 ) . The association between the respiratory illness and the detection of the mammarenavirus was statistically significant ( p<0 . 05 ) ( Supplementary file 1G ) . Nevertheless , probably because of the limited sample sizes , the difference remained statistically significant only in adults over 30 years of age when the analysis was stratified by age groups ( Supplementary file 1H ) . Direct sequencing of the 302bp amplicons of L gene obtained from three of these six respiratory specimens identified Wēnzhōu virus ( Figure 1b ) with a 98–99% similarity to the strain detected in rodents from Veal Renh and 88 . 9–91% with Wēnzhōu virus isolated in China . Based on the absence of contamination of negative controls , and because 2 to 10 nucleotides differences were identified between the patient sequences and those of the positive control of rodent origin ( Supplementary file 1I ) , the positive patient samples are extremely unlikely to be a result of cross-contamination . In addition , the respiratory samples from 2 patients who tested positive by the screening nested RT-PCR were also tested by a qRT-PCR that targets a totally different region of the L gene ( Figure 5 and Supplementary file 1A ) and both samples tested positive with a viral load of 200 and 350 equivalent cDNA copies per reaction , respectively ( or 17 , 900 and 30 , 600 copies/ml of sample in virus transport medium ) . All the 6 patients were additionally tested for 17 common respiratory viruses by RT-PCR ( Buecher et al . , 2010 ) and co-infections were detected in three out of six patients presenting with respiratory symptoms . The clinical data recorded for these patients ( cases 1–6 ) are described in Table 4 . All the human mammarenavirus cases detected by molecular methods originated from Kampong Cham province . Although a total of 19 villages of the Kampong Cham province were included in the community-based study that comprised group 5 , all the 4 cases detected in sub-group 5a originated from Krasang Pul village ( KP ) and were obtained between January and July 2009 . Of note , patients from KP were over-represented ( 32% ) amongst the total number of patients tested in sub-group 5a . The two individuals found mammarenavirus positive in group 6 were from two other villages , Roung Kou ( RK ) and Ampil Kraom ( AK ) , in Kampong Cham province , both of which are situated approximately 8 km distant from KP ( Figure 4 ) . This , in conjunction with the relatively high mammarenavirus seropositivity rates also observed in this province ( Figure 4 ) , suggests that another focus of Wēnzhōu virus exists here . 10 . 7554/eLife . 13135 . 012Figure 5 . Positions of primers used in diagnostic PCR assays . The positions of the primers used for qRT-PCR are included in their names and these positions are based on the sequence of the L gene of Cardamones variant of Wēnzhōu virus . The positions of the primers used in the nested RT-PCR originate from the article of Vieth et al . ( Vieth et al . , 2007 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 13135 . 012
The identification of a novel mammarenavirus in Thai rodents and of a genetic variant of Wēnzhōu virus in Cambodian rodents is the first unequivocal evidence of mammarenaviruses in Southeastern Asia . The identification of the latter virus in Cambodia in individuals of R . exulans sympatric to infected individuals of the known host , R . norvegicus , expands both the geographic and host range of this mammarenavirus species . Although LCMV was not identified in this study ( probably due to the absence of its host the house mouse from large regions of Southeastern Asia ( Aplin et al . , 2003 ) and the study sites in particular ) it is also proposed to have an Asian origin ( Albariño et al . , 2010 ) . Despite this shared geographic heritage , both Wēnzhōu virus and LORV appear to be more closely related to the African LASV/IPPYV group viruses . This is also surprising from the viewpoint of the rodent-mammarenavirus co-evolution hypothesis . Based on this hypothesis a closer relationship would be expected between LASV/IPPYV and LCMV , since the rodent hosts ( Arvicanthis , Mastomys and Praomys species ) ( Childs et al . , 1993; Lecompte et al . , 2005 ) of the African mammarenaviruses are more closely related to Mus spp . than they are to the Asian Rattini tribe ( Lecompte et al . , 2008 ) . As such this finding alongside the recent discovery of divergent arenaviruses in boid snakes ( recently classified as reptarenaviruses , the sister genus to mammarenaviruses ) and more in depth analyses of rodent-mammarenavirus relationships , suggest that the history of the family Arenaviridae is more complex and the theory of co-evolution less robust than initially thought ( Coulibaly-N'Golo et al . , 2011; Irwin et al . , 2012; Stenglein et al . , 2012 ) . Although the host status of rodents of several Rattus spp . could be called into question based on the pathologies observed in some animals experimentally infected with Cardamones variant of Wēnzhōu virus , detrimental effects of infection have been observed in the natural host of another mammarenavirus , Mopeia virus ( Borremans et al . , 2011 ) , and in the current study may have resulted from the unnatural route of infection . In addition , the successful experimental infection and transmission of Wēnzhōu virus in rodents of two of the proposed host species , alongside the detection of this virus in rodents of several Rattus spp . from two geographically remote locations ( Li et al . , 2014 ) , support their status as the reservoir hosts . Due to the relative host specificity of mammarenaviruses , confirmation of the host species of each of these viruses has important implications for their geographic distributions . Many of the species in which Wēnzhōu virus or LORV have been detected are widely distributed ( Aplin et al . , 2003; Ruedi et al . , 1996; Wodzicki and Taylor , 1984 ) . Rodents of these species are also associated with either agriculture ( Bandicota spp . ) or urban settings ( Rattus spp . ) ( Aplin et al . , 2003 ) and as such are likely to be among the few species that may benefit from urbanisation and agricultural intensification . This could ultimately result in increases in both the distribution of their viruses and in their opportunities for zoonotic transmission . Our results suggest that the ecology of both viruses , through their association with peridomestic rodents of several species , mirrors that observed for known zoonotic mammarenaviruses . Like LORV , JUNV is associated with agricultural pest species ( Busch et al . , 2000 ) , whilst Wēnzhōu virus , LASV and LCMV are all hosted by rodents with a preference for human dwellings ( McCormick et al . , 1987; Pocock et al . , 2004 ) . This association with highly utilised human habitats indicates that humans are potentially exposed to both viruses on a regular basis as suggested by the relatively high seroprevalence observed in both the patient and the control group . Therefore , the identification of human Wēnzhōu virus infection is perhaps unsurprising from an epidemiological viewpoint , but nevertheless , represents the first detection of human mammarenavirus infection in Asia . The results from this study suggest that human infection with Wēnzhōu virus , or a closely related virus , occurs widely throughout Cambodia , and may possibly result in disease in some cases . Confirmation of causality however is made problematic by the mild and indistinctive clinical presentation observed , the detection of co-infecting pathogens capable of producing a similar clinical picture in three out of six cases ( although rhinovirus infection is not always symptomatic ) ( Buecher et al . , 2010 ) and the limited number of cases . Although virus was never detected in healthy individuals and all patients with confirmed Wēnzhōu virus infections experienced similar symptoms of acute febrile bronchitis , the non-detection of co-infecting agents in three cases despite extensive testing does not mean that other untested or unknown pathogens were not present . However the clinical picture observed was similar to that reported for milder LCMV infection ( Baum et al . , 1966; Knust et al . , 2014 ) . If like LCMV , Wēnzhōu virus rarely causes distinctive severe disease , then despite the high apparent incidence of the infection indicated by the retrospective serological study , causality is unlikely to be confirmed without further surveys . Although our findings indicate that novel mammarenaviruses are present in Southeastern Asia and that humans appear to be regularly infected in the region , there are several issues that still need to be addressed . A major one was our lack of success in isolating both Cardamones variant of Wēnzhōu virus and LORV in tissue culture . This was surprising as samples clearly contained viable virus as evidenced by successful experimental infection of rats . It was also inhibitive to downstream analyses , as a successful method of isolation could greatly aid in the development of serological tests and in confirming proof of causality in suspected cases of human disease . The reasons behind our inability to isolate these viruses are unknown . It is possible that the viruses may have replicated too slowly to be detected by the methods used but this seems unlikely based on the successful isolation of other arenaviruses under similar conditions ( Gómez et al . , 2011 ) . More likely is the possibility that samples contained viral titres that were too low for successful isolation in the cell lines tested ( Lednicky et al . , 2012 ) or that cell lines were contaminated with Mycoplasma . Isolation attempts were made in cell lines previously used to culture other mammarenaviruses ( e . g . Vero , MDCK , BHK ) , but as Wēnzhōu virus was isolated in canine macrophage DH82 cell line ( Li et al . , 2014 ) , it would be prudent to attempt isolation in this cell line in the future . A serological test highly specific to Wēnzhōu virus is also needed . As such , the specificity of the ELISA test developed during this study needs to be assessed against other mammarenaviruses and potentially optimised further , or an alternative strategy employed . In order to strengthen the case for Wēnzhōu virus as an agent of human disease , future surveys would ideally include a comparison of the seroprevalence in exposed and non-exposed populations , using such a Wēnzhōu virus-specific test . Although attempts were made during this study to obtain sera from individuals not native to the Southeastern Asian region , only limited samples were available . Many of these were from individuals who had been resident in this region for some time and therefore may have been exposed to Southeastern Asian mammarenaviruses . Sourcing and testing of sera from non-endemic areas is therefore needed . Studies aimed at detecting seroconversion in conjunction with molecular testing for viral RNA are also required , with the ultimate goal of isolating the virus in tissue culture or rodents . If Wēnzhōu virus does cause disease , it is also possible that like LASV and LCMV , more severe disease may occur in particular individuals such as pregnant women and immuno-compromised individuals ( Bonthius , 2012; Price et al . , 1988 ) and this too should be examined . Globally infectious diseases remain a major cause of mortality and morbidity resulting in more than 15 million deaths per annum ( Morens et al . , 2004 ) . However it is clear that many of the pathogenic agents responsible still remain undiagnosed , even when they are associated with potentially life threatening conditions . To improve protection we need a better understanding of both the ecology and epidemiology of diseases if we are to succeed in preventing them . Our findings illustrate how a 'One Health' approach in a regional hot-spot for emerging infectious disease can lead to the detection and characterisation of viruses likely associated with previously unrecognized human infections .
No species included in this study are considered to be endangered or threatened and none are included on either the CITES list or the Red List ( IUCN ) . Approval notices for trapping and investigation of rodents were provided by the Ministry of Health Council of Medical Sciences , National Ethics Committee for Health Research ( NHCHR ) in Laos ( number 51/NECHR ) and by the Ethical Committee of Mahidol University , Bangkok , Thailand ( number 0517 . 1116/661 ) and permission to trap in Cambodia was granted by the Cambodian Ministry of Environment . As Cambodia has no ethics committee overseeing animal experimentation , animals were treated in accordance with the guidelines of the American Society of Mammalogists , and within the European Union legislation guidelines ( Directive 86/609/EEC ) . Rodents were trapped in the Cambodian provinces of Mondulkiri ( 12°120’N; 106°890’E ) and Veal Renh ( 10°710’N; 103°820’E ) , the Thai provinces of Buriram ( 14 . 89 N; 103 . 01 E ) , Loei ( 17 . 39 N; 101 . 77 E ) and Nan ( 19 . 15 N; 100 . 83 E ) and the Laotian provinces of Champasak ( 15 . 12 N; 105 . 80 E ) and Luang Prabang ( 19 . 62 N; 102 . 05 E ) as part of the CERoPath ( Community Ecology of Rodent-borne Pathogens ) project ( www . ceropath . org ) . Sites representing a range of habitats with differing degrees of human disturbance were trapped at each location . Two sampling sessions were conducted at each locality with 30 lines of ten traps set over four nights during each session , amounting to 1 , 200 trap nights per session . Locally made , wire live-traps ( approx 40*12*12 cm ) were used at each locality and were baited with cassava , banana or sticky rice . Captured rodents were collected each day , humanely euthanized and blood pellet , serum and organs were harvested , stored temporarily in liquid nitrogen and then transferred to long-term storage at -80ºC . Selected samples were subjected to RT-PCR ( see below ) for evidence of arenavirus infection . Rodent lung samples were homogenised using the MagNA Lyser instrument and bead system ( Roche , Basel , Switzerland ) . Tissue samples were placed in tubes containing ceramic-beads and 500 µl of pre-chilled sterile 1X PBS , subjected to oscillation at 3500 rpm and the lysate centrifuged at 6 , 200rpm for 5 min on a bench-top centrifuge . The RNA from homogenised rodent lung samples , human sera and respiratory specimens was extracted using the QIAmp Viral RNA mini kit ( Qiagen , Hilden , Germany ) as per the manufacturer’s protocol . Rodent samples were screened using a one-step RT-PCR targeting a conserved 395bp region of the L segment , using primers designed for Old World mammarenavirus detection and described previously by Vieth et al . ( Vieth et al . , 2007 ) . To allow greater sensitivity a semi-nested RT-PCR ( 'screening' RT-PCR ) was developed for the human samples , using the Vieth et al . method as the first step , followed by a nested step using primer LVL3359plus ( Vieth et al . , 2007 ) and a primer designed against Cardamones variant of Wēnzhōu virus , CAMN1 ( Figure 5 and Supplementary file 1A ) . Briefly , 1 µL of the first round reaction was added to a PCR mix comprising 13 µl Superscript II 2X reaction buffer ( Life Technologies , Carlsbad , USA ) , 1 µl each of primers LVL3359plus and CAMN1 ( each at 10 µM ) ( Table 1 ) , 1 µL Platinum Taq polymerase ( Life Technologies , Carlsbad , USA ) and molecular grade water to a volume of 25 µl . A product size of 302 bp was generated for the semi-nested assay and addition of this step increased sensitivity by 1 Log-fold ( data not shown ) . In addition a quantitative real-time RT-PCR ( qRT-PCR ) was developed and carried out using the SuperScript III Platinum One-Step qRT-PCR Kit ( Life Technologies , Carlsbad , USA ) and optimized according to the manufacturer’s instructions . A volume of 5 µl of RNA was mixed with 12 . 5 µl of 2X Reaction Mix ( 0 . 4 mM of each dNTP , 3 . 2 mM MgSO4 ) , 2 µM of the forward primer ARV2747-F , 1 µM of each of the reverse primers ARV-2937-R and ARV-3012R , 1 µM of ARV-probe , 0 . 5 µl of RNasin Plus RNase Inhibitor ( 40 U/µl; Promega , Madison , USA ) and 1 µl of SuperScript III Platinum One-Step qRT-PCR enzyme to a final reaction volume of 25 µL . Thermocycling conditions were as follows: reverse transcription at 42°C for 30 min , denaturation at 95°C for 2 min and fluorescence detection for 45 cycles of 95°C for 30 sec , 50°C for 30 sec and 72°C for 1 min . Amplification and detection was performed on a LightCycler 480 II system ( Roche , Basel , Switzerland ) and amplification curves with CT values >35 were considered negative with the threshold line placed above the background signal , intersecting the initial log phase of the curve . In order to quantify mammarenavirus RNA , 395 bp PCR products were generated by conventional RT-PCR using the degenerate primers ARV2747-F and ARV3012-R ( Figure 5 and Supplementary file 1A ) . The cloning process was performed using the TA cloning kit ( Invitrogen , Carlsbad , USA ) according to the manufacturer recommendations . Plasmids containing the sequence of Cardamones variant of Wēnzhōu virus were diluted ( 1 to 1 , 000 , 000 copies/reaction ) and included in each series of quantitative real-time RT-PCR . To obtain enough material for full genome sequencing , filtered supernatant from homogenized lung ( 50 mg in 500 μl PBS ) from an infected , wild-caught Pacific rat ( C0649 ) and from an infected , wild-caught Savile’s bandicoot rat ( R5074 ) were each used to inoculate one adult male Wistar laboratory rat by intra-peritoneal injection ( 500 μl ) . Both rats were sacrificed at 7 days post inoculation ( dpi ) and organs collected ( see below for methodology ) . Total RNA from homogenised lungs were extracted using standard Trizol reagent as per the manufacturer’s instructions ( Life Technologies , Carlsbad , USA ) and confirmed mammarenavirus-positive by RT-PCR ( see above ) . Synthesis of cDNA , amplification , Illumina sequencing ( on HiSeq 2000 ) and bioinformatics analysis were performed as described previously ( Cheval et al . , 2011 ) . Using the program SOAPaligner ( http://soap . genomics . org . cn ) , the mouse ( Mus musculus ) genome 'NCBI37/mm9 assembly' reference genome , was used to filter out rodent genome-derived reads . Remaining reads were assembled using 3 different softwares: Velvet ( www . ebi . ac . uk/~zerbino/velvewww . ebi . ac . uk/~zerbino/velvet ) , SOAPdenovo ( http://soap . genomics . org . cn ) and CLC Genomics Workbench ( www . clcbio . com ) . Overlapping contigs assigned to an 'arenavirus' taxonomy were merged together to form longer sequences . Primers were designed based on the sequences generated by Illumina sequencing and used to obtain the coding complete genome sequences from two further isolates ( C0617 and R4937; Supplementary file 1A ) . To obtain S segment data a previously published protocol designed to generate full S segment amplicons and clones ( Bowen et al . , 2000 ) was used in conjunction with a series of RT-PCRs using the primers designed in this study ( protocols available on request ) and Sanger sequencing of the resultant amplicons . L segment sequence data was obtained using the primers designed in this study in a series of overlapping one step RT-PCRs ( protocols available on request ) and Sanger sequencing . Segments were reconstructed in the software CLC Main Workbench 5 . 5 ( CLC bio A/S , Aarhus , Denmark ) . All genome sequences were submitted to GenBank on 20 February 2013 . Coding complete sequences for both segments of each virus were loaded into the PASC tool , accessible at the National Center for Biotechnology Information ( NCBI ) website ( http://www . ncbi . nlm . nih . gov/sutils/pasc/viridty . cgi ? textpage=overview ) , and analysed using the default parameters . The sequences of New World and Old World mammarenavirus reference strains were retrieved from GenBank . Including the Cardamones variant of Wēnzhōu virus and Loei River virus sequences a total of 28 and 36 mammarenaviruses coding sequences for the L and S segments respectively , were aligned using Muscle ( Edgar , 2004 ) . The nt and aa identity matrices of the coding sequences were calculated by pairwise comparisons using p-distance in MEGA 5 . 2 ( Tamura et al . , 2011 ) . Identity matrices were calculated for the Zinc protein ( Z ) and L coding sequences of L segment and the NP and GPC coding sequences of S segment separately . Jmodeltest ( Posada , 2008 ) was used to select the optimal evolution model by evaluating the selected parameters using the Akaike Information Criterion ( AIC ) . Phylogenetic analyses were performed using maximum likelihood ( ML ) method available in Seaview version 4 . 2 . 5 ( Galtier et al . , 1996; Gouy et al . , 2010 ) using the recommended model GTR+G+I for the coding sequences of L , NP and GPC gene separately . The robustness of nodes was assessed with 1000 bootstrap replicates . All work with infected rodents was carried out in a Bio-safety level 3 animal facilities and adhered to standard European guidelines for animal ethics ( an animal ethics committee does not exist for Cambodia ) . Animals were kept in polyester filter bonneted cages , within a laminar flow Bio-isolator unit . Manipulations were carried out in a class 2 biosafety cabinet inside the BSL3 facilities . Animals were inoculated by intra-peritoneal injection , with 500 µl ( adults ) or 100 µl ( juveniles ) volumes and negative control animals were mock-inoculated with sterile PBS . At the end of each experiment animals were euthanized by exposure to chloroform followed by cervical dislocation . Personal protective equipment worn by the laboratory personnel comprised also a 3M Jupiter Tyvek laminate cape powered respirator ( 3M Jupiter , Diegem , Belgium ) , disposable gowns , shoe covers , and latex rubber gloves . Randomisation and blinding were not performed . Experiments were set up to test for: 1 ) the potential for infection and its estimated duration; 2 ) the potential for horizontal transmission; 3 ) the potential for vertical transmission . Animals used to estimate duration of the infection were housed individually . For horizontal transmission experiments , animals were housed in pairs ( 1 inoculated and 1 un-inoculated individual ) and dams were housed with their pups for vertical transmission experiments and neonatal inoculations . As numbers of animals were limited , experiments used to approximately estimate the duration of infection and potential for horizontal and vertical transmission were conducted at a number of time points with the number of animals available ( see Supplementary file 1C ) . All animals were confirmed negative for mammarenavirus infection by RT-PCR and IFA ( see below for details of method ) before starting experiments . At the end of each experiment , blood was collected by cardiac puncture and serum separated by centrifugation . Brain , peritoneum , bladder , spleen , liver , kidney , lung and heart were collected immediately after death . Sera were tested by IFA and where the volume of blood was sufficient by ELISA ( see below ) and organs were subjected to RT-PCR ( Vieth et al . , 2007 ) and qRT-PCR for selected individuals . Animals were deemed to have been infected if they demonstrated sero-conversion and/or were positive for mammarenavirus RNA . For electron microscopy lung tissue from infected animals were collected , cut into small pieces , fixed in 2 . 5% glutaraldehyde 0 . 1 mol/L phosphate ( Sorensen’s buffer , pH 7 . 4 ) and post-fixed in 2% osmium tetroxide solution . After dehydration and embedding in Epon , semi-thin sections were cut and stained with toluidine blue . Ultra-thin sections were stained with uranyl acetate and lead citrate and then examined with an electron microscope ( Jeol 100 CX II ) . For histopathological analysis organ samples were fixed in RCL2 ( Alphelys , Plaisir , France ) for 24 hr then dehydrated with pure alcohol . Samples were embedded in low melting wax ( polyethylene-glycol distearate; Sigma-Aldrich , St Louis , USA ) , sectioned ( coronally; 5-µm thickness ) and stained with hematoxylin-eosin , periodic acid-Schiff and reticulin . For chromogenic immunohistochemistry human serum from a patient positive by ELISA ( see below ) was incubated overnight at 4°C with rodent tissue sections , in a buffer containing 0 . 5% bovine albumin . Samples were then exposed to a biotinylated donkey anti-human secondary antibody at 1:200 ( Jackson ImmunoResearch Laboratories , Inc . West Grove , PA ) followed by streptavidin-conjugated to horse radish peroxidase revealed by the AEC , a red chromogen at 1:400 ( Vector Lab , Burligame , CA , USA ) . Sections were counterstained with haematoxylin . Negative controls comprised tissue sections incubated with normal polyvalent human immunoglobulins at the same concentration as the human sera used for mammarenavirus detection and incubated in the absence of immunoglobulins . Attempts to isolate the mammarenavirus from infected rodents’ organs was conducted using VERO ( ATCC CRL-1586 ) , MDCK ( ATCC CCL-34 ) and BHK-21 ( ATCC CCL-10 ) cell lines previously described to isolate other mammarenaviruses . Briefly , each organ was homogenized using MagNA lyser instrument ( Roche , Basel , Switzerland ) as describe above and the homogenate was diluted at 1/100 with filtered culture medium ( DMEM; Sigma Aldrich , Steinheim , Germany ) supplemented with 5% of foetal calf serum ( GIBCO ) and 1% of a solution of penicillin-streptomycin ( 10000 units – 10 mg/mL; Sigma Aldrich , Steinheim , Germany ) before inoculation onto each cell line . The cells were incubated at 37°C in a 5% CO2 atmosphere for 7 days and the process was repeated 3 times . At each passage , the presence of the virus was detected by mammarenavirus qRT-PCR as described above . The presence of Mycoplasma contamination was not systematically checked before inoculation . Serological screening of experimentally infected rodents was initially performed by immune fluorescence assay ( IFA ) using LCMV-infected cells due to their availability and because this assay can detect numerous Old World mammarenavirus species . Sera were tested in pooled batches of five and sera from positive pools were then re-tested individually . Sera were diluted to 1:20 in PBS , added in 30 µl volumes to each well and placed at 37ºC for 1 hr . Sera were aspirated from each well and slides were then washed three times in sterile PBS for 5 min each , rinsed in distilled water for 1 min , then allowed to air-dry . A secondary antibody ( Sigma , anti-rat IgG FITC , conjugated in goat ) at a dilution of 1:50 was then added in 30ul volumes to each well . Slides were placed at 37ºC for 1 hr , then washed and rinsed as before . Once dry , PBS containing 30% glycerol was added in small quantities to each slide and a cover-slide placed firmly on top . Slides were visualized with a UV microscope . As the lack of a tissue culture isolate precluded the production of more IFA slides and in an attempt to produce a test more specific to Wēnzhōu virus , once the coding complete genome sequence of Cardamones variant of Wēnzhōu virus was obtained , a Sandwich ELISA ( enzyme-linked immunosorbent assays ) based on the nucleoprotein gene was developed . In order to do this , the nucleoprotein gene of this virus ( GenBank accession number KC669696 ) was cloned into the plasmid pIVEX-MBP , and transformed in BL21 AI E . coli . After incubation and cell lysis , the lysate was purified by IMAC ( Immobilized metal ion affinity chromatography; HisPur Cobalt Resin and purification kit , Thermo Scientific , Waltham , USA ) to separate the fusion protein , ( NP-ARV ) , from the host proteins . The fusion protein was cut at the TEV site , and a second IMAC purification was used to separate NP-ARV from 6His-MBP and 6His-TEV . SDS-PAGE demonstrated that NP-ARV was in the non retained fraction and that the purity was 83 . 3% . The protein was then concentrated on cellulose membrane ( Amicon , Millipore ) to a final concentration of 0 . 8 mg/ml . To obtain monoclonal antibodies ( MAbs ) , BALB/c mice were immunized six times subcutaneously with 10 µg of purified NP-ARV antigen . The first immunization was carried out with complete Freund adjuvant ( CFA ) and the second two weeks later with incomplete Freund adjuvant ( IFA ) . The MAbs exhibited different affinities for different epitopes: G14-95 with very low affinity ( <10–7M ) and H24-20 with good affinity ( 1 . 10–10M ) . Sandwich ELISAs were conducted using MAb H24-20 . This MAb was diluted in dilution buffer ( 1X PBS , 1% skimmed milk powder ) and added to 96-well NUNC-immuno plates with a Polysorp surface ( Nunc , Roskilde , Denmark ) at 1 ug/ml per well . Plates were incubated at 4°C overnight , then washed four times with wash buffer ( 1X PBS , 0 . 05% Tween 20 ) . Plates were blocked with blocking buffer ( 1X PBS , 5% skimmed milk powder ) for 1 hr at 37°C then washed four times with wash buffer . A 100 μl volume of NP-ARV diluted to 0 . 8 ug/ml in PBS 1X was added per well and incubated for 1 hr at 37°C . Plates were then washed four times with wash buffer . Sera ( human or rodent ) diluted 1:100 in dilution buffer were added to wells in 100 µl volumes , incubated for 1 hr at 37°C , and then washed seven times with wash buffer . Peroxidase-labelled goat anti-human IgG ( Biorad , Hercules , USA ) or peroxidase-labelled goat anti-rat IgG ( KPL catalog n# 04-16-02 , Gaithersburg , MD , USA ) was diluted 1:10 , 000 in dilution buffer and 100 µl added to each well . Plates were incubated for 1h at 37°C then washed four times with wash buffer . The substrate solution TMB ( Tetramethyl Benzidine; KPL catalog # 50-76-01 ) was added in 100 μl volumes per well . The reaction was stopped by adding 100 μl of 1N H2SO4 and plates were measured at an absorbance of 450 nm . The means and standard deviations for each plate were calculated using 4 negative control serum samples . The cut-off value for the assay was defined as the mean plus three standard deviations ( sd ) following the formula: cut-off = mean + 3 sd + 10% . Each sample was tested twice in independent assays and when the results led to different conclusions , the sample was retested a third time . To verify the absence of cross-reactivity , human serum positive for anti-LCMV IgM and IgG ( CDC , US ) was tested in four independent series and found to test consistently negative for anti-Wēnzhōu IgG detection . Three positive controls ( OD varying from low to high ) were included in replicates in each plate and the intra-assay variation was considered as acceptable when the ODs of the positive controls were maintained within the ranges of their expected standard deviation values . Seroconversion was defined as when the acute sample was negative and the convalescence sample of the same patient was positive , or when the convalescent sample demonstrated an increase of 50% or more of the OD value of the acute sera . A total of 372 human serum samples and 7 experimentally infected rodent serum samples ( Supplementary file 1F and 1J ) were tested by both IFA and ELISA . The number of samples detected positive and negative by each test were used to calculate coefficients of correlation . Human serum , respiratory and cerebro-spinal fluid samples used in this study were collected previously by the Institute Pasteur in Cambodia ( IPC ) within the framework of several research projects approved by the Cambodian National Ethics Committee for Health Research ( NECHR ) and stored in IPC’s biobank . At the time of sample collection , a written consent form from each patient or their legal guardian was obtained and the NECHR also specifically authorized the use of these stored specimens for the purpose of the present study ( NECHR No . 0205 ) . The use of samples collected for dengue and influenza national surveillance and the use of stored and prospectively collected respiratory samples for the negative control groups were all approved by the NECHR . All the samples were anonymised for the purpose of this study . Samples were tested using the Wēnzhōu virus IgG ELISA or by semi-nested RT-PCR dependent on sample type ( see above ) . Eight groups of human samples were tested for evidence of mammarenavirus infection ( Table 3 ) . Group 1 ( IgG_Dengue-Influenza-like_group 1_eLife_edit . xlsx in Blasdell et al . , 2016 ) comprised acute and convalescent sera from 510 individuals with dengue- or influenza-like illness randomly selected between 2005 and 2010 , originating from Kampong Cham and 11 other provinces . Paired sera comprised acute samples collected before day 4 of fever and convalescent samples collected 7 to 10 days after the onset of fever . Out of the 510 patients tested , only a single sample collected during the acute phase of the febrile episode was available for 98 patients , whilst only a single serum sample collected during the convalescent phase was available for 214 individuals . Paired serum samples , one collected during the acute phase and one during the convalescence phase were available from a further 198 patients . Group 2 ( IgG_Healthy individual_group 2_eLife_edit . xlsx in Blasdell et al . , 2016 ) consisted of sera collected from healthy individuals in a community based dengue seroprevalence study in 2009 in Kampong Cham province . The samples in both groups were tested by Wēnzhōu virus IgG ELISA . Groups 3 to 8 ( PCR_Meningo-encephalitis patients_group 3_eLife_edit . xls , PCR_group 4-8_eLife_edit . xlsx in Blasdell et al . , 2016 ) were all tested for mammarenavirus RNA by 'screening' semi-nested RT-PCR and qRT-PCR . Group 3 comprised 200 patients hospitalized for meningo-encephalitis and who tested negative by PCR and/or serology for the main etiologies of central nervous systems infections usually observed in the country ( Japanese encephalitis virus , dengue viruses , chikungunya virus , herpes simplex virus 1 , influenza A virus , enteroviruses , Nipah virus , measles virus , mumps virus , rubella virus , Streptococcus pneumoniae , Streptococcus suis , Haemophilus influenzae , Neisseria meningitidis , Orientia tsutsugamushi ) . Group 4 consisted of randomly selected sera , obtained during the acute febrile phase from 253 patients . These patients , who originated from different parts of Cambodia , presented between 2009 and 2011 with signs and symptoms suggestive of dengue fever , dengue hemorrhagic fever and dengue shock syndrome . However all tested negative for dengue , Japanese encephalitis and chikungunya virus infections by RT-PCR and serology ( in-house MAC-ELISA and hemagglutination-inhibition assay using antigens derived from the three arboviruses listed above , as described previously [Andries et al . , 2012] ) . Group 5 comprised nasopharyngeal swab samples from 720 individuals presenting with an influenza-like syndrome . This group was further sub-divided into two sub-groups . Subgroup 5a included 328 individuals who tested negative for influenza A and B viruses , respiratory syncytial virus and human metapneumovirus by multiplex RT-PCR ( Buecher et al . , 2010; Arnott et al . , 2011 ) , whilst the 392 individuals in subgroup 5b tested positive for one of these viruses . Group 6 consisted of nasopharyngeal swabs from 279 individuals hospitalised with acute lower respiratory tract infections . Two negative control groups were also tested for mammarenavirus RNA . Group 7 comprised influenza A-negative nasopharyngeal specimens obtained from 266 apparently healthy individuals who had had contact with patients with H5N1 infections , while group 8 included nasopharyngeal swabs collected from 238 randomly selected apparently healthy volunteers seeking anti-rabies vaccination at IPC . All statistical analyses were performed using Stata/SE version 12 . 0 ( StataCorp , TX , USA ) . Significance was assigned at p<0 . 05 for all parameters and 95% of confidence interval was used . Categorical variables between groups were compared by Chi2 test or Fisher’s exact test and t-test or Wilcoxon-Mann-Whitney test were used for continuous variables . | Rodents have long been notorious for spreading disease among humans . Often the animals can carry viruses and transmit them to humans without becoming ill . Certain species thrive in cities and agricultural areas where they come in close contact with humans; this creates many opportunities to spread infection . As humans urbanize and farm larger swaths of previously wild lands , the risk of rodent-transmitted infections increases . As a result , some scientists are working to identify viruses carried by rodents in human settlements and hopefully prevent them from spreading to humans . The mammarenavirsuses are a group of rodent-transmitted viruses that commonly cause illness in people in Africa and Latin America . Each year , one such virus – the Lassa virus –sickens as many as 300 , 000 people in Africa and kills 5 , 000 . So far , only two mammarenaviruses have been found in Asia: one called the Wēnzhōu virus and another called LCMV . However only LCMV is known to cause human illness and many cases of illness caused by mammarenaviruses in Asia may go undetected because they often cause mild symptoms similar to the common cold . Blasdell et al . have now tested lung samples from 20 species of rodents collected at 7 sites in Cambodia , Thailand , and Laos to look for molecules produced by mammarenaviruses . The tests revealed a strain of Wēnzhōu virus circulating in Cambodian rats that often live in urban areas . A new mammarenavirus was also detected in rodents that live in Thai rice fields . However , infecting wild and domestic rodents with the viruses in the laboratory did not cause many noticeable signs of illness . Blasdell et al . then tested samples from Cambodian patients who either had influenza-like symptoms or more serious symptoms that are associated with a condition called Dengue fever ( which is common in the area ) . Some patients with respiratory symptoms tested positive for the Wēnzhōu virus . Because the symptoms are mild and similar to those of other common diseases it is likely that the Wēnzhōu virus may be spreading more widely among humans in Asia . The next challenges are to provide a better estimate of the frequency of this disease in the human population in Asia and to describe the full spectrum of disease that might be associated with this newly discovered infectious disease . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"microbiology",
"and",
"infectious",
"disease"
] | 2016 | Evidence of human infection by a new mammarenavirus endemic to Southeastern Asia |
The basal forebrain cholinergic system modulates neuronal excitability and vascular tone throughout the cerebral cortex and hippocampus . This system is severely affected in Alzheimer's disease ( AD ) , and drug treatment to enhance cholinergic signaling is widely used as symptomatic therapy in AD . Defining the full morphologies of individual basal forebrain cholinergic neurons has , until now , been technically beyond reach due to their large axon arbor sizes . Using genetically-directed sparse labeling , we have characterized the complete morphologies of basal forebrain cholinergic neurons in the mouse . Individual arbors were observed to span multiple cortical columns , and to have >1000 branch points and total axon lengths up to 50 cm . In an AD model , cholinergic axons were slowly lost and there was an accumulation of axon-derived material in discrete puncta . Calculations based on published morphometric data indicate that basal forebrain cholinergic neurons in humans have a mean axon length of ∼100 meters .
The mammalian cerebral cortex and hippocampus are densely innervated by cholinergic fibers that originate in the basal forebrain ( Mesulam , 2004 ) . Most , if not all , cortical and hippocampal neurons respond to cholinergic signals using muscarinic and/or nicotinic acetylcholine receptors localized to pre- and/or postsynaptic sites . Activation of postsynaptic nicotinic receptors leads to transient depolarizing currents with a high calcium:sodium ratio and activation of postsynaptic muscarinic receptors leads to a sustained reduction in potassium currents , both effects producing a general elevation in excitability ( Lucas-Meunier et al . , 2003 ) . Additionally , cholinergic activation of muscarinic receptors in the microvasculature leads to the production of nitric oxide , producing vasodilation ( Hamel , 2004 ) . A wide variety of experiments in primates , cats , and rodents have implicated cholinergic neurotransmission from the basal forebrain to the cortex and hippocampus in attention , memory , and plasticity . In cat primary visual cortex , cholinergic input enhances neuronal responses to a preferred visual stimulus ( Sato et al . , 1987 ) , and in rat motor cortex , cholinergic input promotes more complex motor sequences in response to electrical stimulation ( Berg et al . , 2005 ) . In rat barrel cortex and auditory cortex , cholinergic input is required for experience-dependent synaptic plasticity and circuit reorganization ( Baskerville et al . , 1997; Kilgard and Merzenich , 1998; Zhu and Waite , 1998 ) . In contrast to the high spatial precision of glutamatergic and GABAergic neurotransmission , current evidence indicates that the basal forebrain cholinergic system modulates neuronal excitability and vascular tone over large target areas . The basal forebrain cholinergic system is of special interest because it degenerates in a variety of common neurologic diseases , including Alzheimer's disease ( AD ) and Parkinson's disease , to an extent that correlates with the severity of dementia ( Schliebs and Arendt , 2011 ) . In advanced AD , the relative loss of cholinergic innervation varies by region , with the temporal lobe showing the greatest loss of fibers and the primary sensory , motor , and anterior cingulate cortices showing the least loss ( Geula and Mesulam , 1989 ) . The relative extent of cholinergic fiber loss in different cortical areas appears to correlate inversely with fiber density in the normal brain , suggesting that disease-associated fiber loss progresses at roughly equal rates throughout the cortex and that those regions that began with the fewest fibers are the first to become denuded of cholinergic input . The loss of forebrain cholinergic innervation in AD has stimulated the development of pharmacotherapy to enhance cholinergic signaling as an approach to partially ameliorate cognitive symptoms ( Burns et al . , 2006 ) . From the preceding paragraphs it is clear that an accurate anatomic description of the basal forebrain cholinergic system is important for understanding its function and its susceptibility to degeneration . At present , this description , which derives from retrograde and anterograde filling and from histochemical and immunohistochemical staining , provides a view that is accurate as a statistical picture but is incomplete in one critical respect: the morphologies of individual cholinergic axon arbors are unknown because their extraordinarily large size has , thus far , precluded classical tracer filling and reconstruction . In earlier work , we demonstrated the utility of extremely sparse CreER/loxP labeling methods for visualizing axonal and dendritic morphologies of large neurons , including forebrain cholinergic neurons ( Rotolo et al . , 2008; Badea et al . , 2009 ) . In the present work , we use this approach to visualize and quantify the full 3-dimensional axonal morphologies of individual forebrain cholinergic neurons and to define changes in these arbors in response to disease progression in a mouse model of AD .
In earlier work , we generated an IRES-CreER knock-in in the 3′ untranslated region of the gene coding for choline acetyl transferase ( ChAT; Rotolo et al . , 2008 ) . This allele expresses relatively low levels of CreER and , consequently , shows no recombination of Cre-activated reporters in the absence of tamoxifen or 4-hydroxytamoxifen ( 4HT ) , a prerequisite for visualizing genetically marked neurons at densities of <10 labeled neurons per brain . To visualize large axon arbors in their entirety and to efficiently survey dozens of brains , we chose the highly sensitive histochemical reporter human placental alkaline phosphatase ( AP ) , a GPI-anchored protein that distributes relatively uniformly along dendrites and axons ( Rotolo et al . , 2008 ) . AP histochemistry works efficiently with relatively thick ( 300 µm ) vibratome sections , which minimizes the number of sections required per brain and thereby simplifies the logistics of staining , imaging , and tracing . To minimize background reporter activity , we used an AP reporter knock-in at the Gt ( ROSA ) 26Sor locus ( referred to as R26 ) in which the 3′ half of the AP coding region is in reverse orientation and Cre-mediated recombination restores this segment to the correct orientation ( R26IAP , ‘I’ stands for ‘inverted’; Figure 1A; Badea et al . , 2009 ) . In contrast to standard reporters that are maintained in a repressed state by a loxP-stop-loxP cassette , the R26IAP locus shows undetectable reporter activity prior to Cre-mediated recombination . 10 . 7554/eLife . 02444 . 003Figure 1 . Cholinergic neuron specificity of Cre-mediated recombination . ( A ) Structure of the R26IAP knock-in . In the absence of Cre-mediated recombination , the 3′ half of the AP coding region is inverted in the germline configuration . It assumes the correct orientation following Cre-mediated recombination between inverted loxP sites . ( B ) P30 retina from Chat-IRES-CreER;R26IAP mice treated with 4HT . AP histochemistry labels cholinergic ( starburst ) amacrine cells . Scale bar , 100 µm . ( C–F ) P30 brain from Chat-IRES-CreER;R26IAP mice treated with high dose 4HT at P5 . AP histochemistry labels numerous axons throughout the cortex ( D ) and hippocampus ( F ) , as well as cranial motor neurons ( E ) , the axons of which are seen exiting the brain stem ( red arrows ) . Scale bars in D–F , 200 µm . ( G and H ) Coronal sections of P30 forebrain from Chat-IRES-CreER;R26-LSL-nGFP mice treated with high dose 4HT at P4 . Approximately 50% of cholinergic neurons in the basal forebrain , medial septal nucleus , striatum , and spinal cord ( visualized with ChAT immunohistochemistry ) are GFP+ . Medial to the striatum , a distinctive group of GFP+ cell is ChAT−; these cells presumably expressed Chat ( and , therefore , Cre ) in the early postnatal period and then repress Chat expression in adulthood . In ( H ) , arrows point to ChAT+;GFP− neurons and arrowheads point to ChAT+; GFP+ neurons . Scale bar , 50 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 02444 . 003 The specificity of the Chat-IRES-CreER driver has been documented by Rotolo et al . ( 2008 ) and Badea et al . ( 2009 ) and is demonstrated here with Chat-IRES-CreER;R26IAP mice based on reporter expression in ( 1 ) starburst amacrine cells , the only cholinergic retinal neurons ( Figure 1B ) , ( 2 ) a uniform network of fibers in the cortex and hippocampus , as expected for the axon arbors of forebrain cholinergic neurons ( Figure 1C , D , F ) , and ( 3 ) cranial motor neurons in the brainstem ( Figure 1E ) . Chat-IRES-CreER activation of a nuclear localized GFP reporter ( encoded by a R26-loxP-stop-loxP-nGFP knock-in ) shows co-localization with ChAT immunoreactivity in the basal forebrain , septal nucleus , and ventral spinal cord as expected ( Figure 1G , H ) . Interestingly , a small population of ChAT-negative forebrain cells , located medial to the striatum , shows GFP expression in the adult , implying that these cells transiently express the ChAT gene at the time of 4HT injection [postnatal day ( P ) 4] but not at later times ( Figure 1G ) . A series of 4HT titration experiments with Chat-IRES-CreER;R26IAP mice showed that intraperitoneal ( IP ) injection of 1–5 µg 4HT at P4-5 resulted in ∼10 forebrain cholinergic neurons labeled per brain . Using this protocol , 67 well-separated forebrain cholinergic neurons were imaged and 12 of these neurons–8 from P12 brains and 4 from P30 brains–were traced ( Figures 2 , 3 , 4B , Figure 2—figure supplement 1 ) . Among the traced arbors , nine were in the cortex , two were in the hippocampus , and one was in the olfactory bulb . For each of the remaining 55 neurons , collected between 1 and 12 months of age , we determined the soma location and the boundaries of the arbor territory . 10 . 7554/eLife . 02444 . 004Figure 2 . Axon arbors of forebrain cholinergic neurons from P30 Chat-IRES-CreER;R26IAP mice visualized with sparse Cre-mediated recombination . ( A ) Part of the arbor of a forebrain cholinergic neuron in a P30 hippocampus visualized in a single 300 µm section at three Z-planes and in a Z-stacked image . Bottom , the traced arbor . Scale bar , 200 µm . ( B and C ) Fifteen consecutive 300 µm sagittal sections from a single P30 hemisphere ( C ) with two fully traced AP+ forebrain cholinergic neurons , colored red and green . Black arrows in panels l and o , the two cell bodies . Red arrows in g–n , the proximal axon segment for the red neuron trace . ( B ) An enlarged view of the boxed region of section d in ( C ) . Scale bar in ( B ) , 500 µm ( corrected for tissue shrinkage in BBBA ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02444 . 00410 . 7554/eLife . 02444 . 005Figure 2—figure supplement 1 . Dendrite structure among forebrain cholinergic neurons . ( A ) Left , Z-stack image of a 300 µm sagittal brain section ( ‘section 1’ ) shown in panel ‘o’ of Figure 2C . Middle , trace of AP+ dendrites , part of the soma , and proximal axon arbor for the forebrain cholinergic neuron in section 1 . Right , trace of the remainder of the dendritic arbor and the second part of the soma from the adjacent 300 µm sagittal brain section ( ‘section 2’ ) shown in panel ‘n’ of Figure 2C . Scale bar , 200 µm . ( B ) Total dendrite lengths for five P12 and four P30 forebrain cholinergic neurons . Bars show the mean +/− SD . ( C ) Dendrite arbor volume measured with the polygon method vs dendrite length for the same five P12 and four P30 forebrain cholinergic neurons plotted in panel B . ( D ) Individual dendrite arbor density ( dendrite length divided by dendrite arbor volume ) vs dendrite length for the same five P12 and four P30 forebrain cholinergic neurons plotted in panel B . DOI: http://dx . doi . org/10 . 7554/eLife . 02444 . 00510 . 7554/eLife . 02444 . 006Figure 3 . Axon arbors of forebrain cholinergic neurons from P12 Chat-IRES-CreER;R26IAP mice visualized with sparse Cre-mediated recombination . ( A ) Enlarged view of the boxed region from panel g in ( B ) . Scale bar , 500 µm ( corrected for tissue shrinkage in BBBA ) . ( B ) Fourteen consecutive 300 µm sagittal sections from a single P12 hemisphere ( a–n ) with three traced AP+ cholinergic neurons , colored red , green , and blue . The blue neuron is shown in its entirety , including the cell body and dendrites in the basal forebrain; for the green and red neurons , only the cortical axon arbors are shown . DOI: http://dx . doi . org/10 . 7554/eLife . 02444 . 00610 . 7554/eLife . 02444 . 007Figure 4 . Quantitative analysis of morphologic parameters for cholinergic axon arbors . ( A ) The polygon method for estimating the target area for a single cholinergic axon arbor . Traced axon arbor images are shown for each 300 µm sagittal section from the P30 brain in Figure 2C , panels a–e . A minimal convex polygon has been drawn around each trace , providing an upper estimate of the cortical territory that is directly influenced by the arbor . As seen in panels d and e , the polygon method somewhat overestimates the target area by including regions that are relatively far from the axon . P , posterior . A , anterior . ( B ) Axon arbor locations for the 67 basal forebrain cholinergic neurons analyzed . ( C ) Quantification of tissue volume shrinkage due to dehydration in ethanol and BBBA . ( D ) Schematic of a forebrain cholinergic neuron in a dorsal view of the mouse brain showing the mediolateral cell body and arbor locations , the parameters displayed in panels ( H ) and ( I ) . ( E ) The number of branch points per mm of axon length for the 12 forebrain cholinergic neurons that were traced . These data were obtained from two 300 µm sections per arbor by measuring the total axon length and counting all branch points for the AP+ arbor within each section . ( F ) Scatter plot of arbor volume ( estimated using the polygon method ) vs axon length for the 12 forebrain cholinergic neurons that were traced . ( G ) Scatter plot of axon density ( length divided by arbor volume ) vs axon length for the 12 forebrain cholinergic neurons that were traced . ( H ) Scatter plot of the mediolateral extent ( defined in panel D ) vs arbor volume ( estimated using the polygon method ) for the 55 forebrain cholinergic arbors that were not traced . ( I ) Mediolateral cell body and arbor locations for the 12 traced neurons ( left ) and the 55 untraced neurons ( right ) . Black dots represent cell body location and the vertical bar represents the mediolateral extent of the axon arbor . OB , olfactory bulb . DOI: http://dx . doi . org/10 . 7554/eLife . 02444 . 00710 . 7554/eLife . 02444 . 008Figure 4—figure supplement 1 . Cholinergic axon arbors: Z-stacked images , traces , and branch point locations . ( A–D ) Axon arbors from single cholinergic neurons visualized in 300 µm sagittal sections from olfactory bulb ( A ) , cortex ( B and D ) , and hippocampus ( C ) : Z-stacked images ( left ) , traces ( center ) , and branch points ( right ) . Z-stacked images ( A–C ) are from Chat-IRES-CreER;R26IAP brains at P12 or P30 and image ( D ) is from a Chat-IRES-CreER;R26IAP;APP/PS1 brain at 9 months . Scale bars , 200 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 02444 . 008 At both P12 and P30 , large and complex axon arbors were observed ( Figures 2 and 3 ) . The mean axon length per neuron was 13 cm at P12 ( n = 8; range 3–42 cm ) and 31 cm at P30 ( n = 4; range 11–49 cm ) , and at both ages , the density of branches averaged 4–5 per mm of axon length ( range 3–7 per mm ) , giving a mean of >1000 branch points per arbor at P30 , including occasional branches along the axon's initial segment ( Figure 4E–G , Figure 2—figure supplement 1A , Figure 4—figure supplement 1 ) . In a sample of nine basal forebrain cholinergic neurons , the mean dendritic arbor lengths were 9 . 6 mm at P12 ( n = 5 arbors ) and 11 . 5 mm at P30 ( n = 4 arbors; Figure 2—figure supplement 1 ) . The territory influenced by each cholinergic arbor was estimated by enclosing the axon traces from each 300 µm tissue section with the smallest possible convex polygon , calculating the area of each polygon , and summing the resulting polygonal volumes ( polygon area × 300 µm ) across all of the sections populated by the arbor of interest ( Figure 4A , C ) . Using this measure , individual axon density–defined as mm of axon length per mm3 of polygonal volume for an individual axon—was found to vary by a factor of ∼10 among traced arbors at both P12 ( n = 8 ) and at P30 ( n = 4; Figure 4F , G ) , reflecting significant variation in the size and compactness of cholinergic arbors . The comparison between P12 and P30 implies that there is substantial growth of cholinergic axons after the second week of postnatal life . Although the polygon method somewhat over-estimates the territory influenced by an arbor ( as illustrated in Figure 4A ) , it provides a good measure of the linear extent of the arbor , giving a mean value of ∼2 mm in the adult brain , as seen in a compilation of the mediolateral extents of forebrain cholinergic axon arbors ( Figure 4H , I ) . The coverage factor for forebrain cholinergic axon arbors—defined as the number of arbor territories that encompass any arbitrary point in the cortex and hippocampus—can be calculated based on a mean volume of 1 . 35 mm3 for axon arbor territories at P30 ( n = 4; Figure 4F ) , a total volume of 130 mm3 for the adult mouse cerebral cortex ( 109 mm3 ) and hippocampus ( 21 mm3; Kovacević et al . , 2005 ) , and estimates of the number of forebrain cholinergic neurons of 4500 and 6632 ( Boncristiano et al . , 2002; Perez et al . , 2007 ) . This calculation gives a mean coverage factor of 47–69 . The analogous dendrite coverage factor and the density of dendrites in the basal forebrain regions where cholinergic projection neuron cell bodies reside ( the medial septal nucleus , ventral diagonal band , and horizontal limb of the diagonal band ) can also be calculated based on the volume of these territories ( 0 . 667 mm3; Paxinos and Franklin , 2001 ) , the mean dendrite volume per cholinergic neuron at P30 ( 0 . 0694 mm3; n = 4; Figure 2—figure supplement 1 ) , the mean dendrite length per cholinergic neuron at P30 ( 11 . 5 mm; n = 4; Figure 2—figure supplement 1C , D ) , and the estimated number of forebrain cholinergic neurons noted above . These calculations give a dendrite coverage factor of 470–690 , and a density of cholinergic dendrites in the basal forebrain of 0 . 078–0 . 114 µm/µm3 . Numerous retrograde and anterograde labeling studies have investigated the correlation between the territories targeted by forebrain cholinergic arbors and the locations of the corresponding cell bodies ( e . g . , McKinney et al . , 1983; Saper , 1984; Woolf et al . , 1986; Baskerville et al . , 1993 ) . In our dataset of 67 neurons , there was a clear correlation between soma location and axon arbor position along the mediolateral axis , specifically , cell bodies located more laterally in the basal forebrain give rise to arbors that reside in more lateral cortical , hippocampal , or olfactory bulb territories , supporting the general conclusion that there is a rough topographic map of target territories in the basal forebrain ( Figure 4I ) . As noted in the Introduction , loss of forebrain cholinergic innervation is a prominent feature of AD . To visualize AD pathology at the level of single cholinergic axon arbors , we used APPswe/PS1ΔE9 doubly transgenic mice ( Jankowsky et al . , 2004; referred to hereafter as APP/PS1 ) in which Aβ plaque rapidly accumulates with age in the cortex and hippocampus ( Figure 5—figure supplement 1 ) . This progression is accompanied by microglial reorganization and activation , and premature death ( Figure 5—figure supplement 1 and 2 ) . Analysis of AP+ cholinergic arbors in the cortex and hippocampus of 62 Chat-IRES-CreER;R26IAP;APP/PS1 brains harvested between one and 12 months of age showed fragmentation of axons and a decrease in AP signal strength ( Figure 5A–C ) , consistent with previously described changes in the appearance of ChAT immunoreactive fibers in mouse and human AD brains ( Figure 5—figure supplement 1C , D; Gordon et al . , 2002; Schliebs and Arendt , 2011 ) . There was also a progressive accumulation of large numbers of AP+ and ChAT immunoreactive puncta with diameters up to ∼10 µm in Chat-IRES-CreER;R26IAP;APP/PS1 cortex and hippocampus but not in Chat-IRES-CreER;R26IAP controls ( Figure 5 , Figure 5—figure supplement 1C , D , Figure 5—figure supplement 3A; Boncristiano et al . , 2002 ) . These puncta presumably represent cholinergic axon breakdown products . 10 . 7554/eLife . 02444 . 009Figure 5 . Disruption of cholinergic axon arbors in Chat-IRES-CreER;R26IAP;APP/PS1 mice . ( A ) Upper panel , 300 µm sagittal section of a 12 month old Chat-IRES-CreER;R26IAP;APP/PS1 brain showing part of a single AP+ axon arbor . The olfactory bulb is visible at lower right . Lower panels , three Z-planes enlarged from the region enclosed in the red square in the upper panel . Red arrowheads point to clumps of AP+ material ( puncta ) . Scale bars: upper panel , 500 µm; lower panels , 50 µm . ( B ) Comparison of representative regions from forebrain cholinergic axon arbors in the cortex of Chat-IRES-CreER;R26IAP brains ( WT; left ) and Chat-IRES-CreER;R26IAP;APP/PS1 brains ( right ) , between one and 12 months of age . Structural heterogeneity , including the clumping of AP + material ( puncta ) and loss of AP staining intensity , increases with age in the APP/PS1 background . Scale bar , 50 µm . ( C ) Quantification of AP+ puncta in the cortex and hippocampus of Chat-IRES-CreER;R26IAP ( i . e . , WT ) and Chat-IRES-CreER;R26IAP;APP/PS1 mice at different ages . Puncta appear at 3 months in Chat-IRES-CreER;R26IAP;APP/PS1 mice . The box plots indicate the extreme data points ( top and bottom bars ) , the 25–75% interval ( box ) , and the median ( central line ) . p-values , student's t test . ( D ) Complete tracing of an AP+ cortical cholinergic arbor ( blue ) with the locations of AP + puncta ( red dots ) indicated . Panels a–f show six adjacent 300 µm sagittal sections within which this arbor resides . The three enlarged images above correspond to the boxed regions in panels d–f . Scale bar , 500 µm ( corrected for tissue shrinkage in BBBA ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02444 . 00910 . 7554/eLife . 02444 . 010Figure 5—figure supplement 1 . Aβ deposition , microglial reorganization , and disorganization of cholinergic fibers in the APP/PS1 brain . ( A ) The hippocampus in WT and APP/PS1 mice at 12 months . Numerous Aβ deposits are seen in the APP/PS1 hippocampus , and are associated with migration of microglia , visualized by ionized calcium binding adaptor molecule 1 ( Iba1 ) . Scale bar , 100 µm . ( B ) The cortex of an APP/PS1 mouse at 12 months . Numerous Aβ deposits are present , along with an irregular distribution of microglia . Scale bar , 100 µm . ( C ) Enlarged views of the boxed regions in panel ( B ) . Boxed region ‘a’ lacks large Aβ deposits . Boxed region ‘b’ has numerous large Aβ deposits surrounded by microglia . ChAT immunostaining shows numerous puncta in the region corresponding to Aβ deposits . Scale bar , 25 µm . ( D ) Cholinergic axons in the cortex and hippocampus and cholinergic cell bodies in the basal forebrain in WT and APP/PS1 brains at 12 months . Cholinergic cell bodies in the basal forebrain show little or no difference between the two samples . Cholinergic axons show numerous puncta in the APP/PS1 cortex and hippocampus , some of which are highlighted by red arrows . DOI: http://dx . doi . org/10 . 7554/eLife . 02444 . 01010 . 7554/eLife . 02444 . 011Figure 5—figure supplement 2 . Characterization of APP/PS1 mice: survival and microglial activation . ( A and B ) Sagittal section of a 12 month old WT brain stained with GS-lectin , which visualizes vascular endothelial cells and activated microglia . In the cortex and hippocampus , binding is limited to the vasculature . The boxed region in ( A ) is enlarged in ( B ) . ( C and D ) Sagittal section of a 12 month old APP/PS1 brain stained with GS-lectin as for panels ( A and B ) . In the cortex , hippocampus , and olfactory bulb , GS-lectin binding is observed in the vasculature and in numerous clusters of activated microglia . Activated microglia are not present in other brain regions . The boxed region in ( C ) is enlarged in ( D ) . ( E ) Kaplan–Meier survival curves for Chat-IRES-CreER;R26IAP ( i . e . , WT ) and Chat-IRES-CreER;R26IAP;APP/PS1 mice . DOI: http://dx . doi . org/10 . 7554/eLife . 02444 . 01110 . 7554/eLife . 02444 . 012Figure 5—figure supplement 3 . Tracing AP+ axons , and locating and quantifying AP+ puncta with and without bexarotene treatment . ( A ) Example of a cortical arbor from a Chat-IRES-CreER;R26IAP;APP/PS1 brain at 9 months of age visualized in a single 300 µm sagittal section at three Z-planes ( left three panels ) and in a Z-stacked image ( fourth panel ) . Arrowheads point to some of the larger puncta . Fifth panel , the traced arbor ( blue ) and AP+ puncta ( red ) . Scale bar , 200 µm . ( B ) Induction of ABCA1 in cerebral cortex by bexarotene treatment , assessed by anti-ABCA1 immunoblotting . 7-month old mice received vehicle ( DMSO; n = 3 ) or 100 mg/kg bexarotene in DMSO ( n = 3 ) once daily for 7 days by oral gavage . The six mice were sacrificed on day 8 . ( C ) Quantification of ABCA1 immunoblot signal intensity ( panel B ) and the ratio of liver weight to body weight in mice that received vehicle ( DMSO; ‘veh’ ) or 100 mg/kg bexarotene in DMSO ( ‘bex’ ) once daily for 7 days by oral gavage . Bexarotene treatment induces a rapid hepatomegaly . ( D ) Quantification of AP+ puncta in the cortex and hippocampus of untreated Chat-IRES-CreER;R26IAP ( i . e . , WT ) mice and Chat-IRES-CreER;R26IAP;APP/PS1 mice treated with daily 100 µl gavages of DMSO for 14 days or bexarotene in DMSO for 14 or 23 days . The box plots indicate the extreme data points ( top and bottom bars ) , the 25–75% interval ( box ) , and the median ( central line ) . There is no statistically significant effect of bexarotene treatment on the density of AP+ puncta . p-value , student's t test . DOI: http://dx . doi . org/10 . 7554/eLife . 02444 . 012 Bexarotene ( Targretin ) is a retinoid X receptor ( RXR ) agonist that has been reported to acutely promote ApoE-dependent clearance of soluble Aβ oligomers and to improve cognitive performance in the APP/PS1 mouse model of AD ( Cramer et al . , 2012 ) . Whether bexarotene also promotes clearance of insoluble Aβ deposits ( plaque ) is controversial ( Fitz et al . , 2013; Landreth et al . , 2013; Price et al . , 2013; Tesseur et al . , 2013; Veeraraghavalu et al . , 2013 ) . To determine whether bexarotene treatment alters the destruction of cholinergic axons in the APP/PS1 cortex , nine Chat-IRES-CreER;R26IAP;APP/PS1 littermates were treated with 50–100 µg 4HT IP at P5 and were divided into three groups of three mice at 7 months of age: the first group received daily DMSO ( vehicle ) gavages for 14 days; the second group received daily 100 mg/kg bexarotene gavages for 14 days; and the third group received daily 100 mg/kg bexarotene gavages for 23 days , a regimen that activates RXR in the cerebral cortex as determined by the accumulation of ABCA1 , a known RXR-inducible protein ( Figure 5—figure supplement 3B , C; Schmitz and Langmann , 2005 ) . As observed by others ( Fitz et al . , 2013; Price et al . , 2013; Tesseur et al . , 2013; Veeraraghavalu et al . , 2013 ) , the size and abundance of insoluble Aβ deposits and the number of activated microglia and astroglia around these deposits appeared to be unaltered by bexarotene treatment ( data not shown ) , although we note that the DMSO formulation that we and Veeraraghavalu et al . used differs from the aqueous suspension of micronized bexarotene particles used by Cramer et al . ( 2012 ) ( see also Landreth et al . , 2013 ) . Quantification of the density of AP+ puncta was also unaltered ( Figure 5—figure supplement 3D ) , suggesting that the pathologic processes leading to the production of these puncta was not appreciably modified by the several week bexarotene treatment . The total length of the axons in a P30 mouse forebrain cholinergic neuron arbor–up to 50 cm–is roughly 25 times the linear dimension of the mouse brain . The large size of these arbors suggested that a systematic analysis of arbor sizes among various types of projection neurons might be of general interest ( Table 1 ) . 10 . 7554/eLife . 02444 . 013Table 1 . Axon arbor lengths for diffuse projection neuronsDOI: http://dx . doi . org/10 . 7554/eLife . 02444 . 013Basal forebrain cholinergic neurons ( nucleus basalis of Meynert to cortex ) SpeciesNumber of neuronsAxon density in cortexAxon length in cortexCortical volumeMean axon length/neuronMouse6632 ( a ) 1300 m ( b ) 109 mm3 ( c ) 20 cm4500 ( b ) 1300 m ( b ) 109 mm3 ( c ) 29 cmMouse6632 ( a ) 0 . 020–0 . 044 µm/µm3 ( d; this study ) 109 mm3 ( c ) 33–72 cm4500 ( b ) 0 . 020–0 . 044 µm/µm3 ( d; this study ) 109 mm3 ( c ) 48-107 cmMouse*4 traced neurons following CreER/loxP labeling ( this study ) 31 cmRat7 , 312 ( d ) 0 . 0113 µm/µm3 ( e ) 400 mm3 ( f ) 62 cmChimp315 , 000 ( g ) 0 . 066 µm/µm3 ( h ) 147 cm3 ( i ) 31 mHuman435 , 000 ( g ) 0 . 080 µm/µm3 ( h ) 583 cm3 ( i ) 107 mNigro-striatal dopaminergic neuronsSpeciesNumber of neuronsVaricosities per axon lengthNumber of TH + varicosities in the striatumMean axon length/neuronRat7000 ( j ) 5–7 varicosities/7 µm ( j ) 3 . 4 × 109 per side ( j ) 55–77 cmRat*8 traced neurons following sparse GFP virus infection ( k ) 47 cmSerotonergic neurons ( dorsal raphe nucleus to cortex ) SpeciesNumber of neuronsAxon density in cortexCortical volumemean axon length/neuronRat11 , 500 ( l ) 0 . 023 µm/µm3 ( m ) 400 mm3 ( f ) 80 cm15 , 191 ( n ) 0 . 023 µm/µm3 ( m ) 400 mm3 ( f ) 61 cmHuman80 , 386 ( o ) 0 . 048 µm/µm3 ( r ) 583 cm3 ( i ) 348 m86 , 565 ( p ) 0 . 048 µm/µm3 ( r ) 583 cm3 ( i ) 323 m165 , 000 ( q ) 0 . 048 µm/µm3 ( r ) 583 cm3 ( i ) 170 mHippocampal CA3 pyramidal neuronsRat*1 traced neuron following neurobiotin injection ( s ) 48 cmCutaneous sensory neurons with free endings in back skinMouse*7 traced neurons following CreER/loxP labeling ( t ) 71 cmReferences: ( a ) Perez et al . , 2007; ( b ) Boncristiano et al . , 2002; ( c ) Kovacević et al . , 2005; ( d ) Miettinen et al . , 2002; ( e ) Mechawar et al . , 2000; ( f ) Mengler et al . , 2013; ( g ) Raghanti et al . , 2011; ( h ) Raghanti et al . , 2008a; ( i ) Rilling and Insel , 1999; ( j ) Anden et al . , 1966; ( k ) Matsuda et al . , 2009; ( l ) Descarries et al . , 1982; ( m ) Cunningham et al . , 2005; ( n ) Vertes and Crane , 1997; ( o ) Underwood et al . , 2007; ( p ) Underwood et al . , 1999; ( q ) Baker et al . , 1991; ( r ) Raghanti et al . , 2008b; ( s ) Wittner et al . , 2007; ( t ) Wu et al . , 2012 . The asterisk ( * ) marks experiments in which individual axon arbors were traced . To the best of our knowledge there are only four studies ( including the present one ) in which individual axon arbors from the largest classes of CNS or PNS neurons have been traced and their lengths quantified . These are: ( 1 ) eight nigrostriatal dopaminergic neurons in the rat visualized following sparse infection with a GFP-expressing Sindbis virus ( Matsuda et al . , 2009 ) , ( 2 ) a single CA3 pyramidal neuron in the rat visualized by neurobiotin injection ( Wittner et al . , 2007 ) , ( 3 ) seven cutaneous sensory afferents of the ‘large area , free-ending’ class in mouse skin labeled by sparse CreER activation of an AP reporter ( Wu et al . , 2012 ) , and ( 4 ) the four P30 mouse forebrain cholinergic neurons described here . The mean axon lengths for these four cell types were found to be , respectively: 47 cm ( range: 14–78 cm ) , 48 cm , 71 cm ( range: 64–98 cm ) , and 31 cm ( range: 11–49 cm ) . The validity of the single-cell axon length measurements for dopaminergic and cholinergic neurons can be independently checked with calculations based on the total volume of the target territory , the density of the particular type of axon ( axon length per volume of target territory ) , and the number of neuronal cell bodies giving rise to that type of axon ( Table 1 ) . These population analyses are made possible by the availability of antibodies that localize to different types of axons: anti-ChAT for cholinergic axons ( also visualized with acetylcholine esterase histochemistry ) , anti-tyrosine hydroxylase for striatal dopaminergic axons , and anti-serotonin for serotonergic axons . For example , Anden et al . ( 1966 ) estimated the total length of all dopaminergic axons in the rat striatum to be 7900 meters ( bilaterally ) and the number of mid-brain dopaminergic neurons projecting to the striatum to be 14 , 000 ( bilaterally ) , giving a calculated mean axon length of 56 cm per dopaminergic neuron , in good agreement with the single cell tracing data of Matsuda et al . ( 2009 ) ( Table 1 ) . A similar calculation can be performed for mouse forebrain cholinergic neurons using published estimates of the number of cell bodies in the nucleus basalis of Meynert ( 6632 and 4500 ) , the volume of the mouse cortex as determined by MRI ( 109 mm3 ) , and the total length of cholinergic axons in the mouse cortex ( 1300 m ) , giving individual axon arbor length estimates of 20 cm and 29 cm ( Table 1 ) . ( For these calculations we have omitted the volume of the hippocampus , which in mice represents 19% the volume of the cerebral cortex; its inclusion would modestly increase the estimates of axon length per neuron for those calculations based on axon density and cortical volume . ) As a check on this calculation , we have independently measured the density of cholinergic axons in the mouse motor cortex ( 0 . 044 µm/µm3; Figure 6 ) and also used the cholinergic axon density measured by Descarries et al . ( 2005 ) in parietal cortex ( 0 . 020 µm/µm3 ) to calculate single cholinergic axon arbor lengths of 33–107 cm ( Table 1 ) . Given the limitations of the sampling methods , we consider these estimates of axon length to be in reasonably good agreement with the 11–49 cm range for the four P30 cholinergic neurons traced in the present study . An analogous calculation based on published data for rat forebrain cholinergic neurons gives a mean axon length of 62 cm ( Table 1 ) . 10 . 7554/eLife . 02444 . 014Figure 6 . Quantifying ChAT+ axon density in P30 mouse cortex . ( A ) Coronal section of P30 mouse motor cortex following ChAT immunostaining . The cortical surface is at the top; the base of the cortex is at the bottom . Confocal images of the fluorescently immunostained tissue ( converted to grey scale and inverted ) were captured at Z-plane separations of 2 µm . Z-stacks encompassing planes 3–6 and 7–10 are shown . The traced axons for both sets of Z-stacks are color-coded with axons in stacks 3–6 in green , axons in stacks 7–10 in red , and regions of overlap in blue . Scale bar , 25 µm . ( B ) Axon tracings corresponding to the region in the right panel in ( A ) that is demarcated by the vertical red line . Scale bar , 25 µm . ( C ) ChAT immunostaining corresponding to the region adjacent to the left panel in ( A ) that is demarcated by the vertical black line . Scale bar , 25 µm . ( D ) The red rectangle shows the region of motor cortex analyzed in ( A ) , at approximately Bregma −1 . 06 . DOI: http://dx . doi . org/10 . 7554/eLife . 02444 . 014 The estimates of axon length for large arbors in the rodent brain led us to ask whether it might be possible to perform analogous calculations for neurons in the human CNS . To the best of our knowledge , this simple calculation has not been performed previously . The human data for axon density and neuron counts have been published for forebrain cholinergic neurons and for serotonergic neurons projecting from the dorsal raphe nucleus to the cortex , and cortical volume estimates for humans are available from MRI analyses; forebrain cholinergic neuron data is also available for chimpanzees ( Table 1 ) . These calculations lead to axon length estimates of 107 m and 31 m , respectively , for human and chimpanzee forebrain cholinergic neurons , and an axon length estimate of 170–348 m for human serotonergic neurons . For both neurotransmitter systems , the vast majority of the visualized fibers within the cortex are presumed to derive from projection neurons because the density of cortical cell bodies labeled with anti-ChAT or anti-serotonin antibodies is extremely low ( Raghanti et al . , 2008a , 2008b ) . Even if we allow for a possible under-estimate in neuron counts or a possible over-estimate in axon density measurements by as much as 2–3-fold , the calculations imply that in the human brain these two classes of projection neurons have axons that are , on average , many tens of meters in length .
The large sizes of the cholinergic axon arbors described here are consistent with a spatially diffuse modulatory role for cholinergic transmission in the cortex ( Descarries et al . , 1997; Lucas-Meunier et al . , 2003 ) . In the mouse , cortical columns are approximately 150–300 µm across , the diameter of an individual barrel in the barrel cortex ( Jan et al . , 2008 ) . Since the typical cholinergic axon arbor extends over two millimeters in the plane of the cortex ( Figures 2 and 3 ) , each arbor contacts multiple cortical columns . Although locally induced or locally restricted acetylcholine release is possible , it seems likely that signals originating in the dendrites and leading to action potentials at the soma of a forebrain cholinergic neuron would affect synaptic output over the entire arbor . This line of reasoning implies that cholinergic modulation of cortical and hippocampal function in response to information originating in the basal forebrain is likely to have low spatial resolution . A similar argument can be applied to nigrostriatal dopaminergic signaling ( Matsuda et al . , 2009 ) . The very large axon lengths calculated for human forebrain cholinergic neurons reflect ( 1 ) the enormous evolutionary expansion of the human cerebral cortex , which is ∼5000 times larger than the mouse cerebral cortex , ( 2 ) a more modest expansion in the number of basal forebrain cholinergic neurons , which differ between humans and mice by only a factor of 60–100 , and ( 3 ) a ∼twofold higher density of cholinergic axons in the human cortex ( Table 1 ) . The same line of reasoning applies to the rodent vs human difference in the size of individual serotonergic axon arbors ( Table 1 ) . Extremely large axon arbors present a cell biological challenge , as they require a correspondingly large expenditure of resources for growth , maintenance , and repair , especially as related to membrane synthesis and axonal transport . It is possible that the vulnerability of forebrain cholinergic neurons in the context of AD is related , at least in part , to the large size of their axon arbors . For example , vulnerability might be related to a particular sensitivity of the axonal transport machinery to biochemical perturbations associated with Aβ toxicity . Perhaps more significantly , the requirement that all transportation processes between cell body and axon arbor funnel through a single proximal axon segment suggests that trafficking within this segment may limit the efficacy of cellular responses to axonal damage or stress . The general idea that extreme axon length increases vulnerability to neurodegeneration has been discussed in the context of motor neuron disease ( Cavanagh , 1984; Ferraiuolo et al . , 2011 ) , and it seems reasonable that this concept might apply to a wide variety of neurons with very large axon arbors .
Experiments unrelated to AD were performed with Chat-IRES-CreER/+;R26IAP/+ mice ( referred to in the text as Chat-IRES-CreER;R26IAP ) . For AD experiments , R26IAP/R26IAP mice were crossed to Chat-IRES-CreER/Chat-IRES-CreER;APP/PS1/+ mice to obtain R26IAP/+;Chat-IRES-CreER/+ ( WT control ) and R26IAP/+;Chat-IRES-CreER/+;APP/PS1/+ littermates . The APPswe/PS1ΔE9 line was a gift from Dr Phil Wong ( Johns Hopkins University ) . Cholinergic neuron labeling followed intraperitoneal ( IP ) delivery of 1–5 µg 4-hydroxytamoxifen ( 4HT ) at P4-P5 with analysis at P12 , P30 and later ages , as indicated . A wide range of 4HT doses–from 1 µg to 400 µg– was tested in an initial survey of ∼250 mice to identify the optimal dose for sparse labeling . Mice were handled in accordance with the Institutional Animal Care and Use Committee ( IACUC ) guidelines of the Johns Hopkins Medical Institutions . Mice were deeply anesthetized with ketamine/xylazine and then sacrificed by trans-cardiac perfusion with neutral buffered 10% formalin solution ( Sigma-Aldrich , St . Louis , MO; equivalent to 4% paraformaldehyde ) . Brains or eyes were heated to 70°C for 90 min to inactivate endogenous phosphatase activity . Serial brain sections of 300 µm thickness were produced with a VT1200 vibratome ( Leica , Buffalo Grove , IL ) . AP histochemistry and clearing in 2:1 benzyl benzoate:benzyl alcohol ( BBBA ) were performed as described ( Wu et al . , 2012 ) . For long-term storage , AP-stained brain sections were equilibrated in ethanol and stored at −20°C . The following antibodies were used for immunostaining of 50–100 µm thick floating brain sections: goat anti-ChAT , 1:1000 ( AB143; Millipore , Billerica , MA ) ; rabbit anti-GFP , 1:1000 ( A11122; Invitrogen ) ; mouse anti-β-amyloid ( 6E10 ) , 1:500 ( NE1003; Millipore ) ; rabbit anti-GFAP , 1:1000 ( AB5804; Millipore ) ; rabbit anti-Iba1 , 1:1000 ( 019-19741; Wako , Richmond , VA ) ; mouse anti-GFAP 1:1000 ( MAB360; Millipore ) ; mouse anti-ABCA1 monoclonal antibody HJ1 ( ab66217; Abcam , Cambridge , MA ) ; mouse anti-β-Actin antibody AC-15 ( A5441; Sigma ) ; and rabbit anti-TH , 1:1000 ( AB152; Millipore ) . Secondary antibodies were from Invitrogen . GS-lectin staining used Alexa488-IB4 , 1:1000 ( I21411; Invitrogen , Grand Island , NY ) . Brain sections were incubated in primary antibodies diluted in PBS , 0 . 5% Triton X-100 , 0 . 1 mM CaCl2 ( PBSTC ) + 10% normal goat or donkey serum , washed in PBSTC for 6 hr , and incubated at 4°C overnight in secondary antibodies diluted in PBSTC +10% normal goat or donkey serum . After washing in PBSTC for 4–6 hr , brain sections were mounted in Fluoromount G ( 17984-25; EM Sciences , Hatfield , PA ) . Images were captured on a Zeiss LSM700 confocal microscope and processed with Zen software , ImageJ/Fiji , and Adobe Photoshop . Only brains with fewer than five AP+ neurons/hemisphere were subjected to detailed analysis . For high-resolution analyses , isolated arbors were imaged in bright-field mode at 10X magnification with Z-planes separated by 3 µm . Grey-scale images were captured with a Zeiss Imager Z1 system in montage mode and assembled with Zeiss AxioVision software . Neurites were traced using Neuromantic neuronal tracing freeware ( Darren Myat , http://www . reading . ac . uk/neuromantic ) in semi-automatic mode as described ( Wu et al . , 2012 ) . The total length of an axon arbor was obtained by summing up the lengths of the traces derived from that arbor within each tissue section . ( Following dehydration in ethanol and equilibration in BBBA , 300 µm thick vibratome sections of brain undergo an isotropic volume shrinkage of 3 . 32+/−0 . 17 [Figure 4C] . Axon lengths reported here have been corrected for that shrinkage . ) Branch points were manually scored using ImageJ/Fiji . To estimate arbor volumes , the smallest convex polygon encompassing all axon segments for a given arbor in the Z-stacked image for each 300 µm section was drawn over the image using ImageJ/Fiji , and each polygon area was calculated . Sections that contained only the subcortical axon segment were not included in the polygon analysis . Statistical analyses were performed with Excel and Graph-Pad . Error bars in the figures indicate standard deviation ( SD ) . p-values were calculated with the student's t test . Eight randomly selected 400 µm × 400 µm images of WT or APP/PS1+ axon arbors from each of three mice per genotype and per time point , with or without bexarotene treatments , were acquired with the Zeiss Imager Z1 system as described above . AP deposits >5 pixels in diameter ( images were 620 × 620 pixels ) were manually counted using ImageJ/Fiji software . No AP axon fragmentation or AP deposits were detected in WT mice younger than 8 months . Confocal images at Z-plane intervals of 2 µm within the interior of a 50 µm thick vibratome section were combined in two adjacent Z-stacks of four planes each ( i . e . , 8 µm thickness per Z-stack ) and ChAT+ axon segments were traced using Neuromantic software . To estimate the length of the traced axon segments residing outside of each Z-stack–an artifact that results from the incomplete elimination of out-of-plane signals in the confocal image–the total length of those axon segments that were traced from both of the adjacent Z-stacks was measured , and found to comprise 25 . 4% of the total axon length traced for each Z-stack . As half of this overlap derives from each Z-stack and as this effect occurs on both surfaces of each Z-stack , the corrected axon length for each Z-stack was calculated by subtracting 25 . 4% from the total length of the trace from each Z-stack . Bexarotene was purchased from Sigma-Aldrich ( SML0282 ) and dissolved in DMSO . Mice received daily gavages of 100 µl DMSO with or without bexarotene . Nine Chat-IRES-CreER;R26IAP;APP/PS1 littermates at 7 months of age were divided into three groups of three mice per group: group 1 received vehicle only ( DMSO ) daily for 14 days; group two received 100 mg/kg bexarotene daily for 14 days; and group three received 100 mg/kg bexarotene daily for 23 days . All mice received 50–100 µg 4HT IP at P5 . Mice were perfused and analysed by AP histochemistry and immunofluorescent staining as described above . For ABCA1 immunoblotting , 6 mice at 7 months of age were divided into two groups: 3 mice received vehicle only ( DMSO ) daily for 7 days , and the 3 mice received 100 mg/kg bexarotene daily for 7 days . On the eighth day , the cerebral cortices were homogenized in ∼1 ml PBS with 0 . 5% Triton X-100 , 1 mM phenylmethylsulfonyl fluoride ( PMSF ) , and complete protease inhibitor cocktail tablet ( 11697498001 , Roche , Indianapolis , IN ) , to give a total lysate protein concentration of ∼18 µg/µl . Proteins in SDS sample buffer were loaded without boiling onto a 7 . 5% SDS/polacrylamide gel and immunoblotted with mouse anti-ABCA1 monoclonal antibody HJ1 ( ab66217; Abcam ) . Conveniently , native mouse IgG in the extract runs at lower molecular weight than ABCA1 . Monoclonal Anti-β-Actin antibody AC-15 ( A5441; Sigma ) was used as a loading control . | The human brain is made up of roughly 80 to 100 billion neurons , organized into extensive networks . Each neuron consists of a number of components: a cell body , which contains the nucleus; numerous short protrusions from the cell body called dendrites; and a long thin structure called an axon that carries the electrical signals generated in the cell body and the dendrites to the next neuron in the network . One of the most studied networks in the human brain is the basal forebrain network , which is made up of large neurons that communicate with one another using a chemical transmitter called acetylcholine . This network has a key role in cognition , and its neurons are among the first to degenerate in Alzheimer's disease . However , relatively little is known about the structure of these ‘cholinergic’ neurons because their large size makes them difficult to study using standard techniques . Now , Wu et al . have visualized , for the first time , the complete 3D structure of cholinergic neurons in the mouse forebrain . The mice in question had been genetically modified so that only ten or so of their many thousands of cholinergic neurons expressed a distinctive ‘marker’ protein . This made it possible to distinguish these neurons from surrounding brain tissue in order to visualize their structures . The resulting pictures clearly illustrate the neurons' complexity , with individual axons in adult mice displaying up to 1000 branches . Measurements showed that each cholinergic axon in the mouse brain is roughly 30 centimeters long , even though the brain itself is less than 2 centimeters from front to back . Based on measurements by other researchers , Wu et al . calculated that the axons of single cholinergic neurons in the human brain are about 100 meters long on average . The extreme length and complex branching structure of cholinergic forebrain neurons helps to explain why each neuron is able to modulate the activity of many others in the network . It could also explain their vulnerability to degeneration , as the need to transport materials over such long distances may limit the ability of these neurons to respond to damage . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"cell",
"biology",
"neuroscience"
] | 2014 | Complete morphologies of basal forebrain cholinergic neurons in the mouse |
Termination of messenger RNA translation in Bacteria and Archaea is initiated by release factors ( RFs ) 1 or 2 recognizing a stop codon in the ribosomal A site and releasing the peptide from the P-site transfer RNA . After release , RF-dissociation is facilitated by the G-protein RF3 . Structures of ribosomal complexes with RF1 or RF2 alone or with RF3 alone—RF3 bound to a non-hydrolyzable GTP-analog—have been reported . Here , we present the cryo-EM structure of a post-termination ribosome containing both apo-RF3 and RF1 . The conformation of RF3 is distinct from those of free RF3•GDP and ribosome-bound RF3•GDP ( C/N ) P . Furthermore , the conformation of RF1 differs from those observed in RF3-lacking ribosomal complexes . Our study provides structural keys to the mechanism of guanine nucleotide exchange on RF3 and to an L12-mediated ribosomal recruitment of RF3 . In conjunction with previous observations , our data provide the foundation to structurally characterize the complete action cycle of the G-protein RF3 .
Termination of bacterial messenger RNA ( mRNA ) translation depends on recognition of any one of the three stop codons in the ribosomal A site by either class-1 release factor 1 ( RF1 ) , recognizing UAA and UAG , or 2 ( RF2 ) , recognizing UAA and UGA . RF1 and RF2 induce release of the polypeptide attached to the P-site transfer RNA ( tRNA ) ( for reviews , see Kisselev and Buckingham , 2000; Klaholz , 2011 ) . Subsequent release of the class-1 RF from the post-polypeptide release complex ( RC ) is facilitated by a class-2 release factor , the G-protein RF3 , in a GTP-dependent manner ( Freistroffer et al . , 1997 ) . RF3 , along with initiation factor 2 ( IF2 ) , elongation factor Tu ( EF-Tu ) and elongation factor G ( EF-G ) , constitute a set of four major translational GTPases ( Zavialov and Ehrenberg , 2003 ) . Free RF3 has high affinity to GDP; the RF3•GDP complex dissociates very slowly into RF3 and GDP and , thus , free RF3 in the cytoplasm exists mainly in the GDP-form ( Zavialov et al . , 2001 , 2002 ) . In contrast to EF-Tu , for which GDP exchange to GTP is conferred by the guanine nucleotide exchange factor EF-Ts off the ribosome , the guanine nucleotide exchange factor for RF3 is the ribosome in complex with a class-1 RF ( Zavialov et al . , 2001 ) . GDP-to-GTP exchange occurs rapidly on free IF2 and EF-G in the absence of any exchange factor . As RF3•GDP enters the class-1 RF-bound ribosome , GDP is readily exchanged for GTP , provided the peptide has been hydrolyzed off the P-site tRNA ( Zavialov et al . , 2002 ) . GTP binding is accompanied by conformational changes in both RF3 and the ribosome ( Gao et al . , 2007 ) . The change in ribosome structure promotes rapid release of the class-1 RF followed by GTP hydrolysis and , finally , release of RF3 in the GDP form from the ribosome ( Zavialov et al . , 2001 , 2002 ) . Structurally , termination of mRNA translation has been characterized by cryo-EM ( Rawat et al . , 2006 ) and X-ray crystal ( Petry et al . , 2005; Laurberg et al . , 2008 ) structures of A-site bound class-1 RFs interacting with their cognate stop codons . These structures have , in conjunction with molecular dynamics simulations ( Sund et al . , 2010 ) , contributed to present-day understanding of the principles of stop codon recognition by class-1 RFs and of their activation of the peptidyl transfer center for hydrolysis of the ester bond in peptidyl-tRNA . In addition , the X-ray crystal structure of free RF3•GDP ( Gao et al . , 2007 ) and cryo-EM ( Gao et al . , 2007 ) and crystal ( Jin et al . , 2011; Zhou et al . , 2012 ) structures of post-termination and post class-1 RF release ribosomes in complex with RF3 bound to GTP-analogs are known . The latter complexes indicate large structural changes of RF3 upon GTP binding after release of GDP , and suggest those structural changes to drive the ribosome from its macrostate-1 ( MS-I; i . e . , characterized by lack of intersubunit rotation; for definition of macrostates , see Frank et al . , 2007 ) into a state in which the two ribosomal subunits—30S and 50S—have rotated relative to each other—macrostate-2 ( MS-II ) . Furthermore , these structures of RF3•GDP ( N/C ) P-bound intersubunit-rotated ribosomes are incompatible with class-1 RF binding ( Gao et al . , 2007; Jin et al . , 2011; Zhou et al . , 2012 ) , as previously suggested from biochemical data ( Zavialov and Ehrenberg , 2003 ) . In this study we present cryo-EM density maps of the E . coli ribosomal post-peptide-release complex ( RC; 70S , tRNAfMet and [Met , stop]-mRNA ) in association with RF1 alone and in association with both apo-RF3 and RF1 . Complexes were obtained at a residual concentration of free GDP and in the absence of GTP . We show that the latter complex contains a novel form of RF1 in close interaction with the apo-form of RF3 and furthermore infer a direct interaction of RF3 with ribosomal protein L7/L12 . We use the present data to provide a structural basis for the mechanism of guanine nucleotide exchange on ribosome-bound RF3 and , in conjunction with previous data , to discuss the whole action cycle of RF3 .
We reconstituted ribosomal RCs in vitro by mixing XR7-Met-Stop ( UAA ) mRNA-programmed E . coli ribosomes containing fMet-tRNAfMet with purified RF1 and apo-RF3 ( containing residual guanine nucleotides ) . To determine the occupancy of RF1 and RF3 in the complex , the reaction was subjected to ultracentrifugation through a sucrose cushion followed by gel analysis in which we estimated the amounts of ribosome-bound RF1 and RF3 by comparing their respective band intensities to band intensities of ribosomal protein S1 . The gel showed RF1 and RF3 together on the ribosome in close to 1:1 ratios with S1 in the absence of extra added GDP ( Figure 1A , B ) . When the concentration of extra GDP was increased , the RF3 band decreased gradually and virtually disappeared at 200 μM extra GDP concentration , while the RF1 band remained at 1:1 stoichiometry with S1 ( Figure 1A , B ) . In a second , similar , experiment performed in the absence of RF1 , we detected no binding of RF3 to the ribosome—neither to RC nor to 70S—even with no extra added GDP ( Figure 1C ) . From these experiments we could infer that apo-RF3 formed a stable complex with the RF1-bound ribosome at low concentration of free GDP . However , in the absence of RF1 or at elevated GDP concentration , RF3 did not form a stable complex with the ribosome , in line with previous conclusions based on less direct experimental evidence ( Zavialov et al . , 2001 , 2002 ) . Thus , RF3•GDP does not form a stable complex with the RF1-bound ribosome , and we therefore conclude that the reconstituted ribosomal release complex presented in lane 1 of the gel presented in Figure 1A contains the apo-form of RF3 along with RF1 . 10 . 7554/eLife . 00411 . 003Figure 1 . Occupancy of RF1 and RF3 in the RC–RF1•RF3 complex and the effect of extra added GDP . ( A ) Titrating GDP into a release complex ( RC ) programmed with Met-stop mRNA , and containing fMet-tRNAfMet ( in the P site ) , RF1 and RF3 . The intensity of the band corresponding to RF3 decreased gradually to zero with increasing GDP concentration , but the S1 and RF1 intensities remained unaltered . ( B ) RF3 band intensity relative to S1 and RF1 plotted vs GDP concentration . ( C ) RC ( lanes 1–3 ) and naked 70S ribosome ( lanes 4–6 ) were incubated with either RF1 alone , RF3 alone or with both RF1 and RF3 as control experiments . RF3 could be detectably bound only to RC—and not to 70S— , when RF1 was present in the complex ( lane 3 ) . This confirms that the RF3 bands seen in the gel—in both panels A and C—indeed arise from functional complex formation . DOI: http://dx . doi . org/10 . 7554/eLife . 00411 . 003 The reconstituted RCs were subjected to cryo-EM and single-particle-based reconstruction ( Figure 2—figure supplement 1 ) assisted by image classification . Three classes were identified in total . The first class ( ∼14 , 000 projection images; Figure 2—figure supplement 2 ) yielded a density map of the 70S ribosome in MS-I conformation carrying a P-site tRNA and fragments of an E-site tRNA but with no trace of RF1 or RF3; this class will be disregarded in the following analysis . The second class ( ∼43 , 000 projection images; Figure 2A–C ) yielded a density map of the 70S ribosome in MS-I conformation carrying RF1 in the A site , a tRNA in the P site and trace density of a tRNA in the E site; the third class ( ∼29 , 000 projection images; Figure 2D–F ) yielded a density map of the 70S ribosome in MS-I conformation carrying apo-RF3 , RF1 , a P-site tRNA and trace density of an E-site tRNA . These two maps will be referred to as RC-RF1 and RC-RF1•RF3 , respectively . RF1 and RF3 were identified by their shapes and positions on the ribosome ( Rawat et al . , 2006; Gao et al . , 2007 , respectively ) . 10 . 7554/eLife . 00411 . 004Figure 2 . Termination complexes in association with RF1 alone and with both RF1 and apo-RF3 . ( A ) RC-RF1 ( resolution 8 . 4 Å ) . ( B ) Close-up of RF1 in the map . ( C ) RF1 density from ( B ) . Domains 1 , 2/4 and 3 of RF1 are labeled . ( D ) RC–RF1•RF3 ( resolution 9 . 7 Å ) . ( E ) Close-up of RF1and apo-RF3 in the map . RF1 contacts both L11—L11 contact point is labeled ‘L11 cp’— and apo-RF3 . ( F ) Density of RF1 and RF3 from ( E ) : RF1 domains labeled in black and RF3 domains 1 , 2 and 3 in blue . ( G ) – ( H ) In the RC–RF1•RF3 complex , apo-RF3 does not interact with L11 or L6/SRL . Labels: 30S: small ribosomal subunit ( yellow ) , 50S: large ribosomal subunit ( blue ) , cp: central protuberance , L11: ribosomal protein L11 , sp: spur , bk: beak , L1: L1 stalk , RF1: release factor 1 , RF3: release factor 3 , L6/SRL: position of ribosomal protein L6 and the sarcin/ricin loop , L11-NTD: ribosomal protein L11 N-terminal domain , arc: arc-like density . Density maps obtained from the total data set and of the RC-class not occupied by RF1 or RF3 are presented in Figure 2—figure supplement 1 and Figure 2—figure supplement 2 , respectively . Additional panels of isolated RF1- and RF1 , apo-RF3-densities are presented in Figure 2—figure supplement 3 . FSC curves for resolution assessment of all density maps obtained can be found in Figure 2—figure supplement 4 . DOI: http://dx . doi . org/10 . 7554/eLife . 00411 . 00410 . 7554/eLife . 00411 . 005Figure 2—figure supplement 1 . Density map resulting from refinement of the full data set . Map is displayed at a resolution of 6 . 3 Å in accordance with a Fourier shell correlation ( FSC ) of 0 . 14 between half sets . Labels same as in Figure 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 00411 . 00510 . 7554/eLife . 00411 . 006Figure 2—figure supplement 2 . Density map resulting from refinement of the ‘RC’ class identified in our classification . The map shows the ribosome occupied with a P-site tRNA but no release factors . DOI: http://dx . doi . org/10 . 7554/eLife . 00411 . 00610 . 7554/eLife . 00411 . 007Figure 2—figure supplement 3 . Segmented RF1 and RF1 , apo-RF3 densities . ( A ) , ( C ) Segmented RF1 density from RC-RF1 , compared to the segmented , combined RF1 , apo-RF3 density from RC-RF1•RF3 ( B ) , ( D ) . Density marked by asterisk in ( A ) and ( C ) is no longer present in ( B ) and ( D ) ; instead redistributed density is observed in RF1 domain 1 ( asterisk in [B] and [D] ) upon apo-RF3 binding . The arc of RF3 is labeled in all panels featuring RF3 . L11 cp marks the point of contact between L11 and RF3 in the RC-RF1•RF3 map . RF3 domains labeled in blue; RF1 domains in black . DOI: http://dx . doi . org/10 . 7554/eLife . 00411 . 00710 . 7554/eLife . 00411 . 008Figure 2—figure supplement 4 . FSC curves for density maps obtained from total data , RC , RC-RF1 and RC-RF1•RF3 . Resolutions , determined using the FSC = 0 . 5 cutoff criterion , are 8 . 0 Å ( full data set ) , 8 . 4 Å ( RC-RF1 ) , 9 . 7 Å ( RC-RF1•RF3 ) and 11 . 8 Å ( RC ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00411 . 008 In the RC-RF1 map , domain 3 of RF1 reaches into the peptidyl-transferase center of the 50S ribosomal subunit and the super-domain 2/4 of RF1 occupies the decoding center of the 30S ribosomal subunit . In addition , domain 1 of RF1 contacts the N-terminal domain of protein L11 ( L11-NTD ) and reaches toward the 30S subunit beak ( Figure 2A–C ) . In the RC-RF1•RF3 map , domain 3 of apo-RF3 contacts domain 1 of RF1 and extends toward the spur of the 30S subunit ( Figure 2D–F ) . Domain 3 of apo-RF3 interacts with 30S protein S12 , as observed earlier for RF3 in complex with non-hydrolyzable GTP analogs—GDP ( C/N ) P—bound to the ribosome in the rotated MS-II form in the absence of a class-1 RF ( Gao et al . , 2007; Jin et al . , 2011; Zhou et al . , 2012 ) . However , we observe only a weak interaction between domain 2 of apo-RF3 and 16S rRNA helix 5 , whereas the interaction of RF3•GDP ( C/N ) P with this helix is strong ( Gao et al . , 2007; Jin et al . , 2011; Zhou et al . , 2012 ) . Furthermore , we observe domain 1 of RF1 contacting both domain 3 of RF3 and L11-NTD ( Figure 2D–F and Figure 2—figure supplement 3B , D ) . Comparison of the RF1 conformation in our RC-RF1•RF3 map ( Figure 2D–F and Figure 2—figure supplement 3B , D ) with that in the RC-RF1 map ( Figure 2A–C and Figure 2—figure supplement 3A , C ) reveals a large conformational change of RF1 in response to the binding of apo-RF3 resulting in the formation of a bridge from apo-RF3 via RF1 to L11 ( Figure 2—figure supplement 3 ) . Domain 1 of apo-RF3 is in the vicinity of , but not in contact with , L6/SRL ( Figure 2G , H , Figure 3A ) . In contrast , a direct contact between RF3 and L6/SRL was observed in the post class-1 RF-release RC-RF3•GDP ( C/N ) P maps ( Figure 3B ) ( Klaholz et al . , 2004; Gao et al . , 2007; Jin et al . , 2011; Zhou et al . , 2012 ) , with the ribosome in the rotated MS-II state . Hence , in the apo-RF3 and RF1-containing complex , apo-RF3 is docked to the RC framework only through domain 3 , while in the RC-RF3•GDP ( C/N ) P maps RF3•GDP ( C/N ) P is seen to be firmly docked to RC through contacts involving domains 1 , 2 and 3 ( Klaholz et al . , 2004; Gao et al . , 2007; Jin et al . , 2011; Zhou et al . , 2012 ) . 10 . 7554/eLife . 00411 . 009Figure 3 . Comparison of L11 , L6/SRL , arc ( L12-CTD ) and apo-RF3 in RC-RF1•RF3 and RC-RF3 maps . ( A ) Apo-RF3 is observed in contact with L12-CTD ( arc ) but L12-CTD is not in contact with L11-NTD . Apo-RF3 domain 1 is not in contact with L6/SRL . In contrast ( B ) , RF3•GTP is observed in contact with L6/SRL and in this map ( RC-RF3•GTP ) , L12-CTD is in contact with both RF3 domain 1 and L11-NTD . Both maps are displayed at the resolution of the latter—16 Å—for direct comparison; unsegmented maps were chosen to avoid artifacts on the edges of protein densities and , hence , to present a correct analysis of the interactions occurring . DOI: http://dx . doi . org/10 . 7554/eLife . 00411 . 009 While these observations bear out on the interactions between RF1 and RF3 on the ribosome , an additional finding sheds light on ribosomal recruitment of RF3: Domain 1 of RF3 contains an ‘arc-like’ density ( Figure 2G , H , Figure 2—figure supplement 3B , D ) clearly separated from the density of the NTD of L11 . An ‘arc-like’ density was observed in a similar position , but fully bridging RF3 to L11-NTD in the RC-RF3•GDPNP map ( Figure 3B ) ( Gao et al . , 2007 ) . Moreover , a similar bridge to L11-NTD was observed in EF-G- ( Agrawal et al . , 1998 , 1999 ) , EF-Tu- ( Stark et al . , 1997 ) and IF2-bound ribosome complexes ( Allen et al . , 2005 ) . In the case of EF-G , the bridge was subsequently identified as the CTD of L7/L12 ( henceforth denoted L12; the difference being an N-terminal acetylation in case of L7 ) ( Datta et al . , 2005 ) . The interaction we observe in the present RC-RF1•RF3 map between L12-CTD and apo-RF3 without involvement of L11-NTD has not previously been reported . To interpret the interactions between the RFs and the ribosome in more detail , we used crystal structures of E . coli RF1 ( Graille et al . , 2005; PDB ID 2B3T ) and RF3 ( Gao et al . , 2007; PDB ID 2H5E ) for modeling in our new RC-RF1 and RC-RF1•RF3 maps and the previous RC-RF3•GDPNP map ( Gao et al . , 2007 ) . Unresolved regions in the structures of both RF1 and RF3 were ab initio- or homology-modeled in accordance with the cryo-EM density maps . We performed molecular dynamics flexible fitting ( MDFF ) ( Trabuco et al . , 2009 ) of our atomic RF1 and RF3 models and the X-ray crystal structure of L12-CTD ( Leijonmarck and Liljas , 1987; PDB ID 1CTF ) to the above-mentioned maps ( Figures 4 and 5 ) . Root-mean-square deviation plots describing changes in the structures as well as changes in the cross-correlation coefficient between structures and density maps over the course of the MDFF fits are presented in Figure 4—figure supplement 1 . 10 . 7554/eLife . 00411 . 010Figure 4 . Flexible fitting by MDFF focusing on RF1 conformation and interactions . ( A ) – ( C ) Comparison of E . coli RF1 conformations after MDFF . ( A ) RF1 in RC-RF1 . ( B ) RF1 ( and RF3 ) in RC-RF1•RF3 . ( C ) Superimposition of RF1 from ( A ) and ( B ) . Arrow in ( C ) denotes the conformational change of domain 1 . ( D ) In the RC-RF1 complex , RF1 interacts through helices α2 and α3 with the P-rich 310-helix in the L11-NTD . ( E ) In the RC-RF1•RF3 complex α3 ( RF1 ) interacts with the 310-helix ( L11-NTD ) , while α2 ( RF1 ) interacts with RF3 . α2 and α3 denote helices α2 and α3 in domain 1 of RF1; P-rich 310 denotes the P-rich 310 helix in L11-NTD . Root mean square deviation and cross correlation coefficient plots for our MD flexible fitting are presented in Figure 4—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 00411 . 01010 . 7554/eLife . 00411 . 011Figure 4—figure supplement 1 . RMSD and CCC values for MD flexible fitting . ( A ) Root mean square deviation ( RMSD ) over the course of MD flexible fitting between model in current frame and initial model in the three systems . ( B ) Cross-correlation coefficients ( CCC ) over the course of MD flexible fitting between models and density maps . Red lines indicate the time points at which frames ( models ) were selected for presentation in this study . DOI: http://dx . doi . org/10 . 7554/eLife . 00411 . 01110 . 7554/eLife . 00411 . 012Figure 5 . Flexible fitting by MDFF focusing on RF3 conformation and interaction with RF1 . ( A ) – ( C ) Comparison of E . coli RF3 conformations after MDFF . ( A ) Crystal structure of free RF3•GDP displayed in its simulated density . ( B ) apo-RF3 ( and RF1 ) in RC-RF1•RF3 ( L12-CTD not displayed ) ( C ) RF3 in RC-RF3•GDPNP ( Gao et al . , 2007 ) ( L12-CTD not shown ) . Superimposition ( D ) of closed conformation RF3•GDP ( blue ) , semi-open apo-RF3 ( green ) and closed RF3•GTP ( gray ) . Arrows indicate rotations of domain 3 . ( E ) Close-ups of domain 3 conformations from ( A ) – ( D ) . ( F ) Candidate residues in domain 3 of RF3 for a charge-based interaction with RF1 and 30S protein S12 . ( G ) Candidate residues in RF1 helix α2 for a charge-based interaction with RF3 . RF3 domains labeled in blue; α2 and α3 denote helices α2 and α3 in domain 1 of RF1 . Figure 5—figure supplement 1 shows superimposition of apo-RF3 and RF3•GDPNP . DOI: http://dx . doi . org/10 . 7554/eLife . 00411 . 01210 . 7554/eLife . 00411 . 013Figure 5—figure supplement 1 . Movement of domain 1 in RF3 . Superimposition of apo-RF3 ( green ) and RF3•GDPNP ( gray ) . Domain 1 moves closer ( arrow ) to L6/SRL upon GDPNP recruitment to RF3 to mediate binding of RF3•GDPNP to L6/SRL . DOI: http://dx . doi . org/10 . 7554/eLife . 00411 . 013 In the RC-RF1 map ( Figure 2A ) we observe RF1 in an open conformation bound to the unrotated , MS-I state ribosome ( Figure 4A , C and D ) , in accordance with the results of earlier studies of this complex ( Petry et al . , 2005; Rawat et al . , 2006; Laurberg et al . , 2008 ) . Moreover , our RC-RF1 map conforms to the position of L11 observed in those studies; a position displaced toward the A site when compared to a 70S ribosome with a vacant A site ( Laurberg et al . , 2008 ) . By comparing RF1 fitted to our RC-RF1 ( Figure 4A , D ) and RC-RF1•RF3 ( Figure 4B , E ) maps , the conformational change in domain 1 of RF1 upon RF3 binding can be clearly visualized ( Figure 4C ) . In our RC-RF1 map , helices α2 and α3 in domain 1 are observed in contact with the P-rich 310-helix in L11-NTD ( Figure 4A , D ) ( Petry et al . , 2005; Rawat et al . , 2006; Laurberg et al . , 2008 ) . In the RC-RF1•RF3 map , we observe an RF1-mediated structural bridge between L11-NTD and domain 3 of RF3 ( Figure 4E ) . The stacking between helices α2 and α3 in domain 1 of RF1 is partly disrupted and helix α2 has undergone a rotation by 72° away from L11 ( Figure 4C , D and E ) . Thereby , α2 is positioned within binding distance of domain 3 of RF3 ( Figure 4B , C and E ) while the contact between the tip of α3 and the P-rich 310-helix in L11 is maintained ( Figure 4E ) ( Petry et al . , 2005; Laurberg et al . , 2008 ) . We compared conformations of apo-RF3 and RF3•GTP fitted to their respective maps with the closed conformation of free RF3•GDP ( Figure 5A–C ) . Apo-RF3 in the RC-RF1•RF3 map is observed in a semi-open conformation characterized by a 36° rotation of its domain 3 away from its domain 1 ( Figure 5B , D ) . RF3•GTP in the later stage RC-RF3 map is observed in its open conformation with domain 3 rotated 49° away from domain 1 when compared to the closed conformation of free RF3•GDP ( Figure 5C , D ) . In the semi-open conformation , amino acid stretches R399-Q411 and N458-N467 of RF3 form a cluster of 12 charged or polar amino acids in two flexible loops of domain 3 at the interface with RF1 ( Figure 5E ) . Examining the surface chemistry of helix α2 of RF1 in the region of interaction with RF3 , we find 5 charged amino acids ( H13 , E14 , E17 , E18 and Q20 ) , which are accessible to interaction with RF3 ( Figure 5F ) . To shed further light on the nature of the binding observed between RF1 and apo-RF3 we examined electrostatic surface potentials in the regions mediating their interaction ( Figure 6A–D ) . The surface patch of helix α2 in domain 1 of RF1 that contacts RF3 is negatively charged ( Figure 6A , B ) , whereas the complementary surface patch in RF3 is positively charged ( Figure 6C , D ) . The combined observations from fitting ( Figure 5D , E ) and surface electrostatics ( Figure 6A–D ) suggest that amino acid stretches R399-Q411 and N457-N467 of apo-RF3 are involved in a charge-based interaction with helix α2 of RF1 and residues Q74 , E75 and H76 of protein S12 ( Figure 5D ) . Thus , binding of apo-RF3 to the ribosomal release complex is obtained by formation of a bridge connecting L11-NTD through domain 1 of RF1 with domain 3 of RF3 ( Figures 2E and 4E ) . 10 . 7554/eLife . 00411 . 014Figure 6 . Electrostatic surface potentials of RF1 and RF3 . ( A ) Electrostatic surface potential of RF1 . Red surfaces are electronegative , blue surfaces electropositive . ( B ) Close-up of helix α2 in RF1 , where interaction with RF3 is mediated . Labeled residues ( pointers ) are candidates ( as identified by MDFF ) for a direct interaction with RF3 . The surface potential of helix α2 is electronegative . ( C ) Electrostatic surface potential of RF3 . ( D ) Close-up of the two loops in domain 3 of RF3 posing charged or polar residues for contact with RF1 . The surface potential of RF3 in this region is electropositive , pointing to a charge-based interaction between RF1 and RF3 . In Bacteria expressing RF3 ( E ) the surface potential of RF1 in the α2 helical region ( position of helix α2 indicated by dotted lines ) is overall electronegative . In comparison , this region in class-1 RFs from Bacteria and Archaea not expressing RF3 ( F ) is overall electroneutral . ( G ) RF3 from D . vulgaris displays an electropositive surface similar to what we observe in E . coli RF3 ( position of the flexible loop region indicated by dotted line ) . PDB IDs; T . thermophilus RF1: 3D5A , T . thermophilus RF2: 2WH3 , T . maritima RF1: 1RQ0 , E . coli RF1: current study , modeled/fitted from 2B3T , E . coli RF2: 1Gqe , S . mutans RF1: 1Zbt , D . vulgaris RF3: 3Vqt . DOI: http://dx . doi . org/10 . 7554/eLife . 00411 . 014 As amino acids in these regions of the two release factors are not generally conserved across RF3-expressing species , we investigated whether the electrostatic properties might be conserved ( Figure 6E–G ) . Indeed , we find that in X-ray structures available for class-1 RFs from RF3-containing Bacteria ( E . coli and S . mutans; Figure 6E ) , the surface of helix α2 has an overall electronegative surface potential . In addition , examining the surface potential in RF3 from D . vulgaris ( Figure 6G ) ( Kihira et al . , 2012 ) we find a similar positive surface potential in the same position as we observed for RF3 from E . coli . In contrast , for Bacteria and Archaea lacking RF3 ( T . thermophilus and T . maritima; Figure 6F ) the surface potential of helix α2 is overall electroneutral . These observations suggest the existence of charge-based interactions between RF1 and apo-RF3 . We explored the above-described interaction between apo-RF3 and RF1 in further detail . We generated five RF1 mutants , H13A , E14A , E17A , E18A , and Q20A , and compared the ability of RF3 to recycle these mutants as well as wild-type RF1 during translation termination . In our recycling assay , we mixed RCs containing an MLL tripeptidyl-tRNA in the P site and a UAA stop codon in the A site . The RC was in large excess over RF1 , so that many cycles of RF1 action were required to release MLL tripeptides from all RCs . Consistent with earlier reports ( Freistroffer et al . , 1997; Zavialov et al . , 2001; Gao et al . , 2007 ) , RF3 increased the recycling rate of wild-type RF1 by a factor of ten in a GTP-dependent manner ( Figure 7 ) . The ability of RF3 to recycle wild-type RF1 and RF1 mutants E14A , E17A , and Q20A was similar . In contrast , compared with recycling of wild-type RF1 , recycling of the E18A mutant was reduced threefold and recycling of the H13A mutant was reduced 1 . 4-fold . At the same time , RF3-independent recycling rates of all RF1 variants were indistinguishable , showing that the lower recycling of the E18A and H13A mutants was indeed RF3-mediated . 10 . 7554/eLife . 00411 . 015Figure 7 . MLL release from RC due to RF3-mediated recycling of RF1 ( WT and mutants ) . MLL release from RC due to recycling of RF1 WT and mutants with excess of RF3 over time . The chart shows recycling times for RF1 variants in the presence of RF3 . DOI: http://dx . doi . org/10 . 7554/eLife . 00411 . 015 In our attempt to investigate conservation of the set of amino acids included in our recycling assay , we defined a non-redundant set ( Sethi et al . , 2005 ) of RF1 sequences from Archaea and Bacteria . For the amino acids E14 , E17 and E18 , we observed conservation rates—estimated as occurrence of D or E—of 28% , 40% and 63% , respectively . Q20—no effect in RF3-mediated recycling—and H13 were not conserved . We note that E18—associated with threefold reduction of recycling rate in our assay—is the most conserved amino acid of the candidates . In light of the findings of surface charge conservation and from the mutation studies , we propose a model in which E18 in the α2-region of RF1 is not only involved in establishing the negatively charged surface patch important for interaction with the positively charged counterpart in RF3 but also involved in a specific interaction with RF3 . We attribute the reduction in RF1 recycling rate by the H13A mutation to indirect—steric and/or charge—effects and propose that H13 contributes to optimal orientation of helix α2 including E18 and other residues important for interaction with RF3 . However , a direct interaction between H13 and RF3 cannot be excluded . As mentioned above , we observe apo-RF3 in a semi-open conformation . Previously , the conformation of RF3•GDP has been described as closed ( Gao et al . , 2007 ) and the conformation of RF3•GTP as open ( Gao et al . , 2007; Jin et al . , 2011; Zhou et al . , 2012 ) ( Figure 5A–D ) . The apparent dynamics in RF3 conformation lends insight to the mechanism of nucleotide loading-and-release: in the X-ray structure of RF3•GDP , the β-phosphate group of GDP is stabilized by interaction with the conserved H92 of RF3 ( Figure 8B , E ) ; surprisingly , no Mg2+ ion is participating in this stabilization ( Gao et al . , 2007 ) . H92 is part of switch region 2; switch regions 1 and 2 are known to be flexible regions involved in activation and deactivation of GTPase activity in RF3 . We note that switch region 1 is disordered in the RF3•GDP crystal structure and therefore not resolved . Looking at the position of H92 in semi-open apo-RF3 , we observe H92 retracted from the nucleotide-binding pocket of RF3 ( Figure 8C , E ) . In this retracted position—semi-open RF3—H92 is unlikely to participate in nucleotide stabilization . In the later-stage , open RF3•GTP , H92 is observed in a similarly retracted position where it is not participating in stabilization of GTP ( Figure 8D , E ) . Moreover , coordination of the β-γ phosphates of GTP is performed by a Mg2+ ion , the binding of which is made possible by a well-ordered switch region 1 . The Mg2+ ion is likely to be essential for GTPase activity ( Zhou et al . , 2012 ) . Hence , the release of GDP from RF3 as it changes conformation from closed ( RF3•GDP ) to semi-open ( apo-RF3 ) is likely triggered by the retraction of H92 from the nucleotide-binding pocket . This allows for entry of GTP , which is accompanied by a change in RF3-conformation from semi-open ( apo-RF3 ) to open ( RF3•GTP ) where GTP is stabilized by a Mg2+ ion coordinated by a well-ordered switch region 1 . 10 . 7554/eLife . 00411 . 016Figure 8 . RF3 nucleotide loading-and-release . ( A ) RF3•GTP with the nucleotide-binding pocket inside the red box . ( B ) – ( D ) Nucleotide-binding pockets of RF3•GDP , apo-RF3 and RF3•GTP . In RF3•GDP , GDP is stabilized by interaction with H92 . In apo-RF3 ( B ) and in RF3•GTP ( C ) H92 is retracted away from the nucleotide-binding pocket . Instead , this stabilization is performed by a Mg2+ ion . ( E ) Superimposition of H92 position ( black box in [B]–[D] ) showing H92 in its GDP-stabilizing position ( blue ) and in its retracted positions ( green and gray ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00411 . 016 As mentioned above , we observe an ‘arc-like’ density in contact with domain 1 of apo-RF3 in the RC-RF1•RF3 termination intermediate which we attribute to L12-CTD ( Figure 2D–H; Figure 2—figure supplement 3B , D; Figure 3 and Figure 5B ) . Based on our fitting , the sole interaction between RF3 and L12 involves helix α7 in the G’ subdomain of RF3 and helices α4 and α5 in L12-CTD ( Helgstrand et al . , 2007 ) , while L12-CTD is clearly separated from L11-NTD by a 14-Å gap ( Figure 9A , B and Figure 3A ) . In contrast , in the map of the later-stage RC-RF3•GDPNP ( Figure 9C , D and Figure 3B ) L12-CTD is observed in contact with both L11-NTD and RF3 . This suggests that L12-CTD and RF3 interact prior to formation of a structural bridge with L11-NTD . For RF3 to change from apo-RF3 to RF3•GTP , as observed in the RC-RF3•GDPNP complex , domain 3 must rotate even further away from domain 1 ( i . e . , from 36 to 49° ) , bringing RF3•GDPNP into its fully open conformation ( Figure 5B–D ) . Our fitting predicts that the change from RC-RF1•RF3 to RC-RF3•GDPNP is characterized by a concomitant upward rotation of L12-CTD by ∼45° toward the tip of L11-NTD ( Figure 9E ) . This rotation is accompanied by a ‘hinging-in’ movement of helices α4 and α5 , whereby α4 rotates inward by ∼29° and α5 by ∼11° ( Figure 9F ) , leading to formation of a structural bridge between domain 1 of RF3 and a loop in L11-NTD ( Figure 9C , D ) . Similar flexibility in the α4/α5 region of L12 has been observed in earlier studies ( Leijonmarck and Liljas , 1987; Harms et al . , 2008; Gao et al . , 2009 ) . Along with the change in position and conformation of L12-CTD observed in the RC-RF1•RF3 and RC-RF3•GDPNP complexes , L11 undergoes an upward rotation of ∼7° , thereby disrupting its interaction with helix α3 in domain 1 of RF1 ( Figure 9B , D ) and positioning the tip of the L11-NTD for binding to L12-CTD ( Figure 9D ) . Furthermore , the ribosome conformation changes—primarily by intersubunit rotation—from MS-I to MS-II , and domain 1 of RF3 moves closer , by a 12° rotation of helix α7 , to L6/SRL to form an interaction ( Figure 5—figure supplement 1 ) ( Klaholz et al . , 2004; Gao et al . , 2007; Jin et al . , 2011; Zhou et al . , 2012 ) in accordance with the earlier reported mechanism governing onset of GTPase activity in EF-G ( Clementi et al . , 2010 ) . 10 . 7554/eLife . 00411 . 017Figure 9 . Dynamics of interactions involving RF3 and L12-CTD . ( A ) RC-RF1•RF3 map . ( B ) Close-up of RF1 helix α3 , apo-RF3 , L12-CTD and L11-NTD after fitting . ( C ) RC-RF3•GDPNP map ( Gao et al . , 2007 ) . ( D ) Close-up of RF3•GDPNP , L12-CTD and L11-NTD after fitting . ( E ) L12-CTD position in ( B ) ( cyan ) and ( D ) ( orange ) displayed side-by-side . ( F ) Superimposition of L12-CTD from ( B ) and ( D ) showing the ‘hinging-in’ conformational change observed . DOI: http://dx . doi . org/10 . 7554/eLife . 00411 . 017 To further investigate the nature of L12–CTD interaction with RF3 we examined electrostatic surface properties of both factors in the region of the contact ( Figure 10A–D ) . It is evident , that the region in RF3 occupied by helix α7 constitutes a negatively charged surface patch ( Figure 10A , B ) , whereas the complementary surface occupied by helices α4 and α5 in L12-CTD is positively charged ( Figure 10C , D ) . Examination of electrostatics for the other major translational GTPases ( IF2 , EF-Tu and EF-G ) in the proposed area of interaction shows similar negatively charged patches for potential interaction with L12-CTD ( Figure 10—figure supplement 1A–F ) . 10 . 7554/eLife . 00411 . 018Figure 10 . Electrostatic surface potentials of L12-CTD , RF3 and other major translational GTPases . Red surfaces are electronegative , blue surfaces electropositive . ( A ) The electrostatic surface potential of RF3 ( B ) Close-up of domain 1 in RF3 , displaying the position of helix α7 in the G’ subdomain responsible for binding to L12-CTD ( [C]; close-up in [D] ) . The overall surface potential of the G’ subdomain in RF3 is negative and the overall surface charge of helices α4 and α5 in L12-CTD is positive . Figure 10—figure supplement 1: Surface potentials of IF2 ( A ) and ( B ) , EF-Tu ( C ) and ( D ) , and EF-G ( E ) and ( F ) are all electronegative in domain 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 00411 . 01810 . 7554/eLife . 00411 . 019Figure 10—figure supplement 1 . Surface potentials of IF1 , EF-Tu and EF-G . Surface potentials of IF2 ( A ) and ( B ) , EF-Tu ( C ) and ( D ) , and EF-G ( E ) and ( F ) are all electronegative in domain 1 , where they interact with the electropositive α4/α5 patch of L12-CTD . The close-ups display in ribbon the secondary structures involved in the interactions between L12-CTD and the major translational GTPases ( Kothe et al . , 2004; Allen et al . , 2005; Helgstrand et al . , 2007; Harms et al . , 2008; Gao et al . , 2009 ) . PDB IDs are as follows: E . coli RF3: current study , modeled/fitted from 2H5E , E . coli L12-CTD: 1Ctf , E . coli IF2: 1ZO1 , E . coli EF-Tu: 1DG1 , E . coli EF-G: 2Wrk . DOI: http://dx . doi . org/10 . 7554/eLife . 00411 . 019
The RC-RF1•RF3 structure observed here shows the guanine nucleotide-free apo-form of RF3 in a stable RF1-containing ribosome complex . The existence of such a structure was predicted from biochemical experiments on the action cycle of RF3 ( Zavialov et al . , 2001; Sternberg et al . , 2009 ) , but direct experimental evidence has been lacking . The present gel separation analysis proves that indeed it is the apo-form of RF3 in the RC–RF1•RF3 complex . This analysis shows that RF3•GDP does not form a stable complex with the RC , irrespective of whether or not RF1 is present . Furthermore , formation of a stable ribosomal complex with our RF3 preparation requires the presence of RF1 and is possible only when the concentration of free GDP is low . In the chemical equilibrium between the RC–RF1•RF3 complex and the RC–RF1 complex accompanied by free RF3•GDP , the former complex dominates at low concentrations of free GDP , whereas the latter two dominate at high concentrations of free GDP . Hence , from our gel data ( Figure 1 ) , we infer that the present cryo-EM complex must contain RF1 together with the apo-form of RF3 . It has been shown that the affinity of GTP to free RF3 is at least three orders of magnitude lower than that of GDP and that the average time of spontaneous release of GDP from free RF3 is 30 s ( Zavialov et al . , 2001 ) , a release time that is comparable with the 100 s of GDP from EF-Tu in the absence of EF-Ts ( Ruusala et al . , 1982 ) . The high affinity of GDP to free RF3 , possibly due to multiple contacts between H92 of RF3 and the β-phosphate group of GDP ( Gao et al . , 2007 ) , in combination with the slow dissociation of GDP from RF3 strongly suggests that RF3 exists predominantly in the GDP form in the living cell . Moreover , it has been shown that the class-I RF-bound RC is the guanine nucleotide exchange factor for the G-protein RF3 ( Zavialov et al . , 2001 ) , an observation that suggests that the RC–RF1•RF3 complex contains a form of RF3 from which GDP can readily dissociate and to which GTP can readily associate . That is , the novel complex visualized here is likely to provide the structural basis for the ribosome-dependent guanine nucleotide exchange on RF3 . Accordingly , it is RF3 in the compact GDP form that enters the class-1 RF-containing post-termination ribosome complex from which the peptide chain has dissociated after ester bond hydrolysis . As presented here and in previous work ( Zavialov et al . , 2001; Zavialov et al . , 2002 ) , RF3•GDP has low affinity to the RC . We infer that the semi-open apo-RF3 structure we observe is formed rapidly by its stabilizing contacts with ribosomal protein L12 ( Figure 9 ) and a novel form of RF1 . The latter , we propose , is formed concomitantly with formation of the apo-form of RF3 ( Figure 11 and Video 1 ) . This novel form of RF1 stabilizes the binding of the apo-form of RF3 by contact between helix α2 in RF1-domain 1 and domain 3 of apo-RF3 ( Figure 4E ) . The rapid conformational change of RF3 and the stabilization of RF3’s apo-form make release of GDP and binding of GTP to RF3 a high-probability event and dissociation of RF3•GDP a low-probability event after initial binding of RF3•GDP to the class-1 RF-containing RC . This implicates L12 and RF1 as essential for the efficient recruitment of RF3 to the RC . L12-CTD has been shown capable of interacting with RF3•GDP ( Helgstrand et al . , 2007 ) and it is therefore possible that L12 interacts already with free RF3•GDP ( Figure 11 ) , facilitating the first ribosome-binding step of the factor , but the collection of more experimental data is required to settle this issue . 10 . 7554/eLife . 00411 . 020Figure 11 . Proposed model of translation termination involving intersubunit- , L12- , RF1- and RF3-dynamics . A class-1 RF recognizes its cognate mRNA stop codon in the ribosome and binds in the A site . Here , the class-1 RF mediates release of the nascent protein attached to the P-site tRNA . After nascent protein release , RF3•GDP is recruited to the ribosome . RF3•GDP is in its closed form and does not form a stable complex with the ribosome; it is possible that initital contact between the ribosome and RF3•GDP is mediated by L12-CTD . As RF3 lodges onto the ribosome , GDP is released and apo-RF3 assumes its semi-open conformation . At this point , apo-RF3 is in contact with L12-CTD , the class-1 RF and 30S protein S12 . Upon recruitment of GTP to apo-RF3 , RF3•GTP assumes its open conformation and the ribosome changes from the unrotated macrostate I to the rotated macrostate II and the class-1 RF leaves the complex . In this state , RF3•GTP is in contact with L6/SRL and L12-CTD , already interacting with RF3 , forms a bridge to L11-NTD . This binding state of RF3•GTP marks the onset of RF3’s GTPase activity leading to cleavage of GTP and release of Pi . RF3•GDP dissociates from the ribosomal complex in its closed conformation and the ribosome is ready for subunit recycling . DOI: http://dx . doi . org/10 . 7554/eLife . 00411 . 02010 . 7554/eLife . 00411 . 021Video 1 . Animation of proposed model for RF3-mediated termination of translation . RF1 ( brown ) is positioned in the ribosomal A site in the RC-RF1 complex . Here , its domain 1 makes contact with L11-NTD ( black ) . After initial contact in the cytoplasm between RF3•GDP and L12-CTD , RF3•GDP/L12-CTD lodges onto the ribosome; in this process GDP is released and apo-RF3 adopts its semi-open conformation . RF1 undergoes a conformational change , thereby enabling formation of the stabilizing bridge between L11-NTD , RF1 and apo-RF3 ( RC-RF1•RF3; RF1 in magenta , apo-RF3 in green , L12-CTD in cyan ) . At this point , there is neither a contact between domain 1 of apo-RF3 and L6/SRL nor between L12-CTD and L11 . Next , apo-RF3 recruits GTP , thereby changing to its fully open conformation and locking the ribosome in the rotated ( MS-II ) state . Intersubunit rotation is accompanied by L11 moving away from RF1 and toward RF3•GTP , thereby disrupting its contact with RF1 and triggering the formation of another stabilizing bridge between L11-NTD , L12-CTD and RF3•GTP . Disruption of RF1 interaction with L11 in concert with the ribosome being locked in MS-II is incompatible with continued presence of a class-1 RF in the A site , forcing it to leave the complex ( RC-RF3•GTP; L11 in dark green , L12-CTD in orange , RF3•GTP in gray ) . Moreover , in the rotated RC-RF3•GTP conformation , domain 1 of RF3•GTP contacts ribosomal L6/SRL , marking the onset of GTPase activity . Once GTP is cleaved and Pi is released , RF3•GDP returns to its closed , cytoplasmic state . DOI: http://dx . doi . org/10 . 7554/eLife . 00411 . 021 Conditional on the absence of a peptide chain on the P-site bound tRNA , GTP binding to RF3 induces the open conformation of the factor thereby driving the ribosome into the intersubunit rotated MS-II state ( Klaholz et al . , 2004; Gao et al . , 2007; Jin et al . , 2011; Zhou et al . , 2012 ) . A movement of L11 away from domain 1 of RF1 and toward L12 accompanies this change in ribosome conformation ( Figure 9 ) . In the process , the structural bridge between the P-rich 310-helix in L11 , RF1 domain 1 and RF3 domain 3 ( Figures 2E and 4E ) is broken and replaced by a structural bridge formed between the tip of L11-NTD , L12-CTD and domain 3 of RF3•GTP ( Figure 9C , D and Figure 3B ) , evidently stabilizing RF3•GTP binding to the RC in its MS-II conformation . Concomitantly , RF3•GTP makes contact with L6/SRL on the large ribosomal subunit ( Figure 3 and Gao et al . , 2007; Jin et al . , 2011; Zhou et al . 2012 ) As a result of the movement of L11 , domain 1 of the class-1 RF is no longer in contact with the ribosome . Thermodynamically , this scenario requires a much lower affinity of GTP to the semi-open than to the open RF3 conformation ( Hauryliuk et al . , 2008 ) , implying that the semi-open conformation of RF3 has low affinity to both GDP and GTP in analogy with the low affinity of both GDP and GTP to the EF–Tu•Ts complex ( Ruusala et al . , 1982 ) . Finally , the intersubunit-rotated MSII state leads to a steric clash between the ribosome and the class-1 RF , which results in the release of the class-1 RF from the A site . RF3 is activated for GTP hydrolysis and rapidly leaves the ribosome in the GDP form .
The ribosome ( E . coli , MRE600 ) , XR-7 mRNAs ( Met-stop[UAA] and Met-Leu-Leu-stop[UAA] ) , fMet-tRNAfMet and the initiation and elongation factors were purified following lab protocols ( Antoun et al . , 2004; Huang et al . , 2010 ) . His-tagged RF1 ( kind gift of Valérie Heurgué-Hamard , CNRS , France ) and RF3 were purified to single band using His-trap ( GEHC ) affinity column chromatography . To make RF3 free from extra guanine nucleotides , it was further purified using DEAE-sepharose ion exchange chromatography and gel filtration , by passing through a long superdex-75 column ( 1 . 5 × 100 cm ) in a diluted solution . RF1 mutants ( H13A , E14A , E17A , E18A and Q20A ) were created following the standard protocol for QuikChange Mutagenesis ( Stratagene , Santa Clara , CA ) . The mutations were confirmed with DNA sequencing and the mutant proteins were purified with the same protocol as the wild-type . For recycling of RF1 , an RC containing tripeptidyl 3H-fMet-Leu-Leu tRNALeu in the P site and a UAA stop codon in the A site was prepared as described in ( Bouakaz et al . , 2006 ) with minor modifications . The recycling assay was performed with 50 nM RC and 2 nM RF1 ( wild-type or mutants ) without or with RF3 ( 0 . 5 μM ) in excess following . A ribosomal release complex was first made by incubating E . coli MRE600 ribosomes ( 0 . 5 µM ) , fMet-tRNAfMet ( 1 µM ) and Met-stop ( UAA ) mRNA ( 1 µM ) in 1xHEPES-polymix buffer ( Gao et al . , 2007 ) at 37°C for 10 min . To this , RF1 ( 2 µM ) and RF3 ( 6 µM ) were added and the reaction mix was incubated at 37°C for another 10 min to form the release complex ( RC ) . Occupancy of RF1 and RF3 was checked by running the RC in SDS-PAGE after ultracentrifugation on 30% sucrose cushion ( 20 mM Tris-HCl , 500 mM NH4Cl , 10 . 5 mM MgAcetate , 0 . 5 mM EDTA , 1 . 1 M Sucrose , 3 mM 2-Mercaptoethanol ) . Both RF1 and RF3 produced bands with comparable intensity as the ribosomal protein S1 showing 1:1 incorporation in the release complex ( Figure 1 ) . To study the effect of GDP , similar complexes were formed with increasing concentration of GDP and GTP . Band intensities were digitalized and quantified by the Software UN-SCAN-IT gel TM Version 6 . 1 ( Silk Scientific , Orem , UT ) . To prepare the sample for cryo-EM , the assembled termination complex was diluted in 1xHEPES-polymix buffer to a final concentration of 50 nM . A carbon-coated Quantifoil 2/4 grid ( Quantifoil Micro Tools GmbH , Jena , Germany ) was prepared following standard cryo procedures ( Grassucci et al . , 2007; Sengupta et al . , 2010 ) : Grids were glow-discharged for 25 s in an ( H2 , O2 ) -atmosphere; 15W and 25s using a plasma cleaner ( Gatan Solarus Model 950 Advanced Plasma System; Gatan , Inc . , Warrendale , PA ) to make them hydrophilic . Samples of 4 µl were pipetted onto carbon-coated grids . Grids were blotted in 100% humidity at 4°C for 5 s and plunge-frozen into liquid ethane cooled by liquid nitrogen using a Vitrobot ( FEI , Hillsboro , OR ) . Images were recorded on SO163 film ( Eastman Kodak , Rochester , NY ) using an FEI ( Eindhoven , The Netherlands ) Tecnai Polara operating at 300 kV and a nominal magnification of 59 , 000× under low-dose conditions ( ∼20 e−/Å2 ) by using the automated data collection system AutoEMation ( Lei and Frank , 2005 ) . The 798 film micrographs collected were digitized with a step size of 7 µm on an Imaging Scanner ( Carl Zeiss , Inc . , Jena , Germany ) . The resulting pixel size was 1 . 2 Å on the object scale . After visual inspection and evaluation of the micrographs and their power spectra , 643 were selected for further analysis . Particles were chosen via automated particle selection ( Rath and Frank , 2004 ) followed by manual verification . The total number of particles used was ∼86 , 000 . We employed standard SPIDER scripts in a hierarchical reference-based classification strategy . Initially , the total data set was aligned to a density map of the 70S ribosome in complex with a P-site and an E-site tRNA ( 70S-P , E ) filtered to 20 Å and refined on a course angular grid over several rounds of refinement according to standard SPIDER protocols ( Frank et al . , 1996; Shaikh et al . , 2008 ) which include a 3D projection alignment procedure with correction of the contrast transfer function ( CTF ) using 82 defocus groups , with defocus varying from −1 . 2 to −4 . 6 µm . Next , the projection images were classified on the basis of cross-correlation coefficients with two references , namely ( 70S-P , E ) and an RF1-containing ribosomal complex having P- and E-site tRNAs ( EMDB ID 1184; Rawat et al . , 2006 ) . Two classes were identified ( 70S-P , E: ∼14 , 000 projection images , and EMDB ID 1184: ∼72 , 000 projection images ) . The initial orientation assignments to the ( 70S-P , E ) map were chosen as starting point for refinement of each class . The RF1-containing class was then classified further with the RF1-containing map ( EMDB ID 1184 ) as first reference and a second reference ( ‘RF1/RF3-chimera’ ) generated by adding the segmented RF3 density from the RF3•GDPNP map ( EMDB ID 1302; Gao et al . , 2007 ) to the RF1-containing map ( EMDB ID 1184 ) . Two subclasses of projection images ( EMDB ID 1184: ∼43 , 000 projection images , and RF1/RF3-chimera: ∼29 , 000 projection images ) were identified and refined separately as described above . Following convergence of standard refinement , a second refinement routine was performed in which CTF correction was performed at the 2D projection-level . CTF-corrected projection images were then pooled together and a 3D reconstruction was obtained . Only a subset of the best-aligned particles ( based on cross-correlation with the reference in each view ) was used for reconstruction until the very last rounds of small-angle refinement . The final resolutions for the maps were 8 . 0 Å for the total data set ( Figure 2—figure supplement 1 ) , 11 . 8 Å ( RC; Figure 2—figure supplement 2 ) , 8 . 4 Å ( RC-RF1 ) and 9 . 7 Å ( RC-RF1•RF3 ) , using a Fourier shell correlation ( FSC ) cutoff of 0 . 5 ( Figure2—figure supplement 4 ) . Models of E . coli RF1 and RF3 were built based on their crystal structures ( Graille et al . , 2005; Gao et al . , 2007 ) ( PDB ID: 2B3T and 2H5E , respectively ) . RF1 was modeled in two conformations fitting the cryo-EM maps of RC-RF1 and RC-RF1•RF3 . Missing and unstructured segments , in the crystal structure , as well as segments interacting with the mRNA , and the P-site tRNA , were modeled mostly by homology and in part manually . Residues 70–105 , 227–241 and 285–306 were modeled by homology to the T . thermophilus RF1 crystal structure in complex with the 70S ribosome ( Laurberg et al . , 2008 ) ( PDBID: 3D5A ) , using SWISS-MODEL ( Schwede et al . , 2003; Arnold et al . , 2006; Bordoli et al . , 2009 ) . Residues 57–69 were modeled manually using PyMOL ( The Pymol Molecular graphics system , Version 1 . 3 , Schrödinger , LLC ) to position them in the main chain and then Swiss-Pdb Viewer ( Guex and Peitsch , 1997 ) to modify their torsions in order to fit the density map in a region between helices 2 and 3 that obviously presents different structure from the T . thermophilus RF1 crystal structure . RF3 was modeled interacting with RF1 in the E . coli 70S ribosome by modifying the conformation of RF3 first to roughly fit the corresponding cryo-EM density map . The fit was realized using UCSF Chimera ( Pettersen et al . , 2004 ) by fitting domains I , II and III of the crystal structure as three independent rigid bodies into the RF3 density in the RC-RF1•RF3 map . Subsequently , Swiss-Pdb Viewer was used to accommodate the coils linking different fitted domains . Residues 37–67 , missing from the crystal structure , were modeled de novo in SWISS-MODEL using SYMPRED for secondary structure prediction . Using VMD ( Humphrey et al . , 1996 ) , three systems were prepared for MDFF . One system consisted of RF1 , E . coli ribosomal proteins L11 , L16 and S12 , P-site tRNA , mRNA , the rRNA surrounding scaffold ( 16S rRNA residues 10–25 , 502–542 , 787–797 , 885–928 , 950–1070 , 1190–1232 , 1390–1421 , and 1479–1525 and 23S rRNA residues 1052–1108 , 1903–1969 , 2051–2067 , and 2445–2615 ) and surrounding magnesium cations . The coordinates for the system , except for RF1 , P-site tRNA and mRNA , were taken from the E . coli 70S crystal structure by Berk et al . ( 2006 ) ( PDBID: 2I2P and 2I2T ) . P-site tRNA and mRNA coordinates were taken from the T . thermophilus crystal structure of the 70S ribosome in complex with RF1 ( Laurberg et al . , 2008 ) ( PDBID: 3D5A ) . A second system was created similar to the system mentioned above , the only difference being the addition of RF3 and the CTD of L7/L12 ( Leijonmarck and Liljas , 1987; PDB ID 1CTF ) . The third system consisted of RF3 , the CTD of L7/L12 , E . coli ribosomal proteins L11 and S12 , as well as rRNA segments scaffolding these two proteins ( 16S residues 10–25 , 502–542 , 885–892 , 907–927 , 1390–1418 and 1482–1508 and 23S residues 1052–1108 ) ( Berk et al . , 2006 ) . All systems were embedded in a box of TIP3P water molecules with an extra 12 Å padding in each direction . The system was neutralized by potassium ions and an excess of KCl was added to ∼ 0 . 2 M . All the simulated systems were prepared using CHARMM force field parameters ( Combined CHARMM All-Hydrogen Topology File for CHARMM22 Proteins and CHARMM27 Lipids [MacKerell et al . , 1998] ) . Each system was relaxed by 8000 minimization steps in NAMD ( Phillips et al . , 2005 ) . After relaxation , an additional potential was added to the systems based on cryo-EM segments comprising only the molecules to be simulated , and MDFF simulations were performed . Cryo-EM segments were prepared using the SeggerR plug-in in UCSF Chimera ( Pintilie et al . , 2010 ) . Simulations of RF1- and RF1•RF3-containing complexes were run for 600 ps , whereas the simulation of the RF3-containing complex was terminated after 1 ns . The RF1/RF3 interface in the simulations of RF1•RF3-containing complexes was fine-tuned manually , using PyMOL , in accordance to the outcome of the mutagenesis assays performed . Angular changes between RF3-conformations were estimated based on alignment of amino acid residues 97–105 , 121–133 and 328–343 . Poisson-Boltzmann electrostatic properties for RF1 , RF2 and RF3 models were calculated with the Adaptive Poisson-Boltzman Solver ( Baker et al . , 2001; Dolinsky et al . , 2004 ) . Potentials range from −15 . 91V to 12 . 62V and the results were displayed in VMD using an RWB color scale ranging from –14 to 14 . For the model presented in Figure 11 , we merged density maps of RC-RF1 , RC-RF1•RF3 and RC-RF3 with simulated density maps of L12 . The L12 maps were rendered at a resolution of 11 Å using the NMR structure of isolated L12 ( PDB ID 1RQU ) and the threshold for display was chosen with emphasis on linker visibility . An animation script was generated and ‘Video 1’ was rendered in UCSF Chimera . | Ribosomes are complex molecular machines that join amino acids together to form proteins . The order of amino acids in the protein is specified by a strand of messenger RNA ( mRNA ) , and the process of decoding the mRNA into a string of amino acids is called translation . A ribosome consists of two subunits—one large , one small—that come together at a particular site on the mRNA strand called the translation initiation site . The ribosome then moves along the mRNA—joining together amino acids brought to it by transfer RNA ( tRNA ) —until it reaches a termination site and releases the protein . The ribosome has three sites; the first amino acid to be delivered by a tRNA molecule to the ribosome occupies the site in the middle—also called the P site—and the second amino acid is delivered to the A site . Once the first two amino acids have been joined together , the ribosome moves along the mRNA so that the first amino acid now occupies the third site , called the E or exit site , and the second amino acid occupies the P site , leaving the A site vacant . The third amino acid is then delivered to the A site , and the whole process repeats itself until the ribosome reaches the termination site . Proteins called release factors are responsible for terminating the translation process and releasing the translated string of amino acids , which folds to form a protein . In bacteria this task can by performed by two releases factors , known as RF1 and RF2 . However , the release factor must itself be released to leave the ribosome free to translate another strand of mRNA . Pallesen et al . have used cryo-electron microscopy ( cryo-EM ) to study how a third release factor , RF3 , helps to release RF1 from the ribosome in bacteria . In cells , RF3 usually forms a complex with a molecule called GDP , and the cryo-EM studies show that this molecule is released shortly after the RF3•GDP complex enters the ribosome . Once inside the ribosome , RF3 comes into contact with RF1 and with a protein called L12 that is part of the ribosome . A molecule called GTP—which is well known as a source of energy within cells—then binds to RF3 , and this causes the shape of the ribosome to change . This change of shape results in the release of RF1 and the formation of a new RF3•GDP complex , which then leaves the ribosome . Further work is needed to fully understand the role of L12 in these events , but a detailed understanding of the mechanism for terminating the translation of mRNA by the ribosome is coming into view . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"biochemistry",
"and",
"chemical",
"biology",
"structural",
"biology",
"and",
"molecular",
"biophysics"
] | 2013 | Cryo-EM visualization of the ribosome in termination complex with apo-RF3 and RF1 |
While beige adipocytes have been found to associate with dense sympathetic neurites in mouse inguinal subcutaneous white fat ( iWAT ) , little is known about when and how this patterning is established . Here , we applied whole-tissue imaging to examine the development of sympathetic innervation in iWAT . We found that parenchymal neurites actively grow between postnatal day 6 ( P6 ) and P28 , overlapping with early postnatal beige adipogenesis . Constitutive deletion of Prdm16 in adipocytes led to a significant reduction in early postnatal beige adipocytes and sympathetic density within this window . Using an inducible , adipocyte-specific Prdm16 knockout model , we found that Prdm16 is required for guiding sympathetic growth during early development . Deleting Prdm16 in adult animals , however , did not affect sympathetic structure in iWAT . Together , these findings highlight that beige adipocyte-sympathetic neurite communication is crucial to establish sympathetic structure during the early postnatal period but may be dispensable for its maintenance in mature animals .
The development of adipose tissue was a critical adaptation for our ancestors . White adipose tissue enables the safe storage and rapid mobilization of energy in response to nutritional needs , while brown adipose tissue defends body temperature by dissipating energy as heat ( Rosen and Spiegelman , 2014 ) . In modern times , however , excess high-calorie foods , a sedentary lifestyle , and thermal comfort have contributed to an overexpansion of white fat and a relative paucity of brown fat ( Heymsfield and Wadden , 2017; van Marken Lichtenbelt et al . , 2018 ) . This has resulted in a significant increase in the prevalence of obesity and associated diseases including type 2 diabetes , hypertension , cardiovascular disease , and many types of cancer ( Kopelman , 2000 ) . Obesity now affects over 40% of adults in the United States and over 600 million adults worldwide ( CDC , 2020a; The GBD 2015 Obesity Collaborators , 2017 ) . Excess adiposity is at the center of the leading causes of morbidity and mortality , and obesity-related medical care costs the United States health care system nearly $150 billion each year ( CDC , 2020b ) . Addressing this public health emergency will therefore require new approaches based on a deeper understanding of the tissues and pathways involved in energy homeostasis . The crucial role of adipose tissue in energy balance has driven great interest in investigating this tissue as a target for the treatment of obesity . While white adipocytes store excess energy , thermogenic brown and beige adipocytes convert lipids and glucose into heat , thereby increasing energy expenditure ( Rosen and Spiegelman , 2014 ) . Unlike classical brown adipocytes which are thermogenic in basal conditions , murine beige adipocytes , which resemble human brown adipocytes in their molecular signature ( Shinoda et al . , 2015 ) , reside in white adipose depots and need to be activated by external stimuli such as the sympathetic nervous system to drive thermogenesis ( Rosen and Spiegelman , 2014; Wang and Seale , 2016 ) . Recent studies have shown that activation of thermogenic adipocytes in both rodents and humans is associated with increased whole-body energy expenditure , improved glucose homeostasis , and enhanced insulin sensitivity ( Becher et al . , 2021; Cypess et al . , 2009; Lee et al . , 2014; Seale et al . , 2011; van Marken Lichtenbelt et al . , 2009; Virtanen et al . , 2009 ) , suggesting a new approach to defend against obesity . The sympathetic nervous system plays a key role in enhancing thermogenic function of brown and beige adipocytes . Although located in distinct fat depots , both brown and beige adipocytes are surrounded by dense sympathetic neurites , termed parenchymal innervation ( Blaszkiewicz et al . , 2019; Chi et al . , 2018b; Guilherme et al . , 2019; Jiang et al . , 2017; Wirsen , 1964 ) . Norepinephrine , a neurotransmitter released by these parenchymal neurites , activates β-adrenergic signaling in thermogenic adipocytes , resulting in enhanced thermogenesis and lipolysis ( Cannon and Nedergaard , 2004; Hsieh and Carlson , 1957 ) . The important role of sympathetic stimulation in thermogenesis has driven great interest in understanding the structural and molecular details of sympathetic control of thermogenic adipocytes . Adipocyte-derived factors have been shown to act on the sympathetic nervous system to regulate its structure and activity . Recent studies have identified S100B and TGFβ1 in brown adipocytes as important molecular determinants of sympathetic innervation in brown fat ( Hu et al . , 2020; Zeng et al . , 2019 ) . However , it remains largely unclear how beige adipocytes , which are embedded in white fat depots , modulate their sympathetic innervation . Assisted by a whole-adipose immunolabeling and clearing method , called Adipo-Clear , we recently found that the density of sympathetic parenchymal neurites in close apposition to beige adipocytes is dependent on PRDM16 ( PR domain containing 16 ) , the transcriptional determinant of beige adipocyte identity and function ( Chi et al . , 2018b ) . Specifically , deletion of Prdm16 in adipocytes led to ablation of beige adipocyte function and dramatically reduced parenchymal innervation density , suggesting that beige adipocyte-associated factors regulate the structure of sympathetic innervation . As neural projections and circuits can be regulated during development and by physiological stimuli in adult animals ( Glebova and Ginty , 2005; Holtmaat and Svoboda , 2009 ) , it is important to determine when the sympathetic innervation surrounding beige adipocytes is established . Using 3D whole-tissue imaging , we have begun to decipher the timing of the interactions between sympathetic neurites and beige adipocytes in mouse inguinal subcutaneous white fat ( iWAT ) . We found that sympathetic parenchymal innervation in iWAT actively grows during the early postnatal period . Interestingly , we observed that the establishment of dense parenchymal innervation closely follows the development of early postnatal beige adipocytes . To our surprise , using an inducible , adipocyte-specific Prdm16 knockout mouse model , we found that Prdm16 in beige adipocytes is required for sympathetic axon growth during early development , but not necessary for maintaining sympathetic structure in adulthood .
To better understand adipocyte-sympathetic neurite interactions , we investigated whether the association between beige adipocytes and dense sympathetic innervation is developmentally determined . We first mapped the developmental timing of the sympathetic nervous system in iWAT using Adipo-Clear coupled with light sheet fluorescent imaging . Given that iWAT undergoes active tissue morphogenesis during late embryonic and early postnatal stages ( Wang et al . , 2013 ) , we first performed whole tissue immunostaining and imaging in iWAT isolated from postnatal day ( P ) two mice using an antibody targeting tyrosine hydroxylase ( TH ) , a maker for sympathetic fibers , which acts as the rate-limiting enzyme in the catecholamine biosynthesis pathway . At this stage , adipocytes appeared fully vascularized and organized into distinct lobular structures , as shown by the endothelial cell marker PECAM ( also known as CD31 ) ( Figure 1—figure supplement 1A , C and F ) , consistent with previous reports ( Hong et al . , 2015 ) . While we could detect TH-positive ( TH+ ) signals resembling nerve fascicles as well as fibers wrapping around large blood vessels , dense parenchymal innervation in close apposition to adipocytes , which was reported in adult iWAT ( Chi et al . , 2018b; Jiang et al . , 2017 ) , was not obvious at this age ( Figure 1—figure supplement 1A–F ) . At P6 , more distinct features of sympathetic innervation in iWAT were observed: ( a ) travelling in parallel within nerve fascicles and ( b ) wrapping around main blood vessels in a dense mesh-like morphology ( Figure 1A–C , Figure 1—figure supplement 2A–D ) . Upon further analyzing the innervation pattern across the entire tissue , we observed that these structures were all interconnected to form a continuous sympathetic network . Specifically , we found several convergence points where TH+ nerve fibers within a nerve fascicle deviated from the bundle and merged with the innervation of the central blood vessel ( Figure 1B–C , Figure 1—figure supplement 2A and C , Figure 1—video 1 ) , suggesting that sympathetic fibers leave the nerve fascicle and wrap around the main blood vessel as the first order of innervation . Subsequently , the main blood vessel innervation extended around branching arterioles and venules as the second order of innervation ( Figure 1D , Figure 1—figure supplement 2E–F , Figure 1—video 1 ) . Lastly , discrete nerve fibers became apparent at the terminals of the second-order innervation to project into tissue parenchyma , where adipocytes are located . Notably , the majority of these nerve fibers appeared to follow capillaries to arrive in the tissue parenchyma ( Figure 1E–F , Figure 1—figure supplement 2G–J ) . Although nerve endings were visible in the tissue parenchyma at P6 , we did not observe any extensive innervation surrounding adipocytes . In addition , both the dorsolumbar and inguinal regions of iWAT showed similar innervation patterns at this stage ( Figure 1—figure supplement 3A–B , Figure 1—video 2 ) . The results from P2 and P6 samples indicate that the sympathetic axons in iWAT first grow along the vasculature before reaching the tissue parenchyma , consistent with previous findings showing that developing sympathetic axons follow the vasculature to reach their target organs ( Glebova and Ginty , 2005 ) . Remarkably , adipocyte-innervating neurites became apparent four days later . In the inguinal region , dense parenchymal neurites surrounding adipocytes were first found at P10 , in particular within lobules at the core of this region ( Figure 1G , Figure 1—figure supplement 2K , Figure 1—figure supplement 3D , Figure 1—video 2 ) . At P14 , the number of lobules that contain dense parenchymal neurites dramatically increased , spreading outwards from the core of the inguinal region ( Figure 1H , Figure 1—figure supplement 3F , Figure 1—video 2 ) . From P21 and onwards , more inguinal lobules were found to harbor dense parenchymal innervation ( Figure 1I–J , Figure 1—figure supplement 3H & J , Figure 1—video 2 ) , with the pattern comparable to that of adult iWAT ( Chi et al . , 2018b ) . Interestingly , the emergence of dense parenchymal neurites in the dorsomedial region lagged behind . While parenchymal neurites were detectable in the dorsomedial region at P14 and P21 ( Figure 1H–I , Figure 1—figure supplement 2K–L , Figure 1—figure supplement 3E & G , Figure 1—video 2 ) , we did not observe densely innervated lobules that resemble the adult innervation pattern in this region until P28 ( Figure 1J , Figure 1—figure supplement 3I , Figure 1—video 2 ) . Notably , the dorsolateral region of iWAT remained sparsely innervated relative to the inguinal region and the dorsomedial region throughout the early postnatal period ( Figure 1G–J , Figure 1—figure supplement 2K–L , Figure 1—figure supplement 3A , C , E , G & I , Figure 1—video 2 ) . As our previous findings suggest that beige adipocytes interact with sympathetic projections and modulate the density of sympathetic parenchymal innervation ( Chi et al . , 2018b ) , we next investigated whether early postnatal development of sympathetic innervation may also be regulated by beige adipocytes . We analyzed the localization of beige adipocytes using an antibody against uncoupling protein 1 ( UCP1 ) , a widely accepted marker for thermogenic adipocytes , and compared their distribution in relation to the sympathetic parenchymal innervation in iWAT using whole-tissue imaging . As expected , we observed a strong association between beige adipocytes and parenchymal innervation , even during early postnatal development . Specifically , we found that beige adipocytes first emerge in iWAT of P6 animals that were born and housed at room temperature , as shown by a few UCP1+ adipocytes sparsely distributed in the core of the inguinal region , close to the inguinal lymph node ( Figure 2A–B , Figure 2—figure supplement 1D ) . At P10 , we detected clusters of UCP1+ adipocytes located in distinct lobules in the core of the inguinal region ( Figure 2C–D , Figure 2—figure supplement 1E ) . Four days later , at P14 , the lobules containing UCP1+ adipocytes further expanded from the core ( Figure 2E & G , Figure 2—figure supplement 1F ) . At P21 and P28 , extensive UCP1+ lobules occupied a significant portion of the inguinal region , comparable to the extent of UCP1+ cells only seen in adult animals after cold exposure ( Figure 2H and J , Figure 2—figure supplement 1A , C , G , & H ) . On the other hand , the emergence of UCP1+ adipocytes in the dorsolumbar region again lagged behind . UCP1+ adipocytes in the dorsomedial region first emerged in small clusters at P14 and then as distinct lobules at P21 ( Figure 2E–F & and H–I ) . At P28 , the same region contained a large number of lobules harboring UCP1+ adipocytes ( Figure 2—figure supplement 1A–B ) . Interestingly , the dorsolateral region was devoid of UCP1+ adipocytes at all stages analyzed ( Figure 2A , C , E and H , Figure 2—figure supplement 1A & D–H ) . To obtain a quantitative measure of these early postnatal beige adipocytes , we also examined mRNA levels of brown and beige adipocyte-enriched genes in the inguinal and dorsolumbar regions of iWAT . In line with the imaging results , Ucp1 mRNA expression showed a gradual increase from P6 to P14 in the inguinal region ( Figure 2—figure supplement 2A ) . To our surprise , although we observed more extensive UCP1+ adipocytes in the inguinal region at P21 and P28 by imaging , Ucp1 mRNA levels peaked around P12-P16 , with the expression level at P14 being fourfold and eightfold higher than that of P21 and P28 , respectively ( Figure 2—figure supplement 2A ) . Other thermogenic genes ( Cidea and Cox8b ) also gradually increased their mRNA expression from P6 to P14 , followed by a small downward trend after P21 ( Figure 2—figure supplement 2A ) . These quantitative transcriptional data suggest that the beige adipocytes arising in the inguinal region of iWAT during early postnatal development may exhibit peak thermogenic potential around 2–3 weeks of age and gradually become less active as animals mature . On the other hand , all thermogenic genes showed significantly lower mRNA expression in the dorsolumbar region compared to the inguinal region at most time points ( Figure 2—figure supplement 2A ) . Of note , Prdm16 , the transcriptional coregulator that determines beige adipocyte phenotype , showed a consistent 1 . 5- to twofold increase in mRNA levels in the inguinal relative to the dorsolumbar region across all time points ( Figure 2—figure supplement 2B ) . Other transcriptional regulators of beige adipocyte development , such as Cebpb and Ppargc1a , showed similarly consistent regional differences ( around 1 . 5-fold for Cebpb and twofold for Ppargc1a ) during early development ( Figure 2—figure supplement 2B ) . Furthermore , we did not observe any significant regional differences in markers of adipocyte maturation and function ( Fabp4 , Pparg , and Adipoq ) ( Figure 2—figure supplement 2C ) . Taken together , these data suggest that adipocytes from the two regions of iWAT are equally differentiated , but the inguinal region may harbor more beige progenitor cells or mature adipocytes with the potential to emerge as beige adipocytes . When we overlaid the UCP1 and TH signals , we observed a dramatic overlap between the presence of dense parenchymal innervation and beige adipocytes , particularly from P10 onwards ( Figure 2—figure supplement 3A–J , Figure 1—video 2 ) , strongly suggesting that early postnatal beige adipocytes are associated with the signals enabling sympathetic axon growth . Additionally , as room temperature is considered a mild cold stress to mice , particularly in developing animals that do not have their adult fur pattern , it is possible that early postnatal beige adipocytes arise as a result of cold-induced sympathetic stimulation . When mice were born and raised at a warmer temperature ( 30°C ) , at which cold-induced sympathetic firing is minimized , early postnatal beige adipocytes and sympathetic neurites arise in iWAT with the same patterning as that of room temperature-housed mice ( Figure 2—figure supplement 3K–L ) . In addition , a recent study using genetic sympathetic ablation showed that early postnatal beige adipocytes develop normally in the absence of sympathetic innervation ( Wu et al . , 2020b ) . Together , these data suggest that the development of beige adipocytes is likely not dependent on sympathetic activation , but rather based on a developmentally hard-wired program . We have previously shown that dense parenchymal innervation that localizes to the inguinal region of adult iWAT is significantly reduced by constitutive deletion of Prdm16 in adipocytes ( Chi et al . , 2018b ) . To examine whether early postnatal beige adipocytes and their regulation of dense sympathetic innervation are also dependent on Prdm16 , we analyzed iWAT of adipocyte-specific Prdm16 knockout mice ( Adipoq-Cre; Prdm16lox/lox; hereafter noted as constitutive Prdm16KO or cPrdm16KO mice ) at postnatal days 6 , 14 , and 21 , key time points in the course of beige adipocyte and sympathetic innervation development . At P6 , we observed minimal beige adipocytes and scant parenchymal innervation in both control and cPrdm16KO mice ( Figure 3—figure supplement 1A–H ) , suggesting that the sympathetic nervous system develops similarly in both models prior to the emergence of beige adipocyte clusters . At P14 , the deletion of Prdm16 completely ablated beige adipocytes that normally arise in the inguinal region of control mice , both at the mRNA and protein levels ( Figure 3I , Figure 3—figure supplement 1I , Figure 3—figure supplement 2A–D ) . Correspondingly , the increase seen in parenchymal innervation density in the inguinal region of control mice was not observed in cPrdm16KO mice ( Figure 3A–D , Figure 3—figure supplement 2A–D ) . At P21 , we observed similar ablation of beige adipocytes and lack of growth in parenchymal innervation in cPrdm16KO relative to control samples ( Figure 3E–I , Figure 3—figure supplement 1I , Figure 3—figure supplement 2E–H ) . These results indicate that early postnatal beige adipocytes indeed depend on PRDM16 , the well-characterized transcriptional determinant of brown and beige adipocytes . Importantly , these data strongly suggest that sympathetic axon growth during early iWAT morphogenesis may be regulated by PRDM16-dependent signals . Although PRDM16 is known to be important for beige adipocyte function , it remains possible that ablation of Prdm16 in all adipocytes by Adipoq-Cre altered white adipocyte function and therefore affected sympathetic innervation . To address this , we assessed Prdm16 mRNA and protein levels in the inguinal and dorsolateral regions of iWAT , which are predominantly beige and white regions , respectively ( Figure 2 ) . We performed qPCR on the two regions isolated from control and cPrdm16KO mice at P14 ( Figure 3—figure supplement 2I ) . The control dorsolateral region showed significantly lower expression level of Prdm16 mRNA than the control inguinal region . Importantly , the Prdm16 mRNA level in the control dorsolateral region was indistinguishable from that in cPrdm16KO dorsolateral or inguinal regions , suggesting that the wild-type dorsolateral region naturally expresses very low levels of Prdm16 mRNA with levels indistinguishable from Prdm16 knockout samples . We further assessed PRDM16 protein levels across multiple fat depots of young adult mice ( Figure 3—figure supplement 2J ) . Consistently , the dorsolateral region exhibited a considerably lower level of PRDM16 compared with the inguinal region in wild-type iWAT , while there were no detectable levels of PRDM16 in the iWAT of cPrdm16KO mice or wild-type eWAT . Although there was still a minimal level of PRDM16 protein in the dorsolateral region of iWAT , this may be attributed to the small number of beige adipocytes in this region . Altogether , Prdm16 appears to be minimally expressed in white adipocytes in iWAT , and thus its deletion in white adipocytes is likely to contribute minimally to the changes in sympathetic innervation . Interestingly , although PRDM16 also plays a critical role in brown adipocyte determination and function , deletion of Prdm16 in interscapular brown fat ( iBAT ) does not affect its development or thermogenic function in young adults ( Cohen et al . , 2014; Harms et al . , 2014 ) . Previous studies have shown that the role of PRDM16 in iBAT formation and function is compensated for by PRDM3 , a transcriptional regulator closely related to PRDM16 ( Harms et al . , 2014 ) . Consistent with these findings , we detected similarly extensive sympathetic parenchymal innervation in iBAT of both control and cPrdm16KO mice ( Figure 3—figure supplement 2K–L ) . To further delineate the critical time window for sympathetic innervation patterning in iWAT , we generated an inducible Prdm16 knockout mouse model ( AdipoqrtTA; TRE-Cre; Prdm16lox/lox; hereafter noted as inducible Prdm16KO or iPrdm16KO mice ) , where Prdm16 can be deleted in adipocytes in a doxycycline-dependent manner ( Figure 4A ) . To test whether sympathetic parenchymal innervation may be developmentally determined during a defined time window , doxycycline was delivered to mice from embryonic day ( E ) 14 until P21 , the period of time when both beige adipocytes and parenchymal innervation development become clearly apparent . Following doxycycline treatment , iPrdm16KO and littermate control mice were switched back to chow diet for 2 weeks and subsequently exposed to either room temperature ( RT ) or 8°C for 2 days ( Figure 4B ) . We first confirmed that perinatal doxycycline treatment led to a strong reduction in Prdm16 in adipose depots at the mRNA ( 70% reduction in the inguinal region of iWAT , 65% reduction in iBAT ) and protein levels ( Figure 4—figure supplement 1A–B ) . Both iPrdm16KO ( perinatal , 5 weeks ) and littermate control mice developed similarly , with no difference in body mass ( Figure 4—figure supplement 1C ) . To examine whether the perinatal deletion of Prdm16 led to changes in downstream thermogenic gene expression , we analyzed mRNA levels of thermogenic markers that are induced by cold in wild-type iWAT . Since we have consistently observed that the inguinal region of iWAT is enriched in both early postnatal and cold-inducible beige adipocytes ( Chi et al . , 2018b ) , we hereafter focused on analyzing changes in the inguinal region . As expected , the control inguinal regions exhibited robust increases in the entire panel of thermogenic markers following 2 days of cold exposure ( e . g . 35-fold induction in Ucp1 ) ( Figure 4C , Figure 4—figure supplement 1D ) . In contrast , the cold-induced increases were almost completely blocked following perinatal deletion of Prdm16 ( Figure 4C , Figure 4—figure supplement 1D ) , mirroring our previous findings in constitutive Prdm16KO mice ( Cohen et al . , 2014 ) . On the other hand , the transient Prdm16 deletion did not significantly alter adipocyte differentiation or function , as shown by Pparg , Fabp4 , and Adipoq mRNA levels ( Figure 4—figure supplement 1E ) . We next performed whole-tissue imaging on cleared iWAT of control and iPrdm16KO to assess whether perinatal Prdm16 deletion is sufficient to reproduce the innervation defect seen in cPrdm16KO . While the gross features of sympathetic innervation such as nerve fascicles and blood vessel innervation were preserved in perinatal iPrdm16KO , the parenchymal innervation that localized to the inguinal region of control iWAT was dramatically reduced in perinatal iPrdm16KO ( Figure 4D–I , Figure 4—video 1 ) . To quantitatively assess differences in parenchymal neurite density from the 3D images , we computationally traced and measured the parenchymal neurite lengths in randomly selected tissue volumes contained within lobules in the inguinal region of iWAT ( Figure 4—figure supplement 2A–D ) . When neurite lengths were normalized to volumes of the isolated cuboids , perinatal iPrdm16KO samples showed a 74 . 5% reduction in parenchymal neurite density compared to control samples ( Figure 4J , Figure 4—figure supplement 2F ) . Interestingly , we noticed that the adipocytes in perinatal iPrdm16KO samples also appeared larger in size as outlined by the tissue autofluorescence and vasculature signal ( Figure 4—figure supplement 1F–K ) . It is possible that the decreased neurite density in iPrdm16KO was due to a simple scaling effect; that is , the neurites appear more sparse because the adipocytes are larger in size . To rule out this possibility , we also calculated neurite density by factoring in adipocyte density from each tissue volume , and still observed a significant 52 . 8% decrease in density following perinatal Prdm16 deletion ( Figure 4K , Figure 4—figure supplement 2E–F ) . Similar to our previous findings with the cPrdm16KO model , perinatal Prdm16 knockout in adipocytes did not significantly affect thermogenic or adipogenic markers in iBAT ( Figure 4—figure supplement 3A–B ) . In line with this , the sympathetic parenchymal innervation levels did not appear different between control and perinatal iPrdm16KO iBAT samples ( Figure 4—figure supplement 3C–D ) . To assess whether the innervation defect following perinatal deletion of Prdm16 may be rescued as animals age , we performed the same perinatal deletion of Prdm16 but let animals reach 8 weeks of age on a chow diet , hereafter referred to as iPrdm16KO ( perinatal , 8 weeks ) ( Figure 4—figure supplement 4A ) . As iWAT continues to expand in young adult mice , we reasoned that newly developed adipocytes during iWAT expansion might also affect sympathetic innervation . Consistent with the results from the iPrdm16KO ( perinatal , 5 weeks ) experiment , we observed significantly reduced expression of Prdm16 and other thermogenic genes at the mRNA level in iPrdm16KO ( perinatal , 8 weeks ) relative to control ( Figure 4—figure supplement 4B–D ) . Importantly , TH+ fibers still appeared substantially sparser in iPrdm16KO ( perinatal , 8 weeks ) than control ( Figure 4—figure supplement 4E–J ) . Taken together , these results demonstrate that interactions between beige adipocytes and sympathetic nerve endings during an early critical developmental window are required for establishment of the sympathetic network in iWAT , as perturbations during this window lead to lasting effects on parenchymal innervation density . Next , we assessed whether PRDM16 is also important for maintaining sympathetic parenchymal innervation in adult animals . To that end , we started doxycycline treatment when mice were 8 weeks of age . Following 4 weeks of doxycycline treatment , iPrdm16KO ( adult deletion ) and littermate control mice were placed at either RT or 8°C for 2 days to allow subsequent analysis of the thermogenic gene program ( Figure 5A ) . As expected , doxycycline treatment in adult mice led to a robust reduction in Prdm16 at the mRNA ( 83% in the inguinal region of iWAT at RT ) and protein levels ( Figure 5B , Figure 5—figure supplement 1A ) . No body mass difference was observed between iPrdm16KO ( adult deletion ) and control mice ( Figure 5—figure supplement 1B ) . We also observed significant attenuation of cold-induced upregulation in all thermogenic markers following Prdm16 deletion ( Figure 5B , Figure 5—figure supplement 1C ) . Consistent with our previous findings , Prdm16 deletion did not significantly alter markers of adipocyte differentiation ( Figure 5—figure supplement 1D ) . Interestingly , we found dense sympathetic parenchymal innervation present in the inguinal regions of iWAT from both iPrdm16KO ( adult deletion ) and control samples ( Figure 5C–H , Figure 4—video 1 ) . Quantitative analysis of the parenchymal neurite density ( normalized by volume ) from the 3D images resulted in a small , insignificant decrease ( 17 . 6% ) in neurite density in adult iPrdm16KO relative to control samples ( Figure 5I ) . Furthermore , as the adipocytes appeared larger in adult iPrdm16KO samples ( Figure 5—figure supplement 1E–J ) , we normalized neurite length by adipocyte density . As a result , we observed a reverse relationship with the neurite density in adult iPrdm16KO being slightly higher ( 1 . 23-fold ) than that of controls ( Figure 5J ) . These results suggest that the small density difference when normalized by volume is likely driven by a scaling effect rather than actual neurite remodeling . Importantly , these data indicate that deleting Prdm16 in adipocytes of adult animals causes minimal changes in sympathetic parenchymal neurite density , strongly supporting the early postnatal period being the critical window during which interactions between beige adipocytes and sympathetic nerve endings occur .
Sympathetic innervation plays an important role in regulating two key aspects of adipose tissue function: lipolysis and thermogenesis . Dense sympathetic parenchymal innervation is observed in both iBAT and the inguinal region of iWAT , where beige adipocytes are primarily located; however , sparse innervation is found in eWAT and the dorsolateral region of iWAT , an area of the tissue that is devoid of beige adipocytes even under long term cold stimulation ( Chi et al . , 2018b; Dichamp et al . , 2019; Huesing et al . , 2020; Murano et al . , 2009; Zeng et al . , 2019 ) . The strong association between thermogenic adipocytes and dense sympathetic neurites as well as the association between white adipocytes and sparse innervation suggests that adipocyte type may determine the density of sympathetic parenchymal innervation . Indeed , when beige-to-white adipocyte identity change was achieved by adipocyte-specific deletion of Prdm16 , the density of sympathetic parenchymal innervation was significantly reduced in the inguinal region of iWAT ( Chi et al . , 2018b ) . Although many previous studies have attempted to define the relationships between sympathetic neurite density and beige adipocytes , when and how such relationships are established has remained unclear . Here , we delineated key stages of sympathetic nervous system development in iWAT using whole tissue imaging . Specifically , we observed that sympathetic parenchymal innervation in close apposition to adipocytes is established between P6 and P28 . Importantly , the appearance of UCP1+ beige adipocytes precedes the emergence of dense parenchymal neurites during early postnatal development . We further demonstrated that both early postnatal beige adipocytes and dense parenchymal neurites depend on Prdm16 expression in adipocytes . Using an inducible Prdm16 deletion model , we identified an early critical period during which beige adipocytes modulate sympathetic axon growth . However , Prdm16 deletion in adult mice did not alter the sympathetic structure . Assisted by whole-tissue images , our study carefully examined the growth of sympathetic axons in iWAT . We observed that sympathetic fibers travel in nerve fascicles to arrive at iWAT , and then depart from nerve fascicles to reach main blood vessels within iWAT . By P6 , the sympathetic innervation on blood vessels demonstrated a dense mesh-like structure , resembling that of mature iWAT . However , dense parenchymal innervation surrounding adipocytes , which several studies have characterized in mature iWAT ( Chi et al . , 2018b; Jiang et al . , 2017 ) , is not established at P6 . Instead , we observed sparse discrete sympathetic fibers and small vessels in congruence within the tissue parenchyma , a common phenomenon in sympathetic nerve fiber development where vasculature serves as a guide to direct developing fibers to reach their targets , such as the heart ( Nam et al . , 2013 ) . Subsequently , from P10 until P28 , dense parenchymal innervation becomes obvious where clusters of UCP1+ beige adipocytes are located . Although beige adipocytes emerge in white fat depots of adult mice following cold challenge , we observed strong beige adipogenesis during early postnatal development , similar to the ‘peri-weaning’ beige adipocytes reported in recent studies ( Wu et al . , 2020a; Wu et al . , 2020b ) . Using high-resolution whole tissue imaging focusing on the inguinal region of iWAT , we detected scattered UCP1+ adipocytes at P6 , subsequent emergence of small UCP1+ adipocyte clusters at P10 , and an expansion to almost all lobules from P14 to P28 . On the other hand , beige adipogenesis is delayed in the dorsolumbar region , where small clusters of beige adipocytes are not detected until P14 . It will be of interest to investigate the mechanisms driving the preferential localization of early postnatal beige adipocytes . Although beige adipocyte recruitment in adult mice requires sympathetic stimulation , early postnatal beige adipocytes develop normally in mice born and raised at thermoneutrality with minimal sympathetic activity as well as in mice with genetic sympathetic ablation ( Wu et al . , 2020b ) . Interestingly , we found several genes involved in the transcriptional control of beige adipocyte determination and function to have higher expression levels in the inguinal than the dorsolumbar region throughout early development , suggesting adipocytes or their progenitors in different regions may exhibit unique properties . Additionally , as early postnatal beige adipocytes first emerge close to the core of the inguinal region , other tissue structures such as blood or lymph vessels that are found in the same area ( Figure 1B–C and Figure 5—figure supplement 2A–E ) may play a role in promoting early beige adipogenesis . It is also worth noting that the whole-tissue UCP1+ patterns at P21 and P28 closely resemble that of adult mice following cold exposure . In addition , the thermogenic gene expression of early postnatal beige adipocytes diminishes as mice mature . It is possible that early postnatal beige adipocytes gradually become inactivated during maturation , and cold challenge re-activates the same cells in adult mice . Follow-up studies will be needed to examine the fate of early postnatal beige adipocytes . Of note , emerging studies have described additional beige adipocytes that rely on pathways other than UCP1 to dissipate heat ( Chen et al . , 2019; Ikeda et al . , 2017; Kazak et al . , 2015 ) . Thorough characterization of these beige adipocytes using specific markers will be needed to delineate their development and fate . Using constitutive and inducible adipocyte-specific Prdm16 deletion models , we ablated early postnatal beige adipocyte function and found that this dramatically reduced parenchymal sympathetic neurites in iWAT . Our findings suggest that early postnatal beige adipocytes may express PRDM16-dependent neurotrophic factors that stimulate sympathetic axon growth or downregulate inhibition cues . Recent studies have unveiled important roles of brown adipocyte-derived factors , such as S100B and TGFβ1 , in regulating sympathetic innervation in iBAT ( Hu et al . , 2020; Zeng et al . , 2019 ) . As beige and brown adipocytes share similarities , it is possible that these factors may also affect sympathetic innervation in iWAT . Interestingly , S100b , but not Tgfb1 , showed a regional pattern at the mRNA level during early iWAT development , with higher expression in the inguinal region in a PRDM16-dependent manner ( Figure 5—figure supplement 2F ) . This result suggests that S100B may be one of the potential cues in regulating iWAT sympathetic development . It is also likely that sympathetic neurites are indirectly regulated by additional beige adipocyte-associated cell types , such as immune or stromal cells . Follow-up studies will need to evaluate the role of S100B and other potential factors in modulating sympathetic axon growth during early iWAT morphogenesis . Our studies using an inducible adipocyte-specific Prdm16 deletion model indicated a critical developmental window for the interactions between beige adipocytes and sympathetic nerve terminals . Restricted Prdm16 deletion during early tissue morphogenesis resulted in a lasting reduction in parenchymal neurite density . However , Prdm16 deletion in fully mature mice failed to alter sympathetic neurite density . These results indicate that sympathetic neurites in iWAT respond to signals from beige adipocytes or associated cells during early development . However , when the innervation pattern is fully established , such signals are no longer required for maintaining the innervation level during adulthood . In line with this , we have previously shown that cold-induced beige adipocytes do not promote sympathetic neurite outgrowth in adult iWAT when equivalent tissue regions are compared . Taken together , our data suggest that sympathetic neurite density is regulated by local cues from beige adipocytes during a specific developmental window and exhibits limited plasticity once the pattern is established . Of note , a recent study demonstrated leptin-mediated central regulation of sympathetic innervation in adipose tissue ( Wang et al . , 2020 ) . Specifically , chronic leptin treatment acting on hypothalamic neurons was found to rescue the defect in sympathetic innervation in iBAT and iWAT of adult leptin-deficient mice . It is possible that the central and local regulation of sympathetic innervation in adipose tissue acts with different timing and through distinct mechanisms . Interestingly , the sympathetic axon growth period we observed ( P10-P21 ) largely overlaps with a postnatal leptin surge ( P8-P20 ) reported previously ( Ahima et al . , 1998; Wu et al . , 2020a ) . Future studies are needed to uncouple central ( leptin surge ) and local ( beige adipocyte-associated factors ) effects to fully understand how adipose sympathetic innervation is regulated by each mechanism . It is also worth noting that current studies rely on neurite morphological changes such as length to characterize sympathetic growth or remodeling . As adipocyte size dramatically changes in response to caloric excess or deprivation , sympathetic neurite density may appear different even without active remodeling . A better understanding of sympathetic neurite structural change will be assisted by identifying markers specific to actively remodeling neurites . Thermogenic adipocytes have been demonstrated to provide metabolic benefits that may combat obesity and associated metabolic diseases . As thermogenic adipocytes are primarily induced by sympathetic stimulation , many studies have turned to the sympathetic nervous system in search of novel therapeutic targets for enhancing thermogenic adipocyte function . Our studies here demonstrated a critical developmental window during which beige adipocytes regulate sympathetic neurite density , providing fundamental knowledge about the development of the sympathetic nervous system in mouse subcutaneous white fat and providing a framework for future attempts to target this pathway for therapeutic benefit .
Young wild-type mice of various ages were generated by crossing male and female mice from the C57BL/6J background ( C57BL/6J , JAX 000664 ) obtained from the Jackson Laboratories and maintained in our facilities . The constitutive Prdm16KO ( cPrdm16KO ) mice were generated as previously described ( Cohen et al . , 2014 ) by crossing Adipoq-Cre mice ( JAX 028020 ) with Prdm16lox/lox mice . The inducible Prdm16KO ( iPrdm16KO ) mice were generated by crossing Adipoq-rtTA ( provided by Dr . Philipp E . Scherer ) ( Sun et al . , 2012 ) , TRE-Cre ( B6 . Cg-Tg ( tetO-cre ) 1Jaw/J , JAX 006234 ) , and Prdm16lox/lox mice . All animals in this study were male mice on a pure C57BL/6J background . All mice were maintained on a 12 hr light/dark cycle with free access to food and water . To generate mice born and raised at thermoneutrality , pregnant female mice were housed at 30°C 14 days after vaginal plug formation until the pups reach the indicated ages . All other mice were housed at 23°C . For perinatal Prdm16 deletion , pregnant female mice were fed with a chow diet containing 600 mg/kg doxycycline ( Bio-Serv , S4107 ) 14 days after vaginal plug formation until the pups reach P21 . For adult Prdm16 deletion , the inducible Prdm16KO ( Cre+ and Cre- ) mice were placed on a doxycycline-containing chow diet for the indicated time . All other mice were fed with a standard rodent chow diet . For cold exposure experiments , mice were placed at 8°C for 48 hr with two mice in each cage . Animal care and experimentation were performed according to procedures approved by the Institutional Animal Care and Use Committee at the Rockefeller University . Various regions of iWAT were dissected for qPCR or western blot analyses as illustrated in Figure 1A and Figure 1—figure supplement 2K . After removal of the lymph node , the region between the bottom two dotted lines , guided by the entry of the main blood vessel in the inguinal portion and the upper boundary of the lymph node , was dissected as the inguinal region . The region from the upper boundary of the lymph node to the back was considered as the dorsolumbar region . When indicated , the dorsolumbar region was further divided into dorsomedial and dorsolateral regions by making a cut alongside the blood vessel that travel through the dorsolumbar region . Total RNA was extracted from tissue using TRIzol ( Invitrogen ) along with RNeasy kits ( QIAGEN ) . An RNeasy mini kit was used for adult tissue samples , while an RNeasy micro kit was used for small tissue samples from young mice . For qPCR analysis , RNA was reverse transcribed using the high-capacity cDNA reverse transcription kit ( Applied Biosystems ) . cDNA was used in qPCR reactions containing SYBR-green fluorescent dye ( Applied Biosystems ) . Relative mRNA expression was determined by normalization with Tbp ( TATA-box binding protein ) levels using the ΔΔCt method . The sequences of primers used in this study are listed in Supplementary file 1 . Frozen iWAT and iBAT were minced and homogenized in a hypotonic buffer ( 10 mM HEPES , 10 mM KCl , 1 . 5 mM MgCl2 , 0 . 5 mM DTT , and 1x protease inhibitor cocktail ( cOmplete Mini , Roche ) ) by a dounce homogenizer . Homogenate was incubated on ice for 10 min and then mixed with 1/20 vol of 10% IGEPAL CA-630 ( Sigma-Aldrich , I8896 ) . Samples were then filtered through a 100 μm cell strainer and centrifuged at 1000 x g for 10 min . After centrifugation , lipid and cytoplasmic fractions were removed and nuclear pellets were resuspended in lysis buffer ( 20 mM HEPES , 1 . 5 mM MgCl2 , 0 . 42 M NaCl , 0 . 2 mM EDTA , 0 . 5 mM DTT , 1x protease inhibitor cocktail , and 20% Glycerol ) . Samples were incubated on ice for 30 min and vortexed for 15 s every 10 min during the incubation . After lysis , samples were centrifuged at 20 , 000 x g for 10 min and the supernatant was taken as the nuclear extract . The following antibodies were used in immunoblotting: anti-PRDM16 ( 1:500 , R and D systems , AF6295 ) , anti-Lamin A/C ( 1:2000 , Santa Cruz , sc-376248 ) . Adipo-Clear was performed following a previously published protocol ( Chi et al . , 2018a; Chi et al . , 2018b ) . In this study , primary antibodies including anti-UCP1 ( 1:200 , abcam , ab10983 ) , anti-TH ( 1:200 , Millipore , AB1542 and AB152 ) , anti-CD31 ( 1:200 , R and D systems , AF3628 ) , and anti-LYVE1 ( 1:200 , abcam , ab14917 ) , as well as secondary antibodies conjugated with Alexa-568 and Alexa-647 ( 1:200 , Thermo Fisher Scientific , A10042 , A21448 , A32795 , A11057 ) were used . Immunostaining and imaging with iBAT cryo-sections were dissected from mice perfused and fixed with 1x PBS followed by 4% PFA . Harvested iBAT samples were post-fixed in 4% PFA at 4°C overnight and subsequently washed with 1x PBS for 1 hr at RT three times . Samples were then delipidated and permeabilized as described in the Adipo-Clear protocol ( Chi et al . , 2018a ) . Fully delipidated samples were incubated in 25% sucrose/PBS solution for 2 hr until sinking , and then frozen in Tissue-Tek O . C . T Compound ( Sakura Finetek USA , 4583 ) . Frozen iBAT samples were sectioned into 40 μm slices using a Leica CM3050 S cryostat . Cryo-sections were blocked with PBS/0 . 1% Triton X-100/0 . 05% Tween 20/2 μg/ml heparin ( PtxwH buffer ) containing 3% donkey serum for 1 hr at RT , and then incubated with primary antibodies diluted in PtxwH buffer at RT overnight . Samples were then rinsed in PtxwH buffer for 5 min , 10 min , and 30 min to remove unbound antibodies . Secondary antibodies diluted in PtxwH buffer were then applied to samples at RT for 4 hr . Samples were next rinsed with PtxwH buffer for 5 min , 10 min , and 30 min , followed by 1x PBS for 10 min twice . Finally , samples were immersed in antifade mountant ( ProLong Gold , ThermoFisher Scientific , P10144 ) and sealed with a coverslip . Anti-TH ( 1:200 , Millipore , AB152 ) and Alexa-647 conjugated anti-rabbit secondary ( 1:200 , Invitrogen , A32795 ) antibodies were used for staining cryo-sections . Fluorescently labeled samples were imaged on an inverted LSM 880 NLO laser scanning confocal and multiphoton microscope ( Zeiss ) with a 20X lens ( NA 0 . 8 ) . Whole-tissue iWAT samples were all imaged on a light sheet microscope ( Ultramiscroscope II , LaVision Biotec ) equipped with 1 . 3X and 4X objective lenses and an sCMOs camera ( Andor Neo ) . Images were acquired with the ImspectorPro software ( LaVision BioTec ) . Samples were positioned in an imaging chamber filled with benzyl ether and illuminated from one side by the laser light sheet with 488 , 561 , and 640 nm laser channels . Samples were scanned at a step-size of 4 μm for the 1 . 3x objective and 3 μm for the 4x objective . All images and videos were generated using Imaris software ( version 9 . 5 . 1 , Bitplane ) . 3D tissue reconstruction was generated using the ‘Volume’ function . Maximum intensity projections and optical slices were obtained using the ‘Ortho Slicer’ function . All images were captured using the ‘Snapshot’ tool , while all videos were made using the ‘Animation’ tool . Inguinal regions of iWAT from iKO and control iWAT samples were imaged with the 4X objective lens on the light sheet microscope . Three animals from each group were imaged and analyzed . In each 3D image , we randomly isolated small cuboidal volumes ( 4–8 volumes per sample ) that were completely contained within lobules using the ‘Surfaces’ tool followed by the mask channel option of Imaris . Volumes of the isolated segments were automatically generated by ‘Surfaces’ . To sample parenchymal neurites , we avoided placing volumes in areas that contain nerve bundles or blood vessel innervation . Using the ‘Filament’ tool , we computationally reconstructed parenchymal neurites by automatically tracing the TH signal and calculated the total neurite length within each volume . We presented the ratio of total neurite length ( mm ) by regional volume ( mm3 ) as neurite density within a volume . To adjust for adipocyte size/number , we manually counted adipocyte number as shown by the tissue autofluorescence signal from multiple representative slices within each volume . The average adipocyte number per slice was then multiplied by the height ( z depth ) of that volume to generate a factor representing adipocyte density . The ratio of total neurite length ( mm ) by adipocyte density ( arbitrary unit ) is presented ( Figure 4—figure supplement 2 ) . All statistical analyses were performed using GraphPad Prism 8 ( GraphPad Software , San Diego , CA , USA ) . For gene expression analysis , neurite density quantification , and body weight measurement , we estimated the approximate effect size based on independent preliminary studies . When indicated , an unpaired two-tailed Student’s t test was used to analyze statistical differences . Two-way ANOVA followed by Bonferroni’s multiple comparisons test was applied to determine the statistical differences for the rest of the analyses . The statistical details for each experiment can be found in the figure legends . p Values below 0 . 05 were considered significant throughout the study . | Mammals have two types of fatty tissue: white fat that mainly stores energy , and brown and beige fat , also known as thermogenic fat , which burns energy to generate heat . In humans , brown fat is associated with potent anti-obesity and anti-diabetes effects . A better understanding of how this type of fat develops and functions could lead to therapeutic strategies to treat these conditions . Adult human brown fat is similar to rodent inducible brown fat , also known as beige fat . In adult mice , beige fat cells need stimulation from the environment to form . Cold can lead to the generation of beige fat cells by activating a part of the nervous system known as the sympathetic nervous system . In order for this cold-induced formation of beige fat cells to take place , nerves from the sympathetic nervous system must first innervate the fatty tissue . Beige fat cells themselves are important for establishing this innervation , but it was not well understood when and how this occurs . To study the role of beige fat cells in the establishment of nerve innervation , Chi et al . used genetically modified mice whose beige fat cells are removed when they are treated with an antibiotic called doxycycline . If mice that had not been genetically modified were treated with doxycycline , they developed beige fat cells soon after birth , and these cells shortly became densely innervated by the sympathetic nervous system . However , if the mutant mice were treated with doxycycline around birth , these mice could not make beige fat cells during the treatment and failed to develop dense innervation even when they grew older . These results showed that beige fat cells that form soon after birth are necessary to establish sympathetic nervous system innervation . But are beige fat cells required to maintain this innervation as the mice grow older ? To test this , Chi et al . removed them after the innervation was fully established . These mice maintained their innervation , showing that beige fat cells appear to only be required during the establishment of innervation . Understanding how the sympathetic nervous system establishes its connection to fat so cold can stimulate beige fat formation is a first step to finding new treatments for conditions such as diabetes or obesity . Exploring the timing that underlies the interactions between the sympathetic nervous system and beige fat cells may provide therapeutic targets in this direction . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"developmental",
"biology",
"medicine"
] | 2021 | Early postnatal interactions between beige adipocytes and sympathetic neurites regulate innervation of subcutaneous fat |
Investigations into stem cell-fueled renewal of an organ benefit from an inventory of cell type-specific markers and a deep understanding of the cellular diversity within stem cell niches . Using the adult mouse incisor as a model for a continuously renewing organ , we performed an unbiased analysis of gene co-expression relationships to identify modules of co-expressed genes that represent differentiated cells , transit-amplifying cells , and residents of stem cell niches . Through in vivo lineage tracing , we demonstrated the power of this approach by showing that co-expression module members Lrig1 and Igfbp5 define populations of incisor epithelial and mesenchymal stem cells . We further discovered that two adjacent mesenchymal tissues , the periodontium and dental pulp , are maintained by distinct pools of stem cells . These findings reveal novel mechanisms of incisor renewal and illustrate how gene co-expression analysis of intact biological systems can provide insights into the transcriptional basis of cellular identity .
To maintain homeostasis , adult tissues must replace cells that have completed their life cycle . An emerging model for studying adult mammalian tissue renewal is the rodent incisor , which grows continuously throughout the animal’s life . As with many renewing organs , the differentiated cell types of the rodent incisor have a limited life span and are lost over time . A number of cell types , including the ameloblasts and odontoblasts that secrete the mineralized enamel and dentin , respectively , are constantly generated by progenitors located at the proximal end of the tooth ( Figure 1A , B ) . These cells replenish the tissues that are lost from the distal end of the tooth due to abrasion during gnawing . In the epithelial compartment , stem cell progeny leave the niche , known as the labial cervical loop ( laCL ) , as they begin the process of differentiation , and they then enter a transit-amplifying ( T-A ) zone and proliferate ( Kuang-Hsien Hu et al . , 2014 ) . These cells then differentiate , secrete matrix , and finally undergo apoptosis , all the while gradually advancing towards the distal tip of the organ . This linearity , akin to a conveyor belt , makes the incisor a useful model system to study adult epithelial tissue homeostasis , as tissue renewal occurs in an easily-observed proximo-distal fashion , whereby cells at increasingly advanced stages of maturation are found at progressively more distal locations ( Figure 1A , B ) . The organization of the mesenchyme has been less well-studied than the epithelium , but this tissue also has distinct compartments comprised of progenitors that give rise to various differentiated cell types , including the dentin-producing odontoblasts ( Feng et al . , 2011; Kaukua et al . , 2014; Zhao et al . , 2014 ) . A tissue complex of mesenchymal origin known as the periodontium wraps the incisor growth region and anchors the tooth in the jaw ( Nanci and Bosshardt , 2006 ) . Relatively little is known about the molecular identities of progenitor cells in the incisor or about the signals they use to regulate the production of cell types that are required to maintain homeostasis . The high turnover and short lifespan of differentiated cell types in the incisor indicate that there are active pools of progenitor cells ( Smith and Warshawsky , 1976 , Smith and Warshawsky , 1975 ) . In vivo lineage tracing assays have identified Gli1 and Bmi1-expressing populations of stem cells in both the incisor epithelium and mesenchyme ( Biehs et al . , 2013; Seidel et al . , 2010 ) . Both Gli1 and Bmi1 also mark label-retaining cells ( LRCs ) that divide infrequently and therefore retain BrdU or genetic labels . Within the incisor , LRCs are restricted to the proximal incisor mesenchyme and the proximal part of the laCL and lingual cervical loop ( liCL ) . Additional lineage-tracing studies identified Sox2 as a stem cell marker in the incisor epithelium but not the mesenchyme ( Juuri et al . , 2012 ) . The properties displayed by Gli1- , Bmi1- and Sox2-expressing cells – slow division kinetics , residence in a discrete niche , and contribution to the differentiation of various lineages – are classically considered to be typical of stem cell populations . To date , only these three markers have been found to label incisor stem cells , and thus a major limitation of the incisor model has been a paucity of markers that clearly distinguish its cell types , including progenitor cells . Indeed , this limitation is not unique to the incisor , as the precise cellular composition of most mammalian organs is still unclear . The ability to clearly distinguish cell types and distinct stages of maturation is an essential prerequisite to understanding renewal and regeneration . Gene co-expression analysis is a powerful approach for elucidating transcriptional signatures of distinct cell types in heterogeneous tissue samples ( Oldham et al . , 2008 ) . This approach is based on two straightforward ideas: ( i ) different cell types express different genes , and ( ii ) the relative abundance of a given cell type will vary among heterogeneous tissue samples . Therefore , transcript levels of genes that are specifically and consistently expressed by a cell type will appear highly correlated when measured over a large number of biological replicates . We set out to apply this strategy to the proximal adult mouse incisor region , with the goal of identifying transcriptional signatures of progenitor cells and their descendants . We identified modules of co-expressed genes representing differentiated cells , transit-amplifying cells , and residents of stem cell niches . We further demonstrated the power of this approach by using in vivo lineage tracing to define populations of incisor stem cells , and we discovered that two adjacent mesenchymal tissues , the periodontium and dental pulp , are maintained by distinct pools of stem cells . More generally , our data indicate that this strategy provides a useful analytical framework for deconstructing biological systems by identifying recurrent patterns of transcriptional variation driven by large numbers of cells .
We micro-dissected proximal incisor samples from 94 six-week-old female CD1 wild-type mice; each tissue sample was heterogeneous and consisted of the entire range of cell types present in the proximal incisor region ( Figure 1C ) . We profiled transcriptomes using Illumina Mouse Ref 8 v2 . 0 gene expression BeadChip microarrays , which contain 25 , 697 probes . Following data pre-processing ( Oldham et al . , 2012 ) , we performed genome-wide gene co-expression analysis ( Lui et al . , 2014; Zhang and Horvath , 2005 ) and identified 24 modules of co-expressed genes ( termed A-X , Figure 1D; Figure 1E ) . The characteristic expression pattern of each module was summarized by its first principal component , or module eigengene ( ME ) ( Horvath and Dong , 2008; Oldham et al . , 2008 ) , and verified by selecting several markers among the highest ranked genes and conducting in situ hybridization ( ISH ) analysis . Hierarchical clustering of modules based on eigengene dissimilarity revealed distinct subgroups of modules within the dendrogram ( Figure 1E ) , suggesting broad themes of transcriptional co-variation in the incisor . We quantified the similarity between the expression patterns of individual genes and the eigengenes of co-expression modules by calculating the Weighted Gene Co-expression Network Analysis ( WGCNA ) ( Zhang and Horvath , 2005 ) measure of intramodular connectivity , kME , for all genes with respect to all modules ( Supplementary file 1 ) . kME is defined as the Pearson correlation between the expression pattern of a gene and a ME ( Horvath and Dong , 2008 ) . Intuitively , the ME summarizes the characteristic expression pattern of genes comprising a module , and kME quantifies the extent to which individual genes track this pattern . kME can therefore be used to identify individual genes that best represent a module and mark particular cell types or biological processes ( Oldham et al . , 2008 ) . To determine how these modules mapped to cell types of the incisor , we initially examined the top 15 genes ranked by kME and conducted an extensive ontology search using literature available in PubMed . We also examined genes reported to be expressed during late development and postnatal life in the ‘Gene Expression in Tooth’ database ( http://bite-it . helsinki . fi/ ) . This analysis immediately suggested that Module A ( Figure 1E; Figure 2A ) represents a transcriptional signature of ameloblasts , which are enamel-producing differentiated cells derived from epithelial stem cells in the laCL ( Figure 1B ) . For example , both Lamb3 ( kME . Arank = 4 ) and Lamc2 ( kME . Arank = 10 ) encode subunits of Laminin 5 that are expressed by ameloblasts in developing mouse incisors ( Yoshiba et al . , 2000 ) . Expression of Enam ( kME . A rank = 11 ) is also restricted to ameloblasts ( Kuang-Hsien Hu et al . , 2014 ) . For Lamc2 and Enam , multiple microarray probes targeting these transcripts had kME ranks for Module A within the top 0 . 5% of all probes on the microarray ( Supplementary file 1 ) . Similarly , Ambn and Amelx , which encode enamel matrix proteins that are widely used as ameloblast markers ( Lee et al . , 1996; Snead et al . , 1988 ) , were also strongly associated with Module A ( kME . A ranks of 107 and 118 , respectively; Supplementary file 1 ) . Another gene in this module was Dact2 ( kME . Arank = 192 ) , which encodes a transcription factor-binding protein whose expression has previously been shown to be restricted to the dental epithelium during molar development ( Kettunen et al . , 2010 ) . Further investigation showed that Module C represents a transcriptional signature of odontoblasts , which are the dentin-secreting cells comprising the outer layer of the dental pulp . For example , Bglap1 and Bglap2 were among the top-ranked genes for Module C ( Figure 2B ) , and these are expressed by mesenchymally derived pre-odontoblasts and odontoblasts ( Bronckers et al . , 1987 ) . Phex ( kME . C rank = 1 ) is expressed in developing odontoblasts ( Ruchon et al . , 1998 ) , and Dspp , a known marker of odontoblasts ( Bègue-Kirn et al . , 1998; D'Souza et al . , 1997 ) , was strongly associated with Module C ( kME . C rank = 190 ) . Similarly , Kazald1 , which is expressed by secretory odontoblasts ( James et al . , 2004 ) , was also associated with Module C ( kME . C rank = 266 ) . We next asked if module organization could predict novel markers of distinct cell types , beginning with differentiated cell types such as ameloblasts and odontoblasts . We used immunohistochemistry and ISH to examine expression of genes with high kME for those modules that , to our knowledge , have not been previously implicated in ameloblast or odontoblast biology . As shown in Figure 2C , Tmem108 ( kME . A rank = 1 ) , SOX21 ( kME . A ranks = 5 , 15 , 1302 ) , and StarD10 ( kME . A rank = 8 ) all showed robust and specific expression in the ameloblast lineage . These data demonstrate that Module A consists of genes that are predominantly expressed in the ameloblast lineage of the adult mouse incisor . Next , we investigated expression patterns of genes that were strongly associated with Module C , to determine if the module organization was an effective tool to predict novel markers of odontoblast identity ( Figure 2D ) . As expected , expression of Phex ( kME . C rank = 1 ) was restricted to odontoblasts ( Ruchon et al . , 1998 ) . Expression of Bglap1 , Blgap2 , and Bglap-rs1 ( kME . C ranks = 6 , 9 , 10 , 14 ) was detected with a probe for an mRNA sequence shared by all three genes , confirming specificity to the odontoblast lineage . Expression of Sag ( kME . C rank = 7 ) was also restricted to odontoblasts , while expression of Lox ( kME . C rank = 8 ) showed a more complex pattern , with robust expression in odontoblasts but additional expression in regions where T-A cells are located in epithelium and mesenchyme ( arrow and asterisk in Figure 2D ) . This discrepancy may reflect biological or technical sources of variability such as additional Lox isoforms and/or non-specific targeting by microarray or ISH probes . Overall , these results indicate that Module C consists of genes that are predominantly expressed in the odontoblast lineage . To compare the distributions of predicted expression specificity for ameloblasts ( Z . ameloblast ) and odontoblasts ( Z . odontoblast ) , we generated standardized , genome-wide histograms of kME values for each module ( Figure 2E ) ( Lui et al . , 2014 ) . We observed that known markers of ameloblasts possessed Z . ameloblast >> Z . odontoblast , and vice versa . These results indicate that the expression signatures captured by Modules A and C are both sensitive and specific: canonical markers of each cell type have high kME values for the appropriate module and lower kME values for the inappropriate module . Importantly , the vast majority of genes with the highest kME values for these two modules , which are likely to play central roles in the establishment and maintenance of the functional identities of these cell types ( Lui et al . , 2014; Oldham et al . , 2008 ) have not been characterized in the ameloblast or odontoblast lineages . Taken together , these findings establish that gene co-expression analysis of heterogeneous tissue samples can discern and predict transcriptional signatures of the two principal secretory cell types of the adult mouse incisor . We also identified gene co-expression modules corresponding to other differentiated cell types that are present in this biological system ( Figure 2—figure supplement 1 ) . For example , over-representation analysis with cell type-specific gene sets revealed that Module J was significantly enriched with experimentally validated markers of oligodendrocytes ( p<10−8 , Figure 2—figure supplement 1B–E ) , suggesting that this module corresponds to a transcriptional signature of Schwann cells . Similarly , Module S was significantly enriched with experimentally validated markers of skeletal muscle cells ( p<10−12 , Figure 2—figure supplement 1B ) . It appears that this highly specific module resulted from contamination of a small number of tissue samples by cells from the muscle tissue surrounding the jawbones . These modules corroborate the ability of gene co-expression analysis to isolate distinct transcriptional signatures of differentiated cell types from heterogeneous tissue samples in silico while simultaneously providing a broad perspective on the cellular composition of biological systems . Although some gene co-expression modules clearly corresponded to specific cell lineages , others were less obvious . To determine which cells in the incisor were responsible for producing the transcriptional patterns captured by these modules , we analyzed expression patterns for at least three of the top 15-ranked members of each module . As a starting point , we chose modules that were strongly positively correlated with Modules A and C and therefore clustered in the left branch of the incisor network dendrogram ( Figure 1E ) . Interestingly , expression of Cdkn1a ( kME . B ranks = 2 , 3 , 7 ) , Smox ( kME . Branks = 4 , 5 , 9 , 31 , 68 ) , and Atp2b ( kME . B rank = 8 ) , which were among the highest-ranked members of Module B ( Supplementary file 1 ) , was detected in epithelium-forming cell types of the ameloblast , stratum intermedium , and odontoblast lineages located distal to the laCL ( Figure 2—figure supplement 1G–I ) . Given that a number of genes are expressed in all three lineages during the differentiation and secretory stages of tooth development ( http://bite-it . helsinki . fi/ ) , this finding was not surprising . The expression patterns of these genes , which included factors involved in cell cycle exit , are consistent with the location of Module B in the dendrogram and suggest a close relationship between genes in this module with those that are expressed in the ameloblast and odontoblast lineages . Expression patterns of genes in Modules D and E were similar to those of genes contributing to Module C ( data not shown ) , and genes contributing to Module F , including Tgfbi ( kME . F ranks = 1 , 2 ) , Fgfr3 ( kME . F ranks = 11 , 99 ) , and Nes1 ( kME . F ranks = 28 , 49 ) , were expressed in the distal dental pulp mesenchyme , but not in the proximal-most pulp cells ( Figure 2—figure supplement 1J–L ) . Thus , expression of genes contributing to Modules A-F was predominantly detected in regions distal to the stem cell niches in the epithelial cervical loops and in the mesenchyme located between these epithelial regions . The clustering of Modules A-F may reflect correlated cellular abundance among these regions , as dissections that include a greater representation of the epithelial niches are more likely to include a greater representation of the mesenchyme between them . Functional enrichment analysis indicated that Module U , the second largest co-expression module , was enriched for genes expressed during the mitotic phase of the cell cycle ( p=3 . 1×10−24 ) . We therefore analyzed expression patterns for genes in this module as well as other modules with which it was positively correlated ( Figure 1E , Figure 3A ) . As predicted by the functional enrichment analysis , expression of Pbk ( kME . U rank = 1 ) , Ncaph ( kME . U rank = 2 ) , and Cdca2 ( kME . U rank = 3 ) , the three highest ranked genes contributing to Module U ( Supplementary file 1 ) , was found in actively proliferating transit-amplifying ( T-A ) cells in the proximal incisor ( Figure 3B–D ) , which can be visualized as BrdU-incorporating cells using immunohistochemistry ( Young et al . , 1992 ) . All three genes were expressed in BrdU-positive , T-A cell-containing regions in both the incisor epithelium and mesenchyme . Interestingly , T-A cell-specific expression was not restricted to members of Module U . We also identified a T-A cell signature in co-expression Modules V , W , and X , which were the modules most strongly positively correlated with Module U ( Figure 3E–G , and Figure 3—figure supplement 1I–L ) . Similar to the genes contributing to Module U , expression of genes in Modules V , W , and X labeled proliferating cells in both the epithelium or the mesenchyme in all cases . When we extended our follow-up analyses to modules that clustered within the same branch of the dendrogram , but that were less strongly correlated to Module U , we discovered that Modules N-R also represented T-A cell-specific expression signatures ( Figure 3H–J , and Figure 3—figure supplement 1A–D ) . The surprising finding that T-A cells were represented by multiple closely related modules of co-expressed genes reflects the importance of these actively dividing cells as central executors of incisor renewal as well as the ability of the methodology to unravel detailed gene expression profiles within both epithelial and mesenchymal T-A cell regions . To better understand what the T-A cell-specific modules represent , we analyzed enrichment patterns in these modules for Gene Ontology annotations related to cell proliferation ( Figure 3—figure supplement 2 ) . Whereas Modules T-X were enriched in genes involved in processes characteristic of the M-phase of the cell cycle , Modules N-R were enriched in genes involved in biosynthetic and metabolic processes , suggesting roles during interphase , during which cell growth and DNA replication occur . Together , these data indicate that T-A-specific gene co-expression modules may represent distinct biological processes that are integrated to induce or maintain proliferation . They also suggest that these processes may be temporally segregated in subpopulations of T-A cells . The T-A cell-specific modules provide a platform for elucidating the molecular mechanisms that regulate stem cell progeny during this poorly understood stage of maturation . We next asked if the handful of stem cell markers that have been identified in the adult mouse incisor through in vivo lineage tracing were associated with particular co-expression modules . Gli1 and Bmi1 mark stem cell pools in both the incisor epithelium and mesenchyme , whereas Sox2 exclusively marks epithelial stem cells ( Biehs et al . , 2013; Juuri et al . , 2012; Seidel et al . , 2010 ) . Although Gli1 was not represented by a probe on the microarrays that we used , Ptch1 , which like Gli1 reports active Hedgehog signaling and is expressed in the same pattern in the incisor as Gli1 ( Seidel et al . , 2010 ) , was associated with Module L , and Sox2 was associated with Module H ( Figure 4A , Supplementary file 1 ) . Expression of Bmi1 was detected in all specimens but not associated with a co-expression module . Lgr5 , another gene that recently has been suggested to be expressed by incisor stem cells ( Chang et al . , 2013; Suomalainen and Thesleff , 2010 ) , was not detected by the microarray in any of the samples . These results may reflect the limited sensitivity of microarrays for detecting low-expressed transcripts in rare cell populations in heterogeneous tissue samples . Module H , the co-expression module with which Sox2 was most strongly associated , also contained Tbx1 ( kME . Hrank = 14 ) , which was previously shown to be expressed in the epithelium of developing incisors ( Catón et al . , 2009 ) . We found that Tbx1 expression is also restricted to the epithelium in the adult incisor , and we detected transcripts in the T-A and pre-ameloblast region as well as in several cells in the proximal region of the laCL ( Figure 4B ) where epithelial stem cells reside . Other genes strongly associated with this putative epithelial progenitor module included Epha1 and Prom2 ( Figure 4C , and Figure 4—figure supplement 1A ) ; like Sox2 ( Juuri et al . , 2012 ) and Tbx1 , expression of these genes was restricted to the incisor epithelium and included the proximal regions of the laCL , where we previously detected LRCs . Expression of Epha1 was found predominantly in stellate reticulum ( SR ) cells in the laCL and in more distal cells subtending the SR cells adjacent to this region ( Figure 4—figure supplement 1A ) . In contrast , expression of Prom2 appeared to overlap that of Tbx1 in the T-A cell and pre-ameloblast region as well as in the outer enamel epithelium ( OEE ) of the laCL ( Figure 4B ) . In addition to being a highly-ranked gene for Module H , Prom2 also showed a strong association to the neighboring Module I . Further investigation of Modules G and I , which were strongly positively correlated with Module H and each other ( Figure 1C ) , revealed that multiple genes , including Epha1 and Prom2 , associated strongly with two or all three modules ( Supplementary file 1 ) . Therefore , we extended our expression analysis to include the two neighboring modules and treated the G-H-I group of modules as a clustered unit . Of the genes primarily contributing to Module G , we analyzed the expression patterns of Scnn1b ( kME . G rank = 3 ) , p63 ( kME . G rank = 5 ) , and Nkx2-3 ( kME . G rank = 11 ) , the last of which was previously shown to be required for molar development ( Biben et al . , 2002 ) . All three were expressed predominantly in the T-A and OEE regions of the laCL with no or low levels of expression in the SR ( Figure 4D , and Figure 4—figure supplement 1B ) . Expression was not limited to the CL on the labial side but rather extended distally into both the OEE and differentiated ameloblasts and was also found in the liCL . Isl1 ( kME . G rank = 33 ) , which was previously shown to be expressed exclusively in the incisor epithelium during tooth development ( Mitsiadis et al . , 2003 ) , was expressed in the same pattern in the laCL as p63 , Nkx2-3 , and Scnn1b , with distally extending expression maintained in the OEE and ameloblasts but decreased levels in the pre-ameloblasts ( Figure 3—figure supplement 1C ) . The distally extended expression appears characteristic of Module G genes and is reflected in the strong positive correlation of Module G’s eigengene signature with that of the ameloblast-specific Module A ( Figure 1E ) , whereas the correlations between Modules H and A as well as I and A were weaker ( Figure 1E ) . Expression of Pitx2 ( kME . Irank = 2 ) in the T-A cell and OEE regions of the laCL widely overlapped with the expression domains of Prom2 ( Figure 4C ) . Pitx2 expression appeared to be highest in the apical aspect of the laCL , where Sox2 is expressed at high levels ( Juuri et al . , 2012 ) , and was also present in the SR . In the proximal SR and OEE regions , expression of Pitx2 was detected in LRCs ( Figure 4H–H” ) . However , similar to Tbx1 , Epha1 and Prom2 , expression of Pitx2 was not detected in the liCL epithelium . Shh , which is expressed in epithelial T-A cells , pre-ameloblasts and ameloblasts in the labial incisor epithelium ( Seidel et al . , 2010 ) , was also among the genes contributing to Modules G , H , and I ( kME . Grank = 14 , kME . Hrank = 53 , and kME . Irank = 7 ) , which was surprising given the absence of Shh expression in the proximal laCL . Thus , expression of all analyzed genes contributing to Modules G , H , and I specifically overlapped with T-A cells in the epithelium and , with the exception of Shh , in the proximal laCL . We next focused our attention on modules that appeared to exhibit mesenchymal character . As with Modules G-I , Modules K-M were highly correlated , and several genes showed promiscuity for all three modules ( Figure 1E , Supplementary file 1 ) . Expression analysis of Gpx3 , which was strongly correlated to all three modules ( kME . Krank = 10 , kME . Lrank = 2 , and kME . Mrank = 6 ) , Fbln1 ( kME . L rank = 4 ) , and Igfbp4 ( kME . Lranks = 5 , 12 ) , showed that all three genes are actively transcribed in the mesenchymal tissue in the proximal part of the incisor ( Figure 4E , and Figure 4—figure supplement 1E ) . In the case of Igfbp4 , expression was observed in an additional domain in the proximal laCL epithelium . Similar expression patterns were observed for Pecam1 ( kME . K rank = 13 ) , Scara5 ( kME . Lrank = 15 , kME . Lrank = 1 ) , Igfbp5 ( kME . Krank = 1 ) , and Lrig1 ( kME . Lrank = 10 ) ( Figure 4G – Figure 4—figure supplement 1F ) . Double-labeling experiments using ( i ) antibody staining against BrdU in wild-type animals that were treated with BrdU followed by a long chase period to generate LRCs and ( ii ) an antibody against PECAM1 or ISH to detect Igfbp5 demonstrated that both genes are expressed in LRCs in the incisor mesenchyme and laCL epithelium ( Figure 4J–J” , and Figure 4—figure supplement 1G ) ; the identity of the PECAM1-positive cells in the epithelium is not clear , but as the epithelium is not vascularized , this marker must label a non-vascular cell type in the epithelium . We also noted that expression of genes contributing to Modules K-M was predominantly observed in cells of the periodontium ( Figure 4E–G , and Figure 4—figure supplement 1E–G’ ) , which is the tissue responsible for anchoring the tooth to the alveolar bone . While little is known about the periodontal tissue in the incisor , previous studies that mostly focused on molar teeth identified a small number of genes expressed in periodontal cells ( Supplementary file 2 ) . Comparison with the data obtained from our co-expression analysis showed that most of these factors were associated with Modules K , L , and M , suggesting that this group of modules is driven primarily by gene expression in periodontal cells ( Supplementary file 2 ) . The expression of these markers in the proximal part of the tooth suggested that these modules may be enriched for candidate stem cell markers and genes required for periodontal tissue maintenance . The relative expression patterns of the modules are presented in Figure 4O . To test whether our analysis could enable functional identification of a stem cell population , we next focused on Igfbp5 , which was the highest ranked gene contributing to Module K and was previously found to be increased in LRCs in the hair follicle bulge ( Tumbar et al . , 2004 ) . To establish whether Igfbp5 marks stem cells in the incisor , we generated a tamoxifen-inducible Igfbp5iCre-ERT2 allele that drives expression of Cre recombinase without reducing Igfbp5 expression ( Figure 4—figure supplement 2 ) , and we bred mice carrying both this allele and the Ai14 ( R26RFP ) reporter allele ( Madisen et al . , 2010 ) to perform in vivo lineage tracing ( Figure 4K ) . 24 hr after tamoxifen treatment of the double-heterozygous animals , RFP expression was detected in cells of the periodontium , the liCL and in the proximal portion and OEE of the laCL , as well as in a number of cells in the alveolar nerve that innervates the incisor . Thus , reporter expression was found in regions of the proximal incisor where we detected Igfbp5 mRNA expression ( Figure 4J–J” ) , and we proceeded to perform lineage analyses using the Igfbp5iCre-ERT2 driver . 14 days after treatment with tamoxifen , the number of RFP-positive cells in the incisor mesenchyme surrounding the cervical loops and within the laCL and distally adjacent epithelium was strongly increased when compared to 24 hr post-induction , suggesting that the originally labeled Igfbp5-expressing cells supply new cells in these regions ( Figure 4M ) . A similar distribution of RFP-labeled cells was present 1 month after tamoxifen treatment ( Figure 4N ) . Interestingly , except for a small number of RFP-labeled cells whose morphology and locations were consistent with a neurobiological identity , RFP-positive cells were absent from the inner dental pulp mesenchyme and remained restricted to the periodontal portion of the incisor mesenchyme . These findings suggest that Igfbp5-expressing cells contribute to epithelial and periodontium homeostasis without contributing to maintenance of the pulp compartment . Notably , Lrig1 , a marker of stem cells in the intestine and the skin ( Jensen and Watt , 2006; Powell et al . , 2012 ) not previously identified in incisor stem cells , was represented in Module L . Consistent with this observation , Lrig1 was expressed in the proximal incisor mesenchyme and laCL epithelium but not in pulp cells , odontoblasts or differentiated epithelial cells ( Figure 4G ) . ISH yielded a relatively low signal for Lrig1 when compared to most other genes analyzed for this group of modules , which may suggest transcription at low levels or expression in a subset of cells within the domain . To determine if Lrig1 is expressed by LRCs , we crossed a tamoxifen-inducible Lrig1Cre-ERT2 mouse line ( Powell et al . , 2012 ) with the R26RFP reporter allele . To generate LRCs , Lrig1Cre-ERT/+;R26RFP/+ newborn mice were injected with BrdU and aged for 8 weeks ( Figure 5A ) . Double immunofluorescence assays against BrdU and RFP were performed in specimens from mice chased for 24 hr after tamoxifen administration; the short chase period was used to mark Lrig1-positive cells . This experiment confirmed the presence of Lrig1-expressing LRCs in the proximal laCL epithelium and the most proximal incisor mesenchyme ( Figure 5B–B” ) , indicating that some Lrig1-expressing cells are quiescent . Interestingly , in the mesenchyme-derived portion of the incisor , the BrdU-LRC population was comprised of two neighboring subdomains that appear as stripes in the two-dimensional sections: an Lrig1-negative inner , or distal , region and an Lrig1-positive proximal region ( Figure 5B ) . RFP also marked cells in the alveolar nerve , and occasionally single RFP-positive cells were observed within the liCL and in blood vessels . However , RFP was not detected in differentiated cells of the epithelium or within the pulp mesenchyme . Thus , expression of the Lrig1Cre-ERT2 allele was restricted to regions in the incisor where Lrig1 mRNA is expressed and included some , but not all , BrdU-LRCs ( Figure 4G ) . To definitively determine whether the Lrig1-expressing cells in the laCL and the proximal incisor are stem cells that self-renew and give rise to progeny over an extended period , we followed the fate of RFP-labeled cells in tamoxifen-induced Lrig1Cre-ERT2/+;R26RFP/+ mice over chase periods of different lengths ( Figure 5D ) . As before , 24 hr after treatment with tamoxifen , we detected RFP-positive cells in the LRC-containing regions of the incisor , including the laCL and the mesenchymal tissue surrounding both CLs ( Figure 5E–E” ) ; no RFP was detected in the dental pulp mesenchyme between the CLs or in tissues of uninduced Lrig1Cre-ERT2/+;R26RFP/+ control animals ( data not shown ) . 14 days after the initial labeling , Lrig1-positive cells remained restricted to the proximal laCL epithelium , and RFP-positive cells were absent from the T-A and differentiated cell regions ( Figure 5F–F" ) . Short BrdU chases performed on the lineage tracing specimens revealed that the only proliferative populations were the epithelial and flanking mesenchymal T-A cells , with the periodontium remaining BrdU-negative ( Figure 5—figure supplement 1 ) . Next , we wished to determine the relationship between expression of Lrig1 and Gli1 , an established marker of both epithelial and mesenchymal stem cells ( Seidel et al . , 2010; Zhao et al . , 2014 ) . We found that Lrig1 expression , as reflected by CreER-mediated RFP expression 24 hr after tamoxifen administration , largely overlaps with Gli1 in Lrig1Cre-ERT2/+;R26RFP/+;Gli1lacZ/+mice; however , Lrig1 is more restricted in the SR and absent from the most apical portion of the Gli1lacZ-positive domain of the laCL ( Figure 5C ) . In the mesenchyme , RFP-positive cells were relatively sparse compared to cells expressing Gli1lacZ , and these populations do not co-localize in the dental pulp . In the mesenchyme surrounding the CLs , both double-positive or RFP-positive cells that did not express Gli1lacZ were detected ( Figure 5C’ ) . Thus , Lrig1-positive cells constitute a subdomain of the Gli1-expressing cells in the laCL epithelium , but not in the incisor mesenchyme . Given the apparent overlap of Gli1 and Lrig1 expression , we found it surprising that Lrig1Cre-ERT2/+;R26RFP/+ did not exhibit labeled progeny in differentiated or T-A cells , especially because progeny formation from cells marked by Gli1 , Sox2 , or Bmi1 were detected within days after tamoxifen administration ( Biehs et al . , 2013; Juuri et al . , 2012; Seidel et al . , 2010 ) . However , one month post-induction , a small number of labeled cells was present amongst the T-A cells , as well as in the SR , SI and ameloblasts ( Figure 5G–G” ) . After 6 months , the number of RFP-positive cells in the T-A region and SR had increased , and the majority of ameloblasts , SI cells and OEE cells were labeled ( Figure 5H–H” ) . The extent of RFP labeling in specimens that were analyzed 3 months or 12 months after tamoxifen administration was comparable to the 6 month time point ( data not shown ) . Thus , Lrig1 is expressed by long-lived stem cells located in the OEE and/or SR of the laCL that give rise to cells of all epithelial lineages over long periods of time . The delay in appearance of labeled progeny in the T-A cell region when compared to lineage tracing performed for Gli1 and other markers previously tested ( Biehs et al . , 2013; Juuri et al . , 2012; Seidel et al . , 2010 ) , together with the absence of Lrig1-positive cells from the Gli1-expressing population adjacent to the T-A cell region , suggests that Lrig1 marks a relatively quiescent progenitor pool . In the proximal incisor mesenchyme , the number of RFP-positive cells strongly increased 14 days after induction with tamoxifen ( Figure 5F ) , suggesting that Lrig1-positive cells in this region turn over more rapidly when compared to epithelial cells that express Lrig1 . Similar numbers of RFP-positive cells were observed in the mesenchyme surrounding the CLs in specimens that were analyzed one , 3 , 6 or 12 months post-tamoxifen ( Figure 5G and data not shown ) . Interestingly , RFP-positive cells were absent from the mesenchyme of the dental pulp , the compartment surrounded by the dental epithelium , but rather were restricted to the periodontal tissue surrounding the epithelial cervical loops on the outer surface of the incisor . Together , these data demonstrate that Lrig1 marks two distinct pools of stem cells , one in the incisor epithelium that enables renewal of epithelial cell types , such as ameloblasts , and one in the proximal mesenchyme that produces cells contributing to the periodontium . These fate mapping experiments further revealed that the periodontal mesenchyme in the incisor is maintained by a mesenchymal stem cell pool that is separate from the progenitors that maintain the mesenchymal cells of the dental pulp , pointing to the existence of distinct subpopulations of mesenchymal stem cells in the incisor . We further interrogated renewal of the periodontium by examining Acta2 , a gene that had high rankings for Module L ( which contained Lrig1 ) based on data from three different microarray probes ( kME . Lranks = 120 , 150 , 200; Supplementary file 1 ) . In molars , lineage tracing has shown that Acta2 is expressed by periodontal progenitors ( Roguljic et al . , 2013 ) . As expected from our co-expression data , Acta2 expression was found in the periodontal mesenchyme proximal to the epithelial CLs , but not in mesenchymal cells in the pulp ( Figure 4—figure supplement 3B ) . Transcription was also detected in perivascular cells in both the periodontal and pulp area , and in a small number of cells in the proximal laCL ( Figure 4—figure supplement 3B’ ) . In the periodontal compartment , ACTA2 expression included BrdU LRCs ( Figure 4—figure supplement 3C ) , but no mRNA or protein was detected in a band of cells directly surrounding the laCL and subtending the adjacent OEE on the labial aspect of the incisor ( Figure 4—figure supplement 3B–C’ ) . This Acta2-negative domain contained a number of LRCs and was Lrig1-positive ( Figure 4G – Figure 4—figure supplement 3C’ ) . We next performed lineage tracing in Acta2Cre-ERT2;R26RFP mice ( Wendling et al . , 2009 ) to determine whether Acta2 is also expressed by periodontal progenitors in the incisor . In contrast to the progenitors in molar teeth , progenitors in the incisor produce periodontal fibroblasts that are constantly renewed and move distally at a similar rate as ameloblasts and odontoblasts ( Smith and Warshawsky , 1976 ) . Shortly after tamoxifen administration ( Figure 4—figure supplement 3A ) , RFP-expressing cells were present in areas where we detected Acta2 mRNA and protein ( Figure 4—figure supplement 3D ) . Similar to what we observed when investigating the fate of Lrig1-positive cells , the number of RFP-positive cells in the periodontal mesenchyme increased between the 24 hr and 7 day time points ( Figure 4—figure supplement 3D ) . In contrast , the number of labeled cells detected after longer chase periods appeared to remain constant ( Figure 4—figure supplement 3F ) and RFP-positive cells were still present 6 months after tamoxifen treatment ( data not shown ) . These data are consistent with achievement of a steady state and indicate that Acta2 is expressed by periodontal progenitors in the incisor . To investigate the relationship between Lrig1- and ACTA2-expressing progenitors in the periodontal compartment , we co-labelled cells derived from the Lrig1-positive progenitor pool with ACTA2 . Interestingly , the progeny of Lrig1-positive cells that were labeled three months prior to analysis did not contribute to all regions of the periodontal mesenchyme marked by ACTA2 ( Figure 4—figure supplement 3H ) . Whereas expression overlapped in a domain surrounding the ACTA2-negative inner portion of the periodontium , RFP-positive cells were absent from the outermost layer of periodontal tissue , close to the bone . We also assessed co-localization of Lrig1 descendants with cells expressing N-CAM , a gene that is broadly expressed in the periodontal tissue of the incisor ( Obara and Takeda , 1997 ) and that showed strong membership for Module L ( kME . Lranks = 39 , 206; Supplementary file 1 ) . This analysis confirmed that descendants of Lrig1-expressing cells contributed to the inner but not outer periodontal mesenchyme ( Figure 4—figure supplement 3I ) . In contrast , progeny of cells marked by Acta2 expression were found in the outer portion of the periodontium but absent from the periodontal tissue directly surrounding the incisor epithelium ( Figure 4—figure supplement 3D–G ) . Thus , the inner ( near the tooth ) region of the periodontal mesenchyme is preferentially renewed by Lrig1-expressing progenitors , whereas the outer ( near the bone ) region is renewed by Acta2-expressing progenitors . Together , these data identify subpopulations of periodontal progenitors , marked by Lrig1 and Acta2 expression , that promote renewal of distinct regions of periodontal tissues during incisor homeostasis .
Here we set out to characterize the cellular composition of the adult mouse incisor and identify cell type-specific markers by analyzing patterns of transcriptional co-variation in a large number of biological replicates . The results from this study provide strong evidence that correlated gene expression patterns are driven by variation in the abundance of distinct cell types and cell states . We identified transcriptional signatures driven primarily by ameloblasts , odontoblasts , Schwann cells , and skeletal muscle cells . The ability to detect a transcriptional signature of a cell type through gene co-expression analysis of intact tissue specimens depends on many factors , including the abundance of the cell type , the distinctiveness of its transcriptome , its anatomical distribution with respect to other cell types , the technology platform , the sampling strategy , and the algorithmic approach ( Oldham , 2014 ) . Therefore , different cell types will have different signal to noise ratios . For example , ameloblasts and odontoblasts are abundant , differentiated cell types with distinctive transcriptomes that were easily detected with our strategy . In contrast , Schwann cells are much less abundant but co-express a unique set of genes that allowed them to clearly stand out in the incisor co-expression network . Optimization of some of the factors listed above ( e . g . RNA-seq analysis of a larger number of samples ) should improve the sensitivity of our approach . We also identified transcriptional signatures related to distinct states of cellular differentiation , including the progenitor state , transit-amplifying state , and cell cycle exit at the onset of differentiation . Among the patterns corresponding to differentiation states , we identified modules representing stem cell progeny far along in their maturation process within distinct compartments as well as compartment-specific modules enriched for genes expressed by progenitors . For example , one newly identified progenitor marker , Lrig1 , discriminates subpopulations of Gli1-positive epithelial progenitors that appear to have different capacities to provide daughter cells for replenishing the organ . By providing an unbiased view of the major transcriptional themes in the adult mouse incisor , our study lays the groundwork for future investigations into molecular interactions that are required to establish or maintain the functional identities of distinct cell types and cell states in this model system . Unbiased gene co-expression analysis of intact biological systems also provides a data-driven framework for studying the effects of perturbations , such as blocking a specific molecular pathway or causing injury . By comparing transcriptional co-variation in perturbed and naive states through differential co-expression analysis , identification of relevant phenotypes that affect specific cell types or cell states can be accelerated . Single-cell methods have come of age and hold great promise for gene expression applications . However , technical noise and limited coverage of cells and transcriptomes can constrain the use of this strategy for unbiased characterization of intact biological systems . In contrast , gene co-expression analysis of bulk tissue specimens is a comparatively simple and efficient approach that can reveal the major building blocks of a biological system’s transcriptome by analyzing expression patterns that are derived simultaneously from millions of cells . Thus , these two types of analyses can be seen as complementary with regard to resolution and tissue volume analyzed . From a practical perspective , the presence of genes encoding cell-surface proteins in the co-expression modules we have identified will simplify the isolation , purification , and detailed characterization of individual cell populations using single-cell methods , which will be an important next step . It is also important to note that the microarrays used in this study have limited dynamic range and sample almost exclusively from the protein-coding transcriptome . Therefore , the full picture of gene expression in the adult mouse incisor is likely to be more complex than the initial description presented here . Future surveys that combine deep sequencing of coding and non-coding transcripts from large numbers of intact tissue specimens and single cells will provide a powerful approach for deconstructing the transcriptional architecture of the incisor and other biological systems . In addition , because our studies here exclusively used two-dimensional section analysis , it will be important in the future to analyze proximal incisor gene expression patterns in three dimensions as well . In this study , we discovered several co-expression modules enriched with genes that are predominantly expressed in the LRC-containing regions of the proximal incisor mesenchyme . Amongst the genes with expression profiles most similar to Modules K and L were Igfbp5 and Lrig1 . Lrig1 is expressed in stem cells in the intestine and skin ( Jensen et al . , 2009; Jensen and Watt , 2006; Powell et al . , 2012 ) , and Igfbp5 is transcribed by LRCs in the hair follicle bulge ( Tumbar et al . , 2004 ) . Interestingly , co-labeling with BrdU-LRCs revealed that expression of both Igfbp5 and Lrig1 divides the mesenchymal LRC population into an inner subpopulation that does not express either gene and an outer region where Igfbp5 and Lrig1 expression are found . Lineage tracing for Igfbp5 and Lrig1 revealed that the outer LRC-containing region of the mesenchyme contains progenitors that contribute specifically to periodontal cell lineages but not mesenchymal cells of the dental pulp . This finding demonstrates for the first time that the inner mesenchyme and the periodontal tissue in this continuously growing tooth are maintained by separate pools of progenitors . The extent to which the Igfbp5-positive and Lrig1-positive populations overlap will need to be addressed in the future . Of note , Lrig1-expressing cells were shown to contribute specifically to the upper but not the lower portion of the mesenchymally derived dermis of the skin ( Driskell et al . , 2013 ) . This result is intriguing as it parallels our finding that Lrig1-expressing mesenchymal cells only contribute to the periodontal tissue but not the dental pulp mesenchyme . Moreover , our analysis of cell proliferation in the Lrig1 lineage tracing experiments did not identify a specific periodontal T-A cell population , indicating a relatively slow turnover rate of this tissue . Multiple progenitor pools are known to facilitate coordinated renewal in other organs . For example , in the skin , several stem cell pools exist that contribute , with varying overlap , to the renewal of the interfolliclular epidermis , hair follicles and sebaceous glands ( Jensen et al . , 2009 ) . Although during normal homeostasis the stem cells contribute to their respective compartments , they can be mobilized to regenerate nearby compartments after injury . Whether the Igfbp5- or Lrig1-expressing periodontal progenitors are able to contribute to the dental pulp lineages following injury remains to be determined . Of note , in a small number of Lrig1Cre-ERT/+;R26RFP/+ animals chased for one year after tamoxifen treatment , we observed low-level contribution to the odontoblast lineage ( data not shown ) ; it is possible that this lineage contribution arose as the result of injury . Alternatively , this increase in plasticity could reflect changes in stem cell number or regeneration capacity that occur in the incisor as animals age . To achieve proper homeostatic renewal of an organ composed of several tissues , progeny production by stem cell niches in all tissue compartments must be highly coordinated . Indeed , cell-labeling experiments performed in rodent incisors several decades ago showed that ameloblasts , odontoblasts and periodontal fibroblasts move distally at the same rate ( Beertsen and Hoeben , 1987 ) . The notion that the T-A cell stage serves as an important checkpoint for coordination between tissues is strongly supported by our finding that all factors that contributed to T-A cell specific modules were always expressed in proliferating cells in both the epithelium and mesenchyme and never in only one tissue . Our previous results highlight SHH as a likely signal through which coordination between the different stem cell pools in the incisor is achieved ( Seidel et al . , 2010; Zhao et al . , 2014 ) . These studies showed that stem cells maintaining the epithelial tissues on the labial and lingual aspect of the tooth , the mesenchymal cell types of the dental pulp including the dentin-forming odontoblasts , and the periodontium are all marked by Gli1 expression , a hallmark of responsiveness to HH signaling . Similarly , Bmi1 appears to be expressed by stem cells in all niches of the incisor ( Biehs et al . , 2013 ) , whereas Sox2 and Lrig1 mark stem cells only in a restricted set of compartments ( Figure 5 , ( Juuri et al . , 2012 ) . An interesting open question is whether other members of the SRY-related HMG-box ( SOX ) family of transcription factors are expressed in the liCL , mesenchymal and periodontal stem cell niches and substitute for the function of SOX2 in these regions , or whether SRY-mediated transcriptional regulation is uniquely required for control of progeny formation from epithelial stem cells housed in the laCL . Another intriguing discovery is that distinct sets of stem cells , as defined by a constellation of markers uncovered by module analysis , have different properties in the laCL and give rise to unique cell fates in the mesenchyme . Lrig1 , which encodes a type I transmembrane protein , is specifically expressed by a subset of Gli1-expressing cells in the laCL and by stem cells that give rise to periodontal tissues ( Figure 5 ) . Our lineage tracing analysis further showed that the cells that maintain the lingual incisor epithelium and pulp mesenchyme do not express Lrig1 . An exciting extension to the finding that Lrig1 marks a pool of stem cells in the periodontium would be a comparison with the periodontium around molar roots . In both cases , these tissues develop from the dental follicle , wrap the teeth and anchor them to the jaw . Because molars have a finite growth period , in contrast to the continuous growth of the incisor , it would be of interest to determine how the periodontium is maintained in the molar . Additionally , an important future direction will be establishing the molecular function of LRIG1 in the tooth as well as the source and function of the ligand whose signaling it regulates in this system . In conclusion , the wealth of information gained by our analysis of gene co-expression in the incisor will advance the use of this organ as a model for stem cell-based tissue renewal . More generally , results gained from deconstructing an organ using this transcriptome-focused approach can enable a deeper understanding of the biology of the system . In addition , by comparing gene co-expression networks between different species , important species-specific characteristics of organs can readily be identified . To this end , comparisons of gene co-expression relationships in the developing brains of humans and mice have yielded invaluable insights into transcriptional differences in neural stem cells that have contributed to changes in brain architecture during evolution ( Lui et al . , 2014 ) . Going forward , comparisons of gene co-expression in different organs will help elucidate conserved pathways and general mechanisms that govern tissue renewal from stem cells . Such information can also enhance bioengineering approaches that use stem cells as a starting material for generating tissues for therapeutic purposes .
Mice carrying the Acta2Cre-ERT2 ( Wendling et al . , 2009 ) , Ai14 ( Madisen et al . , 2010 ) , Gli1-lacZ ( Bai et al . , 2002 ) , and Lrig1CreERT2 ( Powell et al . , 2012 ) were maintained and genotyped as previously described . 6-week-old female CD1 mice were purchased from Charles River Laboratories and used for generation of tissue samples for microarray analysis . 6–8 week old animals were used for expression analysis and lineage tracing experiments . For generation of label-retaining cells , neonatal mice were injected daily from P5 to P9 with BrdU ( 5’bromo-2’deoxyuridine ) and aged to 8 weeks . For detection of proliferating cells , BrdU was given in a single injection to adult mice 1 . 5 hr prior to sacrifice . BrdU was administered at 40 µg per gram of bodyweight . Mice were treated with a single dose of 5 mg tamoxifen ( in corn oil ) given by oral gavage in case of lineage tracing studies , and three doses of 10 mg tamoxifen every other day given in case of ablation experiments . Expression and lineage tracing analyses were performed using specimens from at least three different animals , examined for each functional experiment . All animals were maintained at the UCSF vivarium and the UCSF Institutional Animal Care and Use Committee ( IACUC ) approved all experiments performed in this study . The Igfbp5iCreER-T2 line was produced at the Jackson Laboratory . To generate the inducible Cre line , a codon optimized Cre recombinase variant ( iCre ) ( Shimshek et al . , 2002 ) was fused to a modified ligand-binding domain of human estrogen receptor ( ERT2 ) ( Feil et al . , 1997 ) . To allow for expression from the target locus without disruption of the endogenous allele , an internal ribosome entry sequence ( IRES ) was placed 5’ to the start of the iCreERT2 coding sequence and the cassette was inserted into the Igfbp5 sequence after the stop codon . A Frt-flanked neo cassette was placed in the third intron upstream of the final coding exon and the entire construct was flanked by 5’ and 3’ homology arms . A diphtheria toxin cassette was included for negative selection . The construct was linearized and electroporated into C57BL/6J embryonic stem ( ES ) cells , subjected to G418 selection and 37 putative targets were identified by loss of native allele ( LOA ) qPCR screening . Six clones were confirmed by 5’ and 3’ Southern blot , and three clones with normal karyotypes were injected into albino C57BL/6J blastocysts . Chimeras were bred to C57BL/6J mice and screened for germline transmission of the targeted allele . A single clone was pursued and double confirmed by Southern blot . Mandibular incisors of 140 wild-type female CD1 mice were isolated as previously described ( Chavez et al . , 2014 ) , and the tissue region proximal to the first occurrence of mineralized dentin on both labial and lingual aspects of the incisor was isolated and stored in RNA-later ( Ambion ) . The tissue level along the proximo-distal axis was readily identified on both lingual and labial aspect of the incisor with a 5 . 0x magnification based on the color difference between the mineralized tissues . Total RNA was isolated from individual tissue samples using Qiagen’s RNeasy kit according to the manufacturer’s instructions . To improve the total yield in the final step of the protocol , the eluate was run over the microspin column a second time . Only tissues obtained from left lower incisors were used for RNA extraction . RNA quality and quantity were assessed using a Nanodrop Spectrophotometer and a Bioanalyzer assay , and only the 94 samples with a concentration of >20 ng/µl ( average of both assays ) and the highest RNA-integrity scores ( RIN ) of >8 . 5 were used by the microarray facility . RNA concentrations were confirmed utilizing a ribogreen assay , hybridization to Illumina Mouse Ref 8 v2 . 0 gene expression BeadChips performed , and initial data analyzed in R with the SampleNetwork function ( Oldham et al . , 2012 ) , which identifies outlying samples , performs data normalization , and adjusts for batch effects . After removing one outlying sample , data were quantile normalized ( Bolstad et al . , 2003 ) and technical batch effects were assessed . A highly significant batch effect associated with microarray slide was detected and corrected using the ComBat R function ( Johnson et al . , 2007 ) . Gene co-expression modules were identified in R using a four-step approach as previously described ( Molofsky et al . , 2013; Lui et al . , 2014 ) . First , pairwise Pearson correlation coefficients ( cor ) were calculated for all possible pairs of microarray probes ( n = 25 , 697 ) over all samples ( n = 93 ) . Second , probes were clustered using the flashClust implementation of a hierarchical clustering procedure with complete linkage and 1 – cor as a distance measure ( Langfelder and Horvath , 2008 ) . The resulting dendrogram was cut at a static height of ~0 . 594 , corresponding to the top 1% of pairwise correlations for the entire dataset . Third , all clusters consisting of at least 15 members were identified and summarized by their module eigengene ( i . e . the first principal component obtained via singular value decomposition ) ( Horvath and Dong , 2008; Oldham et al . , 2006 ) . Fourth , highly similar modules were merged if their Pearson correlation coefficients exceeded an arbitrary threshold ( 0 . 85 ) . This procedure was performed iteratively such that the pair of modules with the highest correlation >0 . 85 was merged , followed by recalculation of all module eigengenes , followed by recalculation of all correlations , until no pairs of modules exceeded the threshold . Following these steps , 24 co-expression modules were identified . The strength of module membership ( kME ) for each probe on the microarray was determined by calculating the Pearson correlation between its expression pattern over all samples with each module eigengene ( Horvath and Dong , 2008; Oldham et al . , 2008 ) . Module enrichment analysis with curated gene sets was performed using a one-sided Fisher’s exact test in R with gene symbol as a common identifier . Modules were defined as consisting of all unique genes that were positively and maximally correlated with a given module eigengene at a significance threshold of p<8 . 11×10−08 . This threshold corresponds to a Bonferroni-corrected P-value of . 05 / ( the total number of probes X the total number of modules ) . Gene Ontology ( GO ) analysis was performed using The Database for Annotation , Visualization and Integrated Discovery ( DAVID ) ( Dennis et al . , 2003 ) . Enriched GO terms for Biological Processes ( level 5 ) were detected and clustered using the Functional Annotation Clustering tool with default parameters . Sample preparation and decalcification , hematoxylin and eosin staining , immunofluorescence staining and RNA in situ hybridization were performed as previously described ( Seidel et al . , 2010 ) . Primers for the ISH probes were designed using Primer-BLAST ( http://www . ncbi . nlm . nih . gov/tools/primer-blast ) , and a BLAST search within the mouse genomic and transcript database was performed for each primer pair in order to ensure specificity . Primer pairs , probe sizes , as well as gene reference sequences are provided in supplementary files . cDNA from mouse incisors was used as a template for PCR amplification , and fragments of interest were cloned into pGEM-T Easy Vector Systems plasmids ( Promega ) . DH5α cells were transformed with 1 µl of plasmid and plated on LB plates with 100 µg/ml ampicillin . Single colonies were picked for overnight amplification in liquid LB , and plasmid DNA was purified using a Plasmid Miniprep kit ( Qiagen ) . 10 µg of plasmid were linearized using 50 u of the appropriate restriction enzyme , and probes were transcribed using SP6 , T3 , or T7 polymerases together with RNA-DIG labeling mix ( Roche ) . When generating probes to examine the expression of previously identified markers in the incisor system , we noticed that all probes for Bglap1 , Bglap2 and Bglap-rs1 in fact detected transcripts of all three genes , likely as the result of high sequence similarity . Therefore , we designed a probe for in situ hybridization that detects expression of all three genes simultaneously ( expression restricted to the odontoblast lineage in the adult incisor ) . Primer sequences can be found in the Supplementary file 3 . Sections used for immunofluorescence were counterstained with DAPI ( Vector Laboratories ) and mounted in 1% DABCO in glycerol . Signal amplification utilizing the TSA Plus Fluorescein System ( Perkin Elmer , NEL741001KT ) was performed for detection of BrdU and beta-galactosidase following incubation with appropriate biotinylated secondary antibodies . A Leica-TCS SP5 confocal microscope was used for imaging except for detection of p63 in Figure 4—figure supplement 1D . In this case , a Leica DFC500 camera was used with a Leica DM 5000B microscope . For chromatogenic detection of CDKN1A and SOX21 , the same sample preparation and antigen retrieval procedures were followed as for immunofluorescence detection . Overnight incubation with the primary antibody was followed by washes in phosphate-buffered saline ( PBS ) , incubation with the appropriate secondary antibody , washes in PBS , incubation with ABC complex ( VECTASTAIN Elite ABC HRP Kit , Vector Laboratories , PK6100 ) , washes in PBS , signal detection using a DAB Peroxide substrate kit ( Vector Laboratories , SK-4100 ) according to the manufacturer’s instructions , PBS washes and post-fixation in 4% PFA . Slides were mounted with Dako Faramount Aqueous Mounting Media ( Dako , S3025 ) . For visualization of BrdU following in situ hybridization , slides were blocked in 5% bovine serum albumin in PBS with 0 . 1% Tween20 following the color reaction step of the in situ hybridization procedure . Incubation with the primary antibody and subsequent steps were performed as described for detection of CDKN1A and SOX21 . Images were acquired using a Leica DFC500 camera on a Leica DM5000B microscope . Information regarding primary and secondary antibodies can be found in Supplementary file 4 . Images of BrdU and DAPI stained sagittal sections of the cervical loop regions of 3 experimental animals and 3 controls were acquired using a Leica-TCS SP5 confocal microscope . BrdU positive cells of the 5 most central sections per specimen were quantified manually using ImageJ , and a Welch two sample t-test was performed . | To maintain healthy tissues and organs in adult animals , the cells that die or become damaged need to be replaced . This process is made possible by adult stem cells , which can divide to produce more stem cells ( via a process called self-renewal ) or specialize into other types of cells . This means that stem cells can maintain their own population by self-renewal while continually being able to generate specialized cells that replenish tissues and organs . Mouse incisor teeth are useful models to understand how adult organs are regenerated because , unlike human teeth , the incisor teeth of mice and other rodents grow continuously throughout the life of the animal . The tip of the mouse incisor is eroded as the animal eats , resulting in the loss of cells . A group of adult stem cells at the base of the tooth produce new cells that then move to the tip to replace the lost cells . Although virtually all cells in the body have the same set of genes , only small subsets are active in each cell type . It is possible to distinguish cells of different types by their patterns of gene activity . However , little is known about the gene expression patterns that distinguish stem cells and specialized cells in mouse incisors . Using a technique called gene co-expression analysis , Seidel et al . set out to identify all the genes that are active in stem cells and their descendants at the base of the mouse incisor . The experiments reveal the patterns of activity of thousands of genes , providing a clearer picture of the different cell types present and the biological processes at play . Seidel et al . then used other techniques to identify two genes that can be used as markers to identify distinct types of stem cells in the incisor . The next steps following on from this work will be to understand in more detail how stem cells behave in renewing the incisor . In the future , these findings may help guide the use of stem cells in regenerating human teeth and other organs . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"stem",
"cells",
"and",
"regenerative",
"medicine",
"developmental",
"biology"
] | 2017 | Resolving stem and progenitor cells in the adult mouse incisor through gene co-expression analysis |
Retrieving a memory can modify its influence on subsequent behavior . We develop a computational theory of memory modification , according to which modification of a memory trace occurs through classical associative learning , but which memory trace is eligible for modification depends on a structure learning mechanism that discovers the units of association by segmenting the stream of experience into statistically distinct clusters ( latent causes ) . New memories are formed when the structure learning mechanism infers that a new latent cause underlies current sensory observations . By the same token , old memories are modified when old and new sensory observations are inferred to have been generated by the same latent cause . We derive this framework from probabilistic principles , and present a computational implementation . Simulations demonstrate that our model can reproduce the major experimental findings from studies of memory modification in the Pavlovian conditioning literature .
While retrieval-induced memory modification has been documented in a variety of domains—including procedural ( Censor et al . , 2010; Walker et al . , 2003 ) , episodic ( Hupbach et al . , 2007; Karpicke and Roediger , 2008 ) , and instrumental ( Lee et al . , 2006b; Xue et al . , 2012 ) learning—we focus on Pavlovian conditioning , because it offers some of the most elementary and well-studied examples . During the acquisition phase of a typical Pavlovian conditioning experiment , a motivationally neutral conditional stimulus ( CS; e . g . , tone ) is repeatedly paired with a motivationally reinforcing unconditional stimulus ( US; e . g . , a shock ) . This repeated pairing results in the animal producing a conditioned response ( CR; e . g . , freezing ) to the CS . In a subsequent extinction phase , the CS is presented alone , and the animal gradually ceases to produce the CR . A final test phase , after some delay , probes the animal’s long-term memory of the CS-US relationship by presenting the CS alone . In a classic experiment using a Pavlovian fear conditioning task , Misanin et al . ( 1968 ) found that electroconvulsive shock had no effect on a fear memory acquired a day previously; however , if the animal was briefly reexposed to the acquisition cue prior to electroconvulsive shock , the animal subsequently exhibited loss of fear . This finding was followed by numerous similar demonstrations of post-retrieval memory modification ( see Riccio et al . , 2006 , for a historical overview ) . Contemporary neuroscientific interest in this phenomenon was ignited by Nader et al . ( 2000 ) , who showed that retrograde amnesia for an acquired fear memory could be produced by injection of a protein synthesis inhibitor ( PSI ) into the lateral nucleus of the amygdala shortly after reexposure to the acquisition cue . Subsequent studies have provided a detailed neural and behavioral characterization of post-retrieval memory modification , describing a large cast of molecular mechanisms ( Tronson and Taylor , 2007 ) and several boundary conditions on its occurrence ( Dudai , 2012; Duvarci and Nader , 2004; Nader and Hardt , 2009 ) . For instance , it has been shown that stronger and older memories are harder to modify following retrieval ( Suzuki et al . , 2004 ) , and that the modification is cue-specific ( Doyère et al . , 2007 ) . Importantly , there is now evidence that memory modification can be obtained with a purely behavioral procedure . In particular , Monfils et al . ( 2009 ) and Schiller et al . ( 2010 ) showed , in rats and in humans , that reexposing a subject to the cue shortly ( 10 min to 1 hr ) before extinction training is sufficient to reduce conditioned responding at test . This finding presents a deep puzzle for associative learning theory , since the cue reexposure is operationally an extinction trial and hence it is unclear what makes it special . One of our main goals will be to unravel this puzzle , showing how cue reexposure influences probabilistic beliefs about latent causes such that they are eligible for updating by the subsequent extinction training . This body of work has traditionally been understood as probing mechanisms of ‘reconsolidation’ ( Nader et al . , 2000; Przybyslawski and Sara , 1997 ) , under the hypothesis that memory retrieval renders the memory trace unstable , setting in motion a protein-synthesis-dependent process of synaptic stabilization . This process is thought to resemble initial post-learning consolidation , whereby a newly encoded memory gradually becomes resistant to disruption . However , this terminology is heavily theory-laden , and the explanatory adequacy of both consolidation and reconsolidation have been repeatedly questioned ( Ecker et al . , 2015; Miller and Springer , 1973 , Miller and Matzel , 2006 ) . We therefore avoid using this terminology to refer to empirical phenomena , favoring instead the less tendentious ‘post-retrieval memory modification . ’ The relationship of our work to consolidation and reconsolidation will be revisited in the Discussion . Before addressing the key memory modification phenomena , we first situate them within a larger set of issues in Pavlovian conditioning . These problems provide the starting point for our new theory . Classical theories of associative learning , such as the Rescorla-Wagner model ( Rescorla and Wagner , 1972 ) , posit that over the course of acquisition in a Pavlovian conditioning experiment , the animal learns an association between the CS and the US , and the magnitude of the CR reflects the strength of this association . The main weakness of the Rescorla-Wagner model , and many similar models ( Pearce and Bouton , 2001 ) , is its prediction that presenting the CS repeatedly by itself ( ‘extinction’ ) will erase the CS-US association formed during acquisition—in other words , the model predicts that extinction is unlearning . It is widely accepted that this assumption , shared by a large class of models , is wrong ( Delamater , 2004; Dunsmoor et al . , 2015; Gallistel , 2012 ) . Bouton ( 2004 ) reviewed a range of conditioning phenomena in which putatively extinguished associations are recovered in a post-extinction test phase . For example , simply increasing the time between extinction and test is sufficient to increase responding to the extinguished CS , a phenomenon known as spontaneous recovery ( Pavlov , 1927; Rescorla , 2004 ) . Another example is reinstatement: reexposure to the US alone prior to test increases conditioned responding to the CS ( Bouton and Bolles , 1979b; Pavlov , 1927; Rescorla and Heth , 1975 ) . Conditioned responding can also be recovered if the animal is returned to the acquisition context , a phenomenon known as renewal ( Bouton and Bolles , 1979a ) . Bouton ( 1993 ) interpreted the attenuation of responding after extinction in terms of the formation of an extinction memory that competes for retrieval with the acquisition memory; this retroactive interference can be relieved by a change in temporal context or the presence of retrieval cues , thereby leading to recovery of the original CS ( see also Miller and Laborda , 2011 ) . Central to retrieval-based accounts of conditioning is the idea that the associations learned during acquisition are linked to the spatiotemporal context of the acquisition session , and as a result they are largely unaffected by extinction . Likewise , extinction results in learning that is linked to the spatiotemporal context of the extinction session . The manipulations reviewed above are hypothesized to either reinstate elements of the acquisition context ( e . g . , renewal , reinstatement ) or attenuate elements of the extinction context ( e . g . , spontaneous recovery ) . These modifications of contextual elements effectively change the accessibility of particular associative memories . According to this view , modification of the acquisition memory ( in particular , its accessibility ) should occur when the acquisition and extinction phases are linked to the same spatiotemporal context . The major stumbling block is that it is unclear what should constitute a spatiotemporal context: What are its constitutive elements , under what conditions are they invoked , and when should new elements come into play ? Existing theories have operationalized context in several ( not mutually exclusive ) ways: as observable stimuli [e . g . , the conditioning box; Bouton , 1993] , recent stimulus and response history ( Capaldi , 1994 ) , or a random flux of stimulus elements ( Estes , 1950 , 1955 ) . However , no computational implementation has been shown to capture the full range of memory modification phenomena that we discuss below .
In this section , we develop a latent cause theory of Pavlovian conditioning that treats context ( operationalized as the history of sensory data ) as the input into a structure learning system , which outputs a parse of experience into latent causes—hypothetical entities in the environment that govern the distribution of stimulus configurations ( Courville , 2006; Courville et al . , 2006; Gershman et al . , 2010; Gershman and Niv , 2012; Gershman et al . , 2013a , 2015; Soto et al . , 2014 ) . Like the Rescorla-Wagner model ( Figure 1A ) , our theory posits the learning of CS-US associations , but these associations are modulated by the animal’s probabilistic beliefs about latent causes . New causes are inferred when existing causes fail to accurately predict the currently observed CS-US contingency ( Figure 1B ) . This allows the theory to move beyond the ‘extinction=unlearning’ assumption by positing that different latent causes are inferred during acquisition and extinction , and thus two different associations are learned ( see also Redish et al . , 2007 ) . Memory modification arises when CS reexposure provides evidence to the animal that the latent cause assigned to the acquisition phase is once again active , making that cause’s associations eligible for updating ( or disruption by amnestic agents like PSIs ) . 10 . 7554/eLife . 23763 . 003Figure 1 . Model schematic . ( A ) The associative structure underlying the Rescorla-Wagner model . The associative strength between a conditioned stimulus ( CS ) and an unconditioned stimulus ( US ) is encoded by a scalar weight , w , that is updated through learning . ( B ) The associative structure underlying the latent-cause-modulated model . As in the Rescorla-Wagner model , associative strength is encoded by a scalar weight , but in this case there is a collection of such weights , each paired with a different latent cause . The US prediction is a linear combination of weights , modulated by the posterior probability that the corresponding latent cause is active . Alternatively , this model can be understood as consisting of three-way associations between the latent cause , the CS and the US . ( C ) A high-level schematic of the computations in the latent-cause model . Associative learning , in which the associative weights are updated ( using the delta rule ) conditional on the latent-cause posterior , alternates with structure learning , in which the posterior is updated ( using Bayes’ rule ) conditional on the weights . DOI: http://dx . doi . org/10 . 7554/eLife . 23763 . 003 The theory consists of two interacting sub-systems ( Figure 1C ) : an associative-learning system updates a set of CS-US associations using a delta rule ( Rescorla and Wagner , 1972; Sutton and Barto , 1998; Widrow and Hoff , 1960 ) , while a structure learning system updates an approximation of the posterior distribution over latent causes using Bayes’ rule . It is useful to envision the associative-learning system as almost identical to the Rescorla-Wagner model , with the key difference that the system can maintain multiple sets of associations between any CS-US pair ( one for each latent cause; Figure 1B ) instead of just a single set . Given a particular CS configuration ( e . g . , tone in a red box ) , the multiple associations are combined into a single prediction of the US by averaging the US prediction for each cause , weighted by the posterior probability of that cause being active . This posterior probability takes into account not only the conditional probability of the US given the CS configuration , but also the probability of observing the CS configuration itself . In the special case that only a single latent cause is inferred by the structure learning system , the associative learning system’s computations are almost identical to the Rescorla-Wagner model ( see the Materials and methods ) . To infer the posterior distribution over latent causes , the structure learning system makes certain assumptions about the statistics of latent causes ( the animal’s ‘internal model’ ) . Informally , the main assumptions we impute to the animal are summarized by two principles: When combined with a number of auxiliary assumptions , these principles specify a complete generative distribution over stimulus configurations and latent causes—the animal’s internal model . We now describe the theory in greater technical detail . In the section ‘Understanding Extinction and Recovery , ' we walk through a simple example with a single CS . Before modeling specific experimental paradigms , in this section we lay out some general intuitions for how our model deals with extinction and recovery . In previous work ( Gershman et al . , 2010 ) , we argued that the transition from acquisition to extinction involves a dramatic change in the statistics of the animal’s sensory inputs , leading the animal to assign acquisition and extinction trials to different latent causes . The result of this partitioning is that the acquisition associations are not unlearned during extinction , and hence can be later recovered , as is observed experimentally ( Bouton , 2004 ) . Thus , according to our model , the key to persistent reduction in fear is to finesse the animal’s sensory statistics such that the posterior distribution over latent causes favors assigning both acquisition and extinction phases to the same latent cause . One way to understand the factors influencing the posterior distribution over latent causes is in terms of prediction error , the discrepancy between what the animal expects and what it observes . This term typically refers to a US prediction error ( i . e . , was the US predicted or not ? ) , but our analysis applies to CS prediction errors as well: in our framework , anything that is not expected under the current latent cause evokes a prediction error . The prediction error plays two roles in our model: it is an associative learning signal that teaches the animal how to adjust its associative weights , and it is a segmentation signal indicating when a new latent cause is active . When the animal has experienced several CS-US pairs during acquisition , it develops an expectation that is then violated during extinction , producing a prediction error . Learning rules such as Rescorla-Wagner’s are ‘error-correcting’ as they modify associations or values so as to reduce future prediction errors . In our model , however , the prediction error can be reduced in two different ways: either by associative learning ( e . g . , unlearning the CS-US association ) or by structure learning ( e . g . , assigning the extinction trials to a new latent cause ) . Initially , the prior simplicity bias towards a small number of latent causes favors unlearning , but persistent accumulation of these prediction errors over the course of extinction eventually makes the posterior probability of a new cause high . Thus , our framework recapitulates and formalizes the idea that standard acquisition and extinction procedures eventually lead to the formation of two memories or associations , one for CS-US and one for CS-noUS . The trade-off between the effects of prediction errors on associative and structure learning is illustrated in Figure 3 . When prediction errors are small , the posterior probability of the acquisition latent cause is high ( leading to modification of the original memory ) but the amount of CS-US weight change is small as there is little discrepancy between what was predicted and what was observed; if prediction errors are very large , the posterior probability of the acquisition latent cause is low ( leading to formation of a new memory ) , and the change in the weight corresponding to the original memory is again small . In theory , therefore , there should exist an intermediate ‘sweet spot’ for extinction learning where the prediction errors are large enough to induce considerable weight change but small enough to avoid inferring a new latent cause . Later we describe one behavioral paradigm ( the Monfils-Schiller paradigm ) that achieves this sweet spot ( see also Gershman et al . , 2013 ) . 10 . 7554/eLife . 23763 . 005Figure 3 . Cartoon of the model’s predictions for fear extinction . The X-axis represents the size of the prediction error during extinction , and the Y-axis represents the change ( after learning ) in the weight for US prediction for the ‘acquisition latent cause’ ( i . e . , the latent cause inferred by the animal during conditioning ) . DOI: http://dx . doi . org/10 . 7554/eLife . 23763 . 005 To get a feeling for how the model’s response to prediction errors depends on previous experience , consider a simple conditioning paradigm in which a single CS has been paired N times with the US ( 𝒟1:N={xt=1 , rt=1}t=1N ) , with a fixed ITI ( τ ( t ) -τ ( t-1 ) =1 ) . Under most parameter settings , this will result in all the acquisition trials being assigned to a single latent cause ( hence we ignore the cause subscript k in this example and refer to the single cause as the ‘acquisition latent cause’ ) . Now consider what happens when a single extinction trial ( xN+1=1 , rN+1=0 ) is presented . The posterior over latent causes ( Equation 5 ) is proportional to the product of 3 terms: ( 1 ) The prior over latent causes , ( 2 ) the likelihood of the US , and ( 3 ) the likelihood of the CS . The third term plays a negligible role , since the CS does not change across acquisition and extinction , and hence no CS prediction error is generated . As N grows , the prior probability of the acquisition latent cause generating the extinction trial as well increases , due to the simplicity bias of the prior over latent causes . However , associative learning of the weight vector counteracts this effect , since the US expectation , and hence the size of the prediction error due to the absence of the US ( encoded in the likelihood term ) , also grows with N , asymptoting once the US prediction is fully learned . In particular , sensitivity to the US prediction error increases as σr2 decreases ( higher confidence in US predictions ) and α increases ( weaker simplicity bias ) . Parameter-dependence is examined systematically in the next section . In order to understand some of the empirical phenomena described below , we must also explain why spontaneous recovery of the CR after extinction occurs . In our model , this occurs because the posterior probability of the acquisition cause increases as the extinction-test interval is lengthened , due to the temporal compression property of the power law temporal kernel 𝒦 that we chose . As explained above , this kernel has the important property that older timepoints are ‘compressed’ together in memory: latent causes become more equiprobable under the prior as the time between acquisition and test increases . A similar idea was used by Brown et al . ( 2007 ) in their model of episodic memory to explain recency effects in human list learning experiments . Thus , the advantage of the extinction cause over the acquisition cause at test diminishes with the extinction-test interval . One implication of this analysis is that spontaneous recovery should never be complete , since the prior probability of the acquisition cause can never exceed the probability of the extinction cause ( though the ratio of probabilities increases monotonically towards one as the extinction-test interval increases ) ; this appears generally consistent with empirical data ( Rescorla , 2004 ) . There are a few examples of seemingly complete spontaneous recovery ( Quirk , 2002; Brooks and Bouton , 1993; Bouton and Brooks , 1993 ) . This is inconsistent with our theory and would require additional or different mechanisms , but it is currently unclear how common complete spontaneous recovery is , or what factors determine its completeness . While extinction procedures after fear conditioning are , in general , not effective in producing permanent and generalizable reduction of fear , two influential studies ( Monfils et al . , 2009; Schiller et al . , 2010 ) demonstrated that a single reexposure ( ‘retrieval trial’ ) of a CS that had been associated with a shock , 24 hr after acquisition and 10–60 min before extinction , leads to persistent reduction of fear as measured by renewal , reinstatement and spontaneous recovery tests . Importantly , this effect did not require pharmacological interventions such as PSIs , and it was evident in both rodents ( Monfils et al . , 2009 ) and humans ( Schiller et al . , 2010 ) . These studies also revealed that: ( 1 ) reduction of fear in humans is still evident a year later; ( 2 ) the reduction is specific to the cue-reactivated memory; and ( 3 ) increasing the retrieval-extinction interval to 6 hr eliminates the effect . That is , extinction after a retrieval trial is more effective at modifying the original association than regular extinction , but this only holds for extinction sessions administered relatively promptly after the retrieval cue . This latter finding suggests that the retrieval cue engages a time-limited plasticity window , in which extinction operates . These findings have been replicated several times in rodents ( Auchter et al . , 2017; Baker et al . , 2013; Clem and Huganir , 2010; Jones et al . , 2013; Olshavsky et al . , 2013b , 2013a; Ponnusamy et al . , 2016; Rao-Ruiz et al . , 2011 ) and humans ( Agren et al . , 2012; Oyarzún et al . , 2012; Schiller et al . , 2013; Steinfurth et al . , 2014 ) , though the generality of the paradigm remains controversial ( Chan et al . , 2010; Costanzi et al . , 2011; Kindt and Soeter , 2013; Soeter and Kindt , 2011; Kredlow et al . , 2016 ) . It is important to recognize that the so-called ‘retrieval trial’ is operationally no different from an extinction trial—it is a CS presented alone . Essentially , the principal salient difference between the Monfils-Schiller paradigm and regular extinction training is that in the Monfils-Schiller paradigm , the interval between the first and second extinction trials is substantially longer than the intervals between all the other trials . Another difference ( which we address later ) is that in the Monfils-Schiller paradigm , the subject spends the retrieval-extinction interval outside the acquisition context , in its home cage . This phenomenon is thus puzzling for most—if not all—theories of associative learning . What happens during this one interval that dramatically alters later fear memory ? Below we provide a normative computational account of this phenomenon based on our framework for Pavlovian conditioning . We also suggest explanations for some of the inconsistencies across studies . Model simulations of the Monfils-Schiller paradigm are shown in Figure 11 . We simulated three conditions , differing only in the retrieval-extinction interval ( REI ) : No Ret ( REI = 0 , that is , extinction begins with no separate retrieval trial ) , Ret-short ( REI = 3 ) , Ret-long ( REI = 100 ) . Time is measured in arbitrary units here; see the Materials and methods for a description of how these units roughly map on to real time . As observed experimentally , in our simulations all groups ceased responding by the end of extinction . Both Ret-long and No Ret showed spontaneous recovery after a long extinction-test delay . In contrast , in the Ret-short condition there was no spontaneous recovery of fear at test . Examining the posterior distributions over latent causes in the different conditions ( Figure 11B–D ) , we see that the extinction trials were assigned to a new latent cause in the No Ret and Ret-long conditions , but to the acquisition cause in the Ret-short condition . 10 . 7554/eLife . 23763 . 013Figure 11 . Model predictions for the Monfils-Schiller paradigm . ( A ) Simulated conditioned response ( CR ) during acquisition ( Acq; 3 CS-US pairs ) , retrieval ( Ret; 1 CS presentation 24 hr after acquisition , followed by no interval , a short interval , or a long interval before the next phase ) , extinction ( Ext; CS-alone presentations ) and a test phase 24 hr later . Three conditions are shown: No-Ret ( no interval between retrieval and extinction; the ‘Ret’ trial depicted here is the first trial of extinction ) , Ret-short ( retrieval with a short post-retrieval interval ) , and Ret-long ( retrieval with a long post-retrieval interval ) . ( B–D ) The posterior probability distribution over latent causes ( denoted C1 , C2 and C3 ) in each condition . Probabilities for only the top three highest-probability causes are shown . DOI: http://dx . doi . org/10 . 7554/eLife . 23763 . 013 Our theoretical explanation of data from the Monfils-Schiller paradigm rests critically on the ‘rumination’ process ( i . e . , iterative updating according to the EM algorithm ) that occurs in the interval between the first and second extinction trials ( the REI ) . Since there is some probability that the original ( ‘acquisition’ ) latent cause is active during the REI , the first iteration of associative learning in the EM algorithm will reduce the CS-US association for that latent cause . On the next iteration , the model will be even more likely to infer that the original latent cause is active ( since the CS-US association strength is smaller , the prediction error induced by the CS appearing without the US will be even smaller ) . As a result of this increased probability that the original latent cause is active , the CS-US association will be reduced even more . In our model , the number of EM iterations ( up to a maximum of 3 iterations , with one iteration per timestep; see model description ) depends on the length of the REI . More iterations cause the original association to be further weakened after the first retrieval trial , and therefore spontaneous recovery of the original fear memory at test is attenuated ( Figure 12A ) . 10 . 7554/eLife . 23763 . 014Figure 12 . Dynamics of associative and structure learning during the retrieval-extinction interval in the Monfils-Schiller paradigm . ( A ) The X-axis represents the associative weight corresponding to the acquisition latent cause . The Y-axis represents the posterior probability that the acquisition latent cause is active for the retrieval trial . Each numbered square indicates a particular iteration during the retrieval-extinction interval , with '0' indicating the last trial of acquisition . Initially , the prediction error causes the posterior to favor a new latent cause rather than the old acquisition cause , however , over the course of three iterations , incremental reductions in the associative weight pull the posterior probability higher by making the retrieval trial more likely under the acquisition cause . ( B ) As the retrieval-extinction interval grows longer , the probability of assigning the first extinction trial to the acquisition cause changes non-monotonically . Two non-reinforced trials very close in time are likely to come from a new latent cause , thus the posterior probability of the acquisition cause generating these trials starts low . It peaks at a larger retrieval-extinction interval; as this interval increases , the acquisition cause’s associative strength is incrementally reduced , thereby making the extinction trials more likely under the acquisition cause . The curve then gradually diminishes due to the time-sensitive prior that causes temporally separated events to be more likely to be generated by different causes ( Equation 3 ) . Each EM iteration takes a single timestep , and at least 1 EM iteration is always performed , up to a maximum of 3 , depending on the intertrial interval . DOI: http://dx . doi . org/10 . 7554/eLife . 23763 . 014 When the interval is too short ( as in the No Ret condition ) , there is insufficient time ( i . e . , only a single EM iteration ) to reduce the CS-US association and thus later extinction trials are preferentially assigned to a new latent cause rather than the acquisition cause . When the interval is too long ( as in the Ret-long condition ) , although the CS-US association of the acquisition latent cause is reduced during the REI , extinction trials will be preferentially assigned to a new latent cause due to the time-sensitive prior that suggests that events far away in time are generated by different causes . Thus in this condition as well , the original association is not attenuated by the extinction trials any further , and spontaneous recovery of fear occurs at test . It is only in the intermediate condition , the short REI , that the EM iterations reduce the prediction of the US by the acquisition latent cause sufficiently so as to allow later extinction trials to be assigned to this same latent cause , therefore reducing the prediction of the US by this cause even further , effectively ‘erasing’ the fear memory . This nonmonotonic dependence on the REI is shown in Figure 12B . Note that our model predicts that all the boundary conditions discussed earlier should apply to the Monfils-Schiller paradigm . Thus , for example , older memories should be more difficult to disrupt , even with an intermediate REI . We revisit this point in the next section . The importance of iterative adjustment during the retrieval-extinction interval suggests that distracting or occupying animals during the interval should disrupt the Monfils-Schiller effect . For example , our theory predicts that giving rodents a secondary task to perform during the interval will prevent the iterative weakening of the CS-US association of the acquisition cause , leading to assignment of extinction trials to a new latent cause ( as in regular extinction ) and to later recovery of fear . Alternatively , it might be possible to enhance the effect by leaving the animal in the conditioning chamber during the interval; the chamber would serve as a reminder cue , potentially preventing the animal from getting distracted . On the other hand , the conditioning chamber is associated with stress , so rumination may be about shocks rather than their absence . Thus , it might be that rumination in the safe haven of the homecage is most effective at producing subsequent memory modification .
One of the most intriguing recent findings in the memory modification literature was the discovery of a noninvasive behavioral treatment that is effective at attenuating recovery of conditioned fear ( Agren et al . , 2012; Monfils et al . , 2009; Schiller et al . , 2010 ) . Monfils , Schiller and their colleagues demonstrated ( in both rodents and humans ) that performing extinction training within a short interval following a retrieval cue ( an unreinforced CS presentation ) reduced later recovery of fear . The effect was later demonstrated in appetitive learning ( Ma et al . , 2012 ) and contextual fear conditioning ( Flavell et al . , 2011; Rao-Ruiz et al . , 2011 ) . The Monfils-Schiller paradigm has also been applied to drug-associated memory , attenuating drug-seeking in rats and cue-induced heroin craving in human addicts ( Xue et al . , 2012 ) , as well as reducing cocaine-primed reinstatement of conditioned place preference ( Sartor and Aston-Jones , 2014 ) and context-induced reinstatement of alcoholic beer seeking ( Millan et al . , 2013 ) in rats . The Monfils-Schiller paradigm is theoretically tantalizing because it is not a priori clear what is the difference between the retrieval trial and the first trial of any extinction session—why is it that the CS-alone trial in the Monfils-Schiller paradigm acts as a ‘retrieval cue’ , while the first CS-alone trial of a regular extinction session does not ? Previous explanations had suggested that the retrieval cue starts a reconsolidation process , whereas the original ( recalled ) memory is rendered labile , and can be modified while it is being reconsolidated into long term memory . The idea was that the extinction session then modifies this labile memory , permanently rewriting it as a less fearful memory ( Monfils et al . , 2009 ) . However , it is not clear why this should not happen in regular extinction , where the first extinction trial can also be seen as a retrieval cue that initiates a reconsolidation cascade . The effectiveness of this paradigm thus seems to challenge our basic understanding of the interplay between learning and memory processes . Our theory resolves this puzzle by stressing the role of the extended period of learning ( in our model , additional iterations of the EM algorithm ) during the long retrieval-extinction gap , in which the rat is left in its home cage to ‘ruminate’ about its recent experience . Thus our explanation rests not on the existence of a separate reconsolidation process that is invoked by the retrieval trial , but rather on the same learning and memory mechanisms that are at play in acquisition and extinction—the idea that inference about the latent structure of the environment affects whether new information will update an old association , or whether it will be attributed to a new memory ( new latent cause ) . In this sense , according to our theory , the ‘retrieval’ trial is , in fact , not different from any other trial , and perhaps a more accurate nomenclature would be to call the retrieval-extinction interval an ‘updating interval’ rather than focus on a ‘retrieval cue’ . Despite its successes , the effectiveness of the Monfils-Schiller paradigm has been controversial , with several replication failures ( Chan et al . , 2010; Costanzi et al . , 2011; Ishii et al . , 2015; Kindt and Soeter , 2013; Ma et al . , 2012; Soeter and Kindt , 2011 ) . Auber et al . ( 2013 ) described a number of methodological differences between these studies , possibly delineating boundary conditions on the Monfils-Schiller paradigm . Inspired by this suggestion , we showed through simulations that the consequences of several methodological differences ( acquisition-retrieval interval and context similarity ) are indeed predicted by our theory . Nevertheless , important boundary conditions on the length and characteristics of the retrieval-extinction interval remain to be studied; for instance , does it have to be longer than 10 min ( as has been done in previous experiments ) or is the minimum length of this gap more parametrically dependent on the overall pace of new information ( e . g . , the length of the ITIs at acquisition ) . From a neurobiological standpoint , recent work has lent plausibility to the claim that the Monfils-Schiller paradigm erases the CS-US association learned during acquisition . After fear conditioning , there is an upregulation of AMPA receptor trafficking to the post-synaptic membrane at thalamus-amygdala synapses , and memory is impaired if this trafficking is blocked ( Rumpel et al . , 2005 ) , suggesting that changes in post-synaptic AMPA receptor density may be the neural substrate of associative learning in fear conditioning . Monfils et al . ( 2009 ) reported increased phosphorylation of AMPA receptors in the lateral amygdala after the retrieval trial ( a possible correlate of memory labilization ) , and also found that a second CS presented one hour after the first reversed the increase in AMPAr phosphorylation . Clem and Huganir ( 2010 ) found that extinction following retrieval resulted in synaptic removal of calcium-permeable AMPA receptors . The latter finding is significant in that it indicates a reversal of the synaptic changes that occurred during conditioning , supporting the view that the Monfils-Schiller paradigm results in unlearning of the original CS-US association . Furthermore , the Monfils-Schiller paradigm has been shown to induce neural modifications that are distinct from standard extinction ( Lee et al . , 2015; Tedesco et al . , 2014 ) . Our theoretical analysis is consistent with these findings . We showed in simulations that during the retrieval-extinction interval , an associative learning process is engaged ( and continues to be engaged during extinction training ) that decrements the CS-US association , whereas in our model standard extinction engages a structure learning process that assigns the extinction trials to a new latent cause , creating a new memory trace without modifying the original memory . Although we have so far not committed to any specific neural implementation of our model , we believe it fits comfortably into the computational functions of the circuit underlying Pavlovian conditioning . Here we propose a provisional mapping onto this circuit , centering on the amygdala and the ‘hippocampal-VTA loop’ ( Lisman and Grace , 2005 ) connecting the hippocampus and the ventral tegmental area in the midbrain . Our basic proposal is inspired by two lines of research , one on the role of hippocampus in structure learning ( Aggleton et al . , 2007; Gershman et al . , 2010 , 2014 ) , and one on the role of the dopamine system and the amygdala ( Blair et al . , 2001 ) in associative learning . In previous work , we have suggested that the hippocampus is a key brain region involved in partitioning streams of experience into latent causes ( Gershman et al . , 2010 , 2014 ) . This view resonates with earlier models emphasizing the role of the hippocampus in encoding sensory inputs into a statistically compressed latent representation ( Fuhs and Touretzky , 2007; Gluck and Myers , 1993; Levy et al . , 2005 ) . Some of the evidence for this view comes from studies showing that context-specific memories depend on the integrity of the hippocampus ( e . g . , Honey and Good , 1993 ) , indicating that animals without a hippocampus cannot ‘carve nature at its joints’ ( i . e . , partition observations into latent causes; see Gershman and Niv , 2010; Gershman et al . , 2015 ) . Within the current model , we propose that the dentate gyrus ( DG ) activates latent representations of the sensory inputs in area CA3 . Each of these representations corresponds to a latent cause , and their level of activation is proportional to their prior probability ( Equation 3 ) . Mechanistically , these representations may be encoded in attractors by the dense recurrent collaterals that are characteristic of CA3 ( McNaughton and Morris , 1987 ) . An important aspect of our model is that the repertoire of latent causes can expand adaptively . One potential mechanism for creating new attractors is neurogenesis of granule cells in the DG ( Becker , 2005 ) . This account predicts that the role of neurogenesis in creating new attractors should be time-sensitive in a manner comparable to the latent cause prior ( i . e . , it should implement the contiguity principle ) . Consistent with this hypothesis , Aimone et al . ( 2006 ) have suggested that immature granule cells , by virtue of their low activation thresholds , high resting potentials and constant turnover , cause inputs that are distant in time to map onto distinct CA3 representations . Furthermore , evidence suggests that new granule cells die over time if they are not involved in new learning ( Shors et al . , 2012 ) , offering another mechanism by which the contiguity principle could be implemented . Consistent with the idea that neurogenesis supports the partitioning of experience into latent causes , suppression of neurogenesis reduces both behavioral and neural discrimination between similar contexts ( Niibori et al . , 2012 ) . Importantly , many CA3 neurons are temporally selective , responding to individual contexts only if exposures are separated by long temporal intervals , and this selectivity depends on intact neurogenesis ( Rangel et al . , 2014 ) , as one would expect based on the contiguity principle . Our interpretation of neurogenesis predicts that its suppression ( e . g . , by irradiation of the DG ) , will force experiences separated by long temporal gaps to be assigned to the same latent cause , thus eliminating the age-based boundary condition on memory modification ( Alberini , 2007; Milekic and Alberini , 2002; Suzuki et al . , 2004 ) . There is widespread agreement that CS-US associations in auditory fear conditioning are encoded by synapses between the thalamus and the basolateral amygdala ( BLA; McNally et al . , 2011 ) . Accordingly , we suggest that the amygdala transmits a US prediction that is then compared to sensory afferents from the periacqueductal gray region of the midbrain . The resultant prediction error is computed in the ventral tegmental area ( VTA ) and transmitted by dopaminergic projections to both the amygdala and CA1 . Our theory makes the testable prediction that disrupting the neural substrates of associative learning , or potentiating the neural substrates responsible for inferring new latent causes , during the retrieval-extinction interval should block memory updating in the Monfils-Schiller paradigm . Thus , both deactivating the BLA or stimulating the DG ( e . g . , using optogenetic manipulations ) should block memory updating following retrieval . We also predict that the relative balance of activity in these two regions , measured for example using fMRI , should relate to individual differences in conditional responding in the test phase . The role of dopamine in associative learning is well established ( see Glimcher , 2011 for a review ) , and has been specifically implicated in Pavlovian fear conditioning ( Pezze and Feldon , 2004 ) , although the role of dopamine in aversive conditioning is still a matter of controversy ( Mirenowicz and Schultz , 1996; Brooks and Berns , 2013; Cohen et al . , 2012 ) . Dopamine gates synaptic plasticity in the basolateral amygdala ( Bissière et al . , 2003 ) , consistent with its hypothesized role in driving the learning of CS-US associations . We hypothesize that dopaminergic inputs to CA1 have an additional role: influencing the posterior distribution over latent causes . That is , dopamine prediction errors can be used to assess the similarity of current sensory inputs to those expected by the current configuration of latent causes . Large discrepancies will cause the generation of a new latent cause , to account for the current unpredicted sensory input ( see Figure 3 ) . The output of CA1 further feeds back into the VTA by way of the subiculum ( Lisman and Grace , 2005 ) , potentially providing a mechanism by which the posterior distribution over latent causes can modulate the prediction errors , as suggested by our model . In appetitive conditioning experiments , ( Reichelt et al . , 2013 ) have shown that dysregulating dopaminergic activity in the VTA prevented the destabilization of memory by NMDA receptor antagonists ( injected systemically following a retrieval trial ) , consistent with the hypothesis that dopaminergic prediction errors are necessary for memory updating after memory retrieval . It is not known whether this effect is mediated by dopaminergic projections to the hippocampus . A key claim of this paper is that associative and structure learning are coupled: learning about associations depends on structural inferences , and vice versa . Our rational analysis suggested that this coupling can be resolved by alternating between the two forms of learning , using a form of the EM algorithm ( Dempster et al . , 1977; Neal and Hinton , 1998 ) . While we do not believe that this is a literal description of the computational processes underlying learning , it is a useful abstraction for several reasons . First , EM is the standard method in machine learning for dealing with coupled problems of this form—namely , problems in which both latent variables and parameters are unknown . It is also closely related to variational inference algorithms ( see Neal and Hinton , 1998 ) , which have become a workhorse for scalable Bayesian computation . Second , variants of EM have become popular as theories of learning in the brain . For example , Friston ( 2005 ) suggests that it is a basic motif for synaptic plasticity in the cortex , and biologically plausible spiking neuron implementations have been put forth by Deneve ( 2008 ) and Nessler et al . ( 2013 ) . Third , as described in the Appendix , EM reduces to the Rescorla-Wagner model under particular parameter constraints . Thus , it is natural to view the model as a principled generalization of the most well-known account of Pavlovian conditioning . Fourth , the iterative nature of EM plays an important role in our explanation of the Monfils-Schiller effect: the balance between memory formation and modification shifts dynamically over multiple iterations , and we argued that this explains why a short period of quiescence prior to extinction training is crucial for observing the effect . ( Osan et al . , 2011 ) have proposed an autoassociative neural network model of memory modification that explains many of the reported boundary conditions in terms of attractor dynamics ( see Amaral et al . , 2008 also for a related model ) . In this model , acquisition and extinction memories correspond to attractors in the network , formed through Hebbian learning . Given a configuration of sensory inputs , the state of the network evolves towards one of these attractors . The retrieved attractor is then updated through Hebbian learning . In addition , a ‘mismatch-induced degradation’ process adjusts the associative weights that are responsible for the mismatch between the retrieved attractor and the current input pattern ( i . e . , the weights are adjusted to favor the input pattern ) . Mismatch is assumed to accumulate over the course of the input presentation . The degradation process in this model can be viewed as a kind of error-driven learning: When the network does not accurately encode the current input , the weights are adjusted to encode it more accurately in the future . In the case of extinction , this implements a form of unlearning . The relative balance of Hebbian learning and mismatch-induced degradation determines the outcome of extinction training: assuming that the original shock pattern is retrieved at the beginning of extinction , degradation weakens the shock pattern , whereas Hebbian learning strengthens the retrieved shock pattern . Administration of PSIs is modeled as temporarily eliminating the influence of Hebbian plasticity on the weight update . Osan et al . ( 2011 ) showed that their network model could account for a number of the boundary conditions on memory modification described above . For example , they simulated the effect of CS reexposure duration prior to PSI administration ( Eisenberg et al . , 2003; Suzuki et al . , 2004 ) and suggested that post-reexposure PSI administration should have a tangible effect on the shock memory only for short , but not too short reexposure durations ( i . e . , what we modeled as ‘short’ duration in our simulations of the PSI experiments ) : for very short reexposure trials , the shock memory is preferentially retrieved because it has already been encoded in an attractor as a consequence of acquisition ( i . e . , the shock memory is the dominant trace ) . The accumulated mismatch is small , and hence mismatch-induced degradation has little effect on the shock memory . Since the mismatch is close to zero and the effect of PSIs is to turn off Hebbian learning , the net effect of PSI administration following reexposure is no change in the memory . On long reexposure trials , the accumulated mismatch becomes large enough to favor the formation of a new attractor corresponding to the extinction memory ( i . e . , the no-shock memory is the dominant trace ) . In this case , PSI administration will have little effect on the shock memory , because after a sufficiently long duration Hebbian learning is operating on a different attractor . Only in the case of intermediate-length reexposure , mismatch is large enough to induce degradation of the shock attractor , but not large enough to induce the formation of a new , no-shock attractor . The PSI prevents Hebbian learning from compensating for this degradation by strengthening the shock attractor , so the result is a net decrease in the strength of the shock attractor . In addition to the parametric effect of reexposure duration on reconsolidation , Osan et al . ( 2011 ) also simulated the effects of memory strength ( more highly trained memories are resistant to disruption by PSI administration ) , the effects of NMDA receptor agonists ( which have the opposite effects of PSIs ) , and the effects of blocking mismatch-induced degradation ( the amnestic effect of PSI administration is attenuated ) . However , the model of Osan et al . ( 2011 ) is fundamentally limited by the fact that it lacks an explicit representation of time within and between trials . This prevents it from accounting for the results of the Monfils-Schiller paradigm: all the retrieval-extinction intervals should lead to the same behavior ( contrary to the empirical data ) . The lack of temporal representation also prevents it from modeling the effects of memory age on reconsolidation , since there is no mechanism for taking into account the interval between acquisition and reexposure . In contrast , our model explicitly represents temporal distance between observations , making it sensitive to changes in timing . 1212Conceivably , one could incorporate a time-sensitive mechanism into the Osan model by using a ‘temporal context’ signal that drifts slowly over time ( see Sederberg et al . , 2011 ) . Another , related issue with the model of Osan et al . ( 2011 ) is that in order to explain spontaneous recovery , it was necessary to introduce an ad hoc function that governs pattern drift during reexposure . This function—by construction—produces spontaneous recovery , but it is not obvious why pattern drift should follow such a function , and no psychological or neurobiological justification was provided . Nonetheless , an appealing feature of the Osan et al . ( 2011 ) model is its neurobiological plausibility . We know that attractor networks exist in the brain ( e . g . , in area CA3 of the hippocampus ) , and ( in certain circumstances ) support the kinds of learning described above . The model is appealing as it provides a simplified but plausible mapping from computational variables to biological substrates . As we discussed in the previous section , one way to think about latent causes at a neural level is in terms of attractors ( e . g . , in area CA3 ) . Thus , although the formal details of Osan et al . ( 2011 ) differ from our own , there may be neural implementations of the latent cause model that bring it closer to the formalism of the attractor network . However , in its current form , our model is not specified at the same biologically detailed level as the model of Osan et al . ( 2011 ) ; our model makes no distinction between Hebbian plasticity and mismatch-induced degradation , and consequently has nothing to say about pharmacological manipulations that selectively affect one or the other process , for example the disruption of mismatch-induced degradation by inhibitors of the ubiquitin-proteasome cascade ( Lee et al . , 2008 ) . One of the first formal accounts of spontaneous recovery from extinction was developed by Estes , ( 1955 ) . In his stimulus sampling theory , the nominal stimulus is represented by a collection of stimulus elements that change gradually and randomly over time . These stimulus elements enter into association with the US , such that the CR is proportional to the number of conditioned elements . When the CS is presented again at a later time , the CR it elicits will thus depend on the overlap between its current vector of stimulus elements and the vector that was present during conditioning . Extinction reverses the learning process , inactivating the currently active conditioned elements . However , some conditioned elements will not be inactive during the extinction phase ( due to stimulus sampling ) . As the interval between extinction and test increases , these elements will randomly re-enter the stimulus representation , thereby producing spontaneous recovery of the extinguished CR . This theory has since been elaborated in a number of significant ways to accommodate a wide variety of memory phenomena ( Howard , 2014 ) . While stimulus sampling theory , on the surface , appears quite different from our latent cause theory , there are some intriguing connections . The assumption that the same nominal stimulus can have different representations at different times is central to both accounts . Our theory posits latent stimulus elements ( causes ) that change over time , but these elements are not directly observable by the animal; rather , the structure learning system constructs a representation of these elements through Bayesian inference . Knowledge about gradual change is built into the prior through the contiguity principle . Like stimulus sampling theory , our theory views spontaneous recovery as a consequence of the extinction memory’s waning through random fluctuation . Again , this fluctuation is inferred rather than observed . The structure learning system acquires explicit distributional information about the latent causes—information that is absent from the stimulus sampling theory as developed by Estes ( 1955 ) . As a consequence , in our framework the representation of a stimulus contains information about its history and the history of other stimuli that were inferred to have been generated by the same latent cause . Because of the contiguity principle , stimuli that occur nearby in time are likely to have been generated by the same latent cause; this means that the ‘temporal context’ of a stimulus figures prominently in the distributional information stored by the structure learning system . The Temporal Context Model ( TCM; Howard et al . , 2005; Howard and Kahana , 2002; Sederberg et al . , 2008 ) can be viewed as a modern-day elaboration of the Estes stimulus-sampling model; rather than relying on random drift , it maintains a gradually changing ‘context vector’ of recent stimulus history that gets bound to stimulus vectors through Hebbian learning . The context vector can be used to cue retrieval of stimuli from memory ( as in free recall tasks ) , which in turn causes the reinstatement of context bound to the retrieved stimuli . One way to view our latent cause theory is as a particular rationalization of retrieved context models like TCM: the ‘context’ representation corresponds to the posterior over latent causes , retrieving context corresponds to inferring a latent cause , and updating the stimulus-context associations corresponds to updating the sufficient statistics of the posterior ( i . e . , structure learning ) . Indeed , precisely this correspondence was made by Socher et al . ( 2009 ) , where a latent cause model of text corpora was used as the underlying internal model for word lists . The latent cause model extends TCM by positing additional constraints on context drift . For example , in the latent cause model , the diagnosticity of sensory observations matters: a sensory observation that is highly diagnostic of a change in latent causes could have a very large effect on the posterior probabilities that the agent assigns to latent causes ( and thus its ‘context’ , if we consider latent causes to be coextensive with context ) . TCM in its original form does not incorporate any notion of diagnosticity—it merely computes a running average of sensory observations and retrieved contextual information . Bayesian versions of TCM , such as the one developed by Socher et al . ( 2009 ) , could potentially capture effects of diagnosticity , although such effects have not yet been systematically investigated . Connecting our theoretical work with retrieved context models like TCM allows us to make contact with a relevant segment of the human episodic memory literature studying post-retrieval memory modification ( Chan et al . , 2009; Chan and LaPaglia , 2013; Forcato et al . , 2007 , 2010; Hupbach et al . , 2007 , 2009 ) . In one line of research developed by Hupbach and colleagues , the researchers used a list-learning paradigm to show that reminding participants of one list ( A ) shortly before asking them to study a second list ( B ) produced an asymmetric pattern of intrusions at test: participants intruded a large number of items from list B when asked to recall list A , but not vice versa ( Hupbach et al . , 2007 ) . When no reminder was given , participants showed an overall low level of intrusions across list A and list B recall . One interpretation of these findings , in line with reconsolidation accounts of memory modification , is that the reminder caused the memory of list A to become labile , thereby allowing list B items to become incorporated into the list A memory . However , Sederberg et al . ( 2011 ) showed that the findings of Hupbach and colleagues could be accounted for by TCM ( see also Gershman et al . , 2013c for converging neural evidence ) , further suggesting that retrieved context models are relevant to understanding post-retrieval memory modification , but more work is needed to flesh out the correspondences sketched here . Briefly , a latent cause theory might be able to account for the Hupbach results if one assumes that a latent cause associated with list A is retrieved at at the beginning of list B ( analogous to the retrieval of the list A temporal context ) . In developing our theory of memory modification , we have studiously avoided the term ‘reconsolidation’ that appears ubiquitously throughout the literature we have modeled . Reconsolidation , like many concepts in the study of learning , has a dual meaning as both a set of empirical phenomena and as a theoretical hypothesis about the nature of those phenomena ( Rudy et al . , 2006 ) . The theoretical hypothesis is derived from the idea that newly formed memories are initially labile ( sensitive to disruption or modification ) , but over time undergo a protein synthesis-dependent ‘consolidation’ process that converts them into a stable molecular format largely resistant to disruption ( McGaugh , 1966 , 2000 ) . Here we are specifically discussing ‘synaptic consolidation’ that unfolds over seconds to minutes , in contrast to the ‘systems consolidation’ that unfolds over days to months and is hypothesized to involve the transfer of memory from hippocampus to neocortex ( Dudai , 2012 ) . The discovery that post-retrieval PSI administration was effective at disrupting memory long outside the consolidation window ( Nader et al . , 2000 ) inspired the idea that memory retrieval renders memory once again labile , requiring a second phase of consolidation ( named ‘reconsolidation’ ) to stabilize the memory . Like initial consolidation , reconsolidation requires protein synthesis , explaining why PSIs disrupt memory stabilization . We have avoided this terminology for several reasons . First , our theory is formulated at a level of abstraction that does not require commitment to a particular model of synaptic consolidation or reconsolidation . The process by which a memory becomes progressively resistant to disruption can be modeled in various ways ( e . g . , Fusi et al . , 2005; Clopath et al . , 2008; Ziegler et al . , 2015 ) , and it is currently unclear to what extent these biological mechanisms are consistent with normative models of learning ( see Gershman , 2014 , for one attempt at connecting the levels of analysis Gershman , 2014 ) . In particular , our theory does not incorporate an explicit consolidation process; increased resistance to disruption as a function of time arises from the contiguity principle , which implies that beliefs about a latent cause are less likely to be modified by new experience if a long interval has elapsed since the latent cause was believed to be active . Similarly , we do not explicitly model reconsolidation; post-retrieval lability arises from the increased probability that an old latent cause is active once again . A second reason we have avoided the consolidation/reconsolidation terminology is that the underlying theoretical claims face longstanding difficulties . Most theories of synaptic consolidation assume that amnestic agents like PSIs degrade the memory engram , and the post-learning ( or post-retrieval ) consolidation window reflects a period of time during which the trace is vulnerable to degradation . However , as a number of authors have pointed out ( Miller and Springer , 1973 , 2006; Lewis , 1979 ) , amnesia could alternatively arise through disrupted memory retrieval . In other words , the amnestic agent might make a memory harder to retrieve , while sparing the engram . This retrieval-oriented view is consistent with the observation that pre-test reminders ( e . g . , the US or training context ) can cause recovery from amnesia ( Lewis et al . , 1968b; Miller and Springer , 1972; Quartermain et al . , 1972 ) . Another difficulty facing consolidation theory is that the putative consolidation window could be reduced to less than 500 msec ( far shorter than the hypothesized speed of synaptic consolidation ) if animals were familiarized with the learning environment ( Lewis et al . , 1968a , 1969; Miller , 1970 ) . These difficulties inspired a family of retrieval-oriented theories that contrast starkly with storage-oriented consolidation theories ( see Miller and Matzel , 2006; Riccio et al . , 2006 for recent reviews ) . In an influential paper , Lewis et al . ( 1968b ) argued that experimental amnesia results from the impairment of a retrieval pathway rendered labile by reminders . Importantly , this impairment is temporary: a sufficiently salient reminder can activate the impaired retrieval pathway or possibly establish a new retrieval pathway . This idea is compatible with the stimulus sampling framework described in the previous section , where retrieval cues both activate prior memory traces and contribute new stimulus elements to the trace . Another retrieval-oriented theory , advocated by Riccio and colleagues ( Millin et al . , 2001; Riccio et al . , 2006 ) , views experimental amnesia as a state-dependent retrieval impairment . Specifically , the animal’s physiological state is a powerful retrieval cue , so by testing animals in the absence of the amnestic agent ( hence in a different physiological state ) , typical experimental amnesia experiments induce an encoding-retrieval mismatch ( Spear , 1973; Tulving and Thomson , 1973 ) . This idea lead to the counterintuitive prediction , subsequently confirmed , that administration of amnestic agents prior to test would reinstate the impaired memory ( Gisquet-Verrier et al . , 2015; Hinderliter et al . , 1975 ) . The difficulties facing consolidation theory do not necessarily pose problems for our theory , and indeed we showed that our theory predicts the transience of experimental amnesia as well as the reminder effect of pre-test PSI administration . Nonetheless , we see merit in both encoding-oriented and retrieval-oriented theories , since our theory asserts critical roles for both encoding and retrieval processes . In our simulations , we have shown that manipulations can affect both the strength of the CS-US association and also the probability of retrieving that association . One challenge to developing a unified theory of memory modification is that some of the basic facts are still disputed . Some authors have found that Pavlovian contextual fear memories become labile after retrieval ( Debiec et al . , 2002 ) , while others have not ( Biedenkapp and Rudy , 2004 ) , and yet others argue that memory modification is transient ( Frankland et al . , 2006; Power et al . , 2006 ) . A similar situation exists for instrumental memories: some studies have shown that instrumental memories undergo post-retrieval modification ( Fuchs et al . , 2009; Milton et al . , 2008 ) , while others have not ( Hernandez and Kelley , 2004 ) . The literature on post-retrieval modification of human procedural memories has also been recently thrown into doubt ( Hardwicke et al . , 2016 ) . There are many differences between these studies that could account for such discrepancies , including the type of amnestic agent , how the amnestic agent is administered ( systemically or locally ) , the type of reinforcer , and the timing of stimuli . Despite these ambiguities , we have described a number of regularities in the literature and how they can be accounted for by a latent cause theory of conditioning . The theory offers a unifying normative account of memory modification that links learning and memory from first principles .
The EM algorithm , first introduced by Dempster et al . ( 1977 ) , is a method for performing maximum-likelihood parameter estimation in latent variable models . In our model , the latent variables correspond to the vector of latent cause assignments , 𝐳1:t , the parameters correspond to the associative weights , 𝐖 , and the data correspond to the CS-US history , 𝒟1:t={𝐗1:t , 𝐫1:t} , where 𝐗1:t={𝐱1 , … , 𝐱t} and 𝐫1:t={r1 , … , rt} . Let Q ( 𝐳1:t ) be a distribution over 𝐳1:t . The EM algorithm can be understood as performing coordinate ascent on the functional ( 12 ) ℱ ( W , Q ) =∑z1:tQ ( z1:t|𝒟1:t ) logP ( z1:t , 𝒟1:t|W ) =∑z1:tQ ( z1:t|𝒟1:t ) log[P ( 𝒟1:t|z1:t , W ) P ( z1:t ) ] . By Jensen’s inequality , this functional is a lower bound on the log marginal likelihood of the data , logP ( 𝒟1:t|𝐖 ) =log∑𝐳1:tP ( 𝒟1:t , 𝐳1:t|𝐖 ) , which means that maximizing ℱ corresponds to optimizing the internal model to best predict the observed data ( Neal and Hinton , 1998 ) . The EM algorithm alternates between maximizing ℱ ( 𝐖 , Q ) with respect to 𝐖 and Q . Letting n indicate the iteration , {E-step}:Qn+1←argmaxQ ℱ ( Wn , Q ) {M-step}:Wn+1←argmaxW ℱ ( W , Qn+1 ) Alternating the E and M steps repeatedly , ℱ ( 𝐖 , Q ) is guaranteed to converge to a local maximum ( Neal and Hinton , 1998 ) . It can also be shown that ℱ ( 𝐖 , Q ) is maximized with respect to Q ( 𝐳1:t ) when Q=P ( 𝐳1:t|𝒟1:t , 𝐖 ) . Thus , the optimal E-step is exact Bayesian inference over the latent variables 𝐳1:t . There are two challenges facing a biologically and psychologically plausible implementation of this algorithm . First , the E-step is intractable , since it requires summing over an exponentially large number of possible latent cause assignments . Second , both steps involve computations operating on the entire history of observations , whereas a more plausible algorithm is one that operates online , one observation at a time ( Anderson , 1990 ) . Below we summarize an approximate , online form of the algorithm . To reduce notational clutter , we drop the n superscript ( indicating EM iteration ) , and implicitly condition on 𝐖 . The E-step corresponds to calculating the posterior using Bayes’ rule: ( 13 ) qtk=P ( zt=k|𝒟1:t ) =∑𝐳1:t-1P ( 𝒟t|zt=k , 𝒟1:t-1 ) P ( zt=k|𝐳1:t-1 ) ∑j∑𝐳1:t-1P ( 𝒟t|zt=j , 𝒟1:t-1 ) P ( zt=j|𝐳1:t-1 ) . Note that the number of terms in the summation over 𝐳1:t-1 grows exponentially over time; consequently , calculating the posterior exactly is intractable . Following ( Anderson , 1991 ) , we use a ‘local’ maximum a posteriori ( MAP ) approximation see for more discussion ( Sanborn et al . , 2010 ) : ( 14 ) qtk≈P ( 𝒟t|zt=k , 𝐳^1:t-1 , 𝒟1:t-1 ) P ( zt=k|𝐳^1:t-1 ) ∑jP ( 𝒟t|zt=j , 𝐳^1:t-1 , 𝒟1:t-1 ) P ( zt=j|𝐳^1:t-1 ) , where 𝐳^1:t-1 is defined recursively according to: ( 15 ) z^t=argmaxk P ( 𝒟t|zt=k , z^1:t−1 , 𝒟1:t−1 ) P ( zt=k|z^1:t−1 ) . In other words , the local MAP approximation is obtained by replacing the summation over partitions with the sequence of conditionally optimal cluster assignments . Although this is not guaranteed to arrive at the globally optimal partition ( i . e . , the partition maximizing the posterior over all timepoints ) , in our simulations it tends to produce very similar solutions to more elaborate approximations like particle filtering ( Gershman and Niv , 2010; Sanborn et al . , 2010 ) . The local MAP approximation has also been investigated in the statistical literature . Wang and Dunson ( 2011 ) found that it compares favorably to fully Bayesian inference , while being substantially faster . The first term in Equation 15 ( the likelihood ) is derived using standard results in Bayesian statistics ( Bishop , 2006 ) : ( 16 ) P ( 𝒟t|zt=k , 𝐳^1:t-1 , 𝒟1:t-1 ) =𝒩 ( rt;r^tk , σr2 ) ∏d=1D𝒩 ( xtd;x^tkd , νtk2 ) , where ( 17 ) r^tk=∑d=1Dxtdwkd ( 18 ) x^tkd=Ntkx¯tkdNtk+σx2 ( 19 ) νtk2=σx2Ntk+σx2+σx2 . Here Ntk denotes the number of times zτ=k for τ < t and x¯tkd denotes the average cue values for observations assigned to cause k for τ<t . The second term in Equation 15 ( the prior ) is given by the time-sensitive Chinese restaurant process ( Equation 3 ) . The M-step is derived by differentiating ℱ with respect to 𝐖 and then taking a gradient step to increase the lower bound . This corresponds to a form of stochastic gradient ascent , and is in fact remarkably similar to the Rescorla-Wagner learning rule ( see below ) . Its main departure lies in the way it allows the weights to be modulated by a potentially infinite set of latent causes . Because these latent causes are unknown , the animal represents an approximate distribution over causes , 𝐪 ( computed in the E-step ) . The components of the gradient are given by: ( 20 ) [∇ℱ]kd=σr-2xtdδtk , where δtk is given by Equation 7 . To make the similarity to the Rescorla-Wagner model clearer , we absorb the σr-2 factor into the learning rate , η . With two exceptions , we used the following parameter values in all the simulations: α=0 . 1 , η=0 . 3 , σr2=0 . 4 , σx2=1 , θ=0 . 02 , λ=0 . 01 . For modeling the retrieval-extinction data , we treated θ and λ as free parameters , which we fit using least-squares . For simulations of the human data in Figure 13 , we used θ=0 . 0016 and λ=0 . 00008 . Note that θ and λ change only the scaling of the predictions , not their direction; all ordinal relationships are preserved . The CS was modeled as a unit impulse: xtd=1 when the CS is present and 0 otherwise ( similarly for the US ) . Intervals of 24 hr were modeled as 20 time units; intervals of one month were modeled as 200 time units . While the choice of time unit was somewhat arbitrary , our results do not depend strongly on these particular values . In this section we demonstrate a formal correspondence between the classic Rescorla-Wagner model and our model . In the Rescorla-Wagner model , the outcome prediction r^t is , as in our model , parameterized by a linear combinations of the cues 𝐱t and is updated according to the prediction error: ( 21 ) r^t=∑d=1Dwdxtd ( 22 ) δt=rt−r^t ( 23 ) w←w+ηxtδt . The key difference is that in our model , we allow there to be separate weight vectors for each latent cause . When α=0 , the distribution over latent causes reduces to a delta function at a single cause ( since the probability of inferring new latent causes is always 0 ) , and hence there is only a single weight vector . In this case , the two models coincide . | Our memories contain our expectations about the world that we can retrieve to make predictions about the future . For example , most people would expect a chocolate bar to taste good , because they have previously learned to associate chocolate with pleasure . When a surprising event occurs , such as tasting an unpalatable chocolate bar , the brain therefore faces a dilemma . Should it update the existing memory and overwrite the association between chocolate and pleasure ? Or should it create an additional memory ? In the latter case , the brain would form a new association between chocolate and displeasure that competes with , but does not overwrite , the original one between chocolate and pleasure . Previous studies have shown that surprising events tend to create new memories unless the existing memory is briefly reactivated before the surprising event occurs . In other words , retrieving old memories makes them more malleable . Gershman et al . have now developed a computational model for how the brain decides whether to update an old memory or create a new one . The idea at the heart of the model is that the brain will attempt to infer what caused the surprising event . The reason the chocolate bar tastes unpalatable , for example , might be because it was old and had spoiled . Every time the brain infers a new possible cause for a surprising event , it will create an additional memory to store this new set of expectations . In the future we will know that spoiled chocolate bars taste bad . However , if the brain cannot infer a new cause for the surprising event – because , for example , there appears to be nothing unusual about the unpalatable chocolate bar – it will instead opt to update the existing memory . The next time we buy a chocolate bar , we will have slightly lower expectations about how good it will taste . The dilemma of whether to update an existing memory or create a new one thus boils down to the question: is the surprising event the consequence of a new cause or an old one ? This theory implies that retrieving a memory nudges the brain to infer that its associated cause is once again active and , since this is an old cause , it means that the memory will be eligible for updating . Many experiments have been performed on the topic of modifying memories , but this is the first computational model that offers a unifying explanation for the results . The next step is to work out how to apply the model , which is phrased in abstract terms , to networks of neurons that are more biologically realistic . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"neuroscience"
] | 2017 | The computational nature of memory modification |
β‐ and γ‐cytoplasmic actin are nearly indistinguishable in their amino acid sequence , but are encoded by different genes that play non‐redundant biological roles . The key determinants that drive their functional distinction are unknown . Here , we tested the hypothesis that β- and γ-actin functions are defined by their nucleotide , rather than their amino acid sequence , using targeted editing of the mouse genome . Although previous studies have shown that disruption of β-actin gene critically impacts cell migration and mouse embryogenesis , we demonstrate here that generation of a mouse lacking β-actin protein by editing β-actin gene to encode γ-actin protein , and vice versa , does not affect cell migration and/or organism survival . Our data suggest that the essential in vivo function of β-actin is provided by the gene sequence independent of the encoded protein isoform . We propose that this regulation constitutes a global ‘silent code’ mechanism that controls the functional diversity of protein isoforms .
Actin is an essential and abundant intracellular protein that plays a major role in developmental morphogenesis , muscle contraction , cell migration , and cellular homeostasis . Two of the closest related actin isoforms , non-muscle β-actin and γ-actin , are ubiquitously expressed but encoded by different genes , producing nearly identical proteins except for four residues within their N-termini ( Vandekerckhove and Weber , 1978 ) . Notably , β-and γ-actin mRNA coding sequences differ much more significantly , by nearly 13% , due to silent substitutions affecting approximately 40% of their codons ( Erba et al . , 1986 ) . Recent evidence suggests that this difference in mRNA coding sequence affects translation dynamics of the two actin isoforms: β-actin is translated in bursts and accumulates faster than γ-actin ( Buxbaum et al . , 2014; Zhang et al . , 2010 ) . Such faster translation results in their differential post-translational modification by arginylation , which targets only β- but not γ-actin ( Zhang et al . , 2010 ) . Thus , actin isoforms are differentially regulated via changes in their mRNA coding sequence that can affect their translation and post-translationally modified state . Despite their high similarity and abundance in the same cell types , β- and γ-actin play distinct non-redundant biological roles . A body of evidence shows that these actins localize to different parts of the cell and tend to incorporate into different actin cytoskeletal structures ( Dugina et al . , 2009; Kashina , 2006; Otey et al . , 1986 ) . More definitively , several studies show that β-actin knockout in mice results in early embryonic lethality despite proportional up-regulation of other actin isoforms to compensate for the total actin dosage ( Bunnell et al . , 2011; Shawlot et al . , 1998; Shmerling et al . , 2005; Strathdee et al . , 2008; Tondeleir et al . , 2013 , 2014 ) , while γ-actin knockout in mice has a much milder phenotype that does not interfere with animal survival during embryogenesis ( Belyantseva et al . , 2009; Bunnell and Ervasti , 2010 ) . This lethality of β-actin knockout in mice can be rescued by a targeted knock-in of the β-actin cDNA ( Tondeleir et al . , 2012 ) , suggesting that the coding region of this gene is far more important for organism’s survival than any non-coding elements affected by the β-actin knockout . The underlying mechanisms conferring such functional differences to these two nearly identical protein isoforms are unknown . Several distinct features of the two actin isoforms have been proposed as the mechanism conferring unique roles to the two proteins . Some biochemical differences between β- and γ-actin were observed in polymerization assays both in vitro and in vivo ( Bergeron et al . , 2010; Kapustina et al . , 2016; Müller et al . , 2013 ) . In addition , β-actin mRNA , unlike γ-actin mRNA , gets spatially targeted to the cell periphery via zipcode-mediated transport ( Hill and Gunning , 1993; Kislauskis et al . , 1997 ) . Finally , β- and γ-actin actin can differentially regulate gene expression of a subset of cytoskeleton proteins ( Tondeleir et al . , 2013 , 2014 ) . Despite these differences , no study to date has definitively identified the key functional determinants that confer unique functions to the mammalian actin isoforms in organism’s survival , or determined whether these determinants reside at the amino acid level . Here , we used targeted editing of the mouse genome to test the hypothesis that β- and γ-actin functions in vivo are defined by their nucleotide , rather than their amino acid sequence . Although previous studies have shown that disruption of the β-actin gene critically impacts embryonic development and organism survival ( Bunnell et al . , 2011; Shawlot et al . , 1998; Shmerling et al . , 2005; Strathdee et al . , 2008; Tondeleir et al . , 2013 , 2014 ) , we demonstrate here that editing of the β-actin coding sequence to encode γ-actin protein without disruption of the rest of the β-actin gene does not affect mouse survival or produce a visible phenotype at the organismal or cellular level . This result shows that γ-actin protein is functionally capable of substituting for β-actin in the absence of gene disruption . Thus , we demonstrate that the differences in in vivo functions of β- and γ-actin actin are ultimately determined by their nucleotide rather than amino acid sequence .
It has been previously found that knockout of β-actin gene in mice , unlike γ-actin , leads to early embryonic lethality – a result that definitively demonstrates its essential , non-redundant biological function ( Bunnell et al . , 2011 ) . To test whether this essential function of β-actin is defined by its amino acid or nucleotide sequence , we used CRISPR/Cas9-mediated gene editing to introduce five point mutations into the native mouse β-actin gene ( Actb ) , altering it to encode γ-actin protein without changing any of the features of the rest of the gene sequence ( ‘beta-coded gamma actin’ , Figure 1A ) . We termed this edited gene Actbc-g , in which the native β-actin gene is nearly intact ( with five point mutations within the first 10 codons ) and contains the same promoter , as well as the same coding and non-coding elements , but the protein produced from this gene is identical to γ-actin . The outcome is no β-actin protein at all , enabling us to definitively test whether β-actin amino acid sequence , or its nucleotide sequence , is responsible for its essential function in organism’s survival . If intact β-actin amino acid sequence is required , the mutant mice would die in embryogenesis , similarly to the β-actin knockout mice ( Bunnell et al . , 2011; Shawlot et al . , 1998; Shmerling et al . , 2005; Strathdee et al . , 2008; Tondeleir et al . , 2013 , 2014 ) . If the nucleotide sequence also contributes , these mice would be expected to survive longer than the β-actin knockout mice and have an overall milder phenotype . Finally , if the nucleotide sequence is the sole determinant of β-actin function , these mice would have no phenotype at all . This gene editing strategy was successful , generating homozygous mouse mutants that contained no β-actin protein ( Figure 1—figure supplements 1–2 ) . Strikingly , Actbc-g mice appeared completely healthy , viable , and fertile with no signs of deficiencies previously seen in any of the β-actin knockout mouse models . These mice did not exhibit any visible defects in embryogenesis ( Figure 1B ) , and appeared healthy and normal at birth and throughout life ( Figure 1C ) ( observed until approximately 8 months old by the time of the submission of this study for publication ) . These mice also had normal fertility , as evidenced by litter sizes from Actbc-g homozygous breeding pairs that averaged 6 . 4 pups per litter ( ±0 . 38 SEM , n = 9 ) , compared to the average litter size of 6 . 3 pups previously reported for their matching background wild type strain C57BL/6 ( http://www . informatics . jax . org/silver/tables/table4-1 . shtml ) . Thus , this result definitively proves that β-actin nucleotide sequence , rather than its amino acid sequence , determines the essential function of β-actin in vivo . To test for possible milder defects in these mice , we analyzed the overall morphology and appearance of all major organs and body parts in newborn ( P0 ) Actbc-g mice by sagittal sectioning and H&E staining , and found no overall differences or abnormalities between wild type and Actbc-g mice ( Figure 1D ) , suggesting that the embryonic development in these mice occurs normally . Overall , Actbc-g mice appeared completely healthy and normal , suggesting that γ-actin protein encoded by the β-actin gene is fully able to functionally substitute β-actin’s essential role in mouse survival and health . To confirm the replacement of β-actin protein in these mice with the γ-actin protein , we performed quantitative western blots from several tissues where non-muscle actin isoforms are normally expressed at high levels , including brain , kidney , liver , and lungs ( Figure 2 ) . In all these tissues , loss of β-actin protein was accompanied by a prominent increase in γ-actin , without overall changes in total actin levels ( Figure 2 and Figure 2—figure supplement 1 ) . Corresponding changes were also seen on 2D gels from these tissues , run under shallow pH gradient to separate actin isoforms ( Figure 2 and Figure 2—figure supplement 2 ) . We also performed a reciprocal experiment , using CRISPR/Cas9 gene editing to edit the mouse γ-actin gene to encode β-actin protein ( ‘gamma-coded beta actin’ or γc-β-actin , Figure 2—figure supplements 3–4 ) . This strategy was only partially successful , resulting in replacement of the first three nucleotides to convert the N-terminal MEEE sequence of γ-actin into the MDDD sequence of β-actin , while failing to achieve I/V substitution at codon 10 . However , given that the full deletion of γ-actin has a much milder phenotype than β-actin mouse knockout ( Belyantseva et al . , 2009; Bunnell et al . , 2011; Bunnell and Ervasti , 2010 ) , we did not pursue this further and analyzed the partially edited mouse instead . These Actg1c-b mice showed no detectable phenotype . At the same time , they showed disappearance of γ-actin protein and a corresponding increase in β-actin-like protein by western blots ( Figure 2—figure supplement 5 ) . These results suggest that γ-actin in vivo functions , like β-actin , is also defined by its nucleotide , rather than amino acid sequence . Since β-actin has been previously shown to play a major role in directional cell migration , and its knockout in cells leads to severe impairments in their actin cytoskeleton organization and their ability to migrate ( Bunnell et al . , 2011; Tondeleir et al . , 2012 ) , we next analyzed the actin cytoskeleton distribution and directional migration of mouse embryonic fibroblasts ( MEF ) derived from littermate wild type and Actbc-g mice . Despite the complete absence of β-actin protein in these cells , their actin cytoskeleton appeared similar to that of wild type cells . We detected no difference in F-actin levels in these cells ( Figure 3 ) , or in the morphology and appearance of the actin cytoskeleton ( Figure 3 and Figure 4 ) . To test the ability of these cells to migrate , we performed wound healing assays to measure the overall migration rates of the cell monolayers in wild type and Actbc-g . We also measured the directionality of single cell migration on fibronectin-coated dishes . In both assays , no difference was observed between the two cell types ( Figure 5 and Figure 5—videos 1 and 2 ) , confirming that the actin isoform substitution did not result in any significant changes in these cells’ ability to migrate . Thus , our data definitively demonstrate that the essential function of β-actin in vivo is defined by its nucleotide , and not its amino acid sequence . We have previously reported that β- and γ-actin are differentially arginylated , due to differences in their mRNA nucleotide coding sequences ( arising via silent substitutions ) , which leads to different translation dynamics , and the resulting rates of their accumulation in cells ( Buxbaum et al . , 2014; Zhang et al . , 2010 ) . To assess the potential differences in translation dynamics between the actin isoforms , we analyzed the global ribosome profiling data for these genes ( mouse Actb and Actg1 , encoding non-muscle β- and γ-actin , respectively ) available at the GWIPS-Viz genome browser ( http://gwips . ucc . ie; Michel et al . , 2014 ) that aggregates the results of multiple ribosome profiling studies across genomes . Remarkably , composite data from 26 independently performed ribosome profiling studies from different mouse tissues show that the ribosome density on mouse β-actin mRNA is over a thousand fold higher than γ-actin ( average ribosome density over the first 150 codons: 1351 . 607 for β-actin versus 1 . 289 for γ-actin , see also Table 1 and Supplementary file 1 and 2 ) . These data suggest that the translation dynamics of β-actin in vivo is dramatically different from that of γ-actin . We next extended this analysis to the whole family of mouse actin genes and correlated the previously reported phenotypes resulting from mouse knockouts of different actin isoforms with the ribosome density number for each isoform ( Table 1 ) . This analysis shows that members of the actin family have vastly different ribosome densities , with β-actin being by far the highest , while γ-enteric smooth muscle actin is by far the lowest ( ribosome density 0 . 377 ) , suggesting that the intracellular accumulation rate for these two proteins , as well as their translation dynamics , should be vastly different from each other . In support , they also have different mRNA structures in the coding region ( Zhang et al . , 2010 , Supplementary file 3 and 4 ) . Notably , the actin isoforms that tend to become up-regulated in different knockout models are typically those with the closest ribosome density to the isoform that had been knocked out ( Table 1 ) . It appears likely that the success of functional compensation may also be linked to this number , directly or indirectly . For instance , alpha skeletal and alpha smooth muscle actins appear to partially cross-compensate for each other , and these isoforms are also the second and third highest by ribosome density , after β-actin . α-cardiac actin completely rescues knockout of α-skeletal actin , which has the closest ribosome density . In contrast , loss of α-cardiac actin cannot be substituted for by γ-enteric smooth muscle actin , which has a ~10 fold lower ribosome density . These results suggest the actin isoform with similar ribosome density can plausibly compensate for the loss of one of the isoforms . In agreement , given the orders of magnitude difference in ribosome density between β-actin and other actin isoforms , none of the other actin isoforms can compensate for the loss of β-actin . We propose that changes in ribosome density arising from silent substitutions in nucleotide sequence , affect translation dynamics and protein accumulation rates , which in turn regulate functional diversity of actins . We next used sequence analysis and Ribo-seq data profiling to test whether this type of silent substitution dependent protein regulation may potentially be applicable to other closely related protein isoform families in the mouse genome . First , to identify all protein families encoding highly similar proteins , we searched the mouse genome for proteins that are over 90% identical both in length and in their amino acid sequence over their entire length , and are encoded by different genes . This search yielded 741 families in mouse , encoding nearly 4000 different open reading frames . Next , we compared ribosome density numbers for different protein isoforms within these families using published ribosome profiling data found on GWIPS-Viz genome browser , and sorted these families by the normalized standard deviation ( SD ) of difference between ribosome densities for different members within the family . We discarded those families in which the ribosome density on the highest translating gene was less than 1 , suggesting that their relative abundance in the polysome fraction is negligible ( e . g . , the majority of olfactory receptors ) . In the resulting list , we selected the top 100 families with SD of ribosome densities of 1 . 4 and above and classified them by functions ( Supplementary file 1 ) . We also performed the same analysis in two other vertebrate genomes , human and zebrafish ( Supplementary files 2 and 3 ) and cross-referenced the identity of the top families to each other . The top families identified in these searches were similar in all three vertebrate genomes , and contained the actin family as one of the top hits . The other uniformly present top hits included multiple classes of histones and tubulins . Notably , similar analysis of D . melanogaster and C . elegans genomes ( Supplementary files 4 and 5 ) also yielded actins , tubulins , and histones as the top hits . Thus , the occurrence of families of nearly identical proteins with vastly different ribosome densities may be broadly common in higher eukaryotes , supporting our idea that this feature may underlie their functional regulation . We propose that posttranscriptional regulation of highly homologous proteins in seemingly redundant protein families through silent nucleotide-encoded translation dynamics constitutes a novel type of functional regulation that governs some of the most essential functions in eukaryotic genomes . We propose to term this regulation ‘silent code’ , to reflect its origin in silent substitutions within the genetic code .
Our results demonstrate a novel determinant – silent substitutions in nucleotide sequence – that drives the differential functions of non-muscle actin isoforms , a mechanism potentially applicable to other members of the actin family and other highly similar proteins in eukaryotic genomes . In the case of non-muscle actins , our finding resolves decades of controversial studies and reconciles a body of seemingly contradictory results obtained in the attempts to address functional distinction between β- and γ-actin ( Bergeron et al . , 2010; Dugina et al . , 2009; Kapustina et al . , 2016; Patrinostro et al . , 2017 ) . Our data demonstrate that the nucleotide sequence of the gene , rather than the amino acid sequence of the encoded isoform , determines the absolute requirement of β-actin for organism’s survival . Closely related protein isoforms can exhibit functional differences which can be attributed to one or more of the following three sources . First , variations at the amino acid level can cause profound differences in protein function . Second , variations in mRNA properties due to differences in their coding sequence and UTR regions can strongly affect mRNA localization , stability , and translatability via secondary structure and codon usage . Finally , variations in gene intron sequences , promoter , and enhancer regions can contribute to the overall gene regulation , expression levels , and tissue specificity . Contribution of each of these levels to actin isoform function has been extensively investigated in prior studies . While β-actin and γ-actin isoforms share a remarkable conservation at the amino acid level , with just four homologous amino changes at their N-termini , these two proteins have been shown to have slightly different polymerization kinetics ( Bergeron et al . , 2010 ) and to differentially interact with cofilin ( Kapustina et al . , 2016 ) and non-muscle myosin isoforms ( Müller et al . , 2013 ) . At the level of mRNA , actin 3′UTRs are isoform-specific and evolutionarily conserved , suggesting that they play important roles in vivo ( Erba et al . , 1986; Hill and Gunning , 1993 ) . A plethora of literature elucidates the importance of β-actin 3′UTR for the localization of the transcript ( see , e . g . ( Condeelis and Singer , 2005 ) for a comprehensive review ) . It has also been shown that γ-actin mRNA induces a ribosome pausing event , resulting in its slower translation compared to β-actin , a mechanism that drives their differential arginylation ( Zhang et al . , 2010 ) . An alternative poly A site in β-actin mRNA increases its translation levels ( Ghosh et al . , 2008 ) . Finally , at the gene level , several studies point to roles of various regulatory elements in the actin isoforms . γ-actin gene contains a unique and highly conserved intron III ( Lloyd and Gunning , 1993 ) , and an alternatively spliced exon 3a ( Drummond and Friderici , 2013 ) . β-actin exhibits both 3′UTR-dependent ( Lloyd and Gunning , 1993; Lyubimova et al . , 1999 ) and 3′UTR-independent ( Lloyd et al . , 1992; Schevzov et al . , 1992 ) feedback regulation of gene expression . Our prior data showed that silent substitutions in β- and γ-actin coding sequence , amounting to a 13% overall difference , confer changes in the rates of their accumulation in the cell , and in their posttranslational modifications ( Zhang et al . , 2010 ) . We show here , by analysis of publically deposited RiboSeq datasets , that β- and γ-actin mRNAs dramatically differ in ribosome densities , suggesting vastly different translation dynamics and likely different accumulation rates in cells . We propose that these silent substitutions in the nucleotide coding sequence may be key determinants that drive β- and γ-actin function . It is possible , however , that at the gene level additional non-coding elements and the untranslated regions ( UTRs ) of the actin mRNA may also contribute to the functional distinction between the actin isoforms . This possibility requires further investigation . Ribosome profiling experiments show that among the 6 members of the actin family in mice , β-actin by far has the highest representation in the polysome fraction ( Ingolia et al . , 2009 ) . We propose that this confers the cell with the ability to spatially and temporally fine tune translation of β-actin and the local and global rate of its intracellular accumulation ( Buxbaum et al . , 2014; Ströhl et al . , 2017 ) , and thus makes β-actin the most essential of the actin isoforms and cannot be compensated for when deleted . The nearly identical non-muscle γ-actin appears to be more functionally redundant and can be largely compensated for by other actins that have similar ribosome densities ( Perrin and Ervasti , 2010 ) , Table 1 , and Supplementary file 2 and 3 ) . Our findings suggest that actin isoforms with similar ribosome densities and translation dynamics are more likely to compensate for each other’s functions , while β-actin , with the highest-ribosome-density is unique in its functional significance . Elucidating the exact contribution of translation dynamics to actin isoform functions constitutes an exciting direction of future studies . Our result that targeted genome editing of mouse β-actin gene to encode for γ-actin protein leads to no apparent phenotype proves for the first time that the major determinants of β-actin’s essential function in vivo are encoded at the nucleotide level . Some of the existing data point to the possibility that this effect is mediated primarily or exclusively via the coding region , rather than the promoter , the UTR , or the intron sequences . Indeed , deletion of exons 2 and 3 of the β-actin gene that include β-actin translation initiation site but do not encompass the promoter region , UTR , or the non-coding elements in the rest of the gene leads to embryonic lethality ( Bunnell et al . , 2011 ) . Notably , in Actb knockout mice other actin isoforms are up-regulated to compensate for the total actin dosage , but this promoter-mediated up-regulation is insufficient to rescue the phenotype of early embryonic lethality . At the same time , targeted insertion of the human β-actin coding sequence into this region rescues embryonic lethality in these mice , further supporting the idea that the coding sequence plays a key role ( Tondeleir et al . , 2012 ) . Data from our group previously showed that coding sequence drives differences in actin’s posttranslational modifications , one of the forms of functional actin regulation ( Zhang et al . , 2010 ) . While these studies do not fully exclude a potential contribution of non-coding elements , especially those that may be located between exons 2 and 3 in the gene , they strongly support our hypothesis that the coding region is primarily responsible for the uniquely essential role of β-actin in vivo . Elucidating the underlying hierarchy of silent substitutions among other modes of regulation , would further our understating of the various levels of factors affecting the function of different actin isoforms . Despite the fact that non-muscle actin isoform genes have evolutionarily diverged >100 million years ago , they have retained remarkable sequence conservation , far higher than what would be expected if the synonymous substitutions in their coding sequence were completely randomized ( Erba et al . , 1986 ) . This is consistent with our idea that actin isoform coding sequence exists under additional evolutionary pressure , over and above the conservation of amino acid sequence . We propose that at least some of this pressure is aimed to maintain the divergent translation dynamics within the actin family , in order to drive their divergent functions . Our sequence analysis and data mining suggest that this mechanism of homologous protein regulation through ‘silent code’ may be globally applicable to multiple protein families throughout eukaryotic genomes . Notably , one of the top candidate families for this regulation in addition to actin – tubulin – is also a major cytoskeletal protein regulated by multiple posttranslational modifications . Some members of the tubulin family conceivably may be regulated via silent substitutions in their coding sequence affecting translation dynamics , like shown previously for non-muscle actins . While systematic mouse knockout data for the alpha and β− tubulin isoform families is not available , based on the analogy with the actin isoforms , we predict that tubulin β-V ( Tubb5 ) is likely the most essential of the β− tubulin genes , whose knockout likely cannot be substituted by up-regulation of any other β− tubulin . Notably , the two members of the γ− tubulin isoform family , Tubg1 and Tubg2 , have been knocked out in mice , and appear to follow the same trend: knockout of Tubg1 , the tubulin isoform with higher ribosome density , is embryonically lethal , while knockout of Tubg2 with the nearly 200 fold lower ribosome density is not ( Yuba-Kubo et al . , 2005 ) . Further systematic analysis of knockouts of homologous isoforms would enable establishing the universality of the ‘silent code’ .
C57Bl/6 strain was used to generate the gene-edited mice . The donor females were super-ovulated using 5 IU of PMSG followed 48 hr later with 5 IU HCG , after which the females were mated immediately to C57Bl/6 studs . MEGAshort-script T7 transcription kit ( Ambion Thermo Fisher Scientific , Waltham , MA ) was used for in vitro transcription of small guide sgRNA ( gctgcgctggtcgtcgacaaCGG , where CGG is the Protospacer Adjacent Motif , PAM ) as per manufacturer’s protocol . mMESSAGE mMACHINE T7 transcription kit ( Ambion ) was used to synthesize Cas9 mRNA . MEGAclear transcription clean up kit ( Ambion Thermo Fisher Scientific ) was used to purify the synthesized RNAs . About 20 hr post HCG , the CRISPR solution was injected into zygotes via pronuclear injection at a concertation of: Cas9 mRNA: 100 ng/µL , template DNA: 100 ng/µL , and gRNA: 50 ng/µL . The zygotes were further cultured overnight in KSOM media using a 5% CO2 incubator . All the embryos which successfully cleaved to the 2 cell stage were transferred into recipient females via oviduct transfer . A founder female that was mosaic for the mutation was derived and crossed with a wildtype male to derive heterozygotes . One male and female heterozygote from the F1 generation were crossed to produce F2 generation . Two separate litters from the F2 generation produced two wildtype females , five heterozygote males , one heterozygote female , and three homozygote males . Template DNA sequence: 5′CGGCTGTTGGCGGCCCCGAGGTGACTATAGCCTTCTTTTGTGTCTTGATAGTAGTTCGCCATGGAAGAGGAAATCGCTGCGCTGGTCATTGACAACGGCTCCGGCATGTGCAAAGCCGGCTTCGCGGGCGACGATGCTCCCCGGGCTGTA 3′ The Actb gene has an EcoRV site which gets destroyed upon gene editing ( Figure 1—figure supplement 1 ) . We utilized restriction digestion of a PCR product produced from the Actb gene to determine the genotype of the resulting mice . While wild type ( Actb+/+ ) mice gave two bands: 600 bp and 300 bp upon EcoRV digestion , mice homozygous for the mutations ( Actbc-g/Actbc-g ) give a single band at 900 bp . PCR products from mice that are heterozygous for the mutation ( Actb/Actbc-g ) gave three bands upon EcoRV digestion ( Figure 1—figure supplement 1 ) . The results were further verified by sequencing the 5′ end of the Actb gene . In order to verify that β-actin protein was no longer produced , tail samples were lysed and a western blot was carried out using isoform-specific antibodies: mouse anti-β-actin ( Clone 4C2 , EMD Millipore , Burlington , MA , and Clone AC15 , Sigma Aldrich , Burlington , MA ) , mouse anti-γ-actin ( Clone 2C3 EMD Millipore ) . Primary Mouse Embryonic Fibroblasts ( MEFs ) were isolated from the back area of freshly euthanized E12 . 5 mouse embryos by tissue disruption and cultured in DMEM ( Gibco ) supplemented with 10% FBS ( Gibco ) . Cell migration was stimulated by making an infinite scratch wound . The cells were allowed to recover for a period of 2 hr before imaging . Migration rates were measured as the area covered by the edge of the wound in the field of view per unit time using Fiji ( NIH , Bethesda , MD ) . For measurements of directionality in single cell migration assays , primary MEFs were cultured on glass bottom MatTek ( Ashland , MA ) dishes coated with 5 µg/ml Fibronectin at low cell densities . 2 hr after seeding , cells were imaged at 10 min intervals for 4 hr . Single cells were tracked using Metamorph Track Objects ( Molecular Devices , Sunnyvale , CA ) module . The obtained total displacement was divided by total distance of the track to obtain a directionality score shown in Figure 5 . All images for these experiments were acquired on a Nikon Ti microscope with a 10X Phase objective and Andor iXon Ultra 888 EMCCD camera . To quantitate the amount of actin polymer , cells were seeded on coverslips in six well plates at 20 , 000 cells/well overnight and fixed in 4% ( w/v ) PFA at room temperature for 30 min . Cells were then stained with Phalloidin conjugated to AlexaFluor 594 ( Molecular Probes , Eugene , OR ) . Images were acquired using Andor iXon Ultra 888 EMCCD camera at 40X and the total intensity of phalloidin per cell was measured using Metamorph ( Molecular Devices ) . Tissues from 2 month old Actb+/+ and Actbc-g/Actbc-g mice were collected and flash frozen in liquid nitrogen . Brain , Kidney , Liver and Lung tissues were ground and weighed . The samples were lysed directly in 4x SDS sample buffer ( 1:4 w/v ) . Equal volumes of the lysates loaded for SDS-PAGE and 2D gel electrophoresis . Following transfer of the gels , the blots were dried and stained with LI-COR REVRT Total protein stain as per manufacturer's protocol . Images were obtained using an Odyssey scan bed in the 700 nm channel . The blots were then blocked and incubated with primary antibodies for mouse anti-β-actin ( Clone 4C2 , EMD Millipore ) , mouse anti-γ-actin ( Clone 2C3 , EMD Millipore ) , and rabbit anti-pan-actin ( Cytoskeleton , Inc . , Denver , CO ) . Secondary antibodies against mouse and rabbit conjugated to IRDye800 were used to probe the blots and images were acquired in the 800 nm channel using Odyssey scan bed . The total protein intensity was used to account for loading differences and the obtained signals were normalized to the first lane in the blot . 2D gel analysis was performed by Kendrick Laboratories , Inc . ( Madison , WI ) as described by ( Burgess-Cassler et al . , 1989; O'Farrell , 1975 ) using shallow pH gradient for the first dimension to separate actin isoforms ( pH 4–6 , 4–8 ) . NCBI RefSeq mouse genome v . 10 was used for this analysis . GWIPS read density profiles for mouse chromosomes were mapped to the NCBI RefSeq CDS annotations to produce the profiles for mouse mRNAs . Characteristic ribosome density was computed as the average density in the 5’ 150 codons of the CDS . The longest protein encoded in the same locus , as annotated in the genome , was collected from the same locus . Sequences were clustered at 90% identity over 90% of length using blastclust program ( https://www . ncbi . nlm . nih . gov/pubmed/17993672 ) . | Mammalian cells , including human cells , contain high levels of a protein called actin . This protein is essential for many of the processes that organisms use to develop and survive . For example , filaments of actin maintain the shape of cells , and help generate the forces needed for cells to move and divide . As in many other animals , every cell in the human body contains two related actin proteins – known as β-actin and γ-actin . These proteins are made from almost identical amino acid building blocks . Yet the genes that encode these two proteins vary much more . The two actin proteins also play different roles: disrupting the gene for β-actin causes mouse embryos to die , but mice without the gene for γ-actin develop almost like normal . It was not fully understood how these almost identical proteins could perform such different roles . Earlier studies exploring the mechanisms that underlie the unique roles of β- and γ-actin focused on the differences in their amino acid sequences . Now , Vedula , Kurosaka et al . test the hypothesis that the differing roles of these two actin proteins are due to the pronounced differences in the DNA sequences of their genes . A gene-editing technique called CRISPR/Cas9 was used to make small changes to the mouse gene for β-actin so that it coded for the γ-actin protein . As a consequence , these mice did not make any β-actin protein and instead made the γ-actin protein from a mostly intact gene for β-actin . These mice lacking the β-actin protein survived as normal and were fertile . The shape of their organs and the movement of their cells – two other major processes that need β-actin – were also unaffected . Hence , the γ-actin protein can substitute for β-actin when the β-actin gene is intact . These observations imply that it is the DNA sequence of the gene rather than the amino acid sequence of the protein that determines the essential role of β-actin in cell migration and the organism’s survival . The next step will be to see if other proteins work in a similar way . If so , this mechanism might allow scientists to discover new ways to fine-tune how proteins behave in healthy and diseased human cells . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"biochemistry",
"and",
"chemical",
"biology",
"cell",
"biology"
] | 2017 | Diverse functions of homologous actin isoforms are defined by their nucleotide, rather than their amino acid sequence |
The conserved MICOS complex functions as a primary determinant of mitochondrial inner membrane structure . We address the organization and functional roles of MICOS and identify two independent MICOS subcomplexes: Mic27/Mic10/Mic12 , whose assembly is dependent on respiratory complexes and the mitochondrial lipid cardiolipin , and Mic60/Mic19 , which assembles independent of these factors . Our data suggest that MICOS subcomplexes independently localize to cristae junctions and are connected via Mic19 , which functions to regulate subcomplex distribution , and thus , potentially also cristae junction copy number . MICOS subunits have non-redundant functions as the absence of both MICOS subcomplexes results in more severe morphological and respiratory growth defects than deletion of single MICOS subunits or subcomplexes . Mitochondrial defects resulting from MICOS loss are caused by misdistribution of respiratory complexes in the inner membrane . Together , our data are consistent with a model where MICOS , mitochondrial lipids and respiratory complexes coordinately build a functional and correctly shaped mitochondrial inner membrane .
Mitochondria are double membrane-bound organelles whose shape is instrumental to their function in diverse cellular processes such as metabolism and apoptosis ( Labbe et al . , 2014 ) . The inner mitochondrial membrane has the highest density of protein in the cell and is differentiated into three distinct interconnected domains: the boundary region , which are flattened membranes that lie in close apposition to the outer membrane; cristae membranes , which are lamellar invaginations with highly curved edges; and cristae junctions , which are relatively narrow tubules that connect cristae to the boundary membrane and may act as a physical partitioning mechanism that prevents and/or regulates the intermixing of proteins between cristae and the boundary domains ( Mannella et al . , 1994 ) . Compositionally , the boundary membrane is enriched for import and assembly machinery for the mitochondrial proteome ( Vogel et al . , 2006; Wurm and Jakobs , 2006 ) . In contrast , cristae membranes house assembled electron transport chain protein complexes and ATP synthase , which function together to synthesize ATP via oxidative phosphorylation . Specific combinations of electron transport chain complexes further assemble into large mega-Dalton supercomplexes in cristae in a manner dependent on the mitochondrial lipid cardiolipin and act to facilitate electron transport , and likely as diffusion traps to promote their sorting into cristae ( Cruciat et al . , 2000; Zhang et al . , 2002; Pfeiffer et al . , 2003; Acehan et al . , 2011; Lapuente-Brun et al . , 2013; Wilkens et al . , 2013 ) . ATP synthase also assembles in a regulated manner into dimers and dimer oligomers , whose unique structure generates and/or stabilizes the high curvature at the edges of lamellar cristae ( Abrahams et al . , 1994; Paumard et al . , 2002; Strauss et al . , 2008 ) . The proper differentiation of the mitochondrial inner membrane into separate domains is critical for mitochondrial function . Indeed , a hallmark feature of mitochondrial diseases is aberrant mitochondrial ultrastructure and shape ( Zick et al . , 2009 ) . Inner membrane structure is also a critical feature of apoptosis , where cristae junctions are remodeled to facilitate the release of the cell death mediator , cytochrome c ( Scorrano et al . , 2002 ) . Internal mitochondrial architecture may also interface with the external mitochondrial environment , including contacts with the endoplasmic reticulum ( ER ) as nucleoids associate with both mitochondrial cristae and mitochondrial division sites , which are marked by sites of ER contact ( Brown et al . , 2011; Kopek et al . , 2012; Murley et al . , 2013 ) . Thus , proper mitochondrial ultrastructure is critical for a multitude of mitochondrial and cellular functions . The mechanisms by which inner membrane domains are established and maintained are poorly understood . In addition to respiratory supercomplexes , ATP synthase oligomers and mitochondrial lipid composition , the inner membrane fusion dynamin , Mgm1/OPA1 , and scaffolding proteins such as prohibitins have been proposed to play roles in inner membrane structure ( Frezza et al . , 2006; Meeusen et al . , 2006; Merkwirth et al . , 2008 ) . These factors are interdependent—for example , cardiolipin is required for both respiratory supercomplex assembly and Mgm1/OPA1 self-assembly and function and the prohibitins are required to maintain normal mitochondrial lipid homeostasis ( DeVay et al . , 2009; Osman et al . , 2009 ) . This interdependency suggests that inner membrane differentiation is a highly cooperative process , however , exactly how these determinants work together to correctly shape and organize the mitochondrial membrane to ultimately lead to proper respiratory function is also not understood . The recently identified MICOS complex ( previously named MitOS or MINOS ) has been proposed to act as a master regulator/integrator of mitochondrial inner membrane shape and organization ( Harner et al . , 2011; Hoppins et al . , 2011; von der Malsburg et al . , 2011; Alkhaja et al . , 2012; Pfanner et al . , 2014 ) . Consistently , MICOS interacts both physically and functionally with cardiolipin , import machinery , and respiratory complexes ( Rabl et al . , 2009; Hoppins et al . , 2011; Bohnert et al . , 2012; Korner et al . , 2012; Zerbes et al . , 2012; Weber et al . , 2013; Harner et al . , 2014 ) . The MICOS complex is also embedded in the inner membrane with domains facing the intermembrane space that mediate the formation of heterologous structures localized to the inner boundary membrane ( Hoppins et al . , 2011 ) . It is comprised of six core subunits in yeast: Mic60 , Mic10 , Mic19 , Mic27 , Mic26 , and Mic12 , that , with the exception of Mic12 , have mammalian homologs ( Xie et al . , 2007; Mun et al . , 2010; Darshi et al . , 2011; Head et al . , 2011; Alkhaja et al . , 2012; An et al . , 2012; Weber et al . , 2013 ) . Single MICOS subunit deletion causes a characteristic mitochondrial inner membrane morphological defect in cells , consisting of extended , stacked , lamellar inner membranes and a reduction of the number of cristae junctions , with a consequent lamellar mitochondrial shape defect . The common cellular phenotypes of single MICOS subunit deletions indicate that they perform a shared function . However , expression analysis indicates that Mic60 and Mic10 function uniquely as ‘core components’ that direct a hierarchal MICOS assembly as MIC60 deletion causes Mic19 instability and MIC10 deletion causes Mic27 instability ( Harner et al . , 2011; Hoppins et al . , 2011; von der Malsburg et al . , 2011 ) . In addition , specific pairwise combinations of MICOS subunit deletions can produce either positive or negative genetic interactions ( Hoppins et al . , 2011 ) , indicating that although MICOS subunits act cooperatively , they also perform non-redundant roles within mitochondria . For example , Mic60 plays a role in import independently of the MICOS complex ( von der Malsburg et al . , 2011 ) . Here we examine the role of MICOS in the lateral organization of the mitochondrial inner membrane by examining how MICOS is assembled and how MICOS cooperates with the surrounding mitochondrial environment . Using yeast cells deficient for all six components of MICOS , we identify two major MICOS organizing centers , Mic60 and Mic27/Mic10/Mic12 . Our data indicate that Mic27/Mic10/Mic12 assembles at cristae junctions in a respiratory complex- and cardiolipin-dependent manner . In contrast , Mic60 assembles and organizes independently of other MICOS subunits , cardiolipin , and the respiratory machinery , suggesting that Mic60 assemblies may intrinsically mark nascent cristae junctions . MICOS subcomplexes are bridged together by Mic19 , which our data indicates controls the copy number and position of cristae junctions within mitochondria . Together , our findings demonstrate how the MICOS complex works with respiratory complexes and the mitochondria lipid environment to establish inner membrane architecture , organization and function .
To address the functional roles and organization of MICOS , we generated a yeast strain lacking all core MICOS components using a Cre-lox recombination system , which allowed for sequential MICOS subunit gene deletion and selection marker rescue ( Guldener et al . , 1996 ) . Using this approach , we constructed ∆MICOS—a strain containing ‘clean , ’ unmarked deletions of all six MICOS genes . In contrast to single MICOS subunit deletions , ∆MICOS cells displayed a severe respiratory growth defect , as assessed by growth on the non-fermentable carbon source , glycerol ( Figure 1A , right panel ) . Importantly , reintroduction of four of six MICOS genes at their native loci complemented the growth defect , validating the ∆MICOS strain ( Figure 1A ) . The mitochondrial morphology defect of the ∆MICOS strain , as assessed by fluorescence microscopy using the matrix-targeted fluorescent protein , mito-DsRed , was also more penetrant as compared to mitochondria in individual MICOS deletion cells ( Figure 1B ) . Although the predominant lamellar shape phenotype present in single MICOS deletion mutants was observed ( 56% of ∆mic60 cells vs 21% of ∆MICOS cells ) , a majority of ∆MICOS cells also had bulbous mitochondrial structures with a beads on a string appearance , and hollow , large spherical structures that were sometimes continuous with the bulbous structures ( 11% of ∆mic60 cells vs 59% of ∆MICOS cells ) ( Figure 1C ) . These spherical structures are consistent with what have previously observed in a relatively small fraction of ∆mic60 cells ( Itoh et al . , 2013 ) . EM analysis of ∆MICOS cells revealed mitochondrial features analogous to those observed by fluorescence microscopy , including extended lamellar cristae , septated inner membranes , mitochondria with larger diameter structures , and hollow mitochondria , respectively ( Figure 1D ) . 10 . 7554/eLife . 07739 . 003Figure 1 . The MICOS complex is required for oxidative phosphorylation and normal mitochondrial ultrastructure and morphology . ( A ) Serial dilutions of the indicated yeast cells were plated on media containing glucose ( left ) and the non-fermentable carbon source , glyercol ( right ) . ( B ) Mitochondrial morphology in the indicated strains was determined by imaging cells expressing the matrix marker mito-dsRed . Z-projections of confocal fluorescence images are shown , except for the right panel of ∆MICOS , which is a single plane . ( C ) Quantification of mitochondrial morphologies from cells imaged as in ( B ) were categorized . Approximately 100 cells from three independent experiments were quantified and data are represented as mean ± SEM . ( D ) Representative electron microscopy images are shown of chemically fixed yeast cells from the indicated strains . ( E ) Confocal fluorescence microscopy z-projections of cells from the indicated strains expressing mito-DsRed and the functional nucleoid marker Rim1-GFP are shown . The arrow marks aggregation of Rim1-GFP in a ∆MICOS cell . Scale bars: ( B ) 2 μm; ( D ) 500 nm; ( E ) 3 μm . See also Figure 1—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 07739 . 00310 . 7554/eLife . 07739 . 004Figure 1—figure supplement 1 . Rim1 is a functional marker of nucleoids . ( A ) Serial dilutions of the wild type yeast and those expressing Rim1-GFP integrated at the trp1 locus plated on media containing glucose ( left ) and the non-fermentable carbon source , glyercol ( right ) . ( B ) Z-projections of confocal fluorescence microscopy images of wild type rho+ ( top ) and rho0 ( bottom ) cells expressing Rim1-GFP and the mitochondrial matrix marker , mito-DsRed . Scale bar: 3 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 07739 . 004 We examined the distribution of mitochondrial genomes by imaging cells expressing a tagged mitochondrial DNA-binding protein , Rim1-GFP , a functional marker of nucleoids ( Van Dyck et al . , 1992 ) ( Figure 1—figure supplement 1 ) . In wild type and ∆mic60 cells , genomes were relatively uniformly distributed throughout mitochondria , however in ∆mic60 cells , nucleoids localized to the edge of lamellar mitochondrial regions , suggesting they may be excluded from regions of stacked , lamellar inner membranes ( Figure 1E ) . In ∆MICOS cells , while Rim1-marked nucleoids were maintained , a fraction of nucleoids appeared aggregated as compared to in wild type cells ( marked by an arrow in Figure 1E ) . Aggregated nucleoids were mostly observed in spherical mitochondria in ∆MICOS cells , a phenotype previously observed in a relatively smaller fraction of ∆mic60 and ∆mic10 cells ( Itoh et al . , 2013 ) . Our characterization of ∆MICOS cells indicates that , although MICOS subunits have non-redundant functions , they act together to buffer the loss of function caused by deletion of single components . However , these observations point to the critical role of the intact MICOS complex in the maintenance of mitochondrial function , morphology , and distribution of mitochondrial genomes in cells . Members of the MICOS complex are localized in a non-uniform focal-like pattern along the inner membrane in both yeast and mammalian cells ( Hoppins et al . , 2011; Jans et al . , 2013 ) ( Figure 2A ) . These substructures have been postulated to reflect a scaffold-like function of the complex at cristae junctions ( Hoppins et al . , 2011 ) . To gain insight into the molecular basis and functional significance of MICOS organization , we utilized the ∆MICOS strain as a tool to examine the formation of mitochondrial substrctures using functional fluorescent protein tagged versions of MICOS components re-integrated at their respective loci . With the exception of Mic10 , all C-terminal fluorescent protein-tagged MICOS subunits were functional based on their ability to maintain wild type-like tubular mitochondria in a wild type cell background ( Figure 2A ) . In contrast to wild type cells , in ∆MICOS cells the distribution observed for MICOS components Mic27 , Mic19 , Mic12 and Mic26 was uniform , similar to the matrix marker ( Figure 2B ) . Mic60 , however , uniquely localized to discrete foci in a majority of ∆MICOS cells ( Figure 2B , F ) , indicating that Mic60 localizes to substructures independent of its interaction with the MICOS complex , suggesting that Mic60 may have an intrinsic ability to self-assemble . 10 . 7554/eLife . 07739 . 005Figure 2 . Mic60 self-assembles independently and its inner membrane distribution is regulated by Mic19 . ( A ) Z-projections of representative deconvolved fluorescence microscopy images are shown from wild type yeast cells expressing the indicated integrated GFP-tagged MICOS protein and mito-DsRed . The yellow box indicates the region of the cell shown in the inset below . The inset displays a single plane . ( B ) Representative images are shown as in ( A ) for ∆MICOS cells expressing the indicated GFP-tagged MICOS proteins re-integrated at their endogenous loci . The arrows mark sites of Mic60 localization to foci . ( C ) The distribution of Mic60-EGFP in ∆MICOS cells was determined as in ( B ) with the indicated untagged MICOS proteins re-integrated at their endogenous loci . ( D ) Representative Western blot analysis with the indicated antibodies are shown of whole cell lysates ( left panel ) and immunoprecipitation eluates ( IPs; right panel ) from ∆MICOS cells expressing either Mic60-FLAG or Mic60-EGFP at the MIC60 locus , and where indicated , expressing Mic60-EGFP from the ura3 locus using the MIC60 promoter ( pMic60-EGFP ) . G6PDH antibody was used as a loading control . IPs were performed with the indicated antibodies . The asterisk marks a band consistent with the size of IgG heavy chain . ( E ) Western blot analysis with the indicated antibodies of total ( T ) , supernatant ( S ) , and insoluble ( P ) fractions of detergent-solubilized mitochondria isolated from wild type ( left ) or ∆MICOS ( right ) cells expressing Mic60-EGFP and centrifuged at 50 , 000×g for 1 hr , a condition that pellets particles of 60S and greater . ( F ) A graph showing the percentage of cells with detectable Mic60 foci from ∆MICOS cells without and with Mic19 expression as shown in ( B ) and ( C ) . Approximately 75 cells from three independent experiments were quantified and data are represented as mean ± SEM . ( G ) Table describing the number of total spectra and protein coverage for the indicated proteins ( top ) from purifications and mass spectrometry analysis using FLAG antibody from ∆MICOS cell lysate expressing the indicated combinations of Mic60 and Mic19 ( left ) expressed at their endogenous loci . Data shown are the mean of two independent experiments . Scale bars: ( A–B ) 3 μm; ( C ) 2 μm . See also Figure 2—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 07739 . 00510 . 7554/eLife . 07739 . 006Figure 2—figure supplement 1 . Overexpression of Mic60 does not alter its focal localization in ∆MICOS cells . Confocal fluorescence microscopy z-projections of ∆MICOS cells expressing mito-DsRed and Mic60-EGFP expressed at its endogenous locus ( left ) or Mic60-FLAG from the MIC60 locus and pMic60-EGFP expressed from the ura3 locus ( right ) . Arrows mark sites of focal Mic60 . Scale bar: 3 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 07739 . 006 To test this possibility , we performed proteomic analysis of ∆MICOS cells expressing both FLAG-tagged Mic60 at its endogenous locus and EGFP-tagged Mic60 integrated at a second locus driven by its native promoter ( pMic60-EGFP ) . The mild overexpression of Mic60 as a consequence of two genomic copies did not interfere with the ability of pMic60-EGFP to localize to foci ( Figure 2—figure supplement 1 ) . Purification of Mic60 from cross-linked extract from these cells using either α-FLAG or α-GFP antibodies specifically co-purified both tagged Mic60 versions by Western blot analysis ( Figure 2D ) . To rule out the possibility that Mic60 foci represent insoluble aggregates , we solubilized mitochondria isolated from wild type and ∆MICOS cells expressing Mic60-EGFP using the non-ionic detergent , Triton X-100 , under relatively low ionic conditions and subjected extracts to ultracentrifugation . As in wild type cells , Mic60-EGFP expressed in ∆MICOS cells remained soluble , indicating that Mic60 foci represent assembled structures ( Figure 2E ) . These data indicate that Mic60 self-interacts in the absence of other MICOS members and suggest that Mic60 foci in ∆MICOS may represent self-assemblies . To test the relationship of Mic60-labeled structures to MICOS , we examined the effect of expression of other MICOS complex subunits re-integrated at their respective loci on Mic60 localization . Expression of Mic10 , Mic12 , Mic26 , or Mic27 had no apparent effect on Mic60 foci formation or frequency ( Figure 2C ) . In contrast , expression of Mic19 caused Mic60 to become more distributed within mitochondria as compared to the matrix marker , with only 20% of ∆MICOS + Mic19 cells possessing Mic60-EGFP foci as compared to 80% in cells expressing Mic60-EGFP alone ( Figure 2C , F ) . The unique dependence on Mic19 for Mic60 distribution in ∆MICOS cells is consistent with the dependence of Mic19's stability on Mic60 in wild type cells ( Harner et al . , 2011; Hoppins et al . , 2011; von der Malsburg et al . , 2011 ) . To gain insight into the molecular basis of the effect of Mic19 on Mic60 distribution , we performed purifications using α-FLAG antibody and mass spectrometry analysis from cross-linked cell extracts of ∆MICOS strains expressing Mic60-FLAG and untagged Mic19 , or Mic19-FLAG and untagged Mic60 , expressed from their endogenous loci . In each purification , we identified both Mic60 and Mic19 ( Figure 2G ) , suggesting that Mic19 interacts directly with Mic60 independently of other MICOS subunits and that Mic19 , in part , functions within MICOS to regulate the distribution of Mic60 along the inner membrane . Given that Mic60 is able to label discrete structures independently of other MICOS complex members , we tested whether Mic60 was required for the focal distribution of other MICOS subunits . Consistent with the requirement of Mic60 for wild type Mic19 expression , Mic19 in ∆mic60 cells , as detected using fluorescence microscopy , was relatively faint and uniformly labeled the mitochondrial membrane as compared to the matrix marker mito-DsRed ( Figure 3A vs Figure 2A ) . In the absence of Mic60 , Mic26 also localized uniformly , similar to the matrix marker ( Figure 3A ) . In contrast , Mic12 and Mic27 both localized in a non-uniform pattern within mitochondria; this phenotype was most prominent for Mic27 , which localized to distinct and relatively bright punctate structures in a majority of cells ( Figure 3A , marked by arrows , and Figure 3F ) . This suggests that Mic27 , and to a lesser extent Mic12 , localize to mitochondrial sub-structures in a Mic60-independent manner . 10 . 7554/eLife . 07739 . 007Figure 3 . Mic10 , Mic12 , and Mic27 form an independent MICOS subcomplex . ( A ) Confocal fluorescence microscopy z-projections are shown from ∆mic60 cells expressing the indicated GFP-tagged MICOS proteins expressed at their endogenous loci and the mitochondrial matrix marker mito-DsRed . The arrows shown mark the focal localization of Mic27 and Mic12 . ( B ) Localization of Mic27-mCherry compared to the matrix marker mito-GFP was determined in the indicated MICOS deletion cells by confocal fluorescence microscopy as in ( A ) . The arrows mark examples of Mic27 foci . ( C ) Images are shown as in ( A ) for ∆MICOS cells expressing Mic27-EGFP and the indicated untagged MICOS proteins expressed at their endogenous loci . The arrows mark Mic27 foci . ( D ) A graph displaying quantification of the percent of cells as in ( C ) with the indicated number of Mic27-EGFP foci per cell . Approximately 75 cells from three independent experiments were counted and data shown are represented as the mean ± SEM . ( E ) Images are shown as in ( B ) for the indicated yeast cells . ( F ) Quantification as in ( D ) for the cells shown as in ( E ) . ( G ) A graph depicting the number of total spectral counts identified by mass spectrometry of the indicated MICOS proteins from lysates of ∆mic60 cells relative to wild type cells from purifications of Mic27-EGFP with GFP antibody . Data shown are the mean and range of two independent experiments . Scale bars: ( A , B ) 2 μm; ( C , D ) 3 μm . See also Figure 3—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 07739 . 00710 . 7554/eLife . 07739 . 008Figure 3—source data 1 . Table listing spectral count and percent coverage data from mass spectrometry analysis of indicated MICOS protein purifications from the indicated strains . Data from table are presented graphically in Figures 3G , 4C . In some cases , multiple experiments were performed simultaneously with a single control experiment . DOI: http://dx . doi . org/10 . 7554/eLife . 07739 . 008 We further characterized the nature of the Mic27 sub-structures by examining the localization of Mic27 in the absence of each of the MICOS subunits in cells . In the case of ∆mic26 and ∆mic12 cells , Mic27 had similar non-uniform mitochondrial labeling as observed in wild type cells ( Figure 3B ) . In ∆mic10 cells , Mic27 uniformly labeled mitochondria and the fluorescence signal was relatively weaker than in wild type cells , consistent with the observation that Mic10 is required for Mic27 stability . However , in addition to ∆mic60 cells , Mic27 localized to bright foci in ∆mic19 cells , in which there is stable Mic60 expression ( Harner et al . , 2011 ) ( Figure 3B , marked by arrows ) . This observation is consistent with our data suggesting that Mic60 and Mic19 form a complex and function together in a cooperative manner within MICOS ( Figure 2 ) . More significantly , because Mic60 is required for Mic19 stability , it seems likely that the altered localization of Mic27 to more pronounced mitochondrial foci/substructures is a direct consequence of a loss of Mic19 and suggests that these structures represent a second MICOS organizing center . The formation of the Mic27 substructures requires other MICOS subunits as these structures were not observed when Mic27 is expressed alone in ∆MICOS cells ( Figure 2B ) . To determine the minimal MICOS proteins required for Mic27 substructure formation , we re-integrated unmarked MICOS subunits into ∆MICOS cells expressing Mic27-EGFP at its endogenous locus . Consistent with the requirement of Mic10 for Mic27 stability in wild type cells , the addition of Mic10 was sufficient for Mic27 substructures to form , however , their frequency was significantly less than that observed in either ∆mic19 or ∆mic60 cells ( Figure 3C , D ) . Expression of Mic12 in addition to Mic10 , however , supported the formation of Mic27-EGFP substructures at a frequency almost equal to that observed in ∆mic19 or ∆mic60 cells ( compare Figure 3D , 3F ) . Thus , MICOS subunits Mic10 and Mic12 are sufficient for Mic27 substructures to form , suggesting that these three proteins exist as an independent MICOS subcomplex . To further test this idea , we asked whether Mic10 or Mic12 were required for Mic27 substructure formation in ∆mic60 cells . Mic27 substructures were largely undetectable in ∆mic60 ∆mic10 cells , consistent with the requirement of Mic10 for Mic27 stability ( Figure 3E–F ) . We also observed that Mic12 plays a role in Mic27 substructure formation , as the percentage of cells containing Mic27 foci was significantly lower in the absence of Mic12 ( Figure 3F ) . Thus , together our data indicate that Mic10 and Mic12 are both necessary and sufficient for Mic27 substructure assembly and suggest that these proteins form an independent MICOS subcomplex . To directly test this possibility , we purified EGFP-tagged Mic27 from cross-linked lysates of wild type and ∆mic60 cells using α-GFP antibody and identified MICOS subunits by mass spectrometry-based proteomic analysis . Consistent with previous results , proteomic analysis of purifications of Mic27-EGFP in wild type cells robustly identified the entire MICOS complex as assessed by both the number of peptide spectral counts and percent coverage of MICOS subunits ( Hoppins et al . , 2011 ) ( Figure 3—source data 1 ) . To examine Mic27 substructure composition , we directly compared spectral counts of MICOS subunits by normalizing the total number of Mic27 peptide spectral counts purified from ∆mic60 extracts relative to wild type extracts . As expected , purification of Mic27-EGFP from ∆mic60 extracts did not identify Mic60 or Mic19; however , significant relative peptide spectral counts were observed for Mic10 , and to a lesser extent , for Mic12 ( Figure 3G and Figure 3—source data 1 ) . In total , the data indicate that there are two distinct MICOS organizing centers: Mic60 , which may have an intrinsic self-organizational capacity and interacts with Mic19 , and a complex containing Mic27 , which is dependent on Mic12 and Mic10 for assembly . Because of the severe respiratory defect of the ∆MICOS strain and the interrelationship between respiratory complexes and MICOS for generating mitochondrial inner membrane shape ( Hoppins et al . , 2011 ) , we tested whether respiratory complexes are required for Mic60 and Mic27 assembly formation . To block the assembly of mitochondrial respiratory complexes , we generated strains that lack mtDNA ( rho0 ) and examined the distribution of the two MICOS subcomplexes as marked by either Mic60 foci ( in ∆MICOS + Mic60-EGFP cells ) or Mic27 foci ( in ∆mic60 + Mic27-mCherry cells ) . We observed that Mic60 foci in ∆MICOS persisted in a majority of rho0 cells ( foci in 80% of rho+ cells vs 64% of rho0 cells; Figure 4A–B ) . In contrast , although steady state expression level of Mic27 in ∆mic60 rho0 cells was similar to in rho+ cells , Mic27 assemblies were absent in rho0 cells ( foci in 86% of rho+ cells vs 3% of rho0 cells; Figure 4A–B and Figure 4—figure supplement 1 ) . In ∆mic60 rho0 cells , the absence of Mic27 foci corresponded to a loss of its interaction with Mic10 and Mic12 as assessed by mass spectrometry-based proteomic analysis of Mic27 purifications ( Figure 4C and Figure 4—source data 1A ) . The loss of Mic27 subcomplex interactions is due specifically to the combination of mtDNA loss and the absence of Mic60 , as Mic27 co-purified with the entire MICOS complex in wild type rho0 cells as assessed by mass spectrometry analysis of Mic27 purifications ( Figure 4—source data 1B ) . These data reveal that the Mic27/Mic10/Mic12 subcomplex is selectively dependent on the presence of mtDNA , suggesting a critical role of respiratory complexes in the assembly and/or stabilization of this MICOS subcomplex . 10 . 7554/eLife . 07739 . 009Figure 4 . Mic27 assemblies mark cristae junctions . ( A ) Z-projections of confocal fluorescence microscopy images of ( left ) ∆MICOS cells expressing Mic60-EGFP at the MIC60 locus and mito-DsRed or ( right ) ∆mic60 cells expressing Mic27-mCherry at its endogenous locus and mito-GFP . Cells are rho+ or rho0 where indicated . ( B ) Quantification of cells from ( A ) is shown indicating the number of cells with detectable Mic60 or Mic27 foci , as indicated . Approximately 75 cells from three independent experiments were counted and the data shown are represented as the mean ± SEM . Data from rho+ cells for Mic60 and Mic27 are redisplayed from Figures 2F , 3F , respectively . ( C ) A graph depicting the number of total spectral counts identified by mass spectrometry of the indicated MICOS proteins from lysates of ∆mic60 rho+ and rho0 cells relative to wild type cells from purifications of Mic27-EGFP with GFP antibody . Data shown are the mean and range of two independent experiments and data for rho+ cells are redisplayed from Figure 3G . ( D ) Single plane confocal fluorescence microscopy images in 0 . 4 µm steps through a ∆mic60 cell expressing Mic27-EGFP and the Complex III marker Qcr2-mCherry at their endogenous loci and the mitochondrial matrix marker mito-TagBFP . Arrows indicate Mic27 foci identified and analyzed in ( E ) . ( E ) Top row: individual channels and a merged image of a maximum z-projection image of the cell shown in ( D ) . Bottom row: from left to right , a tracing of mitochondria , Qcr2 signal above the threshold detected by the ‘Moments’ algorithm of ImageJ , Mic27 foci identified in the z-projection , and a merged image indicating regions of the mitochondrial perimeter positive for Qcr2 signal and their position relative to Mic27 foci . ( F ) Quantification of the total percentage of mitochondrial perimeter considered positive for Qcr2 signal and the number of Mic27 foci localized to mitochondrial subregions positive for Qcr2 from cells in ( E ) and Figure 4—figure supplement 4 . Scale bars: ( A ) 3 μm; ( D–E ) 1 μm . See also Figure 4—figure supplements 1–5 and Figure 4—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 07739 . 00910 . 7554/eLife . 07739 . 010Figure 4—source data 1 . Tables listing spectral count and percent coverage data from mass spectrometry analysis of indicated MICOS protein purifications from the indicated strains . Data from table ( A ) are presented graphically in Figures 3G , 4C , and this table is redisplayed from Figure 3—source data 1 . Data from table ( B ) are not presented graphically . DOI: http://dx . doi . org/10 . 7554/eLife . 07739 . 01010 . 7554/eLife . 07739 . 011Figure 4—figure supplement 1 . Steady-state Mic27 expression levels are maintained in ∆mic60 rho0 cells compared to rho+ cells . Western blot analysis of whole cell lysates prepared from the indicated strains expressing Mic27-mCherry from its endogenous locus and detected with antibody against mCherry . G6PDH was used as a loading control . DOI: http://dx . doi . org/10 . 7554/eLife . 07739 . 01110 . 7554/eLife . 07739 . 012Figure 4—figure supplement 2 . Individual respiratory complexes are not essential for Mic27 foci formation . ( A ) Z-projections of confocal fluorescence microscopy images of the indicated yeast cells expressing mito-GFP and Mic27-mCherry expressed from its endogenous locus . ( B ) Quantification of the percent of cells with detectable Mic27 foci in the indicated cells as in ( A ) . Data shown are represented as the mean ± SEM from three independent experiments with approximately 75 cells counted per experiment . Data from ∆mic60 cells are redisplayed from Figure 3F . Scale bars: 3 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 07739 . 01210 . 7554/eLife . 07739 . 013Figure 4—figure supplement 3 . Qcr2 and Atp2 are functional markers of Complex III and V , respectively , and localize to sub-mitochondrial regions in ∆mic60 cells . ( A ) Serial dilutions of the yeast cells expressing the indicated markers expressed from their endogenous loci plated on media containing glucose ( left ) and the non-fermentable carbon source , glyercol ( right ) . ( B ) Confocal fluorescence microscopy z-projections of wild type yeast cells expressing mito-GFP as well as Qcr2-mCherry ( left ) or Atp2-mCherry ( right ) expressed at their endogenous loci , respectively . ( C ) As in ( B ) for ∆mic60 cells . ( D–E ) Left: individual channels and a merged image of ∆mic60 cells expressing mito-TagBFP as well as both Qcr2-mCherry and Atp2-EGFP expressed at their endogenous loci . Right: single plane images in 0 . 4 µm steps of the cell shown on the left . Scale bars: ( B , C ) 3 μm; ( D , E ) 1 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 07739 . 01310 . 7554/eLife . 07739 . 014Figure 4—figure supplement 4 . Mic27 assemblies localize adjacent to the cristae marker Qcr2 . ( A–N ) Left: examples of z-projections of confocal fluorescence microscopy images of ∆mic60 cells expressing mito-TagBFP as well as Mic27-EGFP and Qcr2-mCherry expressed from their endogenous loci . Right: tracings of mitochondrial perimeter ( blue ) , Qcr2-positive regions of the perimeter ( red ) , and Mic27 foci ( green ) identified from image on left ( as in Figure 4E ) . ( O ) Quantification of the mitochondrial perimeter , the Qcr2-positive perimeter , the number of Mic27 foci identified , and the number of Mic27 foci adjacent to Qcr2 signal from each example shown in ( A–N ) and Figure 4E . Scale bars: 1 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 07739 . 01410 . 7554/eLife . 07739 . 015Figure 4—figure supplement 5 . Mic27 assemblies localize adjacent to the cristae marker Atp2 . ( A–B ) Left: individual channels and merged images of ∆mic60 cells expressing mito-TagBFP as well as both Atp2-mCherry and Mic27-EGFP expressed at their endogenous loci . Right: single plane images in 0 . 4 µm steps of the cell shown on the left . Arrows indicate the position of Mic27 foci . Scale bars: 1 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 07739 . 015 To test this , we examined the requirement of individual respiratory complexes for Mic27/Mic10/Mic12 subcomplex formation/stability by generating cells deficient for respiratory complex assembly factors for Complex III , IV , or ATP synthase ( ∆cbs1 , ∆mss51 , or ∆atp10 , respectively ) in ∆mic60 cells expressing Mic27-EGFP ( Graef and Nunnari , 2011 ) . Deletion of each respiratory complex assembly factor caused a significant decrease in the percentage of cells containing Mic27 foci ( from 86% to 44–63%; Figure 4—figure supplement 2 ) . This observation indicates that individual respiratory complexes contribute to Mic27/Mic10/Mic12 subcomplex formation/stability , but are not essential . Together , our data indicate that respiratory complexes function in a somewhat redundant manner in the assembly/stability of Mic27/Mic10/Mic12 . The complete absence of respiratory complexes in rho0 cells corresponds to a loss of detectable characteristic cristae as assessed by EM analysis ( Hoppins et al . , 2011 ) . Thus , we considered the possibility that the Mic27/Mic10/Mic12 subcomplex marks cristae junctions . Previously , we observed that cells harboring individual MICOS deletions have a 3–10-fold reduction in the frequency of cristae junctions , but that junction architecture is not altered ( Hoppins et al . , 2011 ) . Thus , the relatively discrete nature and low copy number of Mic27 assemblies in ∆mic60 cells relative to the abundant assemblies in wild type cells is also consistent with its localization at cristae junctions . A direct assessment of whether Mic27-EGFP localizes to cristae junctions in ∆mic60 cells by thin section immuno-EM was not successful due to technical challenges posed by the low copy number of cristae junctions . As an alternative , we utilized fluorescent protein fusions of respiratory complex constituents that selectively localize to cristae ( Wurm and Jakobs , 2006 ) . In wild type cells , functional markers for Complex III ( Qcr2-mCherry ) and ATP synthase ( Atp2-mCherry ) were relatively uniformly distributed as compared to the matrix marker , consistent with previous observations ( Wurm and Jakobs , 2006 ) ( Figure 4—figure supplement 3A–3B ) . In contrast , in ∆mic60 cells , both Qcr2-mCherry and Atp2-mCherry possessed a non-uniform distribution and concentrated in discrete subregions of mitochondria as compared to the matrix marker ( Figure 4—figure supplement 3C ) . In addition , Atp2-EGFP fluorescence signal marked edges of Qcr2-mCherry-labeled subregions of ∆mic60 cells ( Figure 4—figure supplement 3D–3E ) . EM analysis of ATP synthase localization in different cell types indicates that it selectively localizes in multimeric ribbons to the curved edges of cristae ( Strauss et al . , 2008; Acehan et al . , 2011 ) . Thus , the localization patterns of Atp2 in subregions marked by Qcr2 strongly suggests that these are lamellar , stacked cristae as visualized by EM in ∆mic60 cells . We tested whether Mic27 structures were localized adjacent to cristae by examining whether Mic27 foci were spatially linked to Qcr2-labeled subregions in ∆mic60 cells . Significantly , we observed that a majority of Mic27 foci were adjacent to regions labeled by Qcr2 in ∆mic60 cells ( 96% of Mic27 foci were localized close to Qcr2 , which marked less than 40% of the mitochondrial perimeter , n = 15 cells; Figure 4D–F and Figure 4—figure supplement 4 ) . Mic27 foci also were spatially linked to Atp2 labeled subregions in ∆mic60 cells ( Figure 4—figure supplement 5 ) . Taken together our data strongly suggest that independently of Mic60 , at the boundary membrane , the Mic27/Mic10/Mic12 subcomplex is positioned at cristae junctions . We examined the role of the mitochondrial lipid environment on MICOS . In this context , the paralogous Mic27 and Mic26 contain an Apolipoprotein O-like lipid-binding domain , and the mammalian homolog , MIC27/APOOL , has been shown to specifically bind to cardiolipin in vitro ( Weber et al . , 2013 ) . In addition , members of the MICOS complex have a strong negative genetic interaction with members of the cardiolipin synthesis pathway , further suggesting that MICOS and cardiolipin act together to maintain mitochondrial function ( Hoppins et al . , 2011 ) . We therefore asked whether cardiolipin affected the assembly/stability of the Mic27 subcomplex in mitochondria . We examined both the localization of Mic27 and mitochondrial morphology in wild type cells and cells lacking the mitochondrial cardiolipin synthesis enzyme , Crd1 . Deletion of CRD1 led to minor mitochondrial morphology and respiratory growth defects in cells , consistent with published observations ( Jiang et al . , 2000; Chen et al . , 2010 ) ( Figure 5A , D , E ) . In ∆crd1 cells , the non-uniform localization pattern of Mic27 was similar to wild type cells and consistently , mass spectrometry analysis of purifications of Mic27 from ∆crd1 cell extracts indicated that the MICOS complex was intact ( Figure 5A , C , and Figure 5—source data 1 ) . In contrast , we found that in ∆mic60 ∆crd1 cells , Mic27-EGFP foci were largely absent and Mic27 stability was reduced as compared to levels in wild type , ∆mic60 , or ∆crd1 cells ( foci in 87% of ∆mic60 cells vs 2% of ∆mic60 ∆crd1 cells; Figure 5A–B and Figure 5—figure supplement 1A ) . The requirement of cardiolipin for Mic27 stability and Mic27 assemblies was specific , as Mic60 foci formation was not significantly affected in ∆MICOS ∆crd1 cells as compared to ∆MICOS cells ( Figure 5—figure supplement 1B ) . Consistent with loss of Mic27 assemblies , purification and mass spectrometry analysis of Mic27 from cross-linked ∆mic60 ∆crd1 cell extracts indicates a loss of interaction of Mic27 with the MICOS complex ( Figure 5C and Figure 5—source data 1 ) . Together , these data indicate that the association of Mic27 to the Mic27/Mic10/Mic12 subcomplex is dependent on cardiolipin and suggest that cardiolipin is required for its localization at cristae junctions . 10 . 7554/eLife . 07739 . 016Figure 5 . Cardiolipin is required for stability of the Mic10/Mic12/Mic27 subcomplex . ( A ) Images are shown of confocal fluorescence microscopy z-projections of the indicated cells expressing Mic27-EGFP from its endogenous locus and mito-DsRed . ( B ) Quantification of the percent of cells with detectable Mic27 foci in the indicated cells . Data shown are represented as the mean ± SEM from three independent experiments with approximately 75 cells counted per experiment . ( C ) A graph depicting the number of total spectral counts identified by mass spectrometry of the indicated MICOS proteins from lysates of ∆crd1 and ∆mic60 ∆crd1 cells relative to wild type cells from purifications of Mic27-EGFP with GFP antibody . ( D ) Serial dilutions of the indicated yeast cells plated on media containing glucose ( left ) and the non-fermentable carbon source , glyercol ( right ) . ( E ) Quantification of mitochondrial morphologies from the indicated cells expressing mito-DsRed were categorized . Approximately 75–100 cells from three independent experiments were quantified and data are represented as mean ± SEM . Data for wild type and ∆MICOS cells are redisplayed from Figure 1C . Scale bars: 3 μm . See also Figure 5—figure supplement 1 and Figure 5—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 07739 . 01610 . 7554/eLife . 07739 . 017Figure 5—source data 1 . Table listing spectral count and percent coverage data from mass spectrometry analysis of indicated MICOS protein purifications from the indicated strains . Data from table are presented graphically in Figure 5C . In some cases , multiple experiments were performed simultaneously with a single control experiment . DOI: http://dx . doi . org/10 . 7554/eLife . 07739 . 01710 . 7554/eLife . 07739 . 018Figure 5—figure supplement 1 . Interactions between cardiolipin synthesis and MICOS . ( A ) Western blot analysis of whole cell lysates prepared from the indicated strains expressing Mic27-EGFP from its endogenous locus and detected with antibody against GFP . G6PDH was used as a loading control . ( B ) Z-projections of confocal fluorescence microscopy images of yeast cells expressing mito-DsRed and Mic60-EGFP expressed from its endogenous locus in the indicated yeast cells . Arrows mark sites of Mic60 foci formation . ( C ) Serial dilutions of the indicated yeast cells plated on media containing glucose ( left ) and the non-fermentable carbon source , glyercol ( right ) . Scale bars: 2 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 07739 . 018 Because ∆mic60 ∆crd1 cells are deficient for both Mic60 and Mic27 assemblies , we asked whether their growth and mitochondrial morphological characteristics are similar to ∆MICOS cells . We observed that ∆mic60 ∆crd1 and ∆MICOS have comparable growth defects when utilizing glycerol as a non-fermentable carbon source ( Figure 5D ) . In contrast , ∆mic27 ∆crd1 cells did not have a synthetic growth defect on glycerol media ( Figure 5—figure supplement 1C ) . Additionally , deletion of CRD1 in ∆MICOS cells did not further compromise growth on non-fermentable carbon sources ( Figure 5D , compare growth of ∆MICOS ∆crd1 vs ∆MICOS strains ) . Consistently , mitochondrial morphology of ∆mic60 ∆crd1 cells was more severely affected than that observed in ∆mic60 cells and had similar characteristics to those observed in ∆MICOS cells , with a substantial increase in the percent of cells with bulbous morphology ( 9% of ∆mic60 cells compared to 36% ∆mic60 ∆crd1 cells ) and a corresponding decrease in the percent of cells with tubular and lamellar morphology ( Figure 5A , E ) . Together , these data indicate that in cells the absence of both Mic60 and cardiolipin partially phenocopies deletion of the entire MICOS complex and suggest that cardiolipin , in addition to its role in the assembly of respiratory chain supercomplexes , acts to promote the assembly of Mic27 into the Mic27/Mic10/Mic12 subcomplex and its localization to cristae junctions . Importantly , the genetic interaction between CRD1 and MIC10 differs from that of CRD1 and MIC27 , suggesting that there are complex interactions both within and between each MICOS subcomplex and their mitochondrial environment ( Figure 5—figure supplement 1C ) . To understand the role of MICOS in mitochondrial respiratory function , we explored the basis of the ∆MICOS respiratory growth defect . We isolated purified mitochondria from wild type , ∆mic60 , and ∆MICOS strains , and assessed mitochondrial respiratory complex function in vitro . Mitochondrial respiratory complexes and supercomplexes remained largely intact in both ∆mic60 and ∆MICOS mitochondria as assessed by Blue Native PAGE ( BN-PAGE ) analysis relative to mitochondria isolated from wild type cells ( Figure 6A ) . Likewise , both ATP synthase dimerization and activity were unaffected in either single MICOS deletions or ∆MICOS as assessed by an in gel activity assay ( Figure 6C ) . However , despite the presence of mitochondrial respiratory complexes , we detected deficiencies in cytochrome-c oxidase ( Complex IV ) activity specifically in ∆MICOS and not in ∆mic60 ( Figure 6B ) . Using two-dimensional gel analysis , we determined that the composition of respiratory complexes in ∆mic60 and ∆MICOS cells was indistinguishable from wild type cells ( Figure 6D ) . These data indicate that in the absence of the MICOS complex , respiratory complexes can assemble , but do not function optimally . It is possible that in the absence of the MICOS complex , the respiratory defect results from a severe decrease in cristae junctions and consequent disorganization of the inner membrane structure . Alternatively , but not exclusively , MICOS could play a more direct role in the organization of respiratory complexes in the inner membrane . 10 . 7554/eLife . 07739 . 019Figure 6 . Complex III and IV selectively contribute to the ∆MICOS phenotypes . ( A ) A Coomassie-stained Blue Native-PAGE gel of digitonin-solubilized mitochondria isolated from the indicated strains . Labels indicate approximate sizes of the indicated respiratory complexes and supercomplexes . ( B–C ) In-gel activity assays for samples run as in ( A ) for Complex IV activity ( B ) and ATP synthase activity ( C ) . Molecular weight markers are shown on the left . ( D ) Top: Coomassie-stained Blue Native PAGE ( BN-PAGE ) gels of DDM-solubilized mitochondria isolated from the indicated strains . Labels indicate approximate sizes of the indicated respiratory complexes . Bottom: BN-PAGE analysis was followed by SDS-PAGE in a second dimension and silver staining analysis . Molecular weights markers are shown on the left . ( E ) Z-projections of confocal fluorescence microscopy images of the indicated yeast cells expressing mito-DsRed are shown . ( F ) Representative electron microscopy images are shown of chemically fixed ∆MICOS rho0 cells . ( G ) A graph displaying categorization of mitochondrial morphologies from the indicated cells imaged as in ( E ) . Approximately 75–100 cells from three independent experiments were quantified and data are represented as mean ± SEM . Data for ∆MICOS cells are redisplayed from Figure 1C . Scale bars: ( E ) 3 μm; ( F ) 200 nm . DOI: http://dx . doi . org/10 . 7554/eLife . 07739 . 019 To further test this idea , we examined the relationship between respiratory complexes and the mitochondrial morphology defect in ∆MICOS cells . Specifically , we asked whether mitochondrial morphology was altered in ∆MICOS rho0 cells , which lack mitochondrial respiratory complexes . Strikingly , ∆MICOS rho0 cells exhibited relatively normal , tubular mitochondrial morphology as assessed by fluorescence microscopy of matrix-targeted dsRed as compared to ∆MICOS rho+ cells ( 10% rho+ cells vs 95% rho0 cells exhibited tubular mitochondrial morphology; Figure 6E , G ) . As expected , by electron microscopy , the inner membrane of ∆MICOS rho0 cells lacked detectable cristae and were frequently septated , suggesting that mitochondrial inner membrane fusion is compromised likely as a secondary consequence of reduced respiration ( Figure 6F ) ( Sauvanet et al . , 2012 ) . Thus , in the absence of the MICOS complex , expression of respiratory complexes causes inner membrane disorganization , and consequently , severe and diverse abnormal mitochondrial morphologies . Consistently , loss of ATP synthase was previously shown to suppress the predominant lamellar mitochondrial morphology defect of ∆mic60 cells ( Hoppins et al . , 2011 ) . Therefore , we asked whether deletion of individual respiratory complexes alleviated the ∆MICOS mitochondria morphological defect by deleting genes for essential assembly factors of Complex III ( ∆cbs1 ) , Complex IV ( ∆mss51 ) , and ATP synthase ( ∆atp10 ) from ∆MICOS . As in ∆mic60 cells , deletion of ATP synthase assembly selectively suppressed the abnormal mitochondrial lamellar phenotype of ∆MICOS , but the bulbous or spherical morphologies were still present in the cell population . Only 25% of ∆MICOS ∆atp10 cells had normal tubular morphology , compared to 10% in ∆MICOS alone , suggesting distinct morphological phenotypes present in ∆MICOS are caused by separate factors ( Figure 6E , G ) . Consistent with this , deletion of either Complex III or IV assembly factors more significantly alleviated all types of ∆MICOS morphological defects , as ∼75% of ∆MICOS ∆cbs1 or ∆MICOS ∆mss51 cells had normal tubular mitochondrial morphology ( Figure 6E , G ) . These observations , together with the reduced Complex IV activity in ∆MICOS cells , suggest that assembled respiratory Complex III and IV are not properly organized and/or positioned in ∆MICOS , leading to morphological and respiratory growth defects . Our data indicate that , although there is functional redundancy within the MICOS complex , there are functional interrelationships between the Mic60 and Mic27/Mic10/Mic12 MICOS subcomplexes , respiratory complexes , and cardiolipin that are required for proper inner membrane organization , mitochondrial morphology , and respiratory function . To address these interrelationships , we examined how the two MICOS subcomplexes connect and the role of this connection in the proper maintenance of cristae junction number and organization in wild type cells . We considered Mic19 as it has a functional relationship to each MICOS subcomplex; Mic19 disperses Mic60 foci in ∆MICOS and the absence of Mic19 specifically results in the formation of Mic27 focal assemblies in cells ( Figures 2 , 3 ) . These data are consistent with a possible role of Mic19 as an assembly regulator of both MICOS subcomplexes . To test whether Mic19 functions as a connector , we purified tagged Mic60 and Mic27 from cross-linked ∆mic19 extracts and identified interacting partners using mass spectrometry analysis , normalizing the total number of Mic60 or Mic27 peptide spectral counts purified from ∆mic19 extracts relative to wild type extracts . Mass spectrometry analysis of Mic60 purifications indicated that relative to wild type extracts , the absence of Mic19 results in a significant loss of Mic60 interactions with other MICOS components ( Figure 7A and Figure 7—source data 1 ) . By comparison , mass spectrometry analysis of Mic27 purifications from ∆mic19 extracts indicated that Mic27 maintains its interaction with Mic10 relative to wild type extracts ( Figure 7A and Figure 7—source data 1 ) . Together , our data support a model in which the soluble intermembrane space protein Mic19 functions as a primary connector between the two distinct MICOS organizing centers: Mic60 and Mic27/Mic10/Mic12 . 10 . 7554/eLife . 07739 . 020Figure 7 . Mic19 mediates an interaction between the Mic60 and Mic10/Mic12/ Mic27 subcomplexes . ( A ) A graph depicting the number of total spectral counts identified by mass spectrometry of the indicated MICOS proteins from lysates of ∆mic19 cells relative to wild type cells from purifications of Mic60- or Mic27-EGFP with GFP antibody . Data shown are the mean and range of two independent experiments . ( B ) Images are shown of confocal fluorescence microscopy z-projections of ∆mic19 rho+ ( left ) and rho0 ( right ) cells expressing Mic60-EGFP from the MIC60 locus and mito-DsRed . Arrows mark focal localization of Mic60 . ( C ) Z-projections are shown of confocal fluorescence microscopy images of ∆mic19 cells expressing mito-TagBFP as well as Mic60-EGFP and Mic27-mCherry expressed from their endogenous loci . Arrows indicate sites of Mic60 and Mic27 colocalization . ( D ) Single plane confocal fluorescence microscopy images in 0 . 4 µm steps through a ∆mic19 cell expressing Mic60-EGFP and the Complex III marker Qcr2-mCherry at their endogenous loci and the mitochondrial matrix marker mito-TagBFP . Arrows indicate the position of Mic60 foci . ( E ) As in ( D ) for cells expressing Atp2-mCherry . ( F ) A schematic model of the organization and roles of MICOS and its constituent subcomplexes . The model also depicts the interrelationship between MICOS , cardiolipin , and respiratory complexes , and the coordination of these factors in generating mitochondrial inner membrane organization and cristae architecture . Scale bars: ( B ) 3 μm; ( C ) 2 μm; ( D , E ) 1 μm . See also Figure 7—figure supplements 1–2 and Figure 7—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 07739 . 02010 . 7554/eLife . 07739 . 021Figure 7—source data 1 . Table listing spectral count and percent coverage data from mass spectrometry analysis of indicated MICOS proteins . Data are presented graphically in Figure 7A . DOI: http://dx . doi . org/10 . 7554/eLife . 07739 . 02110 . 7554/eLife . 07739 . 022Figure 7—figure supplement 1 . Cardiolipin synthesis is not required for Mic60 foci formation in ∆mic19 cells . Z-projections of confocal fluorescence microscopy images of yeast cells expressing mito-DsRed and Mic60-EGFP expressed from its endogenous locus in the indicated yeast cells . Arrows mark sites of Mic60 foci formation . Scale bars: 2 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 07739 . 02210 . 7554/eLife . 07739 . 023Figure 7—figure supplement 2 . Mic60 assemblies localize adjacent to the cristae markers Qcr2 and Atp2 . ( A–B ) Left: individual channels and merged images of ∆mic19 cells expressing mito-TagBFP as well as both Qcr2-mCherry and Mic60-EGFP expressed at their endogenous loci . Right: single plane images in 0 . 4 µm steps of the cell shown on the left . Arrows indicate the position of Mic60 foci . ( C–D ) As in ( A–B ) for cells expressing Atp2-mCherry . Right panel of ( A ) and ( C ) redisplayed from Figure 7D–E . Scale bars: 1 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 07739 . 023 We next addressed the hierarchical relationship between the Mic60 and Mic27/Mic10/Mic12 MICOS subcomplexes . Mic60 assembles independently of the MICOS complex and its interactions with other MICOS members are significantly reduced in ∆mic19 cells ( Figures 2B , 7A ) . Consistent with these data , we observed that Mic60 localizes to focal assemblies in ∆mic19 cells ( Figure 7B ) . We tested the relationship between Mic60 and Mic27/Mic12/Mic10 assemblies in ∆mic19 cells expressing Mic60-EGFP and Mic27-mCherry . We found that the focal localization of Mic60 and Mic27 overlapped , suggesting Mic60 assemblies also mark cristae junctions ( Figure 7C , see arrows ) . Consistently , Mic60 assemblies in ∆mic19 cells localized adjacent to discrete mitochondrial subregions labeled by the cristae markers Qcr2 and Atp2 , further supporting their localization to cristae junctions ( Figure 7D–E and Figure 7—figure supplement 2 ) . Mic60 assemblies in ∆mic19 cells persist in the absence of mtDNA and cardiolipin synthesis ( Figure 7B and Figure 7—figure supplement 1 ) . Thus , while it is possible that the co-localization of Mic60 with Mic27 in ∆mic19 cells is due to direct interactions between the Mic60 and Mic27/Mic10/Mic12 subcomplexes , the behavior of Mic60 assemblies in ∆mic19 cells is consistent with that of Mic60 assemblies in ∆MICOS cells . These observations suggest that Mic60 structures may independently mark cristae junctions in cells . Their persistence in ∆MICOS rho0 cells , which lack detectable cristae further suggests that Mic60 functions upstream of the Mic27/Mic10/Mic12 subcomplex to position nascent sites of cristae junction formation within mitochondria ( Figures 4A , 6F ) . In the context of our data supporting a role of Mic19 as a MICOS subcomplex regulator and connector , we propose that Mic19 coordinately regulates the assembly of MICOS to dictate the copy number and distribution of cristae in the inner membrane .
Our analysis indicates that MICOS consists of two subcomplexes , which localize to cristae junctions ( Figure 7F ) . Examination of the assembly pathways for each subcomplex has illuminated their relative roles in cristae biogenesis . The Mic27/Mic12/Mic10 subcomplex is dependent on the inner membrane phospholipid cardiolipin and on the expression of respiratory complexes . The second MICOS subcomplex consists of a Mic60 multimer , which assembles in a manner independent of cardiolipin , respiratory complexes , and other MICOS proteins . Although each subcomplex assembles independently , the cardiolipin- and respiratory complex-independent nature of Mic60 assemblies suggest that there is a hierarchical nature to MICOS assembly , where Mic60 functions upstream of the Mic27/Mic10/Mic12 subcomplex in MICOS assembly . In this context , we observed Mic60 assemblies in cells without mtDNA , which lack detectable cristae by electron microscopic analysis . These observations suggest that Mic60 assemblies direct the position of nascent cristae junction sites . The underlying molecular basis for how Mic60 assemblies are positioned within mitochondria is not currently understood , but is likely critical for inner membrane organization . One possibility is that outer membrane-associated determinants direct the biogenesis and/or placement of Mic60 assemblies . Consistent with this , Mic60 independently interacts with components of the mitochondrial import and sorting machinery complex ( SAM/TOB ) through its conserved ‘mitofilin’ domain ( Korner et al . , 2012; Zerbes et al . , 2012 ) . These interactions may allow Mic60 to facilitate proper spacing between the inner boundary and outer membranes and/or to organize these specific outer membrane components relative to inner membrane cristae . Mitochondrial internal determinants may also contribute to the assembly and position of Mic60 , such as the mtDNA-independent replisome , which could function to coordinate cristae biogenesis with nucleoid placement ( Meeusen and Nunnari , 2003 ) . Alternatively , but not exclusively , it is possible that Mic60 assembles at mitochondrial lipid sub-domains positioned , for example , by contacts between mitochondria and the ER ( Kornmann et al . , 2009 ) . The role of the Mic27/Mic12/Mic10 subcomplex is suggested by its dependence on cardiolipin and respiratory complexes . Cardiolipin is also required for the assembly and activity of respiratory complexes and for the formation of respiratory supercomplexes , which are processes that do not depend on MICOS . Thus , our observations suggest that Mic27/Mic12/Mic10 may instead directly sense cardiolipin-rich subdomains of mitochondria to coordinate the non-redundant actions of MICOS and cardiolipin in the respiratory complex-dependent formation of cristae junctions . Consistent with this , Mic26 and Mic27 both share a conserved Apolipoprotein O-like domain , which in other contexts bind to lipids . We could not detect a role for Mic26 in MICOS subcomplex assembly , although its domain architecture suggests it may also function to coordinate MICOS with the mitochondrial lipid environment . In addition , although we failed to generate a functional Mic10 fluorescent protein to directly examine its localization and behavior , the observation that Mic10 is required for Mic27 stability suggests that it plays a central role in directing the formation and function of the Mic27/Mic12/Mic10 subcomplex . The dependence of the Mic27/Mic12/Mic10 subcomplex on respiratory complexes indicates that in addition to direct roles in cristae biogenesis , they also intrinsically contribute to this biogenesis pathway via MICOS assembly . The reciprocal is not true as we find that MICOS is not required for respiratory supercomplex assembly . However , in the absence of MICOS , the activity of cytochrome-c oxidase Complex IV is somewhat reduced and , more importantly , cells lacking MICOS exhibit a severe respiratory defect . One explanation for this severe defect is that MICOS's role in the regulation of cristae copy number is important for producing an optimal distance within cristae between the electron transporting Complex III/IV supercomplexes , localized at lamellar regions , and ATP synthase oligomers localized to the curved edges for efficient oxidative phosphorylation ( Rieger et al . , 2014 ) . Our data also implicate Complex III and IV directly in the nucleation of cristae and in the process of cristae biogenesis . This is supported by the requirement of Complex III and IV for the generation of aberrant mitochondrial morphology in the absence of MICOS ( Figure 6 ) . The basis for this observation may lie in the ability of Complex III and IV to form large cardiolipin-dependent supercomplexes , which could act as diffusion traps for the cristae proteome and lipidome ( Wilkens et al . , 2013 ) . Recent work also implicates a role for the inner membrane fusion dynamin , OPA1 , in the regulation of mitochondrial inner membrane shape and respiratory supercomplex assembly ( Cogliati et al . , 2013 ) . Thus , future work will be required to understand the mechanistic relationships between MICOS and respiratory complexes and OPA1 in the process of inner membrane organization . Our analysis provides insight into the underlying mechanism of cristae copy number and distribution control . A typical wild type mitochondrion possesses many cristae positioned and distributed along the mitochondrial boundary inner membrane . We find that the only soluble MICOS protein , Mic19 functions as the key component directing the inner membrane distribution of each MICOS subcomplex . We propose that through this distributive action and its role as a connector/adaptor , Mic19 regulates the copy number of the MICOS complex and , consequently , the copy number and position of cristae junctions . In this context , it is interesting to note that Mic19 has two mammalian homologs ( MIC19 and MIC25 ) , with conserved coiled-coil-helix motifs , consistent with a role as a key regulatory component of MICOS and suggesting a structural basis for its role as a connector . Our data indicate that the primary function of MICOS is to stabilize , position and control the copy number of cristae junctions to organize the inner membrane into an efficient respiratory machine . Aberrant cristae morphology , very similar to what we observe in cells devoid of MICOS , is a defining feature of a diverse array of human diseases and of aging , and cristae reorganization is a central facet of cytochrome c release in the mediation of apoptotic cell death . Thus , our observations suggest that MICOS will play a significant role in human pathogenesis .
All yeast strains described were constructed in the W303 genetic background ( ade2–1; leu2–3; his3–11 , 15; trp1–1; ura3–1; can1–100 ) . All deletions ( except those used to generate the ∆MICOS strain , see below ) were generated using PCR-based homologous recombination replacing the entire ORF of targeted genes with the kanMX6 , HIS3MX6 , or NatMX6 cassettes ( Longtine et al . , 1998; Schuldiner et al . , 2006 ) . All C-terminal tags were generated using PCR-based targeted homologous recombination using the following cassettes , and are referred to throughout the text using the description in parentheses: GFP ( S65T ) ::HIS3MX6 ( GFP ) , yEGFP::Kan ( EGFP ) , yEGFP::SpHIS5 ( EGFP ) , yEGFP::CaURA3 ( EGFP ) , 3x-yEGFP::CaURA3 ( 3x-EGFP ) , ymCherry::HIS3MX6 ( mCherry ) , ymCherry::Kan ( mCherry ) , and 3xFLAG::HIS3MX6 ( FLAG ) ( Longtine et al . , 1998; Sheff and Thorn , 2004; Hoppins et al . , 2011; Graef et al . , 2013 ) . All yeast transformations were performed by the lithium acetate method and transformants were selected on the appropriate media and verified by PCR , and where appropriate , gene expression . Haploid strains containing multiple tags , deletions , or combinations thereof were generated either by mating and sporulation , or by sequential PCR-based homologous recombination . Rho0 versions of strains were generated by growth in YPD containing 25 μg/ml ethidium bromide for ∼48 hr . The ∆MICOS strain was generated using the Cre-lox P system described previously ( Guldener et al . , 1996 ) . Briefly , the loxP-kanMX-loxP cassette was PCR amplified from pUG6 , integrated into the desired locus by homologous recombination replacing the ORF of the desired MICOS gene , and verified by PCR . The resulting strain was transformed with pSH47 , which expresses the Cre recombinase under the control of a GAL1 promoter . Cre was expressed by growth in YPGal for ∼16 hr . Isolated colonies were tested for loss of G418 resistance and further verified by PCR . Finally , loss of the pSH47 plasmid was selected for by growth on 5-fluoroorotic acid . This process was performed sequentially for each MICOS ORF in the following order: MIC10 , MIC60 , MIC19 , MIC27 , MIC26 , and MIC12 . Untagged MICOS proteins were reintroduced into ∆MICOS in two steps . First , the kanMX6 , TRP1 , or HIS3MX6 selection cassettes were introduced after the stop codon of a MICOS ORF in a wild type strain using PCR-based targeted homologous recombination . Then , genomic DNA was prepared from these strains , and the entire cassette containing the MICOS ORF , selection cassette , and regions upstream and downstream of the ORF was PCR-amplified and reintroduced into the ∆MICOS strain . In the case of reintroduction of tagged MICOS proteins , the same procedure was performed using genomic DNA isolated from the tagged strains described above . pYX142-DsRed ( referred to as ‘mito-DsRed’ ) and pYX142-mtGFP ( referred to as ‘mito-GFP’ ) were previously described ( Westermann and Neupert , 2000; Friedman et al . , 2011 ) . To construct pVT100u-mito-TagBFP ( referred to as ‘mito-TagBFP’ ) , yeast codon optimized TagBFP was synthesized de novo as a gBlock ( IDT ) , digested and inserted into the KpnI/XhoI sites of pVT100u-mtGFP ( Westermann and Neupert , 2000 ) , replacing GFP . All mitochondria plasmid-containing yeast strains were generated by lithium acetate transformation and selection on synthetic dextrose ( SD ) media . To construct pRS306-Mic60-EGFP ( referred to as ‘pMic60-EGFP’ ) , a DNA fragment containing the MIC60 promoter , the Mic60 ORF fused to EGFP , and the ADH terminator was PCR amplified from genomic DNA isolated from a yeast strain expressing Mic60-EGFP at the MIC60 locus . This fragment was digested and cloned into the BamHI/NotI sites of pRS306 ( Sikorski and Hieter , 1989 ) . To construct pRS305-Rim1-GFP ( referred to as ‘Rim1-GFP’ ) , a GFP tag was first integrated at the C-terminus of the Rim1 ORF using pFA6a-GFP ( S65T ) -kanMX6 . A cassette containing the RIM1 promoter , the Rim1 ORF ( including its intron ) , GFP ( S65T ) , and the ADH terminator was PCR amplified from genomic DNA from this strain and cloned into the XhoI/NotI sites of pRS304 . pMic60-EGFP and Rim1-GFP integrate at the ura3-1 and leu2-3 loci , respectively , and were introduced by linearization ( using ApaI for pMic60-EGFP and EcoRV for Rim1-GFP ) and lithium acetate transformation , followed by selection on SD media . Expression was verified by fluorescence microscopy . Cells were grown to exponential phase in YPD , pelleted , and resuspended in water at a concentration of 0 . 5 OD600/ml . 5 μl of 10-fold serial dilutions were plated on YPD and YPEG plates and cells were grown for ∼36 hr ( YPD ) and ∼48 hr ( YPEG ) at 30°C . For all fluorescence microscopy experiments , cells were grown to exponential phase in SD with appropriate plasmid selection , concentrated , and spread on a 3% low-melt agarose pad set on concave microscope slides . Images in Figure 2 were captured with a DeltaVision Real Time ( GE Healthcare Life Sciences; Picastaway , NJ ) microscope using a 60× , 1 . 4 NA objective and a CoolSnap HQ camera ( Photometrics; Tuscon , AZ ) . Z-series were taken with a 0 . 2 μm step size . Images were deconvolved using SoftWoRx software ( GE Healthcare Life Sciences ) , and ImageJ ( NIH; Bethesda , MD ) was used to make linear adjustments to images . For all other fluorescence images , including quantification of morphology and foci , Z-series of cells using a 0 . 2 μm step size were collected using the spinning-disk module of a Marianas SDC Real Time 3D Confocoal-TIRF microscope ( Intelligent Imaging Innovations; Denver , CO ) fitted with a 100× , 1 . 46 NA objective and either a Photometrics QuantEM EMCCD camera or a Hamamatsu ( Japan ) Orca Flash 4 . 0 sCMOS camera . Images were captured with SlideBook ( Intelligent Imaging Innovations ) and linear adjustments were made using ImageJ . Morphological and foci counting analysis was performed manually using ImageJ and for each experiment , at least 75 cells from at least three fields of view were quantified . To quantify the localization of Mic27 foci relative to Qcr2 in cells ( Figure 4D–F and Figure 4—figure supplement 4 ) , cells where Qcr2-mCherry was localized to mitochondrial subregions when compared to mito-TagBFP were chosen blind to Mic27 signal . Single plane images were compared to maximum z-projections to confirm the z-projection was representative of cell appearance . Qcr2 signal was considered positive when it was above an arbitrary threshold using the ‘Moments’ algorithm of ImageJ . Tracings of the mitochondrial perimeter and subregions of the perimeter adjacent to Qcr2 were manually performed . After identifying Qcr2-positive mitochondrial subregions , Mic27 foci were identified blind to Qcr2 signal and then compared to the Qcr2-positive regions of the tracings . Finally , Mic27 foci were scored as adjacent or not to Qcr2 signal in the z-projection and verified in the single plane images . Preparation of cells for morphology was performed as described previously described ( Rieder et al . , 1996 ) . Briefly , cells were pelleted and fixed in 3% glutaraldehyde contained in 0 . 1 M sodium cacodylate , pH 7 . 4 , 5 mM CaCl2 , 5 mM MgCl2 , and 2 . 5% sucrose for 1 hr at 22°C with gentle agitation , spheroplasted , embedded in 2% ultra low temperature agarose ( prepared in water ) , cooled , and subsequently cut into small pieces ( 1 mm3 ) . The cells were then postfixed in 1% OsO4/1% potassium ferrocyanide contained in 0 . 1 M cacodylate/5 mM CaCl2 , pH 7 . 4 , for 30 min at 22°C . The blocks were washed thoroughly 4 times with double-distilled H2O ( ddH2O; 10 min in total ) , transferred to 1% thiocarbohydrazide at 22°C for 3 min , washed in ddH2O ( 4 times for 1 min each ) , and transferred to 1% OsO4/1% potassium ferrocyanide in cacodylate buffer , pH 7 . 4 , for an additional 3 min at 25°C . The cells were washed 4 times with ddH2O ( 15 min in total ) , en bloc stained in Kellenberger's uranyl acetate for 2 hr to overnight , dehydrated through a graded series of ethanol , and subsequently embedded in Spurr resin . Sections were cut on an ultramicrotome ( Ultracut T; Reichert ) , poststained with uranyl acetate and lead citrate , and observed on a transmission electron microscope ( Tecnai 12;FEI; Hillsboro , OR ) at 100 kV . Images were recorded with a digital camera ( Soft Imaging System MegaView III; Olympus; Japan ) , and figures were assembled in Photoshop ( Adobe; San Jose , CA ) with only linear adjustments in contrast and brightness . Cells were grown to exponential phase in YPD , and whole cell lysates of 0 . 25 OD600 cells were obtained by alkaline extraction ( 0 . 255 M NaOH , 1% beta-mercaptoethanol ) followed by precipitation in 9% trichloroacetic acid . Precipitates were washed in acetone , dried , and resuspended in MURB protein sample buffer ( 100 mM MES , pH 7 . 0 , 1% SDS , 3 M urea , 10% β-mercaptoethanol ) . Immunopurifications were performed as previously described ( Hoppins et al . , 2011 ) , except 1000 OD600 cells were used . Briefly , cells were grown to exponential phase in YPD , resuspended in lysis buffer ( 20 mM HEPES pH7 . 4 , 150 mM KOAc , 2 mM Mg ( Ac ) 2 , 1 mM EGTA , 0 . 6 M sorbitol , and 1× Protease Inhibitor Mixture I [EMD Millipore; Billerica , MA] ) , flash-frozen dropwise in liquid N2 , and lysed using a Freezer/Mill ( SPEX; Metuchen , MJ ) . As described , the cell lysate was cleared , cross-linked for 30 min with 1 mM DSP ( Thermo Fisher Scientific; Waltham , MA ) , solubilized with 1% digitonin for 30 min , and pelleted again . The resulting supernatant was used for purifications with either 3 μg α-FLAG antibody ( 1:1000 , Sigma–Aldrich; St . Louis , MO ) and 50 μl μMACS protein G beads ( Miltenyi Biotec; San Diego , CA ) , or with 50 μl μMACS α-GFP MicroBeads ( Miltenyi Biotec ) , and beads were isolated with μ columns and a μMACS separator ( Miltenyi Biotec ) . For Western blot analysis , samples were eluted from beads with MURB sample buffer . For mass spectrometry proteomic analysis , samples were eluted using on-bead trypsin digestion as described previously and submitted to the University of California , Davis , Genome Center Proteomics Core . Sample processing and LC-MS/MS analysis were performed as previously described ( Hoppins et al . , 2011 ) . Raw data from each mass spectrometry run are shown in source data files . For comparative analysis of MICOS interactions in different strain backgrounds , we normalized using the relative number of total immunopurified spectral counts in the mutant extracts to wild type extracts . All graphs of proteomic analyses display the range of values from duplicate experiments . Mitochondria from wild type or ∆MICOS cells expressing Mic60-EGFP tagged at its endogenous locus were isolated by differential centrifugation as described previously ( Hoppins et al . , 2011 ) . Mitochondria were resuspended in lysis buffer ( 20 mM HEPES pH7 . 4 , 0 . 6 M sorbitol , 1 mM DTT , 0 . 1 M NaCl , 1× Protease Inhibitor Mixture I , and 1% Triton X-100 ) and incubated on ice for 30 min , followed by centrifugation at 50 , 000×g for 60 min ( TLA-100 , Beckman Coulter; Brea , CA ) . Total , supernatant , and pellet fractions were resuspended in equivalent volumes of MURB sample buffer , and analyzed by Western blotting . Samples were boiled for 5 min and analyzed by SDS-PAGE , transferred to PVDF or nitrocellulose , and immunoblotted with the following primary antibodies at the indicated concentrations: mouse α-FLAG ( 1:1000 , Sigma–Aldrich ) ; mouse α-GFP ( 1:2000 , UC Davis NeuroMab clone N86/8 ) or ( 1:2000 , Thermo Fisher Scientific clone GF28R ) ; rabbit α-mCherry ( 1:2000 , Thermo Fisher Scientific ) ; rabbit α-G6PDH ( 1:2000 , Sigma–Aldrich ) ; mouse α-Porin ( 1:1000 , Thermo Fisher Scientific ) . The appropriate secondary antibodies conjugated to DyLight 680 and DyLight 800 ( 1:10000 , Thermo Fisher Scientific ) were used and visualized with the Odyssey Infrared Imaging System ( LI-COR; Lincoln , NE ) . Linear adjustments to images were made using Adobe Photoshop . Crude mitochondrial extracts were isolated by differential centrifugations as described previously ( Hoppins et al . , 2011 ) . Blue native electrophoresis ( BN-PAGE ) was performed in 4–16% gradient gels according to the recommendation of the Novex NativePAGE Bis-Tris gel System ( Thermo Fisher Scientific ) . Briefly , 100 µg of isolated mitochondria was solubilized with 6 g digitonin per g mitochondrial protein . The extracts were centrifuged at 4°C for 15 min at 20 , 000×g , and aliquots of the supernatant ( 20 µl ) were immediately loaded on the top of a 4–16% polyacrylamide gel . After electrophoresis , the gel was divided into strips , which were incubated in different solutions at room temperature for 20 min to 1 hr to reveal in gel activity . In order to reveal in gel ATPase activity , gel strips were incubated in a solution of 5 mM ATP , 5 mM MgCl2 , 0 . 05% lead acetate , 50 mM glycine–NaOH pH 8 . 4 . For the cytochrome-c oxidase activity , strips were incubated in the following solution ( diaminobenzidine 0 . 6% [wt/vol] , bovine heart cytochrome c 1 . 2% [wt/vol] , 1 nM catalase , 50 mM H2PO4 , pH:7 ) . The Blue Native/SDS-PAGE two-dimensional analysis was performed as previously described ( Wittig et al . , 2006 ) . Briefly , BN-PAGE was first performed as above except mitochondria were solubilized with 1 g n-Dodecyl β-D-maltoside ( DDM ) per g mitochondrial protein . After BN-PAGE , the gel was divided into strips which were incubated in 1% SDS buffer for 2 hr . The strips were then placed on top of a 16% Tricine-SDS polyacrylamide gel ( Schagger , 2006 ) . After electrophoresis in the second dimension , the gel was stained by classical silver staining . | Structures called mitochondria provide energy that cells need to live and grow . To do this , mitochondria convert energy stored within sugars and other carbon-rich compounds into the energy currency of cells , a molecule called adenosine triphosphate ( called ATP for short ) . Defective mitochondria can cause cells to starve and also cause severe human diseases . A double membrane surrounds each mitochondrion . The outer membrane allows proteins and other substances to enter , while the inner membrane is elaborately folded and contains several groups of proteins—or complexes—including the respiratory complexes that generate ATP . Proper inner membrane folding is critically important . The membrane folds are held in place by structures called cristae junctions , which may also help to restrict proteins to particular areas of the inner membrane . A large inner membrane complex of proteins known as MICOS is important for organizing the inner membrane into folds , although exactly how it does so is not fully understood . MICOS consists of at least six different proteins , most of which are found across yeast and animal species . Friedman et al . have now analyzed how the MICOS complex assembles on the inner membrane in yeast cells using a combination of fluorescence and electron microscopy , proteomics and biochemistry . This revealed that in yeast , MICOS is made up of two independent sub-complexes bridged together by a protein called Mic19 , which additional experiments suggest controls the number and positions of the cristae junctions that hold the folds of the inner membrane in place . As part of the approach to understand MICOS complex organization , Friedman et al . removed the six MICOS proteins from yeast cells . Inside these cells , the inner mitochondrial membrane was misfolded . Furthermore , the respiratory complexes did not work normally and as a consequence the cells were unable to grow normally , suggesting that the correct distribution of respiratory complexes in the inner membrane is important for ATP production and depends on MICOS . These results indicate that MICOS stabilizes the structure of the inner membrane and organizes it into an efficient energy-generating machine . In many human mitochondrial diseases , the inner membrane of mitochondria folds incorrectly , in similar ways to the misfolding seen in the yeast cells that did not contain the MICOS complex . Therefore , the MICOS complex may also influence how these diseases develop . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"cell",
"biology"
] | 2015 | MICOS coordinates with respiratory complexes and lipids to establish mitochondrial inner membrane architecture |
Apoptosis is coupled with recruitment of macrophages for engulfment of dead cells , and with compensatory proliferation of neighboring cells . Yet , this death process is silent , and it does not cause inflammation . The molecular mechanisms underlying anti-inflammatory nature of the apoptotic process remains poorly understood . In this study , we found that the culture supernatant of apoptotic cells activated the macrophages to express anti-inflammatory genes such as Nr4a and Thbs1 . A high level of AMP accumulated in the apoptotic cell supernatant in a Pannexin1-dependent manner . A nucleotidase inhibitor and A2a adenosine receptor antagonist inhibited the apoptotic supernatant-induced gene expression , suggesting AMP was metabolized to adenosine by an ecto-5’-nucleotidase expressed on macrophages , to activate the macrophage A2a adenosine receptor . Intraperitoneal injection of zymosan into Adora2a- or Panx1-deficient mice produced high , sustained levels of inflammatory mediators in the peritoneal lavage . These results indicated that AMP from apoptotic cells suppresses inflammation as a ‘calm down’ signal .
Approximately , 100 , 000 cells undergo apoptosis every second in our bodies , and are engulfed by macrophages and immature dendritic cells ( Fuchs and Steller , 2011 ) . Inefficient clearance of apoptotic cells causes systemic lupus erythematosus ( SLE ) -type autoimmune disease in humans and mice ( Botto et al . , 1998; Hanayama et al . , 2004; Munoz et al . , 2010 ) . Apoptotic cells characteristically expose ‘eat me’ signal ( s ) to macrophages ( Nagata et al . , 2010; Ravichandran , 2011 ) , and one of the primary ‘eat me’ signals is phosphatidylserine ( PtdSer ) . PtdSer is a phospholipid that localizes to the inner leaflet of plasma membranes in healthy cells; however , when cells undergo apoptosis , it is exposed on the outer cell surface in a caspase-dependent manner ( Fadok et al . , 2001; Leventis and Grinstein , 2010; Suzuki et al . , 2013 ) . Macrophages express membrane proteins that function as receptors for PtdSer ( e . g . , Tim-4 , stabllin 2 , and BAI1 ) , or opsonins that bind to PtdSer ( e . g . MFG-E8 , Protein S , and Gas6 ) ( Nagata et al . , 2010 ) . Masking the PtdSer on apoptotic cells prevents their engulfment by macrophages , supporting the idea that PtdSer is an important ‘eat me’ signal . Apoptotic cells also produce ‘find me’ signals to attract macrophages ( Ravichandran , 2011 ) . Lysophosphatidylcholine ( LPC ) ( Lauber et al . , 2003 ) , ATP/UTP ( Elliott et al . , 2009 ) , fractalkine ( Truman et al . , 2008 ) , and sphingosine-1-phosphate ( S1P ) ( Gude et al . , 2008 ) are released from apoptotic cells in a caspase-dependent manner . Macrophages express specific receptors for these signals ( P2Y2 for ATP and UTP , CX3CR1 for fractalkine , and S1PR1 for S1P ) , which may mediate migration to the dying cells ( Ravichandran , 2011 ) . The ‘find me’ signals are thought to prime macrophages for engulfment , for example by enhancing the expression of MFG-E8 ( Miksa et al . , 2007 ) . On the other hand , LPC , ATP/UTP , and S1P are known to cause inflammation ( Trautmann , 2009; Maceyka et al . , 2012; Meyer zu Heringdorf and Jakobs , 2007 ) , which may contradict the anti-inflammatory nature of the apoptotic process . In addition , several groups showed that apoptotic cells produce growth-stimulating molecules such as Wnt3 or prostaglandin E2 ( Chera et al . , 2009; Huang et al . , 2011; Mollereau et al . , 2013 ) , which is also against the silent nature of the process . Thus , how the anti-inflammatory character of the apoptosis is maintained during the process has been unclear . In this study , we found that soluble factors derived from the apoptotic cells activated macrophages to express immediate early response genes including anti-inflammatory genes such as Nuclear receptor 4A ( Nr4a ) and Thrombospondin ( Thbs ) 1 . Using Thbs1 gene expression as a marker , the molecule responsible for the gene expression was identified as AMP . The overexpression of Pannexin 1 in apoptotic cells accelerated the accumulation of AMP in the culture supernatant of apoptotic cells , while a caspase-resistant form of Pannexin 1 inhibited the AMP accumulation . An inhibitor of ecto-5′-nucleotidase blocked the AMP-induced activation of the Thbs1 gene in macrophages , while macrophages from A2a adenosine receptor ( Adora2a ) -deficient mice did not respond to AMP . These results indicated that AMP from apoptotic cells was converted to adenosine by an ecto-5′-nucleotidase on macrophages , and that the adenosine activated a set of genes in macrophages via the A2a adenosine receptor . Mice deficient in Adora2a or Panx1 exhibited prolonged peritonitis following intraperitoneal zymosan injection , suggesting that the AMP released from apoptotic peritoneal cells exerted an anti-inflammatory effect by activating the A2a adenosine receptor .
If apoptotic cells produce ‘danger’ or ‘anti-danger’ signal ( s ) , we rationalized that such signals would activate gene expression in macrophages . To investigate this possibility , we examined the effect of the culture supernatant from apoptotic cells on macrophage gene expression . Mouse WR19L transformants expressing Fas ( W3 cells ) were treated with Fas ligand ( FasL ) for 30 min , washed , and then further incubated for 60 min . Following FasL treatment , more than 90% of the W3 cells were Annexin V positive , and only small percentage were positive for both Annexin V and propidium iodide ( PI ) ( Figure 1—figure supplement 1 ) , indicating that the majority of cells had undergone apoptosis but not necrosis . Mouse bone marrow-derived macrophages ( BMDMs ) were then incubated for 1 hr with the supernatant of FasL-treated W3 cells , and subjected to microarray analysis . As shown in Figure 1A , the mRNA levels of N-myc ( Mycn ) , Nr4a orphan nuclear receptor family members , Egr transcription factors ( Egr2 and Egr3 ) , Thrombospondin 1 ( Thbs1 ) , and Il-1β were 15- to 200-fold higher in the macrophages treated with apoptotic cell supernatant than in the control , untreated macrophages . A real-time RT-PCR analysis confirmed that the supernatants of apoptotic cells but not of healthy cells strongly induced the expression of Nr4a1 , Nr4a2 , and Thbs1 ( Figure 1B ) . When W3 cells were treated with FasL in the presence of Q-VD-OPh , a caspase inhibitor ( Caserta et al . , 2003 ) , the ability of the supernatant to upregulate the Thbs1 gene was abrogated , indicating that the factor ( s ) responsible for upregulating Thbs1 gene were generated in a caspase-dependent manner ( Figure 1C ) . Thbs1 and Nr4a are known to suppress inflammation ( Lopez-Dee et al . , 2011; McMorrow and Murphy , 2011 ) , and a danger signal such as ATP is unlikely to activate these genes . 10 . 7554/eLife . 02172 . 003Figure 1 . Factor ( s ) released from apoptotic cells stimulate gene expression in macrophages . ( A and B ) BMDMs were incubated for 1 hr with medium or with the supernatant of W3 cells that had been treated with ( apoptotic ) or without ( living ) 30 units/ml FasL . RNA from BMDMs was then subjected to microarray analysis . ( A ) Genes whose expression was upregulated more than 10-fold after incubation with the apoptotic cell supernatant are listed . ( B ) Nr4a1 , Nr4a2 , and Thbs1 mRNA levels were quantified by real-time RT-PCR , and normalized to Gapdh mRNA . ( C ) W3 cells were pre-treated with or without 20 μM Q-VD-OPh for 20 min and stimulated with or without 30 units/ml FasL . BMDMs were then incubated for 1 hr with the supernatant of Q-VD-OPh-treated ( + ) or untreated ( − ) living or FasL-treated apoptotic W3 cells , and Thbs1 mRNA levels were determined by real-time RT-PCR . ( D ) BMDMs were incubated with the supernatant of apoptotic W3 cells that had been treated with proteinase K ( proK ) , DNase I or RNase A , and Thbs1 mRNA levels were determined . ( E ) Medium , the culture supernatant of healthy W3 cells ( living ) or apoptotic W3 cells ( apop ) were subjected to ultrafiltration through a 10 kDa-cutoff filter , and the filtrate ( <10 kDa ) and concentrate ( >10 kDa ) were tested for their ability to induce Thbs1 expression in BMDMs . Experiments were performed in triplicates , and the average values are plotted with SD ( bars ) . All experiments were repeated at least twice with BMDM from different mice , and representative data are shown . DOI: http://dx . doi . org/10 . 7554/eLife . 02172 . 00310 . 7554/eLife . 02172 . 004Figure 1—figure supplement 1 . FasL-induced apoptosis in W3 cells . W3 cells treated with or without 30 units/ml FasL for 90 min were stained with a Cy5-labeled Annexin V and PI and analyzed by flow cytometry . The percentage of positively stained cells in each quadrant is indicated . DOI: http://dx . doi . org/10 . 7554/eLife . 02172 . 004 Treatment of the apoptotic cell supernatant with proteinase K ( 50 μg/ml for 60 min ) , DNase I ( 6 U/ml for 60 min ) , or RNase A ( 5 μg/ml for 60 min ) did not prevent its ability to enhance Thbs1 gene expression ( Figure 1D ) , suggesting that the factor ( s ) were not proteins or polynucleotides . When the supernatant was subjected to centrifugal ultrafiltration with a filter with a nominal cutoff of 10 kDa , most of the activity was found in the filtrate , and not in the concentrate ( Figure 1E ) . These results indicated that the molecular weight of the factor ( s ) that activated the macrophages were less than 10 kDa , and may have been present as a free form . To identify the molecule ( s ) present in the apoptotic cell supernatant , the supernatant was subjected to LC-MS analysis . As shown in Figure 2A , various nucleotides and nucleosides such as AMP , UMP , cytidine , inosine , hypoxanthine , and uridine were present in the FasL-treated cell supernatant . 10 . 7554/eLife . 02172 . 005Figure 2 . Identification of the factor in apoptotic cell supernatant that stimulates macrophage gene expression . ( A ) The supernatant of apoptotic W3 cells was subjected to LC-MS analysis . The relative concentration of each compound is represented by the base peak intensity . ( B ) BMDMs were incubated with medium containing 10 μM of the indicated reagents for 1 hr , and Thbs1 mRNA levels were determined by real-time RT-PCR . ( C ) Apoptotic W3 cell supernatant was pretreated with 25 mU/ml apyrase at 37°C for 1 hr . BMDMs were incubated for1 hr with the pretreated or untreated supernatant , and the Thbs1 mRNA levels were determined . BMDMs were also treated with the supernatant in the presence of 10 μM AOPCP or 0 . 9 U/ml adenosine deaminase ( ADA ) , and the Thbs1 mRNA levels were quantified as above . ( D ) BMDMs were incubated with apoptotic cell supernatant ( apop ) or medium supplemented with the indicated concentrations of adenosine ( Ado ) , and the Thbs1 mRNA levels were determined . ( E ) The mRNA levels of the Adora1 , Adora2a , Adora2b , and Adora3 expressed in BMDMs and thio-pMacs were determined by real-time RT-PCR , and normalized to β-actin mRNA . ( F ) BMDMs and thio-pMacs were incubated with W3 apoptotic cell supernatant and adenosine receptor antagonists , 5 nM 8-cyclopentyl-1 , 3-dipropylxanthine ( DPCPX ) ( A1 ) , 10 nM SCH58261 ( A2a ) , 5 μM alloxazine ( A2b ) , or 130 nM VUF5574 ( A3 ) , and Thbs1 mRNA levels were determined . ( G ) , Thio-pMacs from Adora2a+/+ ( control ) or Adora2a−/− mice were incubated with medium , apoptotic or living W3 cell supernatants , and the Nr4a1 and Thbs1 mRNA levels were determined by real-time RT-PCR . Experiments were performed in triplicate , and the average values are plotted with SD ( bars ) . All experiments were repeated at least twice with BMDM or thio-pMacs from different mice , and representative data are shown . DOI: http://dx . doi . org/10 . 7554/eLife . 02172 . 005 To examine which molecule ( s ) in the apoptotic cell supernatant were responsible for the enhanced Thbs1 gene expression , macrophages were incubated with the various compounds found in the apoptotic cell supernatant at 10 μM . The results indicated that AMP , but none of other nucleosides or nucleotides , activated macrophages ( Figure 2B ) . Apyrase , an enzyme that catalyzes the hydrolysis of ATP to AMP , had no effect on the apoptotic cell supernatant activity , indicating that the factor was not ATP or ADP ( Figure 2C ) . In contrast , adenosine-5’-O- ( α , β- methylenediphosphate ) ( AOPCP ) , a non-hydrolyzable ADP analog that inhibits the ecto-5′-nucleotidase , blocked the Thbs1 gene expression induced by the apoptotic cell supernatant ( Figure 2C ) , indicating that adenosine , but not AMP , is responsible for activation of the Thbs1 gene in macrophages . Treatment of the apoptotic cell supernatant with adenosine deaminase ( ADA ) , that converts adenosine to inosine abrogated its ability to activate Thbs1 gene expression ( Figure 2C ) , confirming that adenosine was responsible for the activity . Accordingly , treating BMDMs with adenosine activated the Thbs1 gene in a dose-dependent manner ( Figure 2D ) . Since adenosine was undetectable in the supernatant of the FasL-treated W3 cells , we concluded that AMP in the supernatant of these cells was converted to adenosine by the ecto-5′-nucleotidase expressed on macrophages , resulting in the stimulation of gene expression in macrophages . Extracellular adenosine is recognized by the G-protein coupled adenosine receptor family ( Fredholm , 2007 ) . A real-time RT-PCR analysis indicated that BMDMs and thioglycollate-elicited peritoneal macrophages ( thio-pMacs ) did not express A1 or A3 adenosine receptor ( Adora1 or Adora3 ) , but predominantly expressed the A2b and A2a receptors ( Adora2b and Adora2a ) , respectively ( Figure 2E ) . Accordingly , an A2b antagonist ( alloxazine ) inhibited the apoptotic cell supernatant-induced Thbs1 gene expression in BMDMs , but only weakly inhibited the induced Thbs1 gene expression in thio-pMacs ( Figure 2F ) . An A2a antagonist ( SCH58261 ) had an inhibitory effect on thio-pMacs , but not on BMDMs , whereas antagonists of A1 ( DPCPX ) and A3 ( VUF5574 ) had no effect on thio-pMacs or BMDMs . Adora2a−/− thio-pMacs lost their ability to upregulate the expression of Nr4a1 and Thbs1 genes in response to the apoptotic cell supernatant ( Figure 2G ) , confirming the essential role of the A2a receptor in this process . These results confirmed that adenine nucleotides released from apoptotic cells were converted to adenosine , which in turn activated macrophages by binding to adenosine receptors . Apoptotic cells are reported to release ATP ( Elliott et al . , 2009 ) , and AMP can be generated from ATP by ectonucleotide pyrophosphatase/phosphodiesterase ( E-NTPDase ) ( Bours et al . , 2006 ) . To examine whether these events were involved in the apoptotic supernatant activity , the release of adenine nucleotides from apoptotic cells was followed over time . W3 cells were treated with FasL , and aliquots of the culture supernatant were collected at different time points and analyzed by LC-MS . As shown in Figure 3A , healthy cells secreted low levels of AMP , whereas FasL-induced apoptosis caused a gradual increase of AMP in the supernatant . At 2 hr after the addition of FasL , the AMP level had increased to more than 4 μM in a culture of 1 . 0 × 107 cells/ml . The concentration of ADP also increased , but was less pronounced compared with that of AMP . In contrast , at 2 hr the concentration of ATP was substantially lower , 40 nM or 40 pmoles per 107 cells . In various apoptosis systems , the intracellular ATP level is reported to remain high ( Eguchi et al . , 1997; Bossy-Wetzel et al . , 1998 ) . However , the intracellular concentration of ATP in the FasL-treated W3 cells rapidly decreased , whereas the concentrations of ADP and AMP concomitantly increased ( Figure 3B ) , suggesting that AMP was directly released from apoptotic cells . Exogenously added ATP was slowly degraded in the supernatant of FasL-treated W3 cells ( Figure 3—figure supplement 1 ) providing additional support that AMP was released directly from the apoptotic cells , and not extracellularly converted from ATP . When mouse thymocytes in a culture of 1 × 108 cells/ml were treated with FasL for 90 min to induce apoptosis in approximately 70% of the cells ( 8% were PI positive ) , the culture supernatant was found to contain 3 . 5 μM AMP , whereas the supernatant of healthy cells cultured for 1 hr contained approximately 0 . 2 μM AMP ( Figure 3C ) . Consistent with our finding using W3 cells , the levels of ATP and ADP in the FasL-treated thymocyte were 40 and 100 times lower than that of AMP , respectively . These results further suggest that AMP was directly released from primary thymocytes undergoing apoptosis . 10 . 7554/eLife . 02172 . 006Figure 3 . Release of adenine nucleotides from apoptotic cells . ( A and B ) W3 cells were treated with ( ● , ■ , ▲ ) or without ( ○ , □ , △ ) 120 units/ml FasL , and the concentrations of AMP ( ○ , ● ) , ADP ( □ , ■ ) and ATP ( △ , ▲ ) in the supernatant ( A ) and cells ( B ) were determined by LC-MS or HPLC analysis at the indicated times . ( C ) Mouse thymocytes were treated with ( + ) or without ( − ) 120 units/ml FasL for 90 min , and the concentrations of AMP , ADP , and ATP in the supernatants were determined . Experiments were performed in triplicate , and the average values are plotted with SD ( bars ) . The treatment of cells with FasL and quantification of adenine nucleotides were repeated at least twice , and representative data are shown . DOI: http://dx . doi . org/10 . 7554/eLife . 02172 . 00610 . 7554/eLife . 02172 . 007Figure 3—figure supplement 1 . Minimal degradation of ATP in the apoptotic cell supernatant . ATP was exogenously added to the supernatants of FasL-treated or FasL-untreated W3 cells at a final concentration of 10 μM . At the indicated times , ATP was quantified in triplicate using the luciferase system as described in ‘Materials and methods’ , and the average values are plotted with SD ( bars ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02172 . 007 The AMP release from FasL-treated W3 cells was suppressed by a caspase inhibitor , Q-VD-OPh , in a dose-dependent manner ( Figure 4A ) . Chekeni et al . ( 2010 ) reported that Pannexin 1 is cleaved by caspase in apoptotic cells and mediates the release of ATP from these cells . Pannexins are a family of plasma membrane channel proteins consisting of 3 members ( Baranova et al . , 2004 ) . W3 cells and thymocytes expressed Panx1 , but not Panx2 or Panx3 ( Figure 4—figure supplement 1 ) . A synthetic derivative of glycyrrhetinic acid ( CBX , carbenoxolone ) , an inhibitor of Pannexin and Connexin channels ( D’Hondt et al . , 2009 ) , inhibited the release of adenine nucleotides ( AMP , ADP , and ATP ) from FasL-treated W3 cells , whereas gadolinium chloride ( GdCl3 ) , an inhibitor of the maxi-anion channel , failed to inhibit their release from the apoptotic cells ( Figure 4A ) . Mouse Pannexin 1 contains a caspase 3 recognition sequence ( Asp-Ile-Ile-Asp ) at amino acids 375 to 378 . Therefore , we established W3 cells expressing wild-type or caspase-resistant Pannexin1 ( Figure 4—figure supplement 2 ) . As shown in Figure 4B , W3 transformants expressing wild-type Pannexin 1 released significantly higher levels of adenine nucleotides than the parental W3 cells upon FasL treatment . The caspase 3 recognition sequence of Pannexin 1 was then mutated to a caspase-resistant sequence ( Ala-Ile-Ile-Ala ) . When W3 cells were transformed with the caspase-resistant form of Pannexin 1 , the transformants completely lost their ability to release adenine nucleotides following FasL treatment ( Figure 4B ) , indicating that the caspase cleavage of Pannexin 1 was required for the release of adenine nucleotides . Co-culturing BMDM with FasL-treated W3 cells upregulated the Thbs1 gene expression in BMDMs ( Figure 4C ) , this was not observed when the W3 cells were transformed with the caspase-resistant form of pannexin1 . Furthermore , Panx1−/− thymocytes exhibited reduced AMP release following FasL treatment compared with similarly treated wild-type thymocytes ( Figure 4D ) . Incomplete inhibition of the AMP release by the Panx1−/− thymocytes may have been due to the expression of connexins in mouse thymocytes , or because a significant fraction of the thymocytes had undergone secondary necrosis ( data not shown ) . These results indicated that caspase activation resulted in the opening of Pannexin 1 channels , leading to AMP release . 10 . 7554/eLife . 02172 . 008Figure 4 . Pannexin 1-dependent release of adenine nucleotides from apoptotic cells . ( A ) After pre-treatment with the indicated concentrations of Q-VD-OPh ( QVD ) , carbenoxolone ( CBX ) , or GdCl3 for 20 min , W3 cells were incubated with 120 units/ml FasL for 90 min . The AMP , ADP , and ATP concentrations in the supernatants were determined by LC-MS . ( B ) W3 cells transformed with empty pMXs vector ( ■ , □ ) , or with vectors bearing wild-type ( wt ) ( ● , ○ ) or caspase-resistant mutant ( mt ) ( ▲ , △ ) Pannexin 1 were treated ( ■ , ● , ▲ ) or not treated ( □ , ○ , △ ) with 120 units/ml FasL , and the concentrations of adenine nucleotides in the supernatants were determined at the indicated times . ( C ) The parental W3 cells ( wt ) and W3 cell transformants expressing caspase-resistant form of pannexin 1 ( mt ) were treated with 30 units/ml FasL for 30 min , added to BMDMs , and incubated for 1 hr . The Thbs1 mRNA level in BMDMs was then determined by real-time RT-PCR . ( D ) Thymocytes from Panx1+/+ or Panx1−/− mice were incubated with ( + ) or without ( − ) 120 units/ml FasL , and the concentration of adenine nucleotides in the supernatants were determined at the indicated times . Experiments were performed in triplicate , and the average values are plotted with SD ( bars ) . All experiments were repeated at least twice , and the representative data are shown . DOI: http://dx . doi . org/10 . 7554/eLife . 02172 . 00810 . 7554/eLife . 02172 . 009Figure 4—figure supplement 1 . Panx mRNA expression in W3 cells and mouse thymocytes . Panx1 , Panx2 , and Panx3 mRNA expression in W3 cells and mouse thymocytes was quantified in triplicate by real-time RT-PCR . The average values for each mRNA are expressed as relative values to β-actin mRNA with SD ( bars ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02172 . 00910 . 7554/eLife . 02172 . 010Figure 4—figure supplement 2 . Expression of wild-type and caspase-resistant Pannexin 1 in W3 cell transformants . Expression plasmids containing wild-type and mutant Panx1 genes were introduced into W3 cells . The Panx1 mRNA levels in each transformant were quantified in triplicate by real-time RT-PCR . The average Panx1 mRNA levels were normalized to the endogenous Panx1 mRNA , and expressed as ‘relative expression’ against the endogenous Panx1 with SD ( bars ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02172 . 010 The above results suggested that the AMP released from apoptotic cells is converted to adenosine by macrophages , and stimulates a set of genes in macrophages via activation of adenosine receptor . Among adenosine receptors , the A2 receptors are known to exert an anti-inflammatory effect on macrophages ( Sitkovsky and Ohta , 2005 ) . To confirm that AMP could exert a similar effect on macrophages , mouse resident peritoneal macrophages were treated with zymosan , a glucan that stimulates macrophages via TLR2 ( Takeuchi et al . , 1999 ) , in the presence or absence of AMP . As shown in Figure 5A , zymosan-stimulated macrophages to produce TNFα , a classic pro-inflammatory cytokine , and MIP-2 ( CXCL-2 ) , a critical chemokine for neutrophil migration . The production of these factors was significantly suppressed by the presence of AMP . 10 . 7554/eLife . 02172 . 011Figure 5 . Immunosuppressive function of Pannexin 1 and A2a adenosine receptor in a mouse peritonitis model . ( A ) Mouse resident peritoneal macrophages were incubated with 50 μg/ml zymosan A in the presence ( + ) or absence ( − ) of 100 μM AMP for 16 hr . The TNFα and MIP-2 levels in the culture supernatant were determined in triplicate by ELISA , and the average values are plotted with S . D . ( bars ) . The experiments were performed at least twice , and representative data are shown . ( B ) Peritoneal cells were collected at 3 hr after the injection of zymosan A ( 250 mg/kg ) , and stained with anti-Ly6B . 2 and anti-Ly6G antibodies ( left panel ) . Some of the Ly6B . 2 and Ly6G double-positive cells ( circled in the left panel ) were stained with Cy5-labeled Annexin V ( right panel ) . ( C and D ) After zymosan injection , peritoneal lavage fluid was collected at 2 hr ( n = 5 ) or 6 hr ( n = 7 ) of Panx1+/+ , and at 2 hr ( n = 5 ) or 6 hr ( n = 10 ) of Panx1−/− littermates ( C ) , or 2 hr ( n = 10 ) or 6 hr ( n = 6 ) of Adora2a+/+ or +/− , and at 2 hr ( n = 8 ) or 6 hr ( n = 6 ) of Adora2a−/− mice ( D ) . The TNFα and MIP-2 levels in the fluids were determined by ELISA . The Student’s t test was used for the statistical analysis , and the p values are shown . DOI: http://dx . doi . org/10 . 7554/eLife . 02172 . 011 We next examined the role of AMP secretion from apoptotic cells using the zymosan-induced peritonitis model ( Cash et al . , 2009 ) . 3 hr after the injection of zymosan particles into the mouse peritoneal cavity , 30% of the peritoneal cells were Ly6B . 2 and Ly6G double positive , and 20% of the double-positive cells were Annexin V positive , indicating that neutrophils had infiltrated into the cavities and undergone apoptosis ( Figure 5B ) . 2 hr after the injection of zymosan-particles , the levels of TNFα and MIP-2 in the peritoneal lavage were comparable between the wild-type and Panx1−/− mice . However , by 6 hr the levels of these inflammatory mediators in the peritoneal cavity remained high in Panx1−/− mice , but had returned to normal in the wild-type mice ( Figure 5C ) . These results indicated that Pannexin 1 was required to quickly resolve the zymosan-induced peritonitis . Consistent with the results from Panx1−/− mice , the levels of inflammatory mediators in Adora2a−/− peritoneal lavage fluids were greater than those in the Adora2a+/+ or Adora2a+/− lavage at 6 hr after zymosan injection ( Figure 5D ) . Taken together , these results suggested that apoptotic neutrophils released adenosine nucleotides , most likely AMP , through Pannexin 1 channels during acute peritoneal inflammation . The secreted AMP was , in turn , converted to adenosine by macrophages , leading to the activation of adenosine A2a receptor and the upregulation of Nr4a and Thbs1 gene expression , ensuring the rapid resolution of peritonitis .
Adenine nucleotides act not only as intracellular energy carriers but also as extracellular signaling molecules ( Bours et al . , 2006 ) . For instance , extracellular ATP regulates the microglial response to brain injury via P1 or P2 purinergic receptors ( Davalos et al . , 2005 ) , and adenosine modulates sleep–wake cycles via A2a adenosine receptor ( Urade et al . , 2003 ) . Although the release of adenine nucleotides mainly occurs by exocytosis ( Pascual et al . , 2005 ) , it is also mediated by transport via channel proteins . Connexins , which form gap junctions , can serve as hemichannels , and allow ATP to exit ( Cotrina et al . , 1998; Spray et al . , 2006 ) . Pannexins , which carry 4 transmembrane regions with intracellular N- and C-terminal tails , are structurally similar to connexins and also form hemichannels ( Yen and Saier , 2007 ) . Various physiological and pathological stimuli such as Ca2+-influx , shear stress , and apoptotic cell death , can cause Pannexin 1 channels to open , allowing the passage of molecules smaller than 1 . 0 kDa ( D’Hondt et al . , 2009 ) . Chekeni et al . ( 2010 ) recently showed that in the early stages of apoptosis , caspases cleave Pannexin 1 resulting in the release of ATP , which can serve as a ‘find me’ signal to attract macrophages . In this study , we found that AMP was present at 30- to 100-fold higher concentrations than ATP in the culture supernatants of apoptotic thymocytes and a T cell line . The ATP concentration in healthy cells is approximately 10–50 times higher than that of AMP . However , when cells undergo apoptosis , ATP is quickly hydrolyzed to AMP , while its generating system is inactivated by caspases ( Ricci et al . , 2004 ) . The caspase-cleaved Pannexin channel also contributes to the cellular loss of ATP by allowing ATP to exit cells through the plasma membrane . Thus , the AMP found in apoptotic cell supernatants could have been extracellularly converted from ATP by E-NTPDase . However , the ability of apoptotic or healthy W3 cell supernatants to hydrolyze ATP was very low , suggesting that E-NTPDase was not responsible for the accumulation of AMP in the apoptotic cell supernatants . Thus , it is possible that AMP generated intracellularly in apoptotic cells was released from through these channels . The engulfment of apoptotic cells is a silent process that does not activate inflammation ( Nagata et al . , 2010; Ravichandran , 2011 ) . Since macrophages ingesting apoptotic cells produce TGFβ or IL-10 that inhibits the production of inflammatory cytokines ( Voll et al . , 1997; Fadok et al . , 1998; Lucas et al . , 2006 ) , this interaction and/or engulfment was thought to be responsible for the anti-inflammatory effect . Furthermore , Köröskényi et al . ( 2011 ) recently reported that macrophages engulfing apoptotic cells released adenosine to suppress inflammation . In contrast , we show here that apoptotic cells secrete AMP , which is converted to adenosine , resulting in macrophage activation via adenosine receptors , followed by the upregulation of Thbs1 and Nr4a gene expression . When the apoptotic cells could not release AMP due to the expression of caspase-resistant form of pannexin 1 , the Thbs1 gene expression was not up-regulated . Furthermore , the AMP released from the resident peritoneal macrophages engulfing apoptotic cells was significantly smaller than that released from apoptotic cells ( HY , and SN , data not shown ) . These results suggest that the contribution of the interaction of apoptotic cells with macrophages to anti-inflammatory effect may not be great . Thbs1 is the major activator of TGFβ ( Crawford et al . , 1998 ) , whereas Nr4a family members inhibit the expression of proinflammatory cytokines such as TNFα , IL-8 , and IL-6 in macrophages ( Bonta et al . , 2006 ) , by recruiting a repressor complex to their promoter ( Saijo et al . , 2009 ) . Thus , this may be one mechanism that suppresses inflammation during the apoptosis process . The conversion of AMP to adenosine is catalyzed by CD73 , an ecto-5’-nucleotidase , expressed on the surface of endothelial and immune cells ( Deaglio and Robson , 2011 ) . CD73-null mice spontaneously develop autoimmune diseases such as glomerulitis and peritubular capillaritis , accompanied by enhanced inflammation ( Blume et al . , 2012 ) . They are also susceptible to polymicrobial sepsis induced by cecal ligation and puncture , and exhibit increased mortality and systemic inflammation ( Hasko et al . , 2011 ) . Our results suggest that the substrate for CD73 in these models may be AMP released from apoptotic cells . If AMP cannot be converted to adenosine in the absence CD73 , then activation of the immune system in response to the ‘danger signal ( s ) ’ released from apoptotic cells would not be opposed . We conclude that apoptotic cells themselves contribute to the ‘anti-inflammatory’ nature of the engulfment process by secreting AMP as an immunosuppressive ‘calm-down’ signal . Depending on the types of cells and apoptotic stimuli , intracellular ATP levels remain high ( Bossy-Wetzel et al . , 1998; Eguchi et al . , 1997; Zamaraeva et al . , 2005 ) or rapidly decreases ( Gleiss et al . , 2002; Vander Heiden et al . , 1999 ) during apoptosis . It will be interesting to study whether this mechanism impacts various types of inflammation that are accompanied by apoptosis .
C57BL/6J mice were purchased from Japan SLC . Adenosine receptor A2a ( Adora2a ) knockout mice ( Chen et al . , 1999 ) were from Dr Jian-Fan Chen ( Boston University School of Medicine ) . Pannexin 1 knockout mice ( Panx1tm1a ( KOMP ) Wtsi ) were purchased from the University of California at Davis’ KOMP Repository ( http://www . komp . org/ ) . All mice were housed in a specific pathogen-free facility at Kyoto University , Graduate School of Medicine . The committee of Kyoto University , Graduate School of Medicine , approved our study ( Reference Number: Med Kyo 12 , 107 ) . The leucine-zipper tagged soluble human Fas ligand ( FasL ) was prepared as described previously ( Shiraishi et al . , 2004 ) . One unit of FasL was defined as the dilution that gives a half-maximum response with W3 cells , mouse WR19L cells expressing mouse Fas ( Watanabe-Fukunaga et al . , 1992 ) . Adenosine deaminase , apyrase , DNase I , RNase A and proteinase K were purchased from Calbiochem ( Germany ) , New England BioLabs ( Ipswich , MA ) , Takara Bio ( Japan ) , Ambion ( Austin , TK ) , and Merck ( Germany ) , respectively . A caspase inhibitor , N- ( 2-Quinolyl ) valyl-aspartyl- ( 2 , 6-difluorophenoxy ) methyl Ketone ( Q-VD-OPh ) was from R&D Systems ( Minneapolis , MN ) . Adenosine , AMP , ADP , and ATP were from Nacalai ( Japan ) . Cytidine , inosine , guanine , hypoxanthine , uridine , guanosine , taurine , UMP , CDP-choline CMP , and α , β-methyleneadenosine 5′-diphosphate ( AOPCP ) were from Sigma ( St . Louis , MO ) . Antagonists for A1 ( 8-cyclopentyi-1 , 3-dipropylxanthine ) , A2a ( SCH58261 ) , A2b ( Alloxazine ) , and A3 Adenosine receptor ( VUF5574 ) were from Sigma . Carbenoxolone , an inhibitor of Pannexin channel , was purchased from Sigma . Gadolinium chloride hexahydrate ( GdCl3 . 6H2O ) , an inhibitor of maxi anion channel , was from Aldrich ( St . Louis , MO ) . For mouse bone marrow-derived macrophages ( BMDMs ) , bone marrow cells from female C57BL/6J mice at 8 weeks of age were treated with Buffer EL ( Qiagen ) to remove red blood cells , and cultured for more than 7 days with DMEM containing 10% FCS supplemented with 5% supernatant of CMG14-12 cells producing mouse M-CSF ( Takeshita et al . , 2000 ) . For thioglycollate-elicited peritoneal macrophages ( thio-pMacs ) , mice were injected i . p . with 3% thioglycollate ( Sigma ) . Peritoneal excudate cells were obtained 4 days later , and cultured in DMEM containing 10% FCS . For resident peritoneal macrophages ( rpMacs ) , mouse peritoneal cells were cultured at 37°C for 2 hr in RPMI1640 containing 10% FCS on 12-well plates . Adherent cells were washed once with fresh medium , and used as rpMac . W3 cells were treated at 2 × 106 cells/ml with human FasL at 37°C for 30 min , washed , re-suspended at a concentration of 1 × 107 cells/ml with RPMI containing 1% FCS , and further incubated for 60 min at 37°C . At this stage , 94% cells were Annexin V-positive and PI-negative . The same treatment without FasL generated about 3% Annexin V-positive cells . The culture supernatant of W3 cells was recovered , 0 . 5 ml-aliquots were added to mouse BMDMs ( 2 × 105 cells ) or thio-pMacs ( 5 × 105 cells ) in a 12-well microtiter plate . To study the effect of apoptotic cells , the FasL-treated W3 cells ( 5 × 106 ) were washed , re-suspended in RPMI containing 1% FCS at a concentration of 1 × 107 cells/ml , and added to BMDMs . After incubation at 37°C for 1 hr , the macrophages were washed with PBS , and total RNA for RT-PCR was prepared by using RNeasy mini kit ( Qiagen ) . To study the effect of zymosan on the gene expression , rpMacs ( 6 × 105 cells ) in a 12-well plate were treated with 50 μg/ml of zymosan and 100 μM AMP for 16 hr . The fractionation of the supernatant of apoptotic W3 cells was carried out with a 10 kDa cutoff ( MWCO ) ultrafilter ( Amicon Ultra , Millipore ) . For microarray analysis , RNA was amplified and labeled using GeneChip 3’ IVT Express Kit ( Affymetrix , Santa Clare , CA ) , and the labeled probe was used for hybridization to a Mouse Genome 430 2 . 0 Array GeneChip ( Affymetrix ) . The hybridized signals were detected with an Affymetrix Agilent Microarray scanner , and the array image was analyzed with Microarray Analysis Suit Version 5 . 0 ( Affymetrix ) . Real-time RT-PCR was performed with LightCycler 480 instrument ( Roche , Indianapolis , IN ) after total RNA was reverse transcribed by High Capacity RNA-to-cDNA kit ( Applied Biosystems , Foster City , CA ) . Following primers were used: 5′-ATGCCTCCCCTACCAATCTTC-3′ and 5′-CACCAGTTCCTGGAACTTGGA-3′ for Nr4a1; 5′-TCACCTCCGGTGAGTCTGATC-3′ and 5′-TGCTGGATATGTTGGGTATCATCT-3′ for Nr4a2; 5′-GGTGCTGCAGAATGTGAGGT-3′ and 5′-GCTGGAACCGTTCACCACGT-3′ for Thbs1; 5′-AGCAGGCATCTGAGGGCCCA-3′ and 5′-GAGAGCAATGCCAGCCCCGG -3′ for Gapdh; 5′-TGTTACCAACTGGGACGACA-3′ and 5′-CCATCACAATGCCTGTGGTA -3′ for β-actin; 5′-CATTGGGCCACAGACCTACT-3′ and 5′-CAAGGGAGAGAATCCAGCAG-3′ for Adora1; 5′-CTCTTCTTCGCCTGCTTTGT-3′ and 5′-AATGACAGCACCCAGCAAAT-3′ for Adora2a; 5′-CCTTTGCCATCACCATCAG-3′ and 5′-GTGTCCCAGTGACCAAACCT-3′ for Adora2b; 5′-TCAGCCTGCAAGTCAAGATG-3′ and 5′-CAGCAAAGGCCCAAGAATAG-3′ for Adora3; 5′-GCCAAGAGTGCTCGAGATTT-3′ and 5′-TTCAGGACGCCTGATTTGAT-3′ for Panx1; 5′-CAAGAAGGCCAAGACTGAGG-3′ and 5′-CAGGATGGTGAGAGGGTCAG-3′ for Panx2; 5′-CCTCAGCTCTGACCTGCTGT-3′ and 5′-GAGGAGTAAGAGGGCGTTCC-3′ for Panx3 . The amount of specific mRNA was normalized to Gapdh or β-actin mRNA for each individual sample . TNFα and MIP-2 levels were measured by ELISA kit from BioLegend and R&D Systems , respectively , according to the manufacture’s instruction . W3 cells ( 2 × 106 cells/ml ) were treated with 30 units/ml human FasL for 30 min , washed , re-suspended at a concentration of 1 × 107 cells/ml in serum-free RPMI medium supplemented with ITS-X ( insulin , transferrin and selenous acid; Invitrogen , Carlsbad , CA ) , and further incubated for 1 hr . The supernatant was subjected to LC-MS analysis in Kazusa DNA Research Institute ( Japan ) , using an Agilent 1200 liquid chromatography system ( Agilent technologies , Santa Clare , CA ) connected to a Finnigan LTQ Orbitrap XL mass spectrometer ( Thermo Fisher Scientific , Waltham , MA ) . The chromatography was performed at a constant flow rate of 0 . 5 ml/min with 3–95% linear gradient of acetonitrile in 0 . 1% formic acid , using a TOSOH TSKgel ODS-100V column ( TOSOH , Japan ) . Data were analyzed according to a molecular formula calculation in following databases: ExactMassDB ( http://webs2 . kazusa . or . jp/mfsearcher/exmassdb/ ) , KEGG ( http://www . genome . jp/kegg/ ) , KNApSAcK ( http://kanaya . naist . jp/KNApSAcK/ ) , Flavonoid Viewer ( http://www . metabolome . jp/software/FlavonoidViewer/ ) and LipidMAPS ( http://www . lipidmaps . org/ ) . For quantification of extracellular or intracellular adenine nucleotides , W3 cells ( 1 × 107 cells/ml ) or thymocytes ( 1 × 108 cells/ml ) from 6- to 8-week-old mice were treated with 120 units/ml human FasL at 37°C in RPMI containing 1% FCS . After centrifugation at 1700×g for 4 min , the supernatant of the thymocytes was extracted by chloroform and methanol to remove lipids and proteins , lyophilized and reconstituted with distilled water . The LC-MS analysis was performed using a LC-MS system ( Alliance/ZQ-4000 , Waters , Milford , MA ) . Samples were separated at 45°C with Cosmosil-PAQ column ( 2 . 0 mm × 250 mm; Nacalai ) using 1–45% acetonitrile gradient at a flow rate of 0 . 2 ml/min , and compounds were detected using negative electrospray ionization . The concentration of adenosine , AMP , ADP and ATP was determined by comparing the peak areas with those of standards . In some cases , adenine nucleotides were quantified by HPLC using a Gilson PLC2020 system equipped with a COSMOSIL-PAQ column ( Nacalai ) . The HPLC was performed using acetonitrile gradient ( 1–45% acetonitrile in 30 min ) in the presence of 1 mM dihexyl ammonium acetate ( Tokyo Chemical Industry , Japan ) at a flow rate of 1 ml/min . ATP level was also quantified by luciferase assay system . A full-length cDNA for mouse Panx1 ( GenBank accession number NM_019482 . 2 ) was prepared by RT-PCR from mouse BMDMs , and its authenticity was verified by DNA sequencing . The caspase-recognition sequence , Asp-Ile-Ile-Asp , at amino acid position 375–378 in mouse Panx1 was mutated by recombinant PCR to Ala-Ile-Ile-Ala . The cDNA was Flag-tagged at the C-terminus , and introduced into the pMXs-puro retroviral vector ( Kitamura et al . , 2003 ) . Human 293T cells were co-transfected with the pMXs-puro expression vector , together with pEF-gag-pol and pE-Ampho ( Takara Bio ) . 2 days later , the amphotropic retrovirus in the culture medium was concentrated by centrifugation at 6000×g for 16 hr , and used to infect W3 cells in the presence of 10 μg/ml polybrene . Transformants were selected in medium containing 1 . 0 μg/ml puromycin . To confirm the expression of mouse Panx1 , the cells were permeabilized with 0 . 3% saponin , and stained with FITC-conjugated anti-Flag M2 ( Sigma ) , followed by a flow cytometry with FACSCalibur ( BD Bioscience , San Jose , CA ) . Zymosan-induced peritonitis was performed essentially as described ( Cash et al . , 2009 ) . In brief , female mice ( n = 5–10 per group ) at 8–12 weeks of age were injected i . p . with zymosan A particles ( 250 mg/kg , 0 . 5 ml saline ) . Peritoneal lavage fluid was harvested with 5 ml PBS containing 1% FCS at 2 or 6 hr after the injection . Cells and residual zymosan particles were removed by centrifugation at 900×g for 4 min . Mice were grouped randomly , and studied without being blinded . To detect apoptotic cells , peritoneal cells were suspended in PBS containing 2% FCS , incubated on ice with 1 mg/ml of biotinylated anti-Ly6B . 2 ( AbD Serotec , UK ) , followed by the incubation with 2 μg/ml PE-Cy7-labeled streptavidin ( BD Pharmingen ) and 5 μg/ml FITC-labeled anti-Ly6G ( 1A8; BD Pharmingen ) . After washing with Annexin V binding buffer , cells were stained with 2000-fold diluted Cy5-labeled AnnexinV ( Biovision , Milpitas , CA ) and 500 nM Sytox blue ( Invitrogen ) and analyzed by flow cytometry . | Infections , toxins , and trauma can all injure tissue and cause the cells inside the tissue to die . When a cell dies , the membrane that surrounds it ruptures and its contents spill out , triggering inflammation of the surrounding tissues . This inflammation is part of the body’s efforts to begin the healing process but , if left uncontrolled , inflammation itself can cause further tissue damage . Diseased or damaged cells can also ‘choose’ to kill themselves to protect other healthy cells . This process , which is called apoptosis , also eliminates about 100 , 000 cells that are too old , or just no longer needed , from the human body every second . A cell undergoing apoptosis essentially dismantles itself , and the remains of the cell are packaged up , and cleared away by the white blood cells . Interestingly , this programed cell death releases many of the same molecules as other dying cells , but does so without triggering inflammation . The reason behind this lack of inflammation has not been clear . Now , Yamaguchi et al . have addressed this issue , and shown that cells undergoing apoptosis also release a chemical called adenosine monophosphate ( AMP ) that acts as a ‘calm down’ signal . The AMP is processed by white blood cells to a simpler chemical , which ‘switches on’ various genes in the white blood cells . This leads to the production of proteins that suppress the inflammation that would otherwise be triggered by other molecules released from the cells undergoing apoptosis . The findings of Yamaguchi et al . show how the community of cells in our body is kept in a healthy balance , and in the future , could improve our understanding and the treatment of inflammatory diseases . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Material",
"and",
"methods"
] | [
"cell",
"biology",
"immunology",
"and",
"inflammation"
] | 2014 | Immunosuppression via adenosine receptor activation by adenosine monophosphate released from apoptotic cells |
Auditory sensory outer hair cells are thought to amplify sound-induced basilar membrane vibration through a feedback mechanism to enhance hearing sensitivity . For optimal amplification , the outer hair cell-generated force must act on the basilar membrane at an appropriate time at every cycle . However , the temporal relationship between the outer hair cell-driven reticular lamina vibration and the basilar membrane vibration remains unclear . By measuring sub-nanometer vibrations directly from outer hair cells using a custom-built heterodyne low-coherence interferometer , we demonstrate in living gerbil cochleae that the reticular lamina vibration occurs after , not before , the basilar membrane vibration . Both tone- and click-induced responses indicate that the reticular lamina and basilar membrane vibrate in opposite directions at the cochlear base and they oscillate in phase near the best-frequency location . Our results suggest that outer hair cells enhance hearing sensitivity through a global hydromechanical mechanism , rather than through a local mechanical feedback as commonly supposed .
The exceptional sensitivity of mammalian hearing has been attributed to a micromechanical feedback system inside the cochlea , also called ‘the cochlear amplifier’ or ‘cochlear active process’ ( Dallos et al . , 2008; Davis , 1983; Fettiplace and Hackney , 2006; Hudspeth , 2014; Robles and Ruggero , 2001; Russell et al . , 2007 ) . When sound-induced basilar membrane vibrations deflect hair bundles of the outer hair cells , mechanoelectrical transduction of these cells generates the receptor potential ( Dallos et al . , 1982; Russell and Sellick , 1983 ) . In response to the membrane potential change , mammalian outer hair cells change their length and generate force primarily through the somatic motility driven by the motor protein , prestin ( Ashmore , 2008; Brownell et al . , 1985; Liberman et al . , 2002; Mammano and Ashmore , 1993; Mellado Lagarde et al . , 2008; Ren et al . , 2016a; Santos-Sacchi , 1989; Zheng et al . , 2000 ) . This cellular force is thought to be directly applied to the basilar membrane at its generation location on a cycle-by-cycle basis , consequently amplifying the sound-induced basilar membrane vibration and boosting hearing sensitivity ( Dallos et al . , 2008; de Boer , 1995b; Dong and Olson , 2013; Hudspeth , 2014; Liu and Neely , 2009; Reichenbach and Hudspeth , 2014 ) . For optimal amplification , the cellular force must act on the basilar membrane at an appropriate time at every vibration cycle ( Dallos et al . , 2008; Nilsen and Russell , 1999 ) . Therefore , timing of the cochlear feedback has been a central research topic in the field of auditory neuroscience since the cochlear amplifier was proposed ( Ashmore , 2008; Davis , 1983; Gold , 1948; Gummer et al . , 1996 ) . Because cochlear amplification depends on normal cochlear metabolism and the integrity of cochlear mechanical properties ( Cooper and Rhode , 1992; Fisher et al . , 2012; Lee et al . , 2015; Lee et al . , 2016; Nuttall et al . , 1991; Ren and Nuttall , 2001; Rhode , 1971; Robles and Ruggero , 2001; Ruggero and Rich , 1991; van der Heijden and Versteegh , 2015 ) , the cochlear feedback has to be investigated ultimately in living cochleae . Timing of cochlear feedback was studied in vivo by measuring basilar membrane vibrations at different locations across the width of the basilar membrane in guinea pig ( Nilsen and Russell , 1999 ) . Based on their observation , the authors suggest that forces generated by the outer hair cells directly drive the region of the basilar membrane beneath the Deiters' cells . The temporal relationship between the reticular lamina and basilar membrane vibration was previously observed in guinea pigs using a time-domain optical coherence tomography system as a homodyne interferometer ( Chen et al . , 2011 ) . It was reported that the phase of the reticular lamina vibration led the phase of the basilar membrane vibration at the best frequency ( Figure 5 , Chen et al . , 2011 ) . This phase lead has been thought to ensure the right timing of the outer hair cell force for cochlear amplification . However , the mouse micromechanical data measured using a heterodyne low-coherence interferometer showed no significant phase difference between the reticular lamina and basilar membrane vibration at the best frequency ( Ren et al . , 2016b ) . This discrepancy may have been caused by the animal species and technical differences , that is mouse versus guinea pig and homodyne interferometry versus heterodyne interferometry . Although a number of studies have been conducted recently to measure micromechanical responses in living cochleae ( Gao et al . , 2014; Lee et al . , 2016; Ramamoorthy et al . , 2016; Recio-Spinoso and Oghalai , 2017; Cooper et al . , 2018 ) the timing of the cochlear feedback remains unclear . Since the apical end of outer hair cells is directly connected to the reticular lamina , the reticular lamina vibration can reflect the movement of the outer hair cell under physiological conditions . Therefore , the timing of the cochlear feedback was determined in this study by measuring the latency difference between the reticular lamina and basilar membrane vibration using a custom-built heterodyne low-coherence interferometer ( Hong and Freeman , 2006; Ren et al . , 2016a; Ren et al . , 2016b ) . The present data collected from the gerbil , one of the most commonly used animals for auditory research , demonstrate for the first time that the reticular lamina vibrates after , not before , the basilar membrane vibration .
A representative data set from one of twenty-three sensitive cochleae is presented in Figure 1 . The displacement of the reticular lamina response to 30 dB SPL tones ( 0 dB SPL = 20 µPa ) increased and then decreased with frequency , forming a sharp peak at ~26 kHz ( best frequency , BF ) ( Figure 1A ) . While displacements increased proportionally with sound level at frequencies < 15 kHz , they increased at a much smaller rate near the best frequency . The response peak became broader and shifted toward low frequencies as the sound level increased . In contrast to sharp tuning at 30 dB SPL , the displacement curve at 80 dB SPL showed no response peak . Displacements of the basilar membrane at 30 and 40 dB SPL were ~10 fold smaller than those of the reticular lamina not only near the best frequency but also at lower frequencies ( Figure 1B ) . Basilar membrane response also reached the maximum at ~26 kHz as did the reticular lamina . For ~333 fold sound level increase from 30 to 80 dB SPL , the displacement of the basilar membrane at ~26 kHz increased by ~23 fold compared with only ~3 . 3 fold increase of the reticular lamina displacement at the same frequency . These differences were confirmed by the displacement ratios of the reticular lamina ( Figure 1C ) and basilar membrane ( Figure 1D ) vibration to the stapes vibration . The reticular lamina showed ~40 dB nonlinear compression near the best frequency , which is ~17 dB greater than that of the basilar membrane ( ~23 dB ) . Phase responses of the reticular lamina ( Figure 1E ) are similar to those of the basilar membrane ( Figure 1F ) except for a slightly steeper phase slope at frequencies < 15 kHz . Thus , highly sensitive , sharply tuned nonlinear responses of the reticular lamina and basilar membrane in the gerbil ( Figure 1A–D ) are similar to those measured from the basal turn of mouse cochleae under sensitive conditions ( Ren et al . , 2016b ) . The magnitude relationship between the reticular lamina and basilar membrane vibration is presented in Figure 1G by the ratio of the reticular lamina displacement to the basilar membrane displacement as a function of frequency . All displacement ratios are greater than one , indicating that reticular lamina vibrations were greater than those of the basilar membrane at different frequencies and sound levels . Near the best frequency ( ~26 kHz ) , the displacement ratio was ~10 at 30 dB SPL , decreasing with sound level and becoming ~1 . 5 at 80 dB SPL . In contrast to well documented sharply tuned basilar membrane vibration at the best frequency ( Robles and Ruggero , 2001 ) , the greatest displacement ratios were observed at frequencies far below the best frequency and at high sound levels . The temporal relationship between the reticular lamina and basilar membrane vibration was determined by phase difference as a function of frequency and is presented in Figure 1H . At frequencies < 10 kHz , reticular lamina phase led basilar membrane phase by up to 180 degrees . This phase lead decreased with frequency and the phase difference became slightly negative at the best frequency . The slope of the linear regression line ( thick black dotted ) of the phase difference curve at 70 dB SPL indicates that the latency of the reticular lamina is ~20 µs greater than that of the basilar membrane . Compared to sensitive responses ( red lines in Figure 2A ) , displacements of the reticular lamina vibration decreased dramatically at all frequencies under postmortem conditions ( blue lines in Figure 2A ) , while the basilar membrane vibration decreased only near the best frequency ( blues lines in Figure 2B ) . In contrast to the sharp peak of 30 dB SPL sensitive responses at ~26 kHz , the broad peaks of reticular lamina and basilar membrane postmortem responses shifted toward low frequencies ( blue lines in Figure 2A , B ) . Although the basilar membrane vibration at frequencies < 20 kHz did not decrease significantly , the reticular lamina vibration decreased by >10 fold over the same frequency range . This is unexpected since cochlear amplification has been believed to be effective only near the best frequency ( Robles and Ruggero , 2001 ) . The overlapping blue curves in Figure 2C , D and the equally separated blue curves in Figure 2A , B indicate linear growth for postmortem responses . In contrast to the lack of significant change in basilar membrane phase ( Figure 2F ) , the reticular lamina phase decreased by up to 180 degrees at frequencies < 10 kHz ( Figure 2E ) . The displacement ratio of the reticular lamina to basilar membrane vibration decreased dramatically at all frequencies ( compare blue to red lines in Figure 2G ) . The frequency-dependent phase lead ( red lines in Figure 2H ) was absent under postmortem conditions ( blue lines in Figure 2H ) . Thus , postmortem data in Figure 2 demonstrate that the magnitude and phase differences between the reticular lamina and basilar membrane vibration depend on normal cochlear metabolism . While the reticular lamina and basilar membrane vibration in the gerbil ( Figure 1 ) are qualitatively similar to those in the mouse ( Figure 1C–J , Ren et al . , 2016b ) , there are the following quantitatively differences . Although the measurements were taken from similar longitudinal locations in the basal turn of the cochlea in both species , the best frequencies of the reticular lamina and basilar membrane vibration in the gerbil are significantly lower than those in the mouse . Compared to the mouse postmortem data ( Figure 2G and H , Ren et al . , 2016b ) , the basilar membrane vibrated more than the reticular lamina in postmortem gerbil cochleae ( Figure 2 ) . To determine the latency difference , the phase and magnitude differences between the reticular lamina and basilar membrane vibration were measured as a function of frequency at different sound levels in seven sensitive gerbil cochleae ( Figure 3 ) . For statistical analysis , frequency axes were normalized to the best frequency for each animal . The displacement ratio of the reticular lamina vibration to the basilar membrane vibration was the largest at the low-frequency end and decreased with frequency reaching the lowest level near the best frequency ( Figure 3A , C , E , G ) . The averaged phase data show that the reticular lamina phase led the basilar membrane phase by >135 degrees at low frequencies . This phase lead decreased with frequency and approached zero near the best frequency ( Figure 3B , D , F , H ) . While the magnitude ratio decreased with the sound level near the best frequency ( Figure 3I ) , the phase difference showed no significant change at the same frequency ( Figure 3J ) . Consequently , the latency differences derived from the slope of the phase difference curves show no significant change across stimulus levels ( ANOVA test: F = 0 . 037 , p=0 . 990 ) . This allows us to calculate the latency difference between the reticular lamina and basilar membrane vibration across animals and sound levels . The grouped data from seven gerbils at four different sound levels demonstrate that the latency of the reticular lamina vibration is ~17 . 9 µs greater than that of the basilar membrane ( 17 . 9 ± 0 . 6 µs , n = 28 ) ( Figure 3J ) . To compare this result with that in mice , the latency difference in sensitive mouse cochleae was derived from recently published phase data ( Figure 3F , Ren et al . , 2016b ) using the same experimental procedures as those in the current study . The latency difference between the reticular lamina and basilar membrane vibration in mice ( 12 . 1 ± 0 . 6 µs , n = 5 ) is significantly smaller than that in gerbils ( 17 . 9 ± 0 . 6 µs , n = 28 ) ( t = 7 . 775 , p<0 . 01 , n = 33 ) . The smaller latency difference in mice likely results from higher best frequencies . To confirm latency difference revealed by phase data in Figure 3 , the reticular lamina and basilar membrane response to clicks were measured by recording instantaneous displacement as a function of time at different sound levels . In the sensitive living cochlea , a 10-µs rarefaction click caused a large displacement of the reticular lamina toward the scala tympani indicated by the first negative peak ( red curves in Figure 4A ) at ~0 . 40 ms , while the basilar membrane moved toward the scala vestibuli ( indicated by the first positive peak of blue curves in Figure 4A ) . In addition to the opposite directions , the peak magnitude of the reticular lamina displacement is ~eight fold larger than that of the basilar membrane vibration , which increased proportionally with the sound level . Following the initial opposite movements , both the reticular lamina and basilar membrane oscillated periodically and the vibration magnitudes decreased gradually approaching to their equilibrium positions . To clearly show the temporal relationship , the displacements of the reticular lamina and basilar membrane vibration at 90 dB-p ( 0 dB-p = 20 µPa of peak sound pressure ) were plotted as a function of time with different magnitude scales in Figure 4C . Since the initial peaks indicate the maximum displacements of the reticular lamina and basilar membrane vibration , which effectively stimulate the cochlea , the initial peak times TA and TB ( Figure 4C ) are used to present the latencies of the basilar membrane and reticular lamina response respectively , and the latency difference was determined by TB-TA . Latency difference measured at 90 dB-p in ten sensitive cochleae ( 32 . 6 ± 1 . 5 µs , n = 10 ) demonstrates that the outer hair cell-driven reticular lamina vibration occurs after the basilar membrane vibration . This latency difference is greater than that derived from the phase data in Figure 3 , likely due to the stimulus difference , that is tone verse click . Despite their initial opposite movements , the reticular lamina and basilar membrane moved synchronously in the same direction at ~0 . 51 ms ( TC ) ( Figure 4C ) , because the phase change of the reticular lamina vibration with time is slower than that of the basilar membrane . Similarly , the first period of the reticular lamina ( TRL ) is greater than that of the basilar membrane ( TBM ) ( the low panel in Figure 4A ) indicating that the starting frequency of the reticular lamina is lower than that of the basilar membrane vibration . The magnitude of the reticular lamina vibration decreased dramatically under postmortem conditions and became comparable to that of the basilar membrane ( Figure 4B , D ) . Moreover , the direction of the first peak of the reticular lamina displacement changed from negative to positive , that is from toward the scala tympani to toward the scala vestibuli , which became consistent with the direction of the basilar membrane vibration . These changes indicate that the reticular lamina moves passively following the basilar membrane vibration under postmortem conditions . The data at sound levels below 70 dB SPL were not shown because the initial peak of the basilar membrane vibration was too small to be reliably detected . Thus , the time-domain data in Figure 4 confirm that the reticular lamina vibrates after , not before , the basilar membrane vibration . Moreover , the initially opposite displacements and following synchronous movements of the reticular lamina and the basilar membrane in Figure 4A and C are consistent with the ~180° phase difference at low frequencies and in-phase vibrations at the best frequency in Figure 3 .
This paper reports the first in vivo measurement of the latency difference between the outer hair cell-driven reticular lamina vibration and the basilar membrane vibration . The present data demonstrate that the latency of the reticular lamina vibration is greater than that of the basilar membrane vibration , and there is no significant phase difference between the two structures near the best frequencies . This result is consistent with the mouse data measured using heterodyne interferometry ( Ren et al . , 2016b ) but inconsistent with the guinea pig data , which showed that the phase of the reticular lamina vibration leads the phase of the basilar membrane vibration by ~90° at the best frequency and the phase lead decreases with sound pressure level ( Chen et al . , 2011 ) . The current result also confirms recent studies in the mouse ( Ren et al . , 2016b ) and in the guinea pig ( Recio-Spinoso and Oghalai , 2017 ) that the physiologically vulnerable reticular lamina vibration is significantly greater than the basilar membrane vibration not only at the best frequency but also at low frequencies . Since an ~90° phase lead of the reticular lamina vibration is thought to be required for cochlear feedback to amplify basilar membrane vibration ( Chen et al . , 2011; Gummer et al . , 1996; Nilsen and Russell , 1999; Robles and Ruggero , 2001; Russell and Nilsen , 1997 ) , and since cochlear amplification has been predicted to work only near the best frequency location ( de Boer , 1995a; Dong and Olson , 2013; Liu et al . , 2017; Liu and Neely , 2009; Meaud and Grosh , 2012; Motallebzadeh et al . , 2018; Ni et al . , 2016132016; Ramamoorthy et al . , 2007; Wang et al . , 2016 ) , the current result is inconsistent with the local cochlear feedback hypothesis . Instead , the latency difference between the reticular lamina and basilar membrane vibration found in this study supports the global hydromechanical mechanism for cochlear amplification ( Ren et al . , 2016b ) , which is discussed below . The longitudinal patterns of the reticular lamina and basilar membrane vibrations at the best-frequency ( 26 kHz ) are presented by plotting displacements and phases ( Figure 1A , B , E , F ) as a function of the location along the cochlear length ( blue and red lines in Figure 5A–D ) , which was derived from the stimulus frequency according to the cochlear frequency-location function ( Müller , 1996 ) . While a 30 dB SPL tone-induced response occurred at a < 0 . 2 mm region at the best-frequency location ( Figure 5A , C ) , a 70 dB SPL tone-induced vibration extended from the best-frequency location to the base ( Figure 5B , D ) . The overlapping phase curves near the 2 . 05 mm location ( blue and red lines in Figure 5C , D ) indicate that the reticular lamina and basilar membrane vibrated approximately in the same direction at the best-frequency location . The ~180 degree separation between the blue and red lines near the cochlear base ( Figure 5D ) indicates opposite vibrations of the reticular lamina and basilar membrane . The outer hair cell-driven active movement was estimated by vector subtraction of the basilar membrane vibration from the measured reticular lamina vibration and is presented by green lines in Figure 5A–D . The overlapping blue and green lines in Figure 5A and C indicate that , at low sound levels , reticular lamina vibration is dominated by outer hair cell-driven movement . At 70 dB SPL , the outer hair cell-driven responses saturated near the best-frequency location , indicated by the diverged blue and green lines near the response peak ( Figure 5B ) . Time waveforms of the outer hair cell-driven reticular lamina vibrations ( green ) and the basilar membrane vibrations ( red ) in Figure 5E , F were derived from magnitude and phase in Figure 5A–D at two sequential times . The time difference between solid and dotted green curves is ~6 µs , equivalent to ~57 degree phase difference at 26 kHz . For clearer comparison , basilar membrane time waveforms ( red curves ) were shifted down by 4 nm in Figure 5E and by 50 nm in Figure 5F respectively . When the basilar membrane moves upward to the scala vestibuli near the cochlear base ( red arrow at 0 . 8 mm in Figure 5F ) , depolarized outer hair cells shorten and induce a large downward reticular lamina displacement ( green arrow at 0 . 8 mm in Figure 5F ) . While this movement creates a positive fluid pressure between the reticular lamina and basilar membrane at the cochlear base ( 0 . 7–1 . 2 mm in Figure 5F , G ) , the reticular lamina at locations ~ 1 . 2–1 . 8 mm , however , moves upward ( green arrows near 1 . 6 mm in Figure 5F ) , resulting in a negative fluid pressure inside the cochlear partition . This pressure gradient from the base to an apical location likely results in fluid movement in the apical direction inside the cochlear partition ( blue arrows in Figure 5F , G ) as demonstrated in vitro by electrically stimulating the organ of Corti ( Karavitaki and Mountain , 2007; Zagadou and Mountain , 2012 ) . Since the organ of Corti sits on the basilar membrane , the longitudinal fluid movement can travel forward to the best-frequency location as a result of the basilar membrane traveling wave . Thus , a large population of outer hair cells from a broad cochlear area can change the fluid space between the reticular lamina and basilar membrane at the best frequency location on a cycle-by-cycle basis , consequently enhancing the reticular lamina vibration . In addition , in-phase vibrations of the reticular lamina and basilar membrane result in constructive interference near the best-frequency location ( Figure 5E , F , G ) , which further enhances reticular lamina vibration at the apical end of outer hair cells . The magnitude of the resulting constructive interference decreases as phase differences move into destructive interference regimes at frequencies below and above the best frequency . While this frequency dependent interference may consequently enhance the tuning of the reticular lamina vibration , its effects probably is relatively small due to the large magnitude difference between the reticular lamina and basilar membrane vibration . The interaction between the reticular lamina and basilar membrane at low frequencies may also be involved in two-tone suppression of the auditory nerve or basilar membrane response ( Delgutte , 1990; Ruggero et al . , 1992 ) . It has been shown that the proposed global hydromechanical mechansm is consistent with the observation that auditory nerve activities can be suppressed by stimulating medial olivocochlear efferents or by a low-frequency bias tone not only at the best frequency but also at tail ( low ) frequencies ( Nam and Guinan , 2017; Stankovic and Guinan , 1999 ) . Since the stereocilia bundles of both the inner and outer hair cells are anchored in the hair cell cuticular plates which make up a portion of the reticular lamina , the outer hair cell-driven reticular lamina vibration likely results in fluid movement in the subtectorial space and consequently stimulates inner hair cells . It has been shown in vitro that electrical stimulation of outer hair cells of guinea pig cochleae resulted in a counterphasic motion of the tectorial membrane and inner hair cells at frequencies below 3 kHz ( Nowotny and Gummer , 2006 ) . This result was believed to indicate direct fluid coupling between outer hair cells and inner hair cells through a pulsatile fluid motion . It has also been demonstrated in vitro that outer hair cell stereocilia not only move sideways but also change length in response to sound stimulation ( Hakizimana et al . , 2012 ) . The large bundle deflection was observed when the length change was small , indicating that hair cells are maximally stimulated when the stereocilia length change is minimal . Considering the firm connection of the tallest stereocilia to the tectorial membrane , the stereocilia length change also suggests the interaction between the reticular lamina vibration and the tectorial membrane vibration . Thus , outer hair cells may also play a role in the traveling wave on the tectorial membrane , which has been demonstrated in vitro ( Ghaffari et al . , 2007; Ghaffari et al . , 2010 ) and in vivo ( Dong and Cooper , 2006; Lee et al . , 2015; Lee et al . , 2016; Recio-Spinoso and Oghalai , 2017; Rhode and Cooper , 1996 ) . The interaction between the reticular lamina and tectorial membrane vibration may enhance the stimulus to the inner hair cells and boost hearing sensitivity . Specific mechanisms on how the outer hair cell-driven reticular lamina vibration stimulates inner hair cells , however , remain to be determined experimentally until in vivo micromechanical measurements with cellular spatial resolution become available . The observed additional delay of the reticular lamina vibration was likely caused by mechanoelectrical ( Corey and Hudspeth , 1979 ) and electromechanical ( Brownell et al . , 1985 ) transduction of outer hair cells and mechanical coupling inside the cochlear partition . Although delays for prestin-associated currents in vitro ( Santos-Sacchi and Tan , 2018 ) are longer than the latency difference between the reticular lamina and basilar membrane vibration in vivo , they vary with the membrane voltage of outer hair cells . Voltage excitation away from the resting membrane potential has been shown to have a faster response . Moreover , the depolarized resting potential can minimize the outer hair cell time constant and expanding the bandwidth of the membrane filter by activating voltage-dependent K + conductance ( Johnson et al . , 2011 ) . Mechanical loads on the outer hair cells , such as in vivo condition , could further improve the cell's frequency response ( Iwasa , 2017 ) . Since Deiters' cells are located between the outer hair cells and the basilar membrane , the acoustical and cellular forces to and from the outer hair cells have to be transmitted through Deiters' cells . The soft Deiter's cell soma likely induces delays , which can contribute to the delay difference between the reticular lamina and basilar membrane vibration . This apparently is supported by the postmortem magnitude and phase difference between the reticular lamina and basilar membrane vibration ( Figure 2 ) . When the outer hair cell-generated force is absent under postmortem conditions , the reticular lamina should move passively following the basilar membrane travelling wave with equal magnitude and phase , if the connection between the two structures is rigid . The larger postmortem magnitude and phase difference in gerbils ( Figure 2 ) than those in mice ( Figure 2G and H , Ren et al . , 2016b ) indicate that the mechanical coupling between the reticular lamina and basilar membrane in the gerbil may not be as tight as that in the mouse . Compared to a large time period ( T ) at a low frequency ( f ) ( T = 1/f ) ( such as T = 2 , 000 µs , where f = 500 Hz ) , a 17 . 9-µs latency difference ( Figure 3J ) is negligible and results in an insignificant phase difference . Thus , the ~180 degree phase difference between the reticular lamina and basilar membrane vibration at a low frequency mainly reflects the opposite movements of both ends of outer hair cells ( Brownell et al . , 1985 ) . The same latency difference , however , can result in ~180 degree phase difference at the best frequency , due to the small period ( such as T = 38 µs , where f = 26 kHz ) . This latency-induced phase lag of the reticular lamina vibration , compensating for the ~180 degree phase difference observed at low frequencies , accounts for in-phase vibrations of the reticular lamina and basilar membrane at the best frequency . Therefore , the present in vivo results do not conflict with the in vitro observation that both ends of the cylindrical outer hair cells move in opposite directions at low frequencies ( Brownell et al . , 1985; Santos-Sacchi , 1989 ) . In summary , heterodyne low-coherence interferometry demonstrates in vivo that the outer hair cell-driven reticular lamina vibration occurs after , not before , the basilar membrane vibration . The reticular lamina and basilar membrane move in opposite directions at low frequencies and in phase near the best frequency . This experimental finding conflicts with commonly accepted cochlear local feedback theory and suggests that outer hair cells enhance hearing sensitivity through a global hydromechanical mechanism .
A scanning low-coherence heterodyne interferometer was built based on a scanning laser heterodyne interferometer ( Ren , 2002; Ren , 2004; Ren et al . , 2011 ) by replacing the helium-neon laser with a modified superluminescent diode with related optical and electronic components ( Ren et al . , 2016a; Ren et al . , 2016b ) . Because of the small coherence length , the low-coherence interferometer can measure vibrations with a high axial resolution ( Chen et al . , 2011; Hong and Freeman , 2006; Lee et al . , 2015 ) . The use of low-coherence light and an objective lens with numerical aperture 0 . 42 provides adequate spatial selectivity for measuring the reticular lamina and basilar membrane vibrations in the living cochlea . In addition to its unprecedented sensitivity , this interferometer has a broad dynamic range , high temporal resolution , and low phase noise , due to the use of a 40-MHz carrier for heterodyne detection . In contrast to homodyne interferometers , a heterodyne interferometer can detect vibration directions without 180-degree phase uncertainty ( Hong and Freeman , 2006; Khanna et al . , 1986; Lukashkin et al . , 2005 ) , which ensured the reliability of the phase measurements in this study . Animal anesthesia and surgical procedures were the same as described previously ( Ren , 2002; Ren et al . , 2011; Ren and Nuttall , 2001 ) . Briefly , after about one third of the round window membrane was removed and the opened round window was partially covered with a glass coverslip , the object light from the interferometer was focused on the center of the outer hair cell region of the cochlear partition at the basal turn ( Figure 6A ) . The transverse locations of the basilar membrane and reticular lamina were determined by measuring the backscattered light ( carrier ) level as a function of the transverse location . The locations of the basilar membrane and reticular lamina were indicated by the two peaks of the backscattered light level ( Figure 6B ) and confirmed by the distinct magnitude and phase of the cochlear partition vibrations at the two locations ( Figure 6C , D ) . Cochlear partition vibrations were measured as a function of the transverse location at the best frequency ( 30 kHz ) and at different sound levels ( 0–80 dB SPL ) . When the object beam of the interferometer was focused on the basilar membrane or the reticular lamina , acoustical tones at different frequencies and levels were delivered to the ear canal . The tone frequency was changed from 1 . 8 to 40 . 0 kHz by ~0 . 2 kHz per step . The magnitude and phase of the cochlear partition vibration were measured using a lock-in amplifier ( SR830 , Stanford Research System , Inc . Sunnyvale , CA ) and recorded on a computer . The best frequency was determined by the peak of the basilar membrane displacement as a function of frequency at 30 dB SPL . For recording time waveforms of the reticular lamina and basilar membrane vibration , a 10-µs electrical pulse was generated by a dynamic signal analyzer ( PXI-4461 , National Instruments , Austin , TX ) and used to drive an electrostatic speaker ( EC1 , Tucker-Davis Technologies , Alachua , FL ) . Displacements of the reticular lamina and basilar membrane vibrations were digitized at the rate of 200 , 000 samples per second and averaged synchronously with stimuli for 100 times . Time waveforms from the same time window were plotted to show the temporal relationship between the reticular lamina and basilar membrane vibration ( Figure 4 ) . The reticular lamina and basilar membrane vibration were measured in a random order . Cochlear sensitivity was monitored by continuously recording distortion product otoacoustic emission ( DPOAE ) . The DPOAE at 16 kHz was evoked by two 60 dB SPL tones at 20 and 24 kHz . A cochlea with <5 dB DPOAE decrease was considered sensitive . Postmortem data were collected 10 to 30 min after the animal's death from anesthetic overdose . Stapes vibration was recorded under the same conditions as for the cochlear mechanical measurement . Igor Pro ( Version 7 . 0 . 5 . 2 , WaveMetrics , Lake Oswego , OR ) was used for analyzing data . The frequency responses of the reticular lamina and basilar membrane were presented by displacement and phase as a function of frequency . The transfer functions were estimated by the displacement ratio of the reticular lamina or the basilar membrane to the stapes at different frequencies . Since the phase lag of the cochlear partition vibration ( ϕ ) is a function of latency ( τ ) ( ϕ = 2πfτ , where f is frequency ) , the time relationship between the basilar membrane and reticular lamina vibration were presented by the phase difference ( Δϕ ) ( Δϕ=ϕRL-ϕBM ) , where ϕRL is the reticular lamina phase and ϕBM is the basilar membrane phase at different frequencies . The latency difference ( Δτ ) was derived from the slope of the linear regression line of the phase difference as a function of frequency ( Δϕ/Δf ) ( Δτ=Δϕ/2πΔf ) . The latency difference ( Δτ ) was also determined based on the time waveform of the reticular lamina and basilar membrane response to clicks . Despite the sharp onset of the electrical pulse , the onsets of the reticular lamina and basilar membrane response were distorted due to frequency bandwidth limits of the speaker , the middle ear , and the cochlea , and cannot been measured precisely . Because the first displacement peak of the time waveform indicates the arriving time of the maximum stimulation to the reticular lamina and the basilar membrane , Δτ was determined by the time difference between the first peak of the reticular lamina ( TB ) and that of the basilar membrane ( TA ) , ( Δτ = TB -TA ) ( Figure 4C ) . The longitudinal patterns of the reticular lamina and basilar membrane vibrations were presented by plotting displacement and phase as a function of the longitudinal location . The longitudinal location was derived from the stimulus frequency according to the frequency-location function in gerbil cochleae ( Müller , 1996 ) . The grouped results were presented by mean and standard error calculated across animals . Sound level-dependent latency changes were tested at 50 , 60 , 70 , and 80 dB SPL using one-way ANOVA and p<0 . 05 was considered significantly different . | What is the quietest sound the ear can detect ? All sounds begin as vibrating air molecules , which enter the ear and cause the eardrum to vibrate . We can detect vibrations that move the eardrum by a distance of less than one picometer . That’s one thousandth of a nanometer , or about 100 times smaller than a hydrogen atom . But how does the ear achieve this level of sensitivity ? Vibrations of the eardrum cause three small bones within the middle ear to vibrate . The vibrations then spread to the cochlea , a fluid-filled spiral structure in the inner ear . Tiny hair cells lining the cochlea move as a result of the vibrations . There are two types of hair cells: inner and outer . Outer hair cells amplify the vibrations . It is this amplification that enables us to detect such small movements of the eardrum . Inner hair cells then convert the amplified vibrations into electrical signals , which travel via the auditory nerve to the brain . The bases of outer hair cells are connected to a structure called the basilar membrane , while their tops are anchored to a structure called the reticular lamina . It was generally assumed that outer hair cells amplify vibrations of the basilar membrane via a local positive feedback mechanism that requires the hair cells to vibrate first . But by comparing the timing of reticular lamina and basilar membrane vibrations in gerbils , He et al . show that this is not the case . Outer hair cells vibrate after the basilar membrane , not before . This indicates that outer hair cells use a mechanism other than commonly assumed local feedback to amplify sounds . The results presented by He et al . change our understanding of how the cochlea works , and may help bioengineers to design better hearing aids and cochlea implants . Millions of patients worldwide who suffer from hearing loss may ultimately stand to benefit . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"neuroscience"
] | 2018 | Timing of the reticular lamina and basilar membrane vibration in living gerbil cochleae |
Transcription factor ( TF ) networks determine cell-type identity by establishing and maintaining lineage-specific expression profiles , yet reconstruction of mammalian regulatory network models has been hampered by a lack of comprehensive functional validation of regulatory interactions . Here , we report comprehensive ChIP-Seq , transgenic and reporter gene experimental data that have allowed us to construct an experimentally validated regulatory network model for haematopoietic stem/progenitor cells ( HSPCs ) . Model simulation coupled with subsequent experimental validation using single cell expression profiling revealed potential mechanisms for cell state stabilisation , and also how a leukaemogenic TF fusion protein perturbs key HSPC regulators . The approach presented here should help to improve our understanding of both normal physiological and disease processes .
Tight regulation of gene expression is essential for both the establishment and maintenance of cellular phenotypes within metazoan organisms . The binding of transcription factor proteins ( TFs ) to specific DNA sequence motifs represents the primary step of decoding genetic information into specific gene expression patterns . TF binding sites ( TFBSs ) or motifs are usually short ( 6–10 bp ) , and therefore found just by chance throughout the genome . Functional TFBSs often occur as evolutionarily conserved clusters , which in the case of enhancers can act over long distances , thus necessitating comprehensive analysis of entire gene loci to understand the transcriptional control mechanisms acting at mammalian gene loci . Given the complex regulatory circuitries that arise when the control of multiple genes is considered , transcriptional control is often represented in the form of gene regulatory networks ( GRNs ) , which carry most mechanistic information when constructed from detailed knowledge on the TFs and the cis-regulatory elements with which they interact ( Davidson , 2009; Davidson , 2010; Petricka and Benfey , 2011; Pimanda and Gttgens , 2010; Gottgens , 2015; Schütte et al . , 2012 ) . Importantly , regulatory network models can provide much more than a representation of existing knowledge , because network simulations can reveal possible molecular mechanisms that underlie highly complex biological processes . Boolean modelling approaches have been used to reconstruct core regulatory networks in blood stem cells ( Bonzanni et al . , 2013 ) and myeloid progenitors ( Krumsiek et al . , 2011 ) , but neither of these studies took into account the underlying regulatory structure of the relevant gene regulatory elements . Full gene-regulatory information has been used for an ordinary differential equation-based model ( Narula et al . , 2010; Narula et al . , 2013 ) , but was restricted to a small three-gene core circuit . Large consortia efforts such as ImmGen and FANTOM5 have created comprehensive networks of either regulatory elements or gene signatures important for multipotency and differentiation ( Gazit et al . , 2013; Jojic et al . , 2013 ) . Furthermore , studies looking at gene regulation circuitry in embryonic stem ( ES ) cells have proposed regulatory networks important for ES cell identity ( Dunn et al . , 2014; Zhou et al . , 2007 ) . While the complexities of transcriptional control demand approaches such as network modelling , no single experimental method can provide the complex biological data required for the construction of accurate models . The previously mentioned studies focus their attention on one specific aspect of network modelling and importantly did not combine network analysis with comprehensive functional validation . Given that the key building blocks are gene regulatory sequences and the TFs bound to them , essential information for network reconstruction includes ( i ) comprehensive TF binding data , ( ii ) in vivo validation of the functionality of TF-bound regions as bonafide regulatory elements , and ( iii ) molecular data on the functional consequences of specific TF-binding events ( e . g . activation vs . repression ) . The regulatory network model that we present in this study comprises all of the aforementioned components and is accompanied by functional validation of model predictions .
For the reconstruction of a core GRN model for HSPCs , we focussed on nine major HSPC regulators ( ERG , FLI1 , GATA2 , GFI1B , LYL1 , MEIS1 , PU . 1 , RUNX1 , TAL1 ) , for which genome-wide binding patterns in the murine multipotent progenitor cell line HPC7 have previously been published ( Wilson et al . , 2010 ) . First , we searched the literature to summarise known cis-regulatory regions for the nine TFs that possess haematopoietic activity in transgenic mouse embryos , which recovered a total of 14 regions: Erg+85 ( Wilson et al . , 2009 ) , Fli1-15 ( Beck et al . , 2013 ) , Fli1+12 , Gata2-3 ( Pimanda et al . , 2007 ) , Gata2+3 ( = Gata2+9 . 5 ) ( Wozniak et al . , 2007 ) , Gfi1b+13 , Gfi1b+16 , Gfi1b+17 ( Wilson et al . , 2009; Moignard et al . , 2013 ) , Lyl1 promoter ( Chan et al . , 2007 ) , Spi1-14 ( Wilkinson et al . , 2014 ) , Runx1+23 ( Nottingham et al . , 2007 ) , Tal1-4 ( Gottgens et al . , 2004 ) , Tal1+19 ( Göttgens et al . , 2002 ) and Tal1+40 ( Gottgens et al . , 2010 ) . To extend this partial knowledge of relevant gene regulatory sequences to a comprehensive definition of how these nine TFs might cross-regulate each other , we made use of the genome-wide binding data for the nine TFs ( Wilson et al . , 2010 ) as well as information on acetylation of histone H3 at lysine 27 ( H3K27ac ) ( Calero-Nieto et al . , 2014 ) in the HPC7 blood progenitor cell line . Additional candidate gene regulatory regions for all nine TFs were selected based on the binding of at least three TFs and H3K27ac , since it has been shown previously that transcriptionally active regions are commonly bound by multiple TFs and display H3K27 acetylation ( Hardison and Taylor , 2012 ) . To assign putative candidate regions to a given TF , they had to be located between its respective upstream and downstream flanking genes , i . e . within the gene body itself or its 5’ and 3’ intergenic flanking regions . The Erg gene locus for example contains five candidate cis-regulatory regions based on these criteria , namely Erg+65 , Erg+75 , Erg+85 , Erg+90 and Erg+149 ( Figure 1a ) , of which only the Erg+85 region had previously been tested in transgenic mice ( Wilson et al . , 2009 ) . Inspection of the gene loci of all nine TFs resulted in the identification of 35 candidate cis-regulatory regions ( Figure 1b , Figure 1—figure supplements 1–8 ) . In addition to the 14 haematopoietic enhancers previously published , eight of the 35 new candidate regulatory regions had previously been shown not to possess activity in tissues of the blood system of mouse embryos: Gata2-83 ( Gata2-77 ) , Gfi1b promoter ( Moignard et al . , 2013 ) , Spi1-18 , Spi1 promoter ( Wilkinson et al . , 2014 ) , Runx1 P1 promoter ( Bee et al . , 2009 ) , Tal1-9 , Tal1 promoter ( Sinclair et al . , 1999 ) and Tal1+6 ( Sánchez et al . , 1999 ) . Of the remaining 27 candidate cis-regulatory regions , two coincided with genomic repeat regions ( Runx1-322 and Runx1+1 ) and were excluded from further analysis because mapping of ChIP-Seq reads to such regions is ambiguous . Since a comprehensive understanding of regulatory interactions among the nine HSPC TFs requires in vivo validation of candidate regulatory regions , we next tested the remaining 25 candidate cis-regulatory regions for their ability to mediate reporter gene expression in embryonic sites of definitive haematopoietic cell emergence and colonisation , namely the dorsal aorta and foetal liver of E10 . 5 to E11 . 5 transgenic LacZ-reporter mouse embryos . For the Erg locus , this analysis revealed that in addition to the previously known Erg+85 enhancer , the Erg+65 and Erg+75 regions also displayed activity in the dorsal aorta and/or the foetal liver , while the Erg+90 and Erg+149 regions did not ( Figure 1c ) . Careful inspection of a total of 188 transgenic mouse embryos revealed that nine of the 25 identified regions showed LacZ expression in the dorsal aorta and/or foetal liver ( Figure 1b , Figure 1—figure supplements 1–8 , Figure 1—source data 1 ) . This large-scale transient transgenic screen therefore almost doubled the number of known in vivo validated early haematopoietic regulatory elements for HSPC TFs . 10 . 7554/eLife . 11469 . 003Figure 1 . Identification of haematopoietic active cis-regulatory regions . ( a ) UCSC screenshot of the Erg gene locus for ChIP-Sequencing data for nine haematopoietic TFs ( ERG , FLI1 , GATA2 , GFI1B , LYL1 , MEIS1 , PU . 1 , RUNX1 and TAL1 [Wilson et al . , 2010] ) and for H3K27ac ( Calero-Nieto et al . , 2014 ) in HPC7 cells . Highlighted are all regions of the Erg gene locus that are acetylated at H3K27 and are bound by three or more TFs . Numbers indicate the distance ( in kb ) from the ATG start codon . ( b ) Summary of the identification of candidate cis-regulatory regions for all nine TFs and subsequent analysis in transgenic mouse assays . The inspection of the nine gene loci and the application of the selection criteria ( ≥3 TFs bound and H3K27ac ) identified a total of 49 candidate cis-regulatory regions . The heatmap shows the binding pattern of the nine TFs to all candidate regulatory elements in HPC7 cells: green = bound , grey = unbound . Haematopoietic activity in E11 . 5 transgenic mice is indicated by the font color: black = active , red = not active . Grey indicates genomic repeat regions that were not tested in transgenic mice . Detailed experimental data corresponding to the summary heatmap can be found in Figure 1 and Figure 1—figure supplement 1–8 . ( c ) Haematopoietic activity of the five candidate Erg cis-regulatory regions was determined in E11 . 5 transgenic mouse assays . Shown are X-Gal-stained whole-mount embryos and paraffin sections of the dorsal aorta ( DA , ventral side on the left/top ) and foetal liver ( FL ) , sites of definitive haematopoiesis . Colour coding as in B . DOI: http://dx . doi . org/10 . 7554/eLife . 11469 . 00310 . 7554/eLife . 11469 . 004Figure 1—source data 1 . Number of PCR and LacZ positive transgenic embryos ( E10 . 5–11 . 5 ) for each regulatory region . DOI: http://dx . doi . org/10 . 7554/eLife . 11469 . 00410 . 7554/eLife . 11469 . 005Figure 1—figure supplement 1 . Identification of haematopoietic active cis-regulatory elements for Fli1 . ( a ) The candidate cis-regulatory elements were identified by ChIP-Seq analysis of the TFs ERG , FLI1 , GATA2 , GFI1B , LYL1 , MEIS1 , PU . 1 , RUNX1 and TAL1 as well as H3K27 acetylation in the haematopoietic stem/progenitor cell line HPC7 . Highlighted in pink are the candidate cis-regulatory regions which are bound by at least three of the nine TFs and showed H3K27 acetylation . The numbering represents the direction and distance in kilobases from the start codon ATG ( pro = promoter ) . ( b ) Candidate regions were assayed for haematopoietic enhancer activity in mouse transient transgenic embryos . X-Gal stained whole-mount E11 . 5 embryos and paraffin sections of the dorsal aorta ( DA; longitudinal section , ventral side on the left/top ) and foetal liver ( FL ) are shown for the candidate cis-regulatory regions . Transgenic mouse data are not shown for previously published regions , but relevant publications are listed . DOI: http://dx . doi . org/10 . 7554/eLife . 11469 . 00510 . 7554/eLife . 11469 . 006Figure 1—figure supplement 2 . Identification of haematopoietic active cis-regulatory elements for Gata2 . ( a ) The candidate cis-regulatory elements were identified by ChIP-Seq analysis of the TFs ERG , FLI1 , GATA2 , GFI1B , LYL1 , MEIS1 , PU . 1 , RUNX1 and TAL1 as well as H3K27 acetylation in the haematopoietic stem/progenitor cell line HPC7 . Highlighted in pink are the candidate cis-regulatory regions which are bound by at least three of the nine TFs and showed H3K27 acetylation . The numbering represents the direction and distance in kilobases from the start codon ATG ( pro = promoter ) . ( b ) Candidate regions were assayed for haematopoietic enhancer activity in mouse transient transgenic embryos . X-Gal stained whole-mount E11 . 5 embryos and paraffin sections of the dorsal aorta ( DA; longitudinal section , ventral side on the left/top ) and foetal liver ( FL ) are shown for the candidate cis-regulatory regions . Transgenic mouse data are not shown for previously published regions , but relevant publications are listed . DOI: http://dx . doi . org/10 . 7554/eLife . 11469 . 00610 . 7554/eLife . 11469 . 007Figure 1—figure supplement 3 . Identification of haematopoietic active cis-regulatory elements for Gfi1b . ( a ) The candidate cis-regulatory elements were identified by ChIP-Seq analysis of the TFs ERG , FLI1 , GATA2 , GFI1B , LYL1 , MEIS1 , PU . 1 , RUNX1 and TAL1 as well as H3K27 acetylation in the haematopoietic stem/progenitor cell line HPC7 . Highlighted in pink are the candidate cis-regulatory regions which are bound by at least three of the nine TFs and showed H3K27 acetylation . The numbering represents the direction and distance in kilobases from the start codon ATG ( pro = promoter ) . ( b ) All candidate regions were previously published . Relevant publications are listed . DOI: http://dx . doi . org/10 . 7554/eLife . 11469 . 00710 . 7554/eLife . 11469 . 008Figure 1—figure supplement 4 . Identification of haematopoietic active cis-regulatory elements for Lyl1 . ( a ) The candidate cis-regulatory elements were identified by ChIP-Seq analysis of the TFs ERG , FLI1 , GATA2 , GFI1B , LYL1 , MEIS1 , PU . 1 , RUNX1 and TAL1 as well as H3K27 acetylation in the haematopoietic stem/progenitor cell line HPC7 . Highlighted in pink are the candidate cis-regulatory regions which are bound by at least three of the nine TFs and showed H3K27 acetylation . The numbering represents the direction and distance in kilobases from the start codon ATG ( pro = promoter ) . ( b ) Candidate regions were assayed for haematopoietic enhancer activity in mouse transient transgenic embryos . X-Gal stained whole-mount E11 . 5 embryos and paraffin sections of the dorsal aorta ( DA; longitudinal section , ventral side on the left/top ) and foetal liver ( FL ) are shown for the candidate cis-regulatory regions . Transgenic mouse data are not shown for previously published regions , but relevant publications are listed . DOI: http://dx . doi . org/10 . 7554/eLife . 11469 . 00810 . 7554/eLife . 11469 . 009Figure 1—figure supplement 5 . Identification of haematopoietic active cis-regulatory elements for Meis1 . ( a ) The candidate cis-regulatory elements were identified by ChIP-Seq analysis of the TFs ERG , FLI1 , GATA2 , GFI1B , LYL1 , MEIS1 , PU . 1 , RUNX1 and TAL1 as well as H3K27 acetylation in the haematopoietic stem/progenitor cell line HPC7 . Highlighted in pink are the candidate cis-regulatory regions which are bound by at least three of the nine TFs and showed H3K27 acetylation . The numbering represents the direction and distance in kilobases from the start codon ATG ( pro = promoter ) . ( b ) Candidate regions were assayed for haematopoietic enhancer activity in mouse transient transgenic embryos . X-Gal stained whole-mount E11 . 5 embryos and paraffin sections of the dorsal aorta ( DA; longitudinal section , ventral side on the left/top ) and foetal liver ( FL ) are shown for the candidate cis-regulatory regions . DOI: http://dx . doi . org/10 . 7554/eLife . 11469 . 00910 . 7554/eLife . 11469 . 010Figure 1—figure supplement 6 . Identification of haematopoietic active cis-regulatory elements for Runx1 . ( a ) The candidate cis-regulatory elements were identified by ChIP-Seq analysis of the TFs ERG , FLI1 , GATA2 , GFI1B , LYL1 , MEIS1 , PU . 1 , RUNX1 and TAL1 as well as H3K27 acetylation in the haematopoietic stem/progenitor cell line HPC7 . Highlighted in pink are the candidate cis-regulatory regions which are bound by at least three of the nine TFs and showed H3K27 acetylation . The numbering represents the direction and distance in kilobases from the start codon ATG ( pro = promoter ) . ( b ) E10 embryos and cryosections of the DA ( transverse; ventral down ) and FL are shown . For the Runx1+204 region , a larger 12 kb fragment ( chr16:92 , 620 , 915–92 , 631 , 936 , mm9 ) was used for transient transgenesis , but similar results were obtained with the +204 fragment alone ( data not shown ) . The +24 element was tested in conjunction with the +23 and did not change its tissue specificity ( Bee et al . , 2010 ) . Preliminary data show that the +24 on its own does not mediate robust tissue specific expression of reporter genes . Transgenic mouse data are not shown for previously published regions , but relevant publications are listed . DOI: http://dx . doi . org/10 . 7554/eLife . 11469 . 01010 . 7554/eLife . 11469 . 011Figure 1—figure supplement 7 . Identification of haematopoietic active cis-regulatory elements for Spi1 . ( a ) The candidate cis-regulatory elements were identified by ChIP-Seq analysis of the TFs ERG , FLI1 , GATA2 , GFI1B , LYL1 , MEIS1 , PU . 1 , RUNX1 and TAL1 as well as H3K27 acetylation in the haematopoietic stem/progenitor cell line HPC7 . Highlighted in pink are the candidate cis-regulatory regions which are bound by at least three of the nine TFs and showed H3K27 acetylation . The numbering represents the direction and distance in kilobases from the start codon ATG ( pro = promoter ) . ( b ) All candidate regions were previously published . Relevant publications are listed . DOI: http://dx . doi . org/10 . 7554/eLife . 11469 . 01110 . 7554/eLife . 11469 . 012Figure 1—figure supplement 8 . Identification of haematopoietic active cis-regulatory elements for Tal1 . ( a ) The candidate cis-regulatory elements were identified by ChIP-Seq analysis of the TFs ERG , FLI1 , GATA2 , GFI1B , LYL1 , MEIS1 , PU . 1 , RUNX1 and TAL1 as well as H3K27 acetylation in the haematopoietic stem/progenitor cell line HPC7 . Highlighted in pink are the candidate cis-regulatory regions which are bound by at least three of the nine TFs and showed H3K27 acetylation . The numbering is based on the distance ( in kb ) to promoter 1a . ( b ) All candidate regions were previously published . Relevant publications are listed . DOI: http://dx . doi . org/10 . 7554/eLife . 11469 . 012 Although HPC7 cells are a useful model cell line for the prediction of genomic regions with haematopoietic activity in transgenic mouse assays ( Wilson et al . , 2009 ) , they are refractory to most gene transfer methods and therefore not suitable for functional characterisation of regulatory elements using standard transcriptional assays . By contrast , the 416b myeloid progenitor cell line can be readily transduced by electroporation and therefore represents a candidate cell line for functional dissection of individual regulatory elements . As ChIP-Seq profiles in 416b cells had not been reported previously , we performed ChIP-Seq for H3K27ac and the nine TFs in this cell line ( Figure 2a , Figure 2—figure supplements 1–8 ) . Alongside with our previously published HPC7 data , this new 416b dataset represents the most complete genome-scale TF-binding analysis in haematopoietic progenitor cell lines to date , with all new data being freely accessible under the following GEO accession number GSE69776 and also at http://codex . stemcells . cam . ac . uk/ . Genome-wide TF binding patterns in 416b and HPC7 cells were closely related when compared with binding profiles for the same factors in other haematopoietic lineages ( Figure 2b , Figure 2—source data 1 ) . Inspection of the gene loci for the nine HSPC TFs not only revealed many similarities between 416b and HPC7 cells , but also some differences in TF binding patterns . Overall , TF occupancy at the 23 regions with activity in haematopoietic tissues ( 14 previously published ( Wilson et al . , 2009; Beck et al . , 2013; Pimanda et al . , 2007; Wozniak et al . , 2007; Moignard et al . , 2013; Chan et al . , 2007; Wilkinson et al . , 2014; Nottingham et al . , 2007; Göttgens et al . , 2002; Gottgens et al . , 2004; Gottgens et al . , 2010 ) and 9 newly identified ) does not change between the two cell types in 71% of all cases ( 147 of 207 binding events ) , is gained in 416b cells in 16% ( 33 of 207 ) and lost in 13% ( 27 of 207 ) of cases compared to HPC7 cells ( Figure 2c ) . Next , all 23 elements were filtered to only retain those elements which were bound by at least 3 of the 9 TFs and displayed elevated H3K27ac in HPC7 and 416 cells . This led to the removal of the Gata2-3 , which is not bound by any of the nine TFs in either cell type , Gata2-92 and Gfi1b+13 , which are only bound by one or no TFs in 416b cells , and Fli1-15 , which is not acetylated in 416b cells ( Figure 2c , Figure 2—figure supplements 1–3 ) . Overall , 19 cis-regulatory regions were therefore taken forward as a comprehensively validated set of regions for the reconstruction of an HSPC regulatory network model . 10 . 7554/eLife . 11469 . 013Figure 2 . Comparison of TF binding pattern at haematopoietic active cis-regulatory regions in two haematopoietic progenitor cell lines , HPC7 and 416b . ( a ) UCSC screenshot of the Erg gene locus for ChIP-Sequencing data for nine haematopoietic TFs ( ERG , FLI1 , GATA2 , GFI1B , LYL1 , MEIS1 , PU . 1 , RUNX1 and TAL1 ) and for H3K27ac in 416b cells . Highlighted are those haematopoietic active Erg cis-regulatory regions that were identified based on acetylation of H3K27 and TF binding in HPC7 cells followed by transgenic mouse assays . Numbers indicate the distance ( in kb ) from the ATG start codon . ( b ) Hierarchical clustering of the binding profiles for HPC7 , 416b and other published datasets . The heatmap shows the pairwise correlation coefficient of peak coverage data between the pairs of samples in the row and column . The order of the samples is identical in columns and rows . Details about samples listed can be found in Figure 2—source data 1 . ( c ) Pair-wise analysis of binding of the nine TFs to haematopoietic active cis-regulatory regions of the nine TFs in HPC7 versus 416b cells . Green = bound in both cells types , blue = only bound in 416b cells , orange = only bound in HPC7 cells , grey = not bound in either cell type . DOI: http://dx . doi . org/10 . 7554/eLife . 11469 . 01310 . 7554/eLife . 11469 . 014Figure 2—source data 1 . List of ChIP-Seq samples included in the heatmap in Figure 2b . DOI: http://dx . doi . org/10 . 7554/eLife . 11469 . 01410 . 7554/eLife . 11469 . 015Figure 2—figure supplement 1 . UCSC screenshot for the Fli1 gene locus demonstrating binding patterns for nine key haematopoietic TFs and H3K27ac in 416b cells . Highlighted in pink are cis-regulatory regions that were identified based on the selection criteria ( ≥3 TFs bound and H3K27ac ) in HPC7 cells and were shown to possess haematopoietic activity . The numbering represents the distance ( in kb ) from the start codon ATG . DOI: http://dx . doi . org/10 . 7554/eLife . 11469 . 01510 . 7554/eLife . 11469 . 016Figure 2—figure supplement 2 . UCSC screenshot for the Gata2 gene locus demonstrating binding patterns for nine key haematopoietic TFs and H3K27ac in 416b cells . Highlighted in pink are cis-regulatory regions that were identified based on the selection criteria ( ≥3 TFs bound and H3K27ac ) in HPC7 cells and were shown to possess haematopoietic activity . The numbering represents the distance ( in kb ) from the start codon ATG . DOI: http://dx . doi . org/10 . 7554/eLife . 11469 . 01610 . 7554/eLife . 11469 . 017Figure 2—figure supplement 3 . UCSC screenshot for the Gfi1b gene locus demonstrating binding patterns for nine key haematopoietic TFs and H3K27ac in 416b cells . Highlighted in pink are cis-regulatory regions that were identified based on the selection criteria ( ≥3 TFs bound and H3K27ac ) in HPC7 cells and were shown to possess haematopoietic activity . The numbering represents the distance ( in kb ) from the start codon ATG . DOI: http://dx . doi . org/10 . 7554/eLife . 11469 . 01710 . 7554/eLife . 11469 . 018Figure 2—figure supplement 4 . UCSC screenshot for the Lyl1 gene locus demonstrating binding patterns for nine key haematopoietic TFs and H3K27ac in 416b cells . Highlighted in pink is the promoter ( 'pro' ) that was identified based on the selection criteria ( ≥3 TFs bound and H3K27ac ) in HPC7 cells and was shown to possess haematopoietic activity . The promoter is labelled with 'pro' . DOI: http://dx . doi . org/10 . 7554/eLife . 11469 . 01810 . 7554/eLife . 11469 . 019Figure 2—figure supplement 5 . UCSC screenshot for the Meis1 gene locus demonstrating binding patterns for nine key haematopoietic TFs and H3K27ac in 416b cells . Highlighted in pink is the cis-regulatory region that was identified based on the selection criteria ( ≥3 TFs bound and H3K27ac ) in HPC7 cells and was shown to possess haematopoietic activity . The numbering represents the distance ( in kb ) from the start codon ATG . DOI: http://dx . doi . org/10 . 7554/eLife . 11469 . 01910 . 7554/eLife . 11469 . 020Figure 2—figure supplement 6 . UCSC screenshot for the Runx1 gene locus demonstrating binding patterns for nine key haematopoietic TFs and H3K27ac in 416b cells . Highlighted in pink are cis-regulatory regions that were identified based on the selection criteria ( ≥3 TFs bound and H3K27ac ) in HPC7 cells and were subsequently shown to possess haematopoietic activity . The numbering represents the distance ( in kb ) from the start codon ATG . DOI: http://dx . doi . org/10 . 7554/eLife . 11469 . 02010 . 7554/eLife . 11469 . 021Figure 2—figure supplement 7 . UCSC screenshot for the Spi1 gene locus demonstrating binding patterns for nine key haematopoietic TFs and H3K27ac in 416b cells . Highlighted in pink is the cis-regulatory region that was identified based on the selection criteria ( ≥3 TFs bound and H3K27ac ) in HPC7 cells and was shown to possess haematopoietic activity . The numbering represents the distance ( in kb ) from the start codon ATG . DOI: http://dx . doi . org/10 . 7554/eLife . 11469 . 02110 . 7554/eLife . 11469 . 022Figure 2—figure supplement 8 . UCSC screenshot for the Tal1 gene locus demonstrating binding patterns for nine key haematopoietic TFs and H3K27ac in 416b cells . Highlighted in pink are cis-regulatory regions that were identified based on the selection criteria ( ≥3 TFs bound and H3K27ac ) in HPC7 cells and were shown to possess haematopoietic activity . The numbering represents the distance ( in kb ) from promoter 1a . DOI: http://dx . doi . org/10 . 7554/eLife . 11469 . 022 The reconstruction of a core regulatory network model requires information about the effect of TF binding on gene expression , which can be activating , repressing or non-functional . In order to analyse the effects of all TF-binding events at all 19 regulatory regions , we performed luciferase reporter assays in stably transfected 416b cells . Based on multiple species alignments between five species ( mouse , human , dog , platypus , opossum ) , we identified conserved TFBSs for the nine TFs ( Figure 3a , Figure 3—figure supplements 1–18 , Figure 3—source data 1 ) , and generated mutant constructs for each of the 19 regulatory regions , resulting in 87 reporter constructs that were tested by luciferase assays ( 19 wild-type , 68 mutants ) . To ensure that DNA binding of the TFs was abrogated , the key DNA bases involved in DNA-protein interactions were mutated and the resulting sequences were scanned to ensure that no new binding sites were created ( Lelieveld et al . , 2015 ) . For each of the 19 regulatory regions , the conserved TFBSs were mutated by family , for example , all six Ets sites within the Erg+65 region were mutated simultaneously in one construct , and this element was then treated as the Erg+65_Ets mutant . TFBS mutations reduced or increased activity compared to the wild-type enhancer , or indeed had no significant effect ( Figure 3b , Figure 3—figure supplements 1–18 ) . For instance , at the Erg+65 region , mutation of the six Ets binding sites or the three Gata binding sites reduced luciferase activity , whereas mutation of the three Ebox or the three Gfi motifs increased luciferase activity ( Figure 3b ) . Comparison of the luciferase assay results for all 19 cis-regulatory regions ( Figure 3c ) reveals that for each motif class mutation can result in activation , repression or no-change . This observation even extends to single gene loci , where for example mutation of the Gata site reduced activity of the Erg+65 region , but increased activity of the Erg+85 enhancer ( Figure 3c ) . Taken together , this comprehensive mutagenesis screen highlights the dangers associated with extrapolating TF function simply from ChIP-Seq binding events and thus underlines the importance of functional studies for regulatory network reconstruction . 10 . 7554/eLife . 11469 . 023Figure 3 . TFBS mutagenesis reveals enhancer-dependent effects of TF binding on gene expression . ( a ) Multiple species alignment of mouse ( mm9 ) , human ( hg19 ) , dog ( canFam2 ) , opossum ( monDom5 ) and platypus ( ornAna1 ) sequences for the Erg+65 region . Nucleotides highlighted in black are conserved between all species analysed , nucleotides highlighted in grey are conserved between four of five species . Transcription factor binding sites ( TFBS ) are highlighted in: blue = Ebox , purple = Ets , green = Gata , yellow = Gfi , red = Meis . The nucleotides that were changed to mutate the TFBSs are indicated below the alignment . All binding sites of one motif family ( e . g . all Ebox motifs ) were mutated simultaneously . ( b ) Luciferase assay for the Erg+65 wild-type and mutant enhancer in stably transfected 416b cells . Each bar represents the averages of at least three independent experiments with three to four replicates within each experiment . The results are shown relative to the wild-type enhancer activity , which is set to 100% . Error bars represent the standard error of the mean ( SEM ) . Stars indicate significance: **=p-value <0 . 01 , ***=p-value <0 . 001 . p-values were calculated using t-tests , followed by the Fisher’s method . ( c ) Summary of luciferase assay results for all 19 high-confidence haematopoietic active regulatory regions . Relative luciferase activity is illustrated in shades of blue ( down-regulation ) and red ( up-regulation ) . Crossed-out grey boxes indicate that there is no motif for the TF and/or the TF does not bind to the region . Detailed results and corresponding alignments with highlighted TFBSs and their mutations can be found in Figure 3—figure supplements 1–18 . DOI: http://dx . doi . org/10 . 7554/eLife . 11469 . 02310 . 7554/eLife . 11469 . 024Figure 3—source data 1 . List of TF binding sites and the TFs that bind to them . DOI: http://dx . doi . org/10 . 7554/eLife . 11469 . 02410 . 7554/eLife . 11469 . 025Figure 3—source data 2 . List of co-ordinates and primer sequences for the regulatory regions analysed in this study . DOI: http://dx . doi . org/10 . 7554/eLife . 11469 . 02510 . 7554/eLife . 11469 . 026Figure 3—figure supplement 1 . Multiple species alignment and luciferase assay results for Erg+75 . ( a ) Multiple species alignment ( MSA ) with the following species: mouse ( mm9 ) , human ( hg19 ) , dog ( canFam2 ) , opossum ( monDom5 ) and platypus ( ornAna1 ) . Nucleotides highlighted in black are conserved between all species analysed , nucleotides highlighted in grey are conserved between four of five species . Transcription factor binding sites ( TFBS ) are highlighted in: blue = Ebox , purple = Ets , yellow = Gfi . The nucleotides that were changed to mutate the TFBSs are indicated below the MSA . All conserved binding sites of one motif family ( e . g . all Ebox motifs ) were mutated simultaneously . Where TF binding was observed in ChIP-Seq experiments in 416b cells , but the TFBS was not conserved , the motifs present in the mouse sequence only were mutated . ( b ) For the luciferase reporter assays in stably transfected 416b cells , the averages of at least three independent experiments with three to four replicates within each experiment are shown . Error bars represent the standard error of the mean ( SEM ) . Stars indicate significance: **=p-value <0 . 01 , ***=p-value <0 . 001 . p-values were generated using t-tests , followed by the Fisher’s method and if necessary Stouffer’s z trend . DOI: http://dx . doi . org/10 . 7554/eLife . 11469 . 02610 . 7554/eLife . 11469 . 027Figure 3—figure supplement 2 . Multiple species alignment and luciferase assay results for Erg+85 . ( a ) Multiple species alignment ( MSA ) with the following species: mouse ( mm9 ) , human ( hg19 ) , dog ( canFam2 ) , opossum ( monDom5 ) and platypus ( ornAna1 ) . Nucleotides highlighted in black are conserved between all species analysed , nucleotides highlighted in grey are conserved between four of five species . Transcription factor binding sites ( TFBS ) are highlighted in: blue = Ebox , purple = Ets , green = Gata , yellow = Gfi . The nucleotides that were changed to mutate the TFBSs are indicated below the MSA . All conserved binding sites of one motif family ( e . g . all Ebox motifs ) were mutated simultaneously . ( b ) For the luciferase reporter assays in stably transfected 416b cells , the averages of at least three independent experiments with three to four replicates within each experiment are shown . Error bars represent the standard error of the mean ( SEM ) . Stars indicate significance: ***=p-value <0 . 001 . p-values were generated using t-tests , followed by the Fisher’s method and if necessary Stouffer’s z trend . DOI: http://dx . doi . org/10 . 7554/eLife . 11469 . 02710 . 7554/eLife . 11469 . 028Figure 3—figure supplement 3 . Multiple species alignment and luciferase assay results for Fli1+12 . ( a ) Multiple species alignment ( MSA ) with the following species: mouse ( mm9 ) , human ( hg19 ) , dog ( canFam2 ) , opossum ( monDom5 ) and platypus ( ornAna1 ) . Nucleotides highlighted in black are conserved between all species analysed , nucleotides highlighted in grey are conserved between four of five species . Transcription factor binding sites ( TFBS ) are highlighted in: blue = Ebox , purple = Ets . The nucleotides that were changed to mutate the TFBSs are indicated below the MSA . All conserved binding sites of one motif family ( e . g . all Etsmotifs ) were mutated simultaneously . ( b ) For the luciferase reporter assays in stably transfected 416b cells , the averages of at least three independent experiments with three to four replicates within each experiment are shown . Error bars represent the standard error of the mean ( SEM ) . Stars indicate significance: ***=p-value <0 . 001 . p-values were generated using t-tests , followed by the Fisher’s method and if necessary Stouffer’s z trend . DOI: http://dx . doi . org/10 . 7554/eLife . 11469 . 02810 . 7554/eLife . 11469 . 029Figure 3—figure supplement 4 . Multiple species alignment and luciferase assay results for Gata2-93 . ( a ) Multiple species alignment ( MSA ) with the following species: mouse ( mm9 ) , human ( hg19 ) , dog ( canFam2 ) , opossum ( monDom5 ) and platypus ( ornAna1 ) . Nucleotides highlighted in black are conserved between all species analysed , nucleotides highlighted in grey are conserved between four of five species . Transcription factor binding sites ( TFBS ) are highlighted in: blue = Ebox , purple = Ets , green = Gata , red = Meis , turquoise = Runt . The nucleotides that were changed to mutate the TFBSs are indicated below the MSA . All conserved binding sites of one motif family ( e . g . all Ets motifs ) were mutated simultaneously . ( b ) For the luciferase reporter assays in stably transfected 416b cells , the averages of at least three independent experiments with three to four replicates within each experiment are shown . Error bars represent the standard error of the mean ( SEM ) . Stars indicate significance: **=p-value <0 . 01 , ***=p-value <0 . 001 . p-values were generated using t-tests , followed by the Fisher’s method and if necessary Stouffer’s z trend . DOI: http://dx . doi . org/10 . 7554/eLife . 11469 . 02910 . 7554/eLife . 11469 . 030Figure 3—figure supplement 5 . Multiple species alignment and luciferase assay results for Gata2+3 . ( a ) Multiple species alignment ( MSA ) with the following species: mouse ( mm9 ) , human ( hg19 ) , dog ( canFam2 ) , opossum ( monDom5 ) and platypus ( ornAna1 ) . Nucleotides highlighted in black are conserved between all species analysed , nucleotides highlighted in grey are conserved between four of five species . Transcription factor binding sites ( TFBS ) are highlighted in: blue = Ebox , purple = Ets , green = Gata . The nucleotides that were changed to mutate the TFBSs are indicated below the MSA . All conserved binding sites of one motif family ( e . g . all Ebox motifs ) were mutated simultaneously . ( b ) For the luciferase reporter assays in stably transfected 416b cells , the averages of at least three independent experiments with three to four replicates within each experiment are shown . Error bars represent the standard error of the mean ( SEM ) . Stars indicate significance: ***=p-value <0 . 001 . p-values were generated using t-tests , followed by the Fisher’s method and if necessary Stouffer’s z trend . DOI: http://dx . doi . org/10 . 7554/eLife . 11469 . 03010 . 7554/eLife . 11469 . 031Figure 3—figure supplement 6 . Multiple species alignment and luciferase assay results for Gfi1b+16 . ( a ) Multiple species alignment ( MSA ) with the following species: mouse ( mm9 ) , human ( hg19 ) , dog ( canFam2 ) , opossum ( monDom5 ) and platypus ( ornAna1 ) . Nucleotides highlighted in black are conserved between all species analysed , nucleotides highlighted in grey are conserved between four of five species . Transcription factor binding sites ( TFBS ) are highlighted in: blue = Ebox , purple = Ets , green = Gata , yellow = Gfi , red = Meis , turquoise = Runt . The nucleotides that were changed to mutate the TFBSs are indicated below the MSA . All conserved binding sites of one motif family ( e . g . all Ebox motifs ) were mutated simultaneously . Where TF binding was observed in ChIP-Seq experiments in 416b cells , but the TFBS was not conserved , the motifs present in the mouse sequence only were mutated . ( b ) For the luciferase reporter assays in stably transfected 416b cells , the averages of at least three independent experiments with three to four replicates within each experiment are shown . Error bars represent the standard error of the mean ( SEM ) . Stars indicate significance: **=p-value <0 . 01 , ***=p-value <0 . 001 . p-values were generated using t-tests , followed by the Fisher’s method and if necessary Stouffer’s z trend . DOI: http://dx . doi . org/10 . 7554/eLife . 11469 . 03110 . 7554/eLife . 11469 . 032Figure 3—figure supplement 7 . Multiple species alignment and luciferase assay results for Gfi1b+17 . ( a ) Multiple species alignment ( MSA ) with the following species: mouse ( mm9 ) , human ( hg19 ) , dog ( canFam2 ) , opossum ( monDom5 ) and platypus ( ornAna1 ) . Nucleotides highlighted in black are conserved between all species analysed , nucleotides highlighted in grey are conserved between four of five species . Transcription factor binding sites ( TFBS ) are highlighted in: blue = Ebox , purple = Ets , green = Gata , yellow = Gfi , red = Meis . The nucleotides that were changed to mutate the TFBSs are indicated below the MSA . All conserved binding sites of one motif family ( e . g . all Ets motifs ) were mutated simultaneously . ( b ) For the luciferase reporter assays in stably transfected 416b cells , the averages of at least three independent experiments with three to four replicates within each experiment are shown . Error bars represent the standard error of the mean ( SEM ) . Stars indicate significance: **=p-value <0 . 01 , ***=p-value <0 . 001 . p-values were generated using t-tests , followed by the Fisher’s method and if necessary Stouffer’s z trend . DOI: http://dx . doi . org/10 . 7554/eLife . 11469 . 03210 . 7554/eLife . 11469 . 033Figure 3—figure supplement 8 . Multiple species alignment and luciferase assay results for Lyl1 promoter . ( a ) Multiple species alignment ( MSA ) with the following species: mouse ( mm9 ) , human ( hg19 ) , dog ( canFam2 ) and opossum ( monDom5 ) . Nucleotides highlighted in black are conserved between all species analysed , nucleotides highlighted in grey are conserved between three of four species . Transcription factor binding sites ( TFBS ) are highlighted in: purple = Ets , green = Gata . The nucleotides that were changed to mutate the TFBSs are indicated below the MSA . All conserved binding sites of one motif family ( e . g . all Ets motifs ) were mutated simultaneously . ( b ) For the luciferase reporter assays in stably transfected 416b cells , the averages of at least three independent experiments with three to four replicates within each experiment are shown . Error bars represent the standard error of the mean ( SEM ) . Stars indicate significance: ***=p-value <0 . 001 . p-values were generated using t-tests , followed by the Fisher’s method and if necessary Stouffer’s z trend . DOI: http://dx . doi . org/10 . 7554/eLife . 11469 . 03310 . 7554/eLife . 11469 . 034Figure 3—figure supplement 9 . Multiple species alignment and luciferase assay results for Meis1+48 . ( a ) Multiple species alignment ( MSA ) with the following species: mouse ( mm9 ) , human ( hg19 ) , dog ( canFam2 ) , opossum ( monDom5 ) and platypus ( ornAna1 ) . Nucleotides highlighted in black are conserved between all species analysed , nucleotides highlighted in grey are conserved between four of five species . Transcription factor binding sites ( TFBS ) are highlighted in: purple = Ets , green = Gata , yellow = Gfi , red = Meis . The nucleotides that were changed to mutate the TFBSs are indicated below the MSA . All conserved binding sites of one motif family ( e . g . all Ets motifs ) were mutated simultaneously . ( b ) For the luciferase reporter assays in stably transfected 416b cells , the averages of at least three independent experiments with three to four replicates within each experiment are shown . Error bars represent the standard error of the mean ( SEM ) . Stars indicate significance: ***=p-value <0 . 001 . p-values were generated using t-tests , followed by the Fisher’s method and if necessary Stouffer’s z trend . DOI: http://dx . doi . org/10 . 7554/eLife . 11469 . 03410 . 7554/eLife . 11469 . 035Figure 3—figure supplement 10 . Multiple species alignment and luciferase assay results for Spi1-14 . ( a ) Multiple species alignment ( MSA ) with the following species: mouse ( mm9 ) , human ( hg19 ) , dog ( canFam2 ) , opossum ( monDom5 ) and platypus ( ornAna1 ) . Nucleotides highlighted in black are conserved between all species analysed , nucleotides highlighted in grey are conserved between four of five species . Transcription factor binding sites ( TFBS ) are highlighted in: blue = Ebox , purple = Ets , turquoise = Runt . The nucleotides that were changed to mutate the TFBSs are indicated below the MSA . All conserved binding sites of one motif family ( e . g . all Ets motifs ) were mutated simultaneously . ( b ) For the luciferase reporter assays in stably transfected 416b cells , the averages of at least three independent experiments with three to four replicates within each experiment are shown . Error bars represent the standard error of the mean ( SEM ) . Stars indicate significance: **=p-value <0 . 01 , ***=p-value <0 . 001 . p-values were generated using t-tests , followed by the Fisher’s method and if necessary Stouffer’s z trend . DOI: http://dx . doi . org/10 . 7554/eLife . 11469 . 03510 . 7554/eLife . 11469 . 036Figure 3—figure supplement 11 . Multiple species alignment and luciferase assay results for Runx1-59 . ( a ) Multiple species alignment ( MSA ) with the following species: mouse ( mm9 ) , human ( hg19 ) and dog ( canFam2 ) . Nucleotides highlighted in black are conserved between all species analysed , nucleotides highlighted in grey are conserved between two of three species . Transcription factor binding sites ( TFBS ) are highlighted in: blue = Ebox , purple = Ets , green = Gata , red = Meis . The nucleotides that were changed to mutate the TFBSs are indicated below the MSA . All conserved binding sites of one motif family ( e . g . all Ebox motifs ) were mutated simultaneously . ( b ) For the luciferase reporter assays in stably transfected 416b cells , the averages of at least three independent experiments with three to four replicates within each experiment are shown . Error bars represent the standard error of the mean ( SEM ) . Stars indicate significance: ***=p-value <0 . 001 . p-values were generated using t-tests , followed by the Fisher’s method and if necessary Stouffer’s z trend . DOI: http://dx . doi . org/10 . 7554/eLife . 11469 . 03610 . 7554/eLife . 11469 . 037Figure 3—figure supplement 12 . Multiple species alignment and luciferase assay results for Runx1+3 . ( a ) Multiple species alignment ( MSA ) with the following species: mouse ( mm9 ) , human ( hg19 ) , dog ( canFam2 ) , opossum ( monDom5 ) and platypus ( ornAna1 ) . Nucleotides highlighted in black are conserved between all species analysed , nucleotides highlighted in grey are conserved between four of five species . Transcription factor binding sites ( TFBS ) are highlighted in: blue = Ebox , purple = Ets , green = Gata , yellow = Gfi , red = Meis , turquoise = Runt . The nucleotides that were changed to mutate the TFBSs are indicated below the MSA . All conserved binding sites of one motif family ( e . g . all Ets motifs ) were mutated simultaneously . Where TF binding was observed in ChIP-Seq experiments in 416b cells , but the TFBS was not conserved , the motifs present in the mouse sequence only were mutated . ( b ) For the luciferase reporter assays in stably transfected 416b cells , the averages of at least three independent experiments with three to four replicates within each experiment are shown . Error bars represent the standard error of the mean ( SEM ) . Stars indicate significance: *=p-value <0 . 05 , **=p-value <0 . 01 , ***=p-value <0 . 001 . p-values were generated using t-tests , followed by the Fisher’s method and if necessary Stouffer’s z trend . DOI: http://dx . doi . org/10 . 7554/eLife . 11469 . 03710 . 7554/eLife . 11469 . 038Figure 3—figure supplement 13 . Multiple species alignment and luciferase assay results for Runx1+23 . ( a ) Multiple species alignment ( MSA ) with the following species: mouse ( mm9 ) , human ( hg19 ) , dog ( canFam2 ) and opossum ( monDom5 ) . Nucleotides highlighted in black are conserved between all species analysed , nucleotides highlighted in grey are conserved between three to four species . Transcription factor binding sites ( TFBS ) are highlighted in: blue = Ebox , purple = Ets , green = Gata , red = Meis , turquoise = Runt . The nucleotides that were changed to mutate the TFBSs are indicated below the MSA . All conserved binding sites of one motif family ( e . g . all Ebox motifs ) were mutated simultaneously . ( b ) For the luciferase reporter assays in stably transfected 416b cells , the averages of at least three independent experiments with three to four replicates within each experiment are shown . Error bars represent the standard error of the mean ( SEM ) . Stars indicate significance: *=p-value <0 . 05 , ***=p-value <0 . 001 . p-values were generated using t-tests , followed by the Fisher’s method and if necessary Stouffer’s z trend . DOI: http://dx . doi . org/10 . 7554/eLife . 11469 . 03810 . 7554/eLife . 11469 . 039Figure 3—figure supplement 14 . Multiple species alignment and luciferase assay results for Runx1+110 . ( a ) Multiple species alignment ( MSA ) with the following species: mouse ( mm9 ) , human ( hg19 ) , dog ( canFam2 ) , opossum ( monDom5 ) and platypus ( ornAna1 ) . Nucleotides highlighted in black are conserved between all species analysed , nucleotides highlighted in grey are conserved between four of five species . Transcription factor binding sites ( TFBS ) are highlighted in: blue = Ebox , purple = Ets , green = Gata . The nucleotides that were changed to mutate the TFBSs are indicated below the MSA . All conserved binding sites of one motif family ( e . g . all Ets motifs ) were mutated simultaneously . Where TF binding was observed in ChIP-Seq experiments in 416b cells , but the TFBS was not conserved , the motifs present in the mouse sequence only were mutated . ( b ) For the luciferase reporter assays in stably transfected 416b cells , the averages of at least three independent experiments with three to four replicates within each experiment are shown . Error bars represent the standard error of the mean ( SEM ) . Stars indicate significance: **=p-value <0 . 01 , ***=p-value <0 . 001 . p-values were generated using t-tests , followed by the Fisher’s method and if necessary Stouffer’s z trend . DOI: http://dx . doi . org/10 . 7554/eLife . 11469 . 03910 . 7554/eLife . 11469 . 040Figure 3—figure supplement 15 . Multiple species alignment and luciferase assay results for Runx1+204 . ( a ) Multiple species alignment ( MSA ) with the following species: mouse ( mm9 ) , human ( hg19 ) , dog ( canFam2 ) , opossum ( monDom5 ) and platypus ( ornAna1 ) . Nucleotides highlighted in black are conserved between all species analysed , nucleotides highlighted in grey are conserved between four of five species . Transcription factor binding sites ( TFBS ) are highlighted in: blue = Ebox , purple = Ets , yellow = Gfi , turquoise = Runt . The nucleotides that were changed to mutate the TFBSs are indicated below the MSA . All conserved binding sites of one motif family ( e . g . all Ets motifs ) were mutated simultaneously . Where TF binding was observed in ChIP-Seq experiments in 416b cells , but the TFBS was not conserved , the motifs present in the mouse sequence only were mutated . ( b ) For the luciferase reporter assays in stably transfected 416b cells , the averages of at least three independent experiments with three to four replicates within each experiment are shown . Error bars represent the standard error of the mean ( SEM ) . Stars indicate significance: ***=p-value <0 . 001 . p-values were generated using t-tests , followed by the Fisher’s method and if necessary Stouffer’s z trend . DOI: http://dx . doi . org/10 . 7554/eLife . 11469 . 04010 . 7554/eLife . 11469 . 041Figure 3—figure supplement 16 . Multiple species alignment and luciferase assay results for Tal1-4 . ( a ) Multiple species alignment ( MSA ) with the following species: mouse ( mm9 ) , human ( hg19 ) and dog ( canFam2 ) . Nucleotides highlighted in black are conserved between all species analysed , nucleotides highlighted in grey are conserved between two of three species . Transcription factor binding sites ( TFBS ) are highlighted in: purple = Ets . The nucleotides that were changed to mutate the TFBSs are indicated below the MSA . All conserved binding sites of the Ets motif family were mutated simultaneously . ( b ) For the luciferase reporter assays in stably transfected 416b cells , the averages of at least three independent experiments with three to four replicates within each experiment are shown . Error bars represent the standard error of the mean ( SEM ) . Stars indicate significance: **=p-value <0 . 01 . p-values were generated using t-tests , followed by the Fisher’s method and if necessary Stouffer’s z trend . DOI: http://dx . doi . org/10 . 7554/eLife . 11469 . 04110 . 7554/eLife . 11469 . 042Figure 3—figure supplement 17 . Multiple species alignment and luciferase assay results for Tal1+19 . ( a ) Multiple species alignment ( MSA ) with the following species: mouse ( mm9 ) , human ( hg19 ) , dog ( canFam2 ) and opossum ( monDom5 ) . Nucleotides highlighted in black are conserved between all species analysed , nucleotides highlighted in grey are conserved between three of four species . Transcription factor binding sites ( TFBS ) are highlighted in: purple = Ets . The nucleotides that were changed to mutate the TFBSs are indicated below the MSA . All conserved binding sites of the Ets motif family were mutated simultaneously . ( b ) For the luciferase reporter assays in stably transfected 416b cells , the averages of at least three independent experiments with three to four replicates within each experiment are shown . Error bars represent the standard error of the mean ( SEM ) . Stars indicate significance: ***=p-value <0 . 001 . p-values were generated using t-tests , followed by the Fisher’s method and if necessary Stouffer’s z trend . DOI: http://dx . doi . org/10 . 7554/eLife . 11469 . 04210 . 7554/eLife . 11469 . 043Figure 3—figure supplement 18 . Multiple species alignment and luciferase assay results for Tal1+40 . ( a ) Multiple species alignment ( MSA ) with the following species: mouse ( mm9 ) , human ( hg19 ) and dog ( canFam2 ) . Nucleotides highlighted in black are conserved between all species analysed , nucleotides highlighted in grey are conserved between two of three species . Transcription factor binding sites ( TFBS ) are highlighted in: blue = Ebox , purple = Ets , green = Gata . The nucleotides that were changed to mutate the TFBSs are indicated below the MSA . All conserved binding sites of one motif family ( e . g . all Ebox motifs ) were mutated simultaneously . ( b ) For the luciferase reporter assays in stably transfected 416b cells , the averages of at least three independent experiments with three to four replicates within each experiment are shown . Error bars represent the standard error of the mean ( SEM ) . Stars indicate significance: *=p-value <0 . 05 . p-values were generated using t-tests , followed by the Fisher’s method and if necessary Stouffer’s z trend . DOI: http://dx . doi . org/10 . 7554/eLife . 11469 . 043 We next set out to construct a regulatory network model that incorporates the detailed regulatory information obtained for potential cross-regulation of the nine HSPC TFs obtained in the previous sections ( summarised in Figure 4a ) . We focussed on three categories of causal relationships: ( i ) one or several TFs can bind to a certain type of motif at a given regulatory region , and the probability of a motif being bound depends on the expression levels of the relevant TFs; ( ii ) TFBS mutations at a given regulatory region altered luciferase activities compared to the wild-type , thus capturing the impact of TF binding on the activity of the given regulatory region; ( iii ) individual regulatory regions show varying degrees of activation over baseline controls , which translate into different relative strengths of individual cis-regulatory regions . To incorporate this multi-layered experimental information , we constructed a three-tier dynamic Bayesian network ( DBN ) to jointly represent all those causal relationships ( see Material and Methods and Figure 4b ) . The reconstructed DBN represents a first-order time-homogeneous Markov process , which is a stochastic process where the transition functions are the same throughout all time points , and the conditional probability distribution of future states depends only on the present state ( see Material and Methods ) . The model is calculated so that the expression at t+1 is influenced by the expression at t0; analogously , the expression at t0 is influenced by the expression at t-1 , and so on . Therefore , though the model does not incorporate 'epigenetic memory' , past expression levels directly influence current expression levels . Model execution therefore permits the simulation of gene expression states in single cells over time , as well as the calculation of gene expression distributions for each gene across a population of simulated single cells . 10 . 7554/eLife . 11469 . 044Figure 4 . A three-tier dynamic Bayesian network ( DBN ) incorporating transcriptional regulatory information can recapitulate the HSPC expression state . ( a ) Representation of the complete network diagram generated using the Biotapestry software ( Longabaugh et al . , 2005 ) . ( b ) Schematic diagram describing the DBN which contains three tiers: I . TF binding motifs within regulatory regions , II . cis-regulatory regions influencing the expression levels of the various TFs , and III . genes encoding the TFs . The output of tier III , namely the expression levels of the TF , feed back into the TF binding at the various motifs of tier I . The model therefore is comprised of successive time slices ( t ) . ( c ) Simulation of a single cell over time . The expression levels of all 9 TFs are the same at the beginning ( 0 . 5 ) . The simulation rapidly stabilizes with characteristic TF expression levels . ( d ) Simulation of a cell population by running the model 1000 times . The scale of the x-axis is linear . Each simulation was run as described in ( c ) . DOI: http://dx . doi . org/10 . 7554/eLife . 11469 . 04410 . 7554/eLife . 11469 . 045Figure 4—figure supplement 1 . Simulation of a single cell over time with different expression levels at the beginning . The simulation rapidly stabilizes with characteristic TF expression levels irrespective of the starting conditions . ( a ) The expression levels of all 9 TFs are 0 . 2 at the start of the simulation . ( b ) The expression levels of all 9 TFs are 0 . 8 at the start of the simulation . ( c ) The expression levels for FLI1 , RUNX1 and TAL1 are set to be 0 . 5 at the beginning , with all other TFs not being expressed ( value of 0 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 11469 . 045 Having generated a DBN model incorporating extensive experimental information , we next investigated the expression states following model execution . First , we investigated whether the network model was compatible with the HSPC expression profile from which all the experimental data are derived , namely co-expression of all nine TFs . To this end , model execution was initiated with expression levels for all nine TFs set at the midpoint level of 0 . 5 . A representative single cell modelled over time rapidly adopts characteristic levels of expression for each of the nine genes , with some genes showing perpetual fluctuations ( Figure 4c ) . The same expression levels were reached when the model was initiated with expression starting at 0 . 2 , 0 . 8 or with initially only FLI1 , RUNX1 and TAL1 being expressed at 0 . 5 ( Figure 4—figure supplement 1 ) . We next modelled the overall distribution of the nine TFs as might be seen in a cell population by running 1000 model simulations ( Figure 4d ) . This analysis demonstrated that our model is compatible with co-expression of all nine genes within the same single cell . Moreover , stable expression over time for some genes as well as oscillations around a characteristic mean expression level for other genes suggests that our model may have captured those aspects of HSPC regulatory networks that ensure the maintenance of stem/progenitor cells . The TFs TAL1 and LYL1 are important regulators of adult haematopoiesis , but the deletion of each factor individually has only minor effects on adult HSC function ( Mikkola et al . , 2003; Hall et al . , 2003; Capron et al . , 2006 ) . Combined deletion in adult HSCs however causes a severe phenotype with rapid loss of HSPCs ( Souroullas et al . , 2009 ) . We wanted to investigate to what extent our computer model could recapitulate these known phenotypes through in silico perturbation simulations . To quantify if a change in the expression profile of a given TF was significant , we performed a Wilcoxon rank-sum test . Interestingly , this significance calculation demonstrated that both large and small fold-changes can be significant . Simulated perturbation of just LYL1 caused significant alterations to the expression profiles of Gfi1b , Tal1 , Fli1 and Gata2 , but none of these were associated with a substantial shift in mean expression levels ( Figure 5a , Figure 5—figure supplement 1 ) . Perturbation of just TAL1 caused significant changes to the expression profiles of Runx1 , Gfi1b and Gata2 , and again none of these were associated with a substantial shift in expression levels ( Figure 5b , Figure 5—figure supplement 1 ) . Simultaneous deletion of both factors caused significant changes in gene expression profiles in all TFs except for Fli1 . Unlike for the single TF perturbations , Gata2 and Runx1 showed substantial shifts in expression levels when both LYL1 and TAL1 were simulated to be knocked down ( Figure 5c , Figure 5—figure supplement 1 ) . Of note , the significance calculations highlight that there may be no one perfect way to visualize these small fold-change alterations . We therefore also generated histogram plots as an alternative visualization ( Figure 5—figure supplement 2 ) . 10 . 7554/eLife . 11469 . 046Figure 5 . The DBN recapitulates the consequences of TAL1 and LYL1 single and double perturbations as seen in vivo and in vitro . Computational prediction of gene expression patterns for the nine TFs of interest after perturbation of TAL1 ( a ) , LYL1 ( b ) or both ( c ) . Deletion of TAL1 or LYL1 on their own has no major consequences on the expression levels of the other eight TFs of the gene regulatory network , but simultaneous deletion of both TAL1 and LYL1 caused changes in expression of several genes , mainly a decrease in Gata2 and Runx1 . This major disruption of the core GRN for blood stem/progenitor cells is therefore consistent with TAL1/LYL1 double knockout HSCs showing a much more severe phenotype than the respective single knock-outs . One thousand simulations were run for each perturbation to determine the TFs expression levels in a 'cell population' by selecting expression levels at random time points after reaching its initial steady state . Expression levels of 0 resemble no expression , whereas expression levels of 1 stand for highest expression level that is possible in this system . The scale of the x-axes is linear . ( d ) Gene expression levels measured in single 416b cells transfected with siRNA constructs against Tal1 or a control . The density plots of gene expression levels after perturbation of TAL1 indicate the relative number of cells ( y-axes ) at each expression level ( x-axes ) . The scale of the x-axes is linear . The values indicate the results of the Wilcoxon rank-sum test: alterations to the expression profiles are indicated by the p-value ( statistical significance: p <0 . 001 for computational data and p <0 . 05 for experimental data ) ; substantial shifts in median expression level are indicated by the shift of median ( SOM ) ( SOM >0 . 1 for computational data and >1 for experimental data ) . For details , see Figure 5—figure supplement 1; for full expression data , see Figure 5—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 11469 . 04610 . 7554/eLife . 11469 . 047Figure 5—source data 1 . Raw and normalised data for the single cell gene expression experiments presented in this study . 1 ) TAL1 down-regulation ( related to Figure 5 d ) , 2 ) PU . 1 down-regulation ( related to Figure 6 a ) , 3 ) GFI1B up-regulation ( related to Figure 6b ) and 4 ) AML-ETO9a perturbation ( related to Figure 6 c ) DOI: http://dx . doi . org/10 . 7554/eLife . 11469 . 04710 . 7554/eLife . 11469 . 048Figure 5—figure supplement 1 . Significance tests for the computational and experimental data after TF perturbations . To determine statistical significance the Wilcoxon rank-sum test was used . Alterations to the expression profiles are indicated by the p-value; with statistically significance defined as follows: p <0 . 001 for computational data and p <0 . 05 for experimental data . Significance of a substantial shift in median expression levels are as follows: shift of median >0 . 1 for computational data and >1 for experimental data ( because of different scales ) . If the number for the shift of median is negative , the median of the perturbation data is smaller than that of the wild-type control; if the number is positive , the median of the perturbation is larger than that of the control . For simplicity , all significant changes are highlighted in red ( p-value ) and blue ( shift of median ) . DOI: http://dx . doi . org/10 . 7554/eLife . 11469 . 04810 . 7554/eLife . 11469 . 049Figure 5—figure supplement 2 . Histogram plots showing the gene expression distributions of all nine genes of the network for the perturbations presented in this study . ( a ) LYL1 down-regulation; ( b ) TAL1/SCL down-regulation; ( c ) LYL1 and TAL1/SCL down-regulation; ( d ) PU . 1 down-regulation; ( e ) GFI1B up-regulation; and ( f ) AML-ETO9a simulation . DOI: http://dx . doi . org/10 . 7554/eLife . 11469 . 049 We next wanted to compare model predictions with actual experimental data in the 416b cell line , from which the information for model construction had been derived . Because our DBN model is particularly suited to model the expression states in single cells , we compared predicted and experimentally observed effects of knockdown or overexpression in single cells . To this end , we knocked down the expression of TAL1 in 416b cells by transfecting the cells with siRNA against Tal1 ( siTal1 ) or control siRNA ( siCtrl ) . Forty-eight hours after transfection , gene expression for the nine network genes was analysed in 44 siTal1 treated cells and 41 siCtrl treated cells . Importantly , 29 of 44 cells ( 66% ) transfected with siTal1 showed no expression of Tal1 anymore , demonstrating the successful knockdown ( Figure 5d , Figure 5—source data 1 ) . Down-regulation of TAL1 caused a significant change in the expression profiles of Tal1 , Fli1 and Gfi1b , but a substantial shift of median expression was only observed for Tal1 ( Figure 5—figure supplement 1 ) . Experimental validation therefore confirmed the occurrence of statistically significant small-fold changes in expression profiles following single TF knockdown , although there was no perfect match between the genes affected in the model and experiment . To extend comparisons between model predictions and experimental validation , we investigated the consequences of knocking down the expression of PU . 1 and overexpressing GFI1B . Complete removal of PU . 1 in silico after the model had reached its initial steady state had no effect on the expression levels of the other TFs ( Figure 6a ) . To investigate whether the model prediction is comparable to experimental data obtained from single cells , single cell gene expression analysis using the Fluidigm Biomark HD platform was performed using 416b cells transduced with shRNA against PU . 1 ( shPU . 1 ) or luciferase ( shluc ) . Three days after transduction , 121 shPU . 1 and 123 shluc transduced single cells were analysed for their expression of Spi1 and the other eight TFs of the network . 18 shPU . 1-transduced cells ( 15% ) showed a complete loss of Spi1 , and expression of Spi1 in the remaining cells was markedly reduced compared to the control cells ( shluc ) ( Figure 6a , Figure 5—source data 1 ) , highlighting the efficiency of the PU . 1 knockdown . Spi1 , Runx1 , Erg and Fli1 showed a significant change in expression profiles after the depletion of PU . 1 , but this involved a substantial shift in median expression levels only for Spi1 and Runx1 ( Figure 5—figure supplement 1 ) . Expression profiles of the remaining five TFs did not change as a result of reduced PU . 1 levels ( Figure 6a , Figure 5—source data 1 ) , therefore mostly confirming the model prediction . 10 . 7554/eLife . 11469 . 050Figure 6 . The DBN captures the transcriptional consequences of network perturbations . Left panel: Computational prediction of gene expression after perturbation of specific TFs . 1000 simulations were run for each perturbation to determine expression levels in a 'cell population' ( expression at 0 resembles no expression , whereas expression of 1 represents the highest possible expression level ) . The scale of the x-axes is linear . Right panel: Density plots of gene expression levels in single 416b cells after perturbation of specific TFs indicating the relative number of cells at each expression level . The scale of the x-axes is linear . The values indicate the results of the Wilcoxon rank-sum test: alterations to the expression profiles are indicated by the p-value ( statistical significance: p<0 . 001 for computational data and p<0 . 05 for experimental data ) ; substantial shifts in median expression level are indicated by the shift of median ( SOM ) ( SOM >0 . 1 for computational data and >1 for experimental data ) . For details , see Figure 5—figure supplement 1 . ( a ) PU . 1 down-regulation: ( Left ) Computational prediction of gene expression after PU . 1 knockdown ( Spi1 was set to 0 after reaching its initial steady state ) . ( Right ) Gene expression levels measured in single 416b cells transduced with shRNA constructs against shluc ( wild-type ) or shPU . 1 ( PU . 1 knockdown ) . ( b ) GFI1B overexpression: ( Left ) Computational prediction of gene expression after overexpression of GFI1B ( Gfi1b was set to 1 after reaching its initial steady state ) . ( Right ) Gene expression levels in single 416b cells transduced with a GFI1B-expressing vector compared to an empty vector control ( wild-type ) . ( c ) Consequences of the AML-ETO9a oncogene: ( Left ) Computational prediction of gene expression patterns after introducing the dominant-negative effect of the AML-ETO9a oncogene ( Runx1 was fixed at the maximum value of 1 after reaching its initial steady state and in addition all Runt binding sites were set to have a repressive effect ) . ( Right ) Gene expression levels measured in single 416b cells transduced with an AML-ETO9a expressing vector fused to mCherry . mCherry positive cells were compared to mCherry negative cells ( wild-type ) . DOI: http://dx . doi . org/10 . 7554/eLife . 11469 . 05010 . 7554/eLife . 11469 . 051Figure 6—source data 1 . Summary of all computational simulations for perturbations of one or two TFs . The results for a total of 162 simulations are shown . The data can be accessed using the embedded hyperlinks . The y-axes show the number of cells and the x-axes the relative expression level . Blue curves represent wild-type data and red curves represent perturbation data . DOI: http://dx . doi . org/10 . 7554/eLife . 11469 . 051 Next , we modelled GFI1B overexpression in silico by increasing the expression level of Gfi1b to the maximum value after the model had reached its initial steady state which led to a significant change in the expression profiles of Gfi1b , Meis1 , Erg and Runx1 , although only Gfi1b and Meis1 showed a substantial shift in median expression levels ( Figure 6b , Figure 5—figure supplement 1 , Figure 5—source data 1 ) . Expression profiles of the other five TFs were unaltered . Single cell gene expression analysis of 90 single 416b cells transduced with a GFI1B-expressing vector and 104 single 416b cells transduced with an empty control vector showed a significant increase in the expression of Gfi1b and a significant alteration to the expression profile of Erg , but only the changes to Gfi1b involved a substantial shift in median expression levels . No significant expression changes were seen in any of the other seven network genes ( Figure 6b ) . Both PU . 1 and GFI1B perturbation studies therefore emphasize the resilience of the HSPC TF network to single TF perturbation . Moreover , our in silico model reflects this , thus suggesting that the comprehensive experimental information used to construct the network model has allowed us to capture key mechanistic aspects of HSPC regulation . Of note , there were no short-term major expression changes immediately after the perturbation in the in silico simulations for the three single TF perturbations described above . For completeness , we performed in silico modelling for all permutations of single TF knockdown / overexpression as well as all pairwise combinations of all 9 TFs analysed ( a total of 162 simulations , Figure 6—source data 1 ) . As the TF network described above is relatively stable to single TF perturbations , we set out to test whether a simulation that mimics the situation present in leukaemic cells can influence the expression states of the nine TFs in our network . The Aml-Eto9a translocation is amongst the most frequent mutations in AML ( reviewed in [Licht , 2001] ) . The resulting fusion protein is thought to act in a dominant-negative manner to repress RUNX1 target genes . To simulate the leukaemic scenario caused by AML-ETO expression , we fixed the level of Runx1 to be the maximum value 1 and at the same time converted all activating inputs of RUNX1 to inhibiting inputs in our DBN model . Interestingly , this simulation of a 'leukaemic' perturbation caused significant expression changes to all eight of the core HSPC TFs ( Figure 6c ) . To compare the AML-ETO simulation results with experimental data , we utilised a doxycycline-inducible expression system to generate 416b cells with inducible expression of AML1-ETO fused to a mCherry reporter via a self-cleaving 2A peptide spacer . Following doxycycline induction , 56 single mCherry positive and 122 single mCherry negative 416b cells were analysed by single cell gene expression . Significant gene expression changes can be seen in six of the nine core HSPC TFs ( all except Tal1 , Erg and Gata2 ) thus highlighting significant overlap between predictions and experimental validation , although there are also notable differences between model predictions and the experimental data ( see for example Gata2; Figure 6c , Figure 5—figure supplement 1 , Figure 5—source data 1 ) . These results demonstrate that our new HSPC network model can capture many gene expression changes caused by ectopic expression of a leukaemia oncogene as well as providing a useful model for normal HSPC transcriptional regulation . The inability of any model to completely recapitulate experimental data is not unexpected . Possible reasons in our case may include more complex activities of the onco-fusion protein than would be captured by our assumption that its 'only' function is as a straightforward dominant-negative effect , or the fact that the computational model is a closed system of only the 9 network TFs , whereas the experimental single cell perturbation is subject to possible knock-on consequences from gene changes outside of the 9 TF network .
Transcription factor networks are widely recognised as key determinants of cell type identity . Since the functionality of such regulatory networks is ultimately encoded in the genome , the logic that governs interactions between network components should be identifiable , and in due course allow for the construction of network models that are capable of capturing the behaviour of complex biological processes . However , the construction of such network models has so far been severely restricted because the identification and subsequent functional characterisation of mammalian regulatory sequences represent major challenges , and the connectivity and interaction rules within regulatory networks can be highly complex . Here , we report a comprehensive mammalian transcriptional network model that is fully grounded in experimental data . Model simulation coupled with subsequent experimental validation using sophisticated single cell transcriptional assays revealed the mechanistic basis for cell state stability within a haematopoietic progenitor model cell line , and also how a leukaemogenic TF fusion protein can perturb the expression of a subset of key blood stem cell regulators . Pictorial representations of putative network models are commonly shown in publications reporting ChIP-Seq TF binding datasets ( Tijssen et al . , 2011 ) . However , due to the lack of experimental underpinning , such representations are simple images that do not encode any of the underlying gene regulatory logic , and importantly therefore cannot provide executable computational models that can be used to simulate biological systems . Although the experimentally-grounded network model shows good agreement with the relative expression states of the nine TFs for the wild-type as well as the perturbation data , model predictions are not correct in all cases . Apart from the obvious caveat that any computer model is an abstraction of reality and therefore will not be correct in every detail , it also needs to be stressed that we treat the nine TFs as an isolated module for the computer simulations , and therefore could not account for possible influences by additional genes that may affect single cell gene expression measurements in the perturbation experiments . Statistical significance calculations demonstrated that both the computer model and the experimental data showed significant changes in gene expression profiles that were associated with minimal fold-change alterations to median expression levels . Such alterations to expression profiles were prevalent in both single and double gene perturbations , whereas substantial shifts in median expression were mostly restricted to the double perturbations ( and also the AML-ETO oncogene overexpression ) . This observation suggests that ( i ) our approach has the capacity to reveal the aspects of the fine-grained nature of biological networks , and ( ii ) the network presented in this study is largely resistant to perturbations of individual TFs in terms of substantial fold-change alterations in median expression levels . We believe that it may well be possible that the statistically significant small-fold changes in HSPC network genes may be responsible for the mild phenotypes seen when major HSPC regulators are deleted in adult HSPCs . Tal1-/- mice for example are not viable because TAL1 is absolutely required for embryonic blood development ( Shivdasani et al . , 1995 ) , yet deletion of TAL1 in adult HSCs only causes minor phenotypes ( Mikkola et al . , 2003 ) . Another noteworthy observation is that it would have been impossible to detect the statistically significant yet small fold-changes using conventional expression profiling , because they only become apparent following the statistical analysis of expression distributions generated by assaying lots of single cells . More generally , it is important to acknowledge that the question of how close the present model comes to capturing the underlying biological processes can only be revealed through much more exhaustive experimental validation studies . A potential caveat for network reconstruction based on identification of regulatory elements comes from the difficulties associated with capturing negative regulatory elements . As shown elegantly for CD4 and CD8 gene silencers in the lymphoid lineage , TFs involved in the early repression of a locus are not required for the maintenance of the silenced state ( Taniuchi et al . , 2002a; Taniuchi et al . , 2002b ) . Identification of negative regulatory inputs may therefore require an expansion of datasets to look across sequential developmental stages . It will therefore be important in the future to extend the work presented here to include additional HSPC regulators as well as additional stages along the haematopoietic differentiation hierarchy . Of note , TF-mediated cellular programming experiments have demonstrated that the modules of 3–4 TFs are able to confer cell-type specific transcriptional programmes ( Takahashi and Yamanaka , 2006; Graf and Enver , 2009; Batta et al . , 2014; Riddell et al . , 2014 ) , consistent with the notion that a network composed of nine key HSPC regulators is able to capture useful information about HSPC regulatory programmes . One of the most striking observations of the regulatory network defined here is the high degree to which the HSPC expression state is stabilised . As such , this model is different from previous experimentally-grounded transcriptional regulatory network models ( Peter and Davidson , 2011 ) . These earlier model organism networks have inherent forward momentum , where the model captures the progression through successive embryonic developmental stages characterised by distinct expression states . The model reported here is based on and validated with data from haematopoietic progenitor cell lines , which can differentiate ( Pinto do O et al . , 1998; Dexter et al . , 1979 ) , but can also be maintained in stable self-renewing conditions . A recent study by Busch and colleagues tracked labelled Tie2+ HSCs in the bone marrow , and showed that haematopoietic progenitors in vivo are also characterised by a substantial self-renewal capability , therefore highlighting the stable state in which they can reside for several months ( Busch et al . , 2015 ) . The observed stability of the HSPC expression state presented here is therefore likely to capture aspects of the regulatory mechanisms maintaining the steady state of primary haematopoietic progenitor cells , a notion reinforced further by the fact that our model is based on in vivo validated regulatory elements . The two types of models therefore accurately capture the properties of the distinct biological processes , e . g . driving developmental progression on the one hand , and maintaining a given cellular state on the other . Different design principles are likely to be at play , with feed-forward loops representing key building blocks of early developmental GRNs , while the network described here shows an abundance of auto-regulatory feedback loops and partially redundant enhancer elements , both of which may serve to stabilise a given cellular state . Of particular interest may be the organisation of the Runx1 gene locus , where RUNX1 protein provides positive feedback at some , and negative feedback at other HSPC enhancers . Given that these different enhancers employ overlapping yet distinct sets of upstream regulators , it is tempting to speculate that such an arrangement not only stabilises a given expression level , but also provides the means to either up- or down-regulate RUNX1 expression in response to diverse external stimuli that may act on specific RUNX1 co-factors at either the repressing or activating RUNX1 binding events . Taken together , we report widely applicable experimental and computational strategies for generating fully validated regulatory network models in complex mammalian systems . We furthermore demonstrate how such a model derived for blood stem/progenitor cells reveals mechanisms for stabilisation of the progenitor cell state , and can be utilised to analyse core network perturbations caused by leukaemic oncogenes .
The mouse myeloid progenitor 416b cell line ( Dexter et al . , 1979 ) was received from Chester Beatty lab and confirmed to be mycoplasma free . The cells were cultured in RPMI with 10% FCS and 1% Penicillin/Streptomycin . ChIP assays were performed as previously described ( Wilson et al . , 2009; Calero-Nieto et al . , 2014 ) , amplified using the Illumina TruSeq ChIP Sample Prep Kit and sequenced using the Illumina HiSeq 2500 System following the manufacturer’s instructions . Sequencing reads were mapped to the mm10 mouse reference genome using Bowtie2 ( Langmead and Salzberg , 2012 ) , converted to a density plot and displayed as UCSC genome browser custom tracks . Peaks were called using MACS2 software ( Zhang et al . , 2008 ) . Mapped reads were converted to density plots and displayed as UCSC genome browser custom tracks . The raw and processed ChIP‐Seq data have been submitted to the NCBI Gene Expression Omnibus ( www . ncbi . nlm . nih . gov/geo ) and assigned the identifier GSE69776 . A binary binding matrix was created using in-house scripts , clustered using the dice coefficient and a heatmap was plotted using gplots in R to compare newly generated ChIP-Seq data with previously published data ( Sanchez-Castillo et al . , 2015 ) . Genomic fragments spanning the candidate cis-regulatory regions were generated by PCR or ordered as gBlocks ( Life Technologies GmbH , Germany ) and cloned downstream of the LacZ gene in an hsp68LacZ ( Runx1 constructs ) or SVLacZ ( all other constructs ) reporter vector . Coordinates of candidate chromosomal regions and corresponding primer sequences are given in Figure 3—source data 2 . For Runx1 , E10 mouse transient transgenic embryos carrying LacZ enhancer-reporter constructs were generated by pronuclear injection of ( C57BL/6 x CBA ) /F2 zygotes following standard procedures . Transgenic embryos were identified by LacZ-specific PCR on genomic DNA isolated from yolk sac ( 5’-GCAGATGCACGGTTACGATG-3’; 5’-GTGGCAACATGGAAATCGCTG-3’ ) . Xgal staining and cryostat sectioning were performed as previously described ( Nottingham et al . , 2007 ) . Embryos were photographed using a Leica MZFLIII microscope , Leica DFC 300F digital camera ( Leica Microsystems , Milton Keynes , UK ) and Openlab software ( Improvision , Coventry , UK ) and sections were examined using a Nikon Eclipse E600 microscope ( Nikon , Japan ) equipped with 20x and 40x Nomarski objectives . Photographs were taken using a Nikon DXM 1200c Digital Camera ( Nikon , Tokyo , Japan ) . E11 . 5 transient transgenic embryos of all other candidate cis-regulatory regions were generated by Cyagen Biosciences Inc ( Guangzhou , China ) . Whole-mount embryos were stained with 5-bromo-4-chloro-3-indolyl-β-d-galactopyranoside ( X-Gal ) for β-galactosidase expression and photographed using a Nikon Digital Sight DS-FL1 camera attached to a Nikon SM7800 microscope ( Nikon , Kingston-upon-Thames , UK ) . Candidate transgenic mouse embryos with LacZ staining in haematopoietic tissues were subsequently embedded in paraffin , stained with 0 . 1% ( w/v ) Neutral Red and cut into 6 μm deep longitudinal sections . Images of sections were acquired with a Pixera Penguin 600CL camera attached to an Olympus BX51 microscope . All images were processed using Adobe Photoshop ( Adobe systems Europe , Uxbridge , United Kingdom ) . Wild-type and mutant DNA fragments for candidate regulatory regions were either cloned using standard recombinant DNA techniques , ordered as gBlocks ( Life Technologies ) or obtained from GeneArt by Life Technolgies . DNA fragments were cloned into pGL2 basic or pGL2 promoter vectors from Promega using restriction enzymes or by Gibson Assembly . TFBSs for the nine TFs of interest ( corresponding DNA sequences are listed in Figure 3—source data 1 ) were identified based on multiple species alignments between five species ( mouse , human , dog , platypus , opossum ) . Where a region contained multiple instances of the same motif , a single mutant construct with all relevant motifs mutated simultaneously was generated ( for generated point mutations check Figure 3a and Figure 3—figure supplements 1–18 ) . Where TF binding was observed in ChIP-Seq experiments in 416b cells , but the TFBS was not conserved , the motifs present in the mouse sequence were mutated . Stable transfections of the 416b cell line were performed using 10 μg reporter construct , 2 μg neomycin resistance plasmid and 1x107 416b cells in 180 μl culture medium per pulse . The sample was electroporated at 220 V and capacitance of 900 μF using the GenePulser Xcell Electroporation System ( Bio-Rad , United Kingdom ) . Immediately after transfection , the sample was split into four culture plates . Twenty-four hours after transfection Geneticin G418 ( Gibco by Life Technologies ) at a final concentration of 0 . 75 mg/ml was applied to the culture to select for transfected cells . The activity of the luciferase reporter constructs was measured 12–16 days after transfection by using a FLUOstar OPTIMA luminometer ( BMG LABTECH , United Kingdom ) . The luciferase activity was normalised to the cell number and presented as relative activity compared to the wild-type construct . All assays were performed at least three times in quadruplicates . The TAL1 knockdown was performed using pools of siRNA against Tal1 ( Dharmacon , United Kingdom ) which were transfected into 416b cells . Briefly , 1 x 106 cells were electroporated with either a control or Tal1 siRNA . Forty-eight hours after transfection , cells were sorted into 96-well PCR plates containing lysis buffer using the BD Influx Cell Sorter . The PU . 1 knockdown was performed as previously described ( Calero-Nieto et al . , 2014 ) . The MigR1-Gfi1b retroviral expression vector and the corresponding empty vector control ( Xu and Kee , 2007 ) were used for GFI1B overexpression . Two million 416b cells were transduced with the above listed vectors by adding viral supernatant and 4 μg/ml polybrene to the cells , followed by centrifugation at 900 x g for 90 min at 32°C and incubation with 5% CO2 at 32°C . Half of the media was then replaced with fresh culture media , and cells were incubated at 37°C with 5% CO2 . Forty-eight hours after transduction , GFP+ cells for each cell population were sorted into 96-well PCR plates containing lysis buffer using the BD Influx Cell Sorter . To induce AML1-ETO9a expression , the 416b cell line was co-transfected with: 1 ) a plasmid containing the tetracycline transcription silencer ( tTS ) , the tetracycline transactivator ( rtTA ) and blasticidine resistance under the control of the EF1α promoter; 2 ) a plasmid containing the entire Aml-Eto9a cDNA ( obtained from vector MigR1-AE9a , Addgene no . 12433 ) in frame with a F2A element and the mCherry protein under the control of a tetracycline responsive element; and 3 ) transposase PL623 ( Wang et al . , 2011 ) ( kindly donated by Pentao Liu , Sanger Institute , Cambridge ) to promote simultaneous stable integration of the two constructs described above . After 6 days of culture without selection , cells were incubated with 1 µg/ml of Doxycycline for 24 hr and then stained with DAPI . mCherry positive and negative cells that did not stain with DAPI were sorted into 96-well PCR plates containing lysis buffer using the BD Influx Cell Sorter . Single cell gene expression analysis was performed using the Fluidigm BioMark platform followed by bioinformatics analysis as previously described ( Moignard et al . , 2013 ) . All cells that express less than 48% of genes assayed were removed from the analysis for PU . 1 knockdown and GFI1B overexpression , all cells expressing less than 56% of genes assayed were removed from the TAL1 knockdown and all cells that express less than 44% of genes assayed were removed from the analysis for the AML-ETO9a induction . Importantly , this thresholding resulted in the removal of similar numbers of cells in both the perturbation and control arms of the experiments . The raw data as well as the normalised data ( normalised to Ubc and Polr2a ) of the gene expression analysis are listed in Figure 5—source data 1 ) . The first-order DBN shown in Figure 4b was established on the basis of regulatory information summarized in Figure 4a . The DBN essentially presents a discrete-time stochastic process that has the Markov property , i . e . the state of the process at the next time point depends purely on its state at the current time point . Also note that this is a time-homogeneous ( or time-invariant ) DBN , where the transition functions/matrices are the same throughout all time points . To specify parameters of the DBN , we defined a motif family at a specific regulatory region as a unique binary variable; with value '1' indicating that no motif of a motif family is bound at the specific region and value '2' indicating that at least one motif of the motif family is bound by a TF at this region . We assumed that any of the following three factors can lead to a higher probability of a motif being bound by a TF and therefore taking the value 2: ( i ) more motifs of the same type present within a regulatory region; ( ii ) multiple TFs that can bind to the same motif , such as TAL1 and LYL1 both binding to Eboxes; ( iii ) higher expression levels of the TFs . The probabilities were thus calculated based on these three sources of information ( see below for an example ) . We next defined that every regulatory region was a continuous variable on the close interval [0 , 1] , and its value was determined by the accumulated effects of all motifs present within the regulatory region . Finally , the expression levels of the nine TFs were also defined as continuous variables ranging from 0 to 0 . 8 , and their expression levels were determined by the accumulated activities of the relevant regulatory regions . Considering that variables in the top tier of the DBN are binary whereas those in the middle and bottom tiers are continuous , we found conditional linear Gaussian distribution ( Koller and Friedman , 2009 ) to be an appropriate generic representation of the intra-slice conditional probability distributions . Specifically , the regression coefficient of a regulatory region on a motif family was estimated by normalizing the logarithmic deviation of luciferase activity , where deviation refers to the change of luciferase activity between the wild-type and the mutated ( one motif family at a time ) regulatory region ( see below for a demonstration ) . Using the logarithmic deviation allowed us to account for the differences in effect sizes of various motif families by rescaling the differences to a comparable range . Similarly , for each of the nine genes , the regression coefficient of its expression level on a relevant regulatory region was estimated by normalizing the logarithmic deviation of luciferase activity , where deviation refers to the change of luciferase activity compared to the empty vector controls . All Matlab source codes are available at https://github . com/Huange and also at http://burrn-sim . stemcells . cam . ac . uk/ . Significance for the results of the luciferase reporter assays was calculated by combining the p-values of each experiment ( generated by using the t-test function in Excel ) using the Fisher’s method , followed by the calculation of Stouffer’s z trend if necessary . Significance tests for changes in TF expression levels caused by TF perturbations ( both computational and experimental ) were evaluated by Wilcoxon rank-sum tests . | Blood stem cells and blood progenitor cells replenish a person’s entire blood system throughout their life and are crucial for survival . The stem cells have the potential to become any type of blood cell – including white blood cells and red blood cells – while the progenitor cells are slightly more restricted in the types of blood cell they can become . It is important to understand how the balance of cell types is maintained because , in cancers of the blood ( also known as leukaemias ) , this organisation is lost and some cells proliferate abnormally . Almost all of a person’s cells will contain the same genetic information , but different cell types arise when different genes are switched on or off . The genes encoding proteins called transcription factors are particularly important because the proteins can control – either by activating or repressing – many other genes . Importantly , some of these genes will encode other transcription factors , meaning that these proteins essentially work together in networks . Schütte et al . have now combined extensive biochemical experiments with computational modelling to study some of the transcription factors that define blood stem cells and blood progenitor cells in mice . Firstly , nine transcription factors , which were already known to be important in blood stem cells , were thoroughly studied in mouse cells that could be grown in the laboratory . These experiments provided an overall view of which other genes these transcription factors control . Additional targeted investigations of the nine transcription factors then revealed how these proteins act in combination to activate or repress their respective activities . With this information , Schütte et al . built a computational model , which accurately reproduced how real mouse blood stem and progenitor cells behave when , for example , a transcription factor is deleted . Furthermore , the model could also predict what happens in single cells if the amounts of the transcription factors change . Lastly , Schütte et al . studied a common type of leukaemia . The model showed that the mutations that occur in this cancer change the finely tuned balance of the nine transcription factors; this may explain why leukaemia cells behave abnormally . In future these models could be extended to more transcription factors and other cell types and cancers . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"chromosomes",
"and",
"gene",
"expression",
"computational",
"and",
"systems",
"biology"
] | 2016 | An experimentally validated network of nine haematopoietic transcription factors reveals mechanisms of cell state stability |
Over 80% of multiple-tested siRNAs and shRNAs targeting CD95 or CD95 ligand ( CD95L ) induce a form of cell death characterized by simultaneous activation of multiple cell death pathways preferentially killing transformed and cancer stem cells . We now show these si/shRNAs kill cancer cells through canonical RNAi by targeting the 3’UTR of critical survival genes in a unique form of off-target effect we call DISE ( death induced by survival gene elimination ) . Drosha and Dicer-deficient cells , devoid of most miRNAs , are hypersensitive to DISE , suggesting cellular miRNAs protect cells from this form of cell death . By testing 4666 shRNAs derived from the CD95 and CD95L mRNA sequences and an unrelated control gene , Venus , we have identified many toxic sequences - most of them located in the open reading frame of CD95L . We propose that specific toxic RNAi-active sequences present in the genome can kill cancer cells .
One of the most popular methods utilized to reduce gene expression in cells is RNA interference ( RNAi ) . RNAi has been used in several studies to identify genes critical for the survival of human cancer cell lines ( Cowley et al . , 2014; Hadji et al . , 2014; Hart et al . , 2014; Morgens et al . , 2016; Wang et al . , 2015 ) . During RNAi , gene expression is inhibited by small interfering ( si ) RNAs , small hairpin ( sh ) RNAs or micro ( mi ) RNAs . miRNAs are generated as primary transcripts in the nucleus where they undergo processing to pre-miRNAs by the Drosha-DGCR8 complex before being exported to the cytosol by exportin 5 ( Ha and Kim , 2014; Krol et al . , 2010 ) . Once in the cytosol , pre-miRNAs and shRNAs are cleaved by Dicer , a type III RNase that functions in complex with TRBP , generating 21–23 nucleotide long fragments of double-stranded RNA ( dsRNA ) that have two nucleotide 3' overhangs ( Zamore et al . , 2000 ) . DsRNA fragments or chemically synthesized double-stranded siRNAs are loaded into the RNA-induced silencing complex ( RISC ) as single-stranded RNAs ( the guide RNA ) ( Siomi and Siomi , 2009 ) . A near-perfect complementarity between the guide strand of the si/miRNA and the target mRNA sequence results in cleavage of the mRNA ( Pratt and MacRae , 2009 ) . Incomplete complementarity results in inhibition of protein translation and contributes to mRNA degradation ( Guo et al . , 2010 ) . mRNA targeting is mostly determined by the seed sequence , positions 2-7/8 of the guide strand , which is fully complementary to the seed match in the 3'UTR of targeted mRNAs . Similar to miRNAs , although not fully explored , siRNAs and shRNAs also target multiple mRNAs besides the mRNAs they were designed to silence—a phenomenon commonly referred to as off-target effect ( OTE ) —that is generally sought to be avoided ( Birmingham et al . , 2006; Jackson et al . , 2006; Lin et al . , 2005 ) . The death receptor CD95 ( Fas/APO-1 ) mediates induction of apoptosis when bound by its cognate ligand CD95L , most prominently in the context of the immune system ( Krammer , 2000 ) . However , more recently , it has become apparent that the CD95/CD95L system has multiple tumor-promoting activities ( Peter et al . , 2007 ) . CD95 signaling promotes cell growth ( Chen et al . , 2010 ) , increases motility and invasiveness of cancer cells ( Barnhart et al . , 2004; Kleber et al . , 2008 ) , and promotes cancer stemness ( Ceppi et al . , 2014; Drachsler et al . , 2016; Qadir et al . , 2017 ) . In fact , we reported tumors barely grew in vivo when the CD95 gene was deleted ( Chen et al . , 2010; Hadji et al . , 2014 ) . Therefore , it appeared consistent that multiple shRNAs and siRNAs targeting either CD95 or CD95L slowed down cancer cell growth ( Chen et al . , 2010 ) and engaged a distinct form of cell death characterized by the activation of multiple cell death pathways ( Hadji et al . , 2014 ) . This unique form of cell death cannot be inhibited by conventional cell death or signaling pathway inhibitors or by knockdown of any single gene in the human genome ( Hadji et al . , 2014 ) ; it preferentially affects transformed cells ( Hadji et al . , 2014 ) including cancer stem cells ( Ceppi et al . , 2014 ) . Here , we report that loading of CD95 and CD95L-derived sequences ( si/shRNAs targeting CD95 or CD95L ) into the RISC elicits a distinct form of cell death that results from the targeting of multiple survival genes in a unique form of OTE .
More than 80% of multiple-tested shRNAs or siRNAs designed to target either CD95 or CD95L were toxic to multiple cancer cells ( Hadji et al . , 2014 ) . We have now extended this analysis to Dicer substrate 27mer DsiRNAs designed to target CD95L ( Figure 1—figure supplement 1A , [Kim et al . , 2005] ) . All five DsiRNAs displayed toxicity when introduced into HeyA8 cells at 5 nM ( Figure 1—figure supplement 1B ) reinforcing our previous observation that the majority of CD95 and CD95L targeting si/shRNAs are toxic to cancer cells . We also analyzed a data set of a genome-wide analysis of 216 cells infected with a pooled library of the TRC shRNAs ( Cowley et al . , 2014 ) . Most of the shRNAs we have tested were found to be depleted in the infected cell lines included in this study . The following shRNAs were found to be depleted in the listed percentage of the 216 cell lines tested: shL4 ( 99 . 5% ) , shL1 ( 96 . 8% ) , shR6 ( 88 . 9% ) , shR7 ( 75% ) , shR3 ( 71 . 8% ) , shL2 ( 67 . 1% ) , shR5 ( 38 . 4% ) , shL5 ( 26 . 4% ) , and shR8 ( 21 . 3% ) ( Figure 1—figure supplement 1C ) . Consistent with our data , shL1 and shR6 were found to be two of the most toxic shRNAs . Again in this independent analysis , the majority of tested shRNAs ( 67% ) targeting either CD95 or CD95L killed more than half of all tested cancer cell lines . Interestingly , a more recent RNAi screen did not report toxicity after expressing shRNAs against CD95 or CD95L ( Morgens et al . , 2016 ) . The authors of this study used a second-generation shRNA platform based on a miR-30 backbone . To determine the source of the discrepancy in the data , we generated miR-30-based Tet-inducible versions of some of our most toxic shRNAs ( shL1 , shL3 , shL4 , shR5 , shR6 , and shR7 , Figure 1—figure supplement 2A ) and found none of them to be highly toxic to HeyA8 cells ( Figure 1—figure supplement 2B ) . To determine their knockdown efficiency , we induced their expression in cells carrying sensor plasmids in which the fluorophore Venus was linked to either the CD95L or CD95 open reading frame ( ORF ) . Expression of most of these miR-30-based shRNAs also did not efficiently silence Venus expression ( Figure 1—figure supplement 2C ) . In contrast , two of our most toxic shRNAs shL3 and shR6 when expressed in the Tet-inducible pTIP vector not only killed HeyA8 cells , but also very efficiently suppressed Venus fluorescence in cells expressing the targeted Venus sensor ( Figure 1—figure supplement 2D ) . These data suggest that the levels of shRNAs produced from the miR-30-based vector may not be sufficient to be toxic to the cancer cells . Because expression levels of shRNAs are difficult to titer , we used siRNAs to determine the concentration of the toxic CD95L-derived siL3 required to kill HeyA8 cells ( Figure 1—figure supplement 2E ) . Growth was effectively blocked ( and cells died , data not shown ) when siL3 was transfected at 1 nM—a concentration well below the commonly used and recommended siRNA concentration of 5–50 nM ) —but not at 0 . 1 nM . These data suggest that this form of toxicity does not require high amounts of si- or shRNAs; however , the low expression we achieved from the miR-30 based shRNA vectors was not enough to effectively induce the toxicity . Because these miR-30-based shRNA vectors were developed to reduce off-target effects , the toxicity of CD95 and CD95L-targeting si/shRNAs described by us and others could be due to an OTE . While this was a plausible explanation , the high percentage of toxic si/shRNAs derived from CD95 and CD95L seemed to exclude a standard OTE and pointed at a survival activity of CD95 and CD95L . We therefore tested whether exogenously added recombinant CD95L protein could protect cells from the toxicity of CD95L-derived shRNAs . When NB7 cells were incubated with different concentrations of a soluble form of CD95L ( S2 ) , toxicity exerted by shL1 was not affected ( Figure 1A , left panel ) . NB7 neuroblastoma cells were chosen for these experiments because they lack expression of caspase-8 ( Teitz et al . , 2000 ) and hence are completely resistant to the apoptosis-inducing effects of CD95L . An ostensible moderate and dose-dependent protection was detected when cells were treated with a highly active leucine-zipper tagged CD95L ( LzCD95L ) ( Figure 1A , center panel ) . However , this effect is likely due to the growth-promoting activities of soluble CD95L , which also significantly affected the growth of the cells expressing a scrambled control shRNA ( seen for both S2 and LzCD95L ) . The recombinant LzCD95L protein was active , as demonstrated by its apoptosis-inducing capacity in CD95 apoptosis-sensitive MCF-7 cells ( Figure 1A , right panel ) . To test whether CD95L or CD95 proteins could protect cancer cells from death , we introduced silent mutations into the targeted sites of three very toxic shRNAs: shL1 , shL3 ( both targeting CD95L ) and shR6 ( targeting CD95 ) . We first introduced eight silent mutations into the sites targeted by either shL1 or shL3 ( Figure 1B ) and expressed these proteins in NB7 cells ( Figure 1C ) . Both mutant constructs were highly resistant to knockdown by their cognate shRNA but still sensitive to knockdown by the other targeting shRNA ( Figure 1C ) . Overexpression of these shRNA-resistant versions of the CD95L ORF did not protect the cells from shL1 or shL3 , respectively ( Figure 1D ) . Interestingly , expression of full length CD95L slowed down the growth of the NB7 cells right after infection with the lentivirus despite the absence of caspase-8 ( data not shown ) . Infection with shRNAs was therefore performed 9 days after introducing CD95L when the cells had recovered and expressed significant CD95L protein levels ( Figure 1C ) . We then mutated the CD95 mRNA in the targeted site of shR6 ( Figure 1E ) . Neither expression of wild-type ( wt ) nor mutated ( MUT ) CD95 in MCF-7 cells ( Figure 1F ) reduced the toxicity when cells were infected with the pLKO-shR6 or another toxic lentiviral shRNA , pLKO-shR7 ( Figure 1G ) . These data suggested that neither exogenously added recombinant CD95L or exogenously expressed CD95L or CD95 protein can protect cells from toxic shRNAs derived from these genes . To determine whether we could prevent cancer cells from dying by this form of cell death by deleting the endogenous targeted sites , we used CRISPR/Cas9 gene-editing to excise sites targeted by different shRNAs and siRNAs in both alleles of the CD95 and CD95L genes . We first deleted a 41 nt piece of the CD95L gene in 293T cells , that contained the target site for shL3 ( Figure 2A and C ) . While internal primers could not detect CD95L mRNA in three tested clones , primers outside of the deleted area did detect CD95L mRNA ( Figure 2D , and data not shown ) . Three clones with this shL3 Δ41 deletion were pooled and tested for toxicity by shL3 expressed from a Tet-inducible plasmid ( pTIP-shL3 ) . Compared to a pool of control cells transfected only with the Cas9 plasmid , the 293T shL3 Δ41 cells were equally sensitive to the toxic shRNA ( Figure 2G ) . This was also observed when the clones were tested individually ( data not shown ) . To exclude the possibility that shL3 was inducing cell death due to a unique activity of shL3 and/or 293T cells , we deleted the same 41 nt in CD95L in the ovarian cancer cell line HeyA8; We also generated HeyA8 clones in which we either removed a 64 nt region containing the target site for the siRNA siL3 in the CD95L coding sequence or a 227 nt region containing the target site for shR6 in CD95 ( Figure 2A and B and Figure 2—figure supplement 1 ) . In all cases , homozygous deletions were generated ( Figure 2E ) . To confirm the deletion of the shR6 target site , we infected HeyA8 cells treated with the Cas9 plasmid only and HeyA8 with a homozygous deletion of the shR6 site with shR6 and , as positive controls , with shR2 ( targeting the CD95 ORF ) and shR6' ( targeting the CD95 3'UTR ) . Five days after infection , CD95 mRNA was quantified by real time PCR using a primer located outside the 227 bp deletion ( Figure 2F ) . The mutated CD95 mRNA was still detectable in the shR6 Δ227 cells . While shR2 and shR6' ( both targeting outside the deleted region ) caused knockdown of CD95 mRNA in both the Cas9 control and the shR6 Δ227 cells , shR6 could only reduce mRNA expression in the Cas9 control cells . These data document that HeyA8 CD95 shR6 Δ227 cells no longer harbor the sequence targeted by shR6 . Now having HeyA8 cells lacking one of three RNAi-targeted sites in either CD95 or CD95L , we could test the role of the CD95 and CD95L gene products in protecting HeyA8 cells from the death induced by either shRNA ( shL3 and shR6 , two different vectors: pLKO or the Tet inducible pTIP ) or the siRNA siL3 . In all cases , the shRNA or siRNA that targeted the deleted region was still fully toxic to the target-site deleted cells ( Figure 2H and I ) . We saw efficient growth reduction and cell death in siL3 site-deleted cells transfected with as little as 1 nM siL3 ( Figure 2I , and data not shown ) . These data firmly establish that cells were not dying due to the knockdown of either CD95 or CD95L . shRNAs and early generation naked siRNAs showed general toxicity when introduced in large amounts , presumably by eliciting an interferon ( IFN ) response ( Marques and Williams , 2005 ) or by saturating the RISC ( Grimm et al . , 2006 ) . However , both chemically modified siRNAs at very low concentrations and lentiviral shRNAs at an MOI <1 were still toxic ( data not shown ) . We therefore decided to test whether the observed toxicity involved canonical RNAi and activity of the RISC . To test shRNAs or siRNAs targeting CD95L , we introduced the Venus-CD95L sensor ( inset in Figure 3A , right panel ) into HeyA8 CD95 protein k . o . cells we had generated in the process of deleting the shR6 site ( Figure 2—figure supplement 1 , clone #2 was used for the following studies; see figure legend for strategy and characterization of the clones ) . While double-stranded ( ds ) -siL3 effectively silenced Venus expression and induced toxicity , neither the sense nor the antisense single-stranded ( ss ) RNAs significantly decreased Venus expression or induced toxicity ( Figure 3A ) . In addition , no activity was found when ds-siL3 , synthesized as deoxyribo-oligonucleotides , was transfected into the cells ( Figure 3B ) . Using this type of analysis , we tested a number of modified siRNAs for RNAi activity and toxicity . For siRNAs to be fully active , they require 3' overhangs on both strands ( Bernstein et al . , 2001 ) . Converting siL3 to a blunt-end duplex resulted in substantial loss of RNAi activity and toxicity ( Figure 3C ) . Due to the topology of the RISC , siRNA activity is decreased by modification of the 5’ end of the antisense/guide strand ( Chiu and Rana , 2003 ) . To test whether cell death induced by siL3 would be affected by a bulky modification , we placed a Cy5 moiety at any of the four possible ends of the siL3 duplex . Only when the siL3 duplex carried a 5’ modification in the guide strand did it prevent RNAi activity and toxicity; modifications in the three other positions had no effect ( Figure 3C ) . This was confirmed for another siRNA , siL2 . To test whether the toxicity of siL3 required association with a macromolecular complex , which would be consistent with RISC involvement , we performed a competition experiment . HeyA8 cells were transfected with 10 nM of siL3 , and a mutated nontoxic oligonucleotide , siL3MUT , was titered in ( Figure 3D ) . siL3MUT reduced the growth inhibitory activity of siL3 in a dose-dependent fashion suggesting that siL3 and siL3MUT compete for the same binding site in the cells , pointing at involvement of the RISC . To determine involvement of RNAi pathway components in the toxicity of CD95 and CD95L-derived sequences , we tested HCT116 cells deficient for either Drosha or Dicer ( Kim et al . , 2016 ) . Growth of parental HCT116 cells was impaired after infection with shL3 or shR6 viruses ( Figure 3E , left panel ) . Consistent with the requirement of Dicer to process shRNAs , Dicer-/- cells were completely resistant to the toxic shRNAs ( Figure 3E , center panel ) . This was also supported by the inability of shR6 to silence CD95 protein expression in these cells ( Figure 3F ) . Dicer-/- cells were not resistant to toxic siRNAs as these cells died when transfected with siL3 , which is consistent with mature siRNAs not needing further processing by Dicer ( Figure 3G , center panel ) . Interestingly , Drosha-/- cells were hypersensitive to the two toxic shRNAs ( Figure 3E , right panel , p<0 . 0001 , according to a polynomial fitting model ) , and shR6 efficiently knocked down CD95 expression in Drosha-/- cells ( Figure 3F ) . Both Drosha-/- and Dicer-/- cells were much more susceptible to the toxicity induced by siL3 than parental cells ( Figure 3G , center and right panel , p<0 . 0001 , according to a polynomial fitting model ) . The hypersensitivity of the Drosha-/- cells to toxic si/shRNAs and of Dicer-/- cells to toxic siRNAs can be explained by Drosha-/- and Dicer-/- cells allowing much more efficient uptake of mature toxic RNAi-active species into the RISC because they are almost completely devoid of competing endogenous miRNAs ( Kim et al . , 2016 ) . To determine the contribution of the siRNA seed sequence to their toxicity , we generated a set of chimeric siRNAs in which we systematically replaced nucleotides of the toxic siL3 siRNA with nucleotides of a nontoxic scrambled siRNA . We did this starting either from the seed end or from the opposite end ( Figure 3H ) . HeyA8 cells expressing both the Venus-CD95L sensor ( to monitor level of knockdown ) and a Nuc-Red plasmid to fluorescently label nuclei ( to monitor the effects on cell growth ) were transfected with 5 nM of the chimeric siRNAs; total green fluorescence and the number of red fluorescent nuclei were quantified over time . The siL3 control transfected cells showed an almost complete suppression of the green fluorescence and high toxicity . In the top panel of Figure 3H , the data are summarized in which siL3 nucleotides were stepwise replaced with siScr nucleotides from the seed sequence end . Both RNAi and toxicity were profoundly reduced when three of the terminal siL3 nucleotides were replaced with the siScr nucleotides in those positions , suggesting the seed region ( 6mer highlighted in blue ) is critical for both activities . Consistently , as shown in the bottom panel of Figure 3H , when siL3 nucleotides were replaced with siScr nucleotides from the non-seed end , neither RNAi nor the toxicity was diminished until replacements affected residues in the seed region . These data suggest the 6mer seed sequence of siL3 was critical for both RNAi activity and its toxicity . A general OTE by RNAi has been described ( Birmingham et al . , 2006; Jackson et al . , 2006; Lin et al . , 2005 ) . However , this was been reported to cause toxicity in most cases , and the targeted mRNAs were difficult to predict ( Birmingham et al . , 2006 ) . The fact that 22 of the tested CD95 and CD95L-targeting sh- and si/DsiRNAs were toxic to many cancer cells evoking similar morphological and biological responses ( Hadji et al . , 2014 ) generated a conundrum: Could an OTE trigger a specific biology ? To test this , we expressed two toxic shRNAs - one targeting CD95L ( shL3 ) and one targeting CD95 ( shR6 ) - in cells lacking their respective target sequences and subjected the RNA isolated from these cells to an RNA-Seq analysis . In order to detect effects that were independent of cell type , delivery method of the shRNA , or targeted gene , we expressed shL3 in 293T ( ΔshL3 ) cells using the Tet-inducible vector pTIP and shR6 in HeyA8 ( ΔshR6 ) cells using the pLKO vector . In each case , changes in RNA abundance were compared to cells expressing a non-targeting shRNA in matching vectors . Total RNA was harvested in all cases at either the 50 hr time point ( before the onset of cell death ) or at the 100 hr time point ( during cell death ) ( Figure 4A ) . To achieve high stringency , the data were then analyzed in two ways: first , using a conventional alignment-based analysis to identify genes for which the mRNA changed more than 1 . 5-fold ( and an adjusted p-value of less than 0 . 05 ) and second , by a read-based method , in which we first identified all reads that changed >1 . 5 fold and then subjected each read to a BLAST search to identify the gene it was derived from . Only RNAs that were detected by both methods were considered ( Supplementary file 1 ) . The combination of the analyses resulted in one mRNA that was upregulated and 11 mRNAs that were downregulated ( Figure 4B ) . Using an arrayed qPCR approach , most of these detected mRNA changes were validated for both cell lines ( Figure 4—figure supplement 1A ) . Interestingly , for nine of the eleven genes , published data suggest they are either highly upregulated in cancer and/or critical for the survival of cancer cells , as their inhibition or knockdown resulted in either growth reduction or induction of various forms of cell death ( see legend of Figure 4—figure supplement 1 for details ) . Significantly , six of these eleven downregulated genes were recently identified in two independent genome-wide lethality screens to be critical for cancer cell survival ( Blomen et al . , 2015; Wang et al . , 2015 ) ( Figure 4B and Figure 4—figure supplement 1B ) ( Supplementary file 2 ) . Considering these two screens only identified 6 . 6% of human genes to be critical for cell survival , we found a significant enrichment ( 54 . 5% , p-value=3×10−6 according to binomial distribution ) of these survival genes among the genes downregulated during the cell death induced by either shL3 or shR6 . All six survival genes are either highly amplified or mutated in human cancers ( Figure 4—figure supplement 2A ) . In addition to these six genes , GNB1 and HIST1H1C were reported to be required fitness genes in a recent high-resolution CRISPR-based screen ( Hart et al . , 2015 ) . A kinetic analysis showed most of the deregulated mRNAs were downregulated early with a significant effect already at 14 hr , more than two days before the onset of cell death ( Figure 4—figure supplement 1C and data not shown ) . This suggested the cells were dying because of the silencing of multiple critical survival genes , providing an explanation for why multiple cell death pathways were activated . We therefore call this type of cell death DISE ( for Death Induced by Survival gene Elimination ) . To confirm some of the downregulated genes were also critical survival genes for HeyA8 cells , we transfected HeyA8 cells with siRNA SmartPools targeting each of the eleven genes . Individual knockdown of seven of the targeted genes resulted in reduced cell growth when compared to cells transfected with a pool of scrambled siRNAs ( Figure 4C ) . To mimic the effect of the CD95 and CD95L-derived shRNAs , we treated HeyA8 cells with a combination of siRNA pools targeting these seven genes . Remarkably , 1 nM of this siRNA mixture ( 35 . 7 pM of each individual siRNA ) was sufficient to effectively reduce growth of the cells ( Figure 4—figure supplement 2B ) and also cause substantial cell death ( Figure 4—figure supplement 2C ) , suggesting it is possible to kill cancer cells with very small amounts of siRNAs targeting a network of these survival genes . To test the generality of this phenomenon , we inducibly expressed another CD95L derived shRNA , shL1 , in 293T cells using the pTIP vector , and transfected HeyA8 cells with 25 nM siL3 . We subjected the cells to RNA-Seq analysis 100 hrs and 48 hrs after addition of Dox or after transfection , respectively . To determine whether survival genes were downregulated in all cases of sh/siRNA-induced cell death , we used a list of 1882 survival genes and 423 genes not required for survival ( nonsurvival genes ) recently identified in a CRISPR lethality screen ( Supplementary file 2 ) . We subjected the four ranked RNA-Seq data sets to a gene set enrichment analysis using the two gene sets ( Figure 4D ) . In all cases , survival genes were significantly enriched towards the top of the ranked lists ( most downregulated ) . In contrast , nonsurvival genes were not enriched . One interesting feature of DISE that emerged was the substantial loss of histones . Of the 16 genes that were significantly downregulated in cells treated with any of the four sh/siRNAs , 12 were histones ( Figure 4E ) . While it might be expected that dying cells would downregulate highly expressed genes such as histones , we believe that losing histones is a specific aspect of DISE because a detailed analysis revealed the downregulated histones were not the most highly expressed genes in these cells ( Figure 4—figure supplement 3 ) . In addition , almost as many genes with similarly high expression were found to be upregulated in cells after DISE induction . A Metascape analysis revealed genes involved in mitotic cell cycle , DNA conformation change , and macromolecular complex assembly were among the most significantly downregulated across all cells in which DISE was induced by any of the four sh/siRNAs ( Figure 4F ) . These GO clusters are consistent with DISE being a form of mitotic catastrophe with cells unable to survive cell division ( Hadji et al . , 2014 ) and suggest a general degradation of macromolecular complexes . To test whether the toxic shRNAs directly targeted genes through canonical RNAi , we subjected the two gene lists obtained from the RNA-Seq analysis ( the cell lines treated with either shL3 or shR6 at the 50 hr time point ) to a Sylamer analysis ( van Dongen et al . , 2008 ) designed to find an enrichment of miRNA/siRNA-targeted sites in the 3'UTR of a list of genes ranked according to fold downregulation ( Figure 5A ) . This analysis identified a strong enrichment of the cognate seed match for shL3 and shR6 in cells treated with either of these two shRNAs . The analyses with cells treated with shRNAs for 100 hrs looked similar but less significant , suggesting early targeting by the shRNAs followed by secondary events ( data not shown ) . Enrichment in 6mers and 8mers were both detected ( only 8mers shown ) in the 3'UTRs but not the ORF of the ranked genes ( data not shown ) . Interestingly , the seed matches detected by the Sylamer analysis were shifted by one nucleotide from the expected seed match based on the 21mer coded by the lentivirus . RNA-Seq analysis performed for the small RNA fraction confirmed in all cases ( shScr and shL3 in pTIP , and shScr and shR6 in pLKO ) , the shRNAs in the cells were cleaved in a way resulting in the predominant formation of an siRNA shifted one nucleotide away from the shRNA loop region ( black arrow heads in Figure 5—figure supplement 1A ) . This allowed us to design toxic mature siRNAs based on the sequences of shL3 and shR6 . These shRNA-to-siRNA converts were toxic to HeyA8 cells ( Figure 5—figure supplement 1B ) confirming that the observed toxicity was not limited to the TRC shRNA platform , but based on a sequence-specific activity of the si/shRNAs . The generalizability of the Sylamer results for shL3 and shR6 was tested with cells treated with either shL1 or siL3 . In both cases , when the ranked RNA Seq data were subjected to a Sylamer analysis , the seed matches of the si/shRNA introduced were again significantly enriched in the 3'UTR of downregulated RNAs ( Figure 5—figure supplement 2 ) . In none of the Sylamer analyses of the four data sets , did we see enrichment of seed matches in the 3'UTRs of downregulated RNAs that matched the passenger strand . In all cases , the only significantly enriched sequences matched the seed sequences in the guide strand of the si/shRNAs we introduced . Our data suggested that DISE inducing si/shRNAs caused an early loss of survival genes , and at the same time downregulated RNAs through canonical RNAi targeting their 3'UTR . However , it was not clear whether the most highly downregulated survival genes were targeted in their 3'UTR by RNAi-active sequences . We determined as little as six nucleotides dictated whether an siRNA killed cancer cells ( see Figure 3H ) . 10 of the 11 downregulated genes identified in the RNA-Seq analysis described in Figure 4A and B contained multiple 6mer seed matches for either shL3 and/or shR6 ( Figure 5B ) . It is therefore likely the two shRNAs , shL3 and shR6 , killed cells by targeting a network of genes enriched in critical survival genes through RNAi . The only gene without an shL3 or shR6 seed match was HIST1H1C . Interestingly , only four of the histones downregulated in cells after treatment with any of the four tested si/shRNAs had a 3'UTR ( underlined in Figure 4E ) suggesting that most histones were not directly targeted by the si/shRNAs . Using arrayed qPCR , we tested whether other toxic shRNAs targeting either CD95 or CD95L also caused downregulation of some of the 11 genes silenced by shL3 and shR6 . HeyA8 cells were transfected with the toxic siRNA siL3 ( RNA harvested at 80 hrs ) or the toxic shRNAs shL1 , shL3 or shR7 ( RNA harvested at 100 hrs ) . While shL1 did not have much of an effect on the expression of these genes , shR7 caused downregulation of 7 of 11 of the same genes targeted by shL3 even though the 6mer seed matches of the two shRNAs are very different ( CTTTGT for shL3 and GGAGGA for shR7 ) ( Figure 4—figure supplement 1D ) . To determine whether preferential targeting of survival genes was responsible for the death of the cells , we tested whether there was an association between the presence or absence of a predicted seed match in the 3'UTR for the si/shRNA introduced and whether a gene would be downregulated ( >1 . 5 fold downregulated , p<0 . 05 ) among survival genes using the Fisher's Exact test ( Figure 5C ) . In almost all cases , this analysis revealed that survival genes containing a predicted seed match in their 3'UTR were statistically more likely to be downregulated than survival genes without such a motif . The analysis with shL1 treated cells did not reach statistical significance , likely due to the fact that this shRNA was found to be very toxic and the 100 hr time point may have been too late to observe evidence of significant targeting . This interpretation is supported by the observation that the significance for both shL3 and shR6 to target survival genes was higher at 50 hrs when compared to the 100 hr time points ( Figure 5C ) and that the Sylamer analysis of the shL1 treated cells was less significant after 100 hrs of treatment than any of the other Sylamer analyses ( Figure 5—figure supplement 2 ) . Now that we had established that the toxicity of the studied shRNAs involved targeting of survival genes rather than CD95 or CD95L , we had to assume that when studying a larger set of shRNAs that the level of knockdown of the targeted genes and the toxicity were not strictly correlated . This was confirmed for the TRC shRNAs targeting the ORF or 3'UTR of CD95 in CD95 high expressing HeyA8 cells ( Figure 5—figure supplement 3 ) . While some of the toxic shRNAs efficiently silenced CD95 ( i . e . shR6 and shR2 ) few did not ( i . e . shR5 ) . In summary , our analyses suggest that cells die by DISE due to an early and selective silencing of survival genes through targeting seed matches in their 3'UTR followed by the downregulation of histones . The majority of commercially available si- , Dsi- , and shRNAs targeting either CD95 or CD95L were highly toxic to cancer cells . We therefore asked whether these two genes contained additional sequences with similar activity . To test all shRNAs derived from either CD95L or CD95 , we synthesized all possible shRNAs , 21 nucleotides long , present in the ORF or the 3'UTR of either CD95L or CD95 starting with the first 21 nucleotides after the start codon , and then shifting the sequence by one nucleotide along the entire ORF and 3'UTR ( Figure 6A ) . We also included shRNAs from a gene not expressed in mammalian cells and not expected to contain toxic sequences , Venus . All 4666 oligonucleotides ( 700 Venus , 825 CD95L ORF , 837 CD95L 3'UTR , 987 CD95 ORF , and 1317 CD95 3'UTR shRNAs ) were cloned into the Tet-inducible pTIP vector ( Figure 6B ) as five individual pools . We first tested the activity of each individual pool to be toxic and to target the Venus sensor protein ( fused to either the ORF of CD95 or CD95L ) . NB7 cells were again used because of their resistance to the Venus-CD95L sensor , which was found to be slightly toxic to CD95 apoptosis competent cells . NB7-Venus-CD95L cells infected with the Venus-targeting shRNA pool showed some reduction in fluorescence when Dox was added , however , the shRNA pool derived from the CD95L ORF was much more active in knocking down Venus ( Figure 6—figure supplement 1A ) . No significant green fluorescence reduction was detected in cells after infection with the shRNA pool derived from the CD95L 3'UTR since the targeted sequences were not part of the sensor . Similar results were obtained when NB7-Venus-CD95 cells were infected with the Venus , CD95 ORF , and CD95 3'UTR targeting shRNA pools . To determine their ability to reduce cell growth ( as a surrogate marker for toxicity ) , we infected NB7 parental cells with each of the five pools ( parental cells were used for this experiment to avoid a possible sponge effect by expressing either CD95L or CD95 sequences that were part of the Venus sensors ) . Interestingly , the pool of 700 shRNAs derived from Venus did not cause any toxicity ( Figure 6—figure supplement 1B ) . In contrast , the pool of the shRNAs derived from CD95L significantly slowed down growth , while no toxicity was observed when cells were infected with the pool of shRNAs derived from the CD95L 3'UTR . In the case of CD95 , both the shRNAs derived from the ORF and the 3'UTR showed some toxicity . However , the shRNAs derived from the 3'UTR caused greater toxicity compared to those derived from the ORF . The data suggest that overall the shRNAs derived from the CD95L ORF and the CD95 3'UTR contain the most toxic sequences . To determine the toxicity of each of the shRNAs in the pools , NB7 cells were infected with the libraries of shRNA viruses ( MOI <1 ) , and after puromycin selection cells were pooled 1:1:1 ( Venus ORF/CD95L ORF/CD95L 3'UTR pools or Venus ORF/CD95 ORF/CD95 3'UTR pools ) to allow for competition between shRNAs when Dox was added ( Figure 6B ) . Cells were cultured for 9 days with and without Dox to allow for cell death to occur . To identify depleted shRNAs , shRNA barcodes were detected through next generation sequencing of PCR products to determine the relative abundance of each shRNA in three pools: 1 ) the cloned plasmid libraries , 2 ) cells after infection and culture for 9 days without Dox , and 3 ) cells infected and cultured with Dox for 9 days . A total of 71 , 168 , 032 reads were detected containing a complete sequence of one of the cloned shRNAs . Virtually all shRNAs were substantially represented in the cloned plasmids ( Supplementary file 3 ) . The shRNAs in the CD95L pool ( comprised of the Venus , CD95L ORF , and CD95L 3’UTR subpools ) and the CD95 pool ( comprised of the Venus , CD95 ORF , and CD95 3’UTR subpools ) were ranked from highest ( most toxic ) to lowest underrepresentation . During this and subsequent analyses , we noticed in many cases , Dox addition did cause a reduction of shRNAs , indicating an increase in toxicity; however , in other instances , infection alone and without the addition of Dox was toxic . This effect was likely due to the well-described leakiness of the Tet-on system ( Pham et al . , 2008 ) , which we confirmed for shR6 in NB7 cells ( Figure 6—figure supplement 2A ) . To capture all toxic shRNAs , we therefore decided to split the analysis into two halves: 1 ) the changes in abundance after infection compared to the composition in the plasmid pool ( infection -Dox ) and 2 ) the changes in abundance after Dox addition compared to the infected –Dox cells ( infection +Dox ) . In subsequent analyses , shRNAs underrepresented after infection are either boxed ( Figure 6C ) or shown ( Figures 6D and 7B and Figure 7—figure supplement 1B ) in blue and the ones underrepresented after Dox addition are either boxed or shown in orange . The results for all shRNAs are shown in Figure 6—figure supplement 2B . Grey dots represent all shRNAs and red dots represent only the ones that were significantly underrepresented at least 5-fold . Interestingly , the highest abundance of downregulated shRNAs was found in the CD95L ORF and the CD95 3'UTR pools of shRNAs , which is consistent with the increased toxicity observed when NB7 cells were infected with either of these two pools individually ( see Figure 6—figure supplement 1B ) . The shRNAs of these two toxic pools were highly enriched in the underrepresented shRNAs in the two pooled experiments ( CD95L and CD95 ) . Their toxicity was also evident when all shRNAs in each pool ( 2362 shRNAs in the CD95L and 3004 shRNAs in the CD95 pool ) were ranked according to the highest fold downregulation ( Figure 6C ) . The three subpools in each experiment are shown separately . Thus , again this analysis identified the ORF of CD95L and the 3'UTR of CD95 as the subpool in each analysis with the highest enrichment of underrepresented shRNAs ( Figure 6C ) . This analysis allowed us to describe the toxicity landscape of CD95L and CD95 ORFs and their 3'UTRs ( Figure 6D ) . All shRNAs significantly underrepresented at least five-fold ( red dots in Figure 6—figure supplement 2B ) are shown along the CD95L pool ( Figure 6D , left ) and the CD95 pool ( Figure 6D , right ) sequences . For both CD95L and CD95 , toxic shRNAs localized into distinct clusters . The highest density of toxic sequences was found in the stretch of RNA that codes for the intracellular domain of CD95L ( underlined in green in Figure 6D ) . Our data suggest toxic shRNAs derived from either CD95L or CD95 kill cancer cells by targeting a network of genes critical for survival through canonical RNAi . Therefore , we wondered how many 8mer seed sequences derived from these toxic shRNAs would have corresponding seed matches in the 3'UTR of critical survival genes in the human genome . Would it be possible to predict with some certainty in an in silico analysis what shRNAs would be toxic to cells ? To calculate such a hypothetical toxicity index , we used the ranked CRISPR data set ( Wang et al . , 2015 ) with 1882 survival genes ( SGs ) and 423 nonSGs . Based on our RNA-Seq analyses , we hypothesized the survival genes contained more putative seed matches for toxic shRNAs in their 3'UTRs than the nonsurvival genes ( Figure 7A , left ) and that the number of seed matches in the 3'UTRs of survival genes divided by the number of seed matches in the 3'UTR of nonsurvival genes would , to some extent , predict toxicity of an si/shRNA ( Figure 7A , right ) . To establish a Toxicity Index ( TI ) for each shRNA , we first gathered 3’UTR sequences for 1846 of the survival genes and 416 of the nonsurvival genes . We then generated a list containing a normalized ratio of occurrences of every possible 8mer seed match in the 3'UTRs of the survival and non-survival gene groups . This resulted in a ratio for each of the 65 , 536 possible 8mer combinations ( Supplementary file 4 ) , the TI . We then assigned to each of the 4666 shRNAs in our screen its TI , and ranked each pool within the two experiments of our screen according to the highest TI ( red stippled lines in Figure 7B ) . We then further separated the shRNAs into two groups: those that were toxic just after infection and those toxic after addition of Dox ( Figure 7B , Supplementary file 5 ) . In each ranked list , we could now assess whether the experimentally determined toxicity of shRNAs correlated with the in silico predicted TI . Remarkably , the highest enrichment of toxic shRNAs was found amongst those with higher TI for the subpool of shRNAs targeting the CD95L ORF followed by shRNAs in the subpool targeting the CD95 3'UTR . To confirm the significance of this finding , we repeated the analysis 10 , 000 times by randomly assigning 8mers and their associated TIs to the two shRNA pools and again sorted the data from highest to lowest TI . The reported p-values were calculated based on these permutated datasets using Mann-Whitney U tests . We noticed that survival genes tend to be more highly expressed than nonsurvival genes ( data not shown ) . To address the question whether toxic si/shRNAs only target survival genes or all genes that are highly expressed , we recalculated the TI based on a set of 850 highly expressed and expression matched survival and nonsurvival genes ( Figure 7—figure supplement 1A ) . This alternative TI tracked slightly less well with the toxic shRNAs we identified , but the enrichment of toxic shRNAs towards the top of the list ranked according to the new TI was still statistically significant ( Figure 7—figure supplement 1B ) . This analysis demonstrates survival genes contain more seed matches for toxic shRNAs in their 3'UTR than nonsurvival genes regardless of the expression level . This suggests , to a certain extent , it is possible to predict the experimental toxicity of shRNAs based on the in silico calculated TI . Our data suggest DISE results from a sequence-specific off-target activity that depends on the presence of certain seed matches in the 3'UTR of survival genes . Thus , DISE inducing RISC associated small RNAs behave in manner similar to miRNAs . This raised the question whether these seed matches have special properties . While we did not find a sequence motif that was present in all toxic si/shRNAs , we did find that sequence composition , specifically GC content , which has been reported to affect the specificity of shRNAs ( Gu et al . , 2014; Ui-Tei et al . , 2004 ) , correlated with the toxicity of shRNAs . When the GC content of the 6mer seed sequences of all underrepresented shRNAs detected in the shRNA screen across the CD95L ORF was plotted , we found a significant correlation between the GC content and higher toxicity ( indicated by underrepresentation ) ( Figure 7C and D ) . This correlation was even more pronounced when plotting GC content versus the 6mer toxicity index ( Supplementary file 4 ) ( Figure 7E ) . While not an absolute requirement , higher GC content made shRNAs more toxic , consistent with reports demonstrating that shRNAs with high GC content in the seed region showed decreased on-target and increased off-target activity ( Gu et al . , 2014; Ui-Tei et al . , 2004 ) . In summary , our data suggest that si- and/or shRNAs with certain seed sequences are toxic to cancer cells by targeting critical survival genes through an RNAi mechanism independent of both Drosha and Dicer . Furthermore , the data suggest high miRNA content , presumably through competing for occupancy in the RISC , might render cells less sensitive to DISE .
There are a number of rules that have been elucidated for designing si/shRNAs ( Bramsen et al . , 2009 ) to avoid undesired effects such as OTE ( Petri and Meister , 2013 ) , general toxicity due to the presence of toxic sequence motifs ( Fedorov et al . , 2006; Petri and Meister , 2013 ) , poisoning/saturating of the RISC ( Grimm et al . , 2006 ) , or evocation of an IFN response ( Marques and Williams , 2005 ) . The following arguments and evidence support our prediction that DISE is a manifestation of a novel , functionally important , conserved mechanism of genome regulation , and not the result of one of the above-mentioned effects: Our data provide strong evidence that the toxicity observed is a sequence-specific event caused by seed matches present in the targets of the toxic si/shRNAs rather than by a toxic motif enriched in all toxic si/shRNAs ( i . e . the UGGC motif described before [Fedorov et al . , 2006] ) . We did find a correlation between the toxicity of shRNAs ( both predicted by the TI and experimentally determined in the shRNA screen ) and the GC content in their seed region . While this correlation was significant , it was not a requirement as some of the most toxic si- and shRNAs had a low 8mer seed GC content ( shL3 , 25%; shR6 , 25%; siL3 , 37 . 5% ) . Our data suggest that survival genes may contain different types of seed matches ( based on base composition or sequence ) when compared to nonsurvival genes . Such a distinction has indeed been described before ( Stark et al . , 2005 ) . In a study in Drosophila , it was determined that survival genes are depleted of seed matches targeted by highly expressed miRNAs . These authors concluded that evolution must have selected against the presence of seed matches for highly expressed miRNAs in the 3'UTR of survival genes . It is therefore not surprising that a gene ontology ( GO ) analysis of all miRNA targets ( the ‘targets’ ) in this study described these genes as being involved in development and differentiation ( Stark et al . , 2005 ) . In contrast , genes not targeted by miRNAs ( the ‘antitargets’ ) grouped in GO clusters that were consistent with cell survival ( Stark et al . , 2005 ) . A similar phenomenon was also shown in mammalian cells; genes with fewer miRNA target sites , as predicted by Targetscan , contained distinct enriched GO terms from those enriched in genes with many predicted target sites . The genes with fewer sites were enriched in GO terms like ribosomal subunits and respiratory chain , whereas target-heavy genes were more enriched in regulatory-related GO terms ( Zare et al . , 2014 ) . It is possible the DISE-inducing si/shRNAs carry seed sequences that preferentially target seed matches present in the 3’UTRs of the ‘anti-targets’ . However , as our data on the miR-30 based shRNAs suggest , DISE-inducing shRNAs must be expressed at a certain level to be toxic . Part of our data was generated using a widely used first generation stem loop shRNA platform , the TRC library . The TRC shRNAs have recently been found to be prone to cause OTE . Gu et al . showed that the loop design of this system results in imprecise Dicer cleavage and , consequently , the production of different mature small-RNA species that increase passenger loading , one major source of OTE ( Gu et al . , 2012 ) . More recently , it was reported that most guide RNAs derived from the TRC hairpin were shifted by 4 nt 3' of the expected 5' start site ( Watanabe et al . , 2016 ) . While we did see a shift in processing of these stem loop shRNAs , we did not see such a high level of imprecision in the cleavage of our toxic shRNAs . In fact , 99 . 4% of the shR6 guide RNAs started at the same nucleotide position ( Figure 5—figure supplement 1A ) . The majority of the processing of both our pTIP and pLKO-based shRNAs was shifted by one nucleotide ( Figure 5—figure supplement 1A ) . This shift was consistent with the defined seed matches that were detected in the Sylamer analyses . In general , one major seed match was detected with one other minor species ( this was less obvious for shL1 , Figure 5—figure supplement 2 ) . Furthermore , all four Sylamer analyses only detected enrichments in the 3'UTR of downregulated mRNAs that were consistent with only the guide strand targeting the mRNA and not the passenger strand . In all cases , including in cells transfected with the siRNA siL3 , the primary enriched sequence motifs were either 7 , or 8mers present in the 3'UTR of the targeted mRNAs . Our data on DISE are consistent with a number of properties of RNAi OTE that have previously been reported . Similar to DISE , OTE-mediated silencing requires a 6/7nt seed sequence of complementarity ( Birmingham et al . , 2006; Jackson et al . , 2006; Lin et al . , 2005 ) and it targets mRNAs in the 3'UTR ( Birmingham et al . , 2006 ) . Our data on shRNAs , siRNAs , and DsiRNAs suggest that DISE is not limited to one platform and requires sequence-specific targeting . This conclusion is also consistent with a previous report that suggested that sequence-dependent off-target transcript regulation is independent of the delivery method ( Jackson et al . , 2006 ) . The authors found the same enrichment of 6mers and 7mers in 3'UTRs of targeted mRNAs for siRNAs and shRNAs ( Jackson et al . , 2006 ) . We previously reported that DicerExo5-/- HCT116 cells ( with deleted Exon 5 ) were at least as sensitive to induction of DISE ( by either shL3 or shR6 ) than wt cells suggesting that Dicer deficient cells could be killed by DISE ( Hadji et al . , 2014 ) . It has been reported that these Dicer deficient cells are hypomorphs ( Ting et al . , 2008 ) and indeed , we detected low residual Dicer expression by western blotting ( Hadji et al . , 2014 ) . We have now revisited this issue with HCT116 cells rendered completely deficient for Dicer using CRISPR/Cas9 gene editing ( Kim et al . , 2016 ) . The fact that these Dicer-/- cells were now completely resistant to the toxic effects of shL3 or shR6 demonstrates the complete absence of Dicer protein and activity . Similar to the Drosha-/- cells , in the absence of mature miRNAs , which seem to attenuate DISE , Dicer-/- cells are hypersensitive to DISE induced by siRNAs . We are proposing an entirely new concept of killing cancer cells that is based on the toxicity of CD95 and CD95L derived small RNAs . Naturally , there are many open questions such as: We interpret the hypersensitivity of both Drosha-/- and Dicer-/- cells to DISE in the following way: Most of the small RNAs in the cells that are loaded into the RISC are miRNAs . Using AGO pull-down experiments , we determined 98 . 4% of AGO- associated RNAs in HCT116 cells to be miRNAs ( 99 . 3% in HeyA8 cells , data not shown ) . It was recently reported that Drosha-/- cells showed a reduction of miRNA content from roughly 70–80% to 5–6% , and Dicer-/- cells showed a reduction down to 14–21% ( Kim et al . , 2016 ) . Since neither Drosha-/- nor Dicer-/- cells express reduced AGO2 protein levels ( see inset in Figure 3E ) , it is reasonable to assume that their RISC can take up many more of the toxic DISE inducing RNAs than the RISC in wt cells explaining the super toxicity of both DISE inducing si/shRNAs and CD95L mRNAs in these cells . We previously showed expression of either shL3 and shR6 induced DISE in immortalized normal ovarian fibroblasts much more efficiently than in matching nonimmortalized cells ( Hadji et al . , 2014 ) , suggesting that this form of cell death preferentially affects transformed cells . Our data now provide an interesting model to explain the higher sensitivity of cancer cells to DISE when compared to normal cells . It is well documented that cancer cells in general have global downregulation of miRNAs when compared to normal tissues ( Lu et al . , 2005 ) . This might free up the RISC for DISE-inducing RNAs and would imply that miRNAs may protect normal cells from DISE . Overall , our data allow us to predict that any small RNA with DISE-inducing RNAi activity that does not require Dicer processing can kill cancer cells regardless of Dicer or Drosha status . In fact , in an accompanying manuscript , we demonstrate that DISE can be triggered in vivo to treat ovarian cancer in mouse xenografts by delivering CD95L-derived siRNAs using nanoparticles ( Murmann et al . , 2017 ) . No toxicity was observed in the treated mice . These data suggest that it might be possible to develop a novel form of cancer therapy based on the DISE OTE mechanism .
Primary antibodies for Western blot: anti-β-actin antibody ( Santa Cruz #sc-47778 , RRID:AB_626632 ) , anti-human CD95L ( BD Biosciences #556387 , RRID:AB_396402 ) , and anti-human CD95 ( Santa Cruz #sc-715 , RRID:AB_2100386 ) , anti-human AGO2 ( Abcam #AB186733 , RRID:AB_2713978 ) , anti-human Drosha ( Cell Signaling #3364 , RRID:AB_2238644 ) , and anti-Dicer ( Cell Signaling #3363 , RRID:AB_2093073 ) . Secondary antibodies for Western blot: Goat anti-rabbit; IgG-HRP ( Southern Biotech #SB-4030–05 , RRID:AB_2687483 and Cell Signaling #7074 , RRID:AB_2099233 ) and Goat anti-mouse; IgG1-HRP; ( Southern BioTech #1070–05 , RRID:AB_2650509 ) . Conjugated antibody isotype control for CD95 surface staining were FITC-mouse anti-human CD95 ( BD Biosciences #556640 , RRID:AB_396506 ) and FITC-mouse IgG1 , ĸ isotype control ( BD Biosciences #551954 , RRID:AB_394297 ) . Recombinant soluble S2 CD95L and leucine-zipper tagged ( Lz ) CD95L were described before ( Algeciras-Schimnich et al . , 2003 ) . Reagents used: propidium iodide ( Sigma-Aldrich #P4864 ) , puromycin ( Sigma-Aldrich #P9620 ) , G418 ( Affymetrix #11379 ) , doxycycline ( Dox ) ( Sigma-Aldrich #9891 ) , Lipofectamine 2000 ( ThermoFisher Scientific #11668027 ) , and Lipofectamine RNAiMAX ( ThermoFisher Scientific #13778150 ) . The ovarian cancer cell line HeyA8 ( RRID:CVCL_8878 ) , the neuroblastoma cell line NB7 ( RRID:CVCL_8824 ) , and the breast cancer cell line MCF-7 ( RRID:CVCL_0031 ) were grown in RPMI 1640 medium ( Cellgro #10–040 CM ) , 10% heat-inactivated FBS ( Sigma-Aldrich ) , 1% L-glutamine ( Mediatech Inc ) , and 1% penicillin/streptomycin ( Mediatech Inc ) . The human embryonic kidney cell line 293T ( RRID:CVCL_0063 ) and Phoenix AMPHO ( RRID:CVCL_H716 ) cells were cultured in DMEM ( Cellgro #10–013 CM ) , 10% heat-inactivated FBS , 1% L-Glutamine , and 1% penicillin/streptomycin . HCT116 Drosha-/- and Dicer-/- cells were generated by Narry Kim ( Kim et al . , 2016 ) . HCT116 parental ( cat#HC19023 , RRID:CVCL_0291 ) , a Drosha-/- clone ( clone #40 , cat#HC19020 ) and two Dicer-/- clones ( clone #43 , cat#HC19023 and clone #45 , cat#HC19024 ) were purchased from Korean Collection for Type Cultures ( KCTC ) . All HCT116 cells were cultured in McCoy’s medium ( ATCC , cat#30–2007 ) , 10% heat-inactivated FBS , 1% L-Glutamine , and 1% penicillin/streptomycin . All cell lines were authenticated using STR profiling and tested monthly for mycoplasm using PlasmoTest ( Invitrogen ) . All lentiviruses were generated in 293T cells using pCMV-dR8 . 9 and pMD . G packaging plasmids . Retroviruses were generated in Phoenix AMPHO cells using the VSVg packaging plasmid . NB7 cells overexpressing wild type and mutant CD95L cDNAs used in Figure 1C and D were generated by infecting cells seeded at 50 , 000 to 100 , 000 cells per well on a 6-well plate with empty pLenti , pLenti-CD95L-WT , pLenti-CD95L-L1MUT , and pLenti-CD95L-L3MUT ( described below ) with 8 μg/ml polybrene . Selection was done with 3 μg/ml puromycin for at least 48 hr . MCF-7 cells overexpressing CD95 cDNAs used in Figure 1F were generated by seeding cells at 50 , 000 per well in a 6-well plate followed by infection with pLNCX2-CD95 or pLNCX2-CD95R6MUT ( described below ) in the presence of 8 μg/ml polybrene . Selection was done with 200 μg/ml G418 48 hrs after infection for 2 weeks . The HeyA8 cells used in Figure 3D carried a lentiviral Venus-siL3 sensor vector ( Murmann et al . , 2017 ) and were infected with NucLight Red lentivirus ( Essen Bioscience #4476 ) with 8 μg/ml polybrene and selected with 3 μg/ml puromycin and sorted for high Venus expression 48 hr later . HeyA8 ΔshR6 clone #2 sensor cells used in Figure 3A–3C were infected with lentiviruses generated from the Venus-CD95L sensor vector ( described below ) to over-express the Venus-CD95L chimeric transcript . Cells were sorted for high Venus expression 48 hr later . NB7 cells over-expressing either the Venus-CD95L sensor or the Venus-CD95 sensor ( described below ) used in Figure 6—figure supplement 1A were similarly generated . The Venus-CD95L ORF and Venus-CD95 ORF ( full length ) sensor vectors were created by sub-cloning the Venus-CD95L or the Venus-CD95 inserts ( synthesized as a minigene by IDT with flanking XbaI RE site on the 5’ end and EcoRI RE site at the 3’ end in the pIDTblue vector ) , which are composed of the Venus ORF followed by either the CD95L ORF ( accession number NM_000639 . 2 ) or the CD95 ORF ( accession number BC012479 . 1 ) as an artificial 3’UTR ( both lacking the A in the start codon ) , respectively , into the modified CD510B vector ( Ceppi et al . , 2014 ) using XbaI and EcoRI . Ligation was done with T4 DNA ligase . The pLNCX2-CD95R6MUT vector was synthesized by replacing a 403 bp fragment of the CD95 ORF insert from the pLNCX2-CD95-WT vector ( Hadji et al . , 2014 ) with a corresponding 403 bp fragment that had eight silent mutation substitutions at the shR6 site ( 5’-GTGTCGCTGTAAACCAAACTT - > 5’-ATGTCGCTGCAAGCCCAATTT-3’ ) using BstXI ( NEB #R0113 ) and BamHI ( NEB #R3136 ) restriction enzymes ( mutant insert was synthesized in a pIDTblue vector with 5’ end BstXI site and 3’ end BamHI RE site ) . Dox-inducible vectors expressing shRNAs were constructed by subcloning an annealed double-stranded DNA insert containing the sequence encoding the shRNA hairpin ( sense strand: 5’-TGGCTTTATATATCTCCCTATCAGTGATAGAGATCGNNNNNNNNNNNNNNNNNNNNNCTCGAGnnnnnnnnnnnnnnnnnnnnnTTTTTGTACCGAGCTCGGATCCACTAGTCCAGTGTGGGCATGCTGCGTTGACATTGATT-3’ ) into the pTIP-shR6 vector ( Hadji et al . , 2014 ) . BsaBI ( NEB #R0537 ) and SphI-HF ( NEB #R3182 ) were used to digest both the pTIP-shR6 vector ( to excise the shR6 insert ) and the double-stranded shRNA DNA cassette insert followed by ligation with T4 DNA ligase . The template oligos were purchased from IDT . The poly-N represents the two 21 bp sequences that transcribe for the sense ( N ) and antisense ( n ) shRNA . miR-30 based shRNAs were generated by The Gene Editing and Screening Core , at Memorial Sloan Kettering , NY , by converting the 21mers expressed in the pLKO and pTIP vectors into 22mers followed by cloning into the Dox-inducible LT3REPIR vector as described ( Dow et al . , 2012 ) . A vector expressing an shRNA against Renilla luciferase was used as control ( Dow et al . , 2012 ) . We identified two gRNAs that target upstream and downstream of the site to be deleted . These gRNAs were expected to result in the deletion of a DNA piece just large enough to remove the target site . The CRISPR gRNA scaffold gene blocks were from IDT and consisted of the DNA sequence 5’-TGTACAAAAAAGCAGGCTTTAAAGGAACCAATTCAGTCGACTGGATCCGGTACCAAGGTCGGGCAGGAAGAGGGCCTATTTCCCATGATTCCTTCATATTTGCATATACGATACAAGGCTGTTAGAGAGATAATTAGAATTAATTTGACTGTAAACACAAAGATATTAGTACAAAATACGTGACGTAGAAAGTAATAATTTCTTGGGTAGTTTGCAGTTTTAAAATTATGTTTTAAAATGGACTATCATATGCTTACCGTAACTTGAAAGTATTTCGATTTCTTGGCTTTATATATCTTGTGGAAAGGACGAAACACCGNNNNNNNNNNNNNNNNNNNGTTTTAGAGCTAGAAATAGCAAGTTAAAATAAGGCTAGTCCGTTATCAACTTGAAAAAGTGGCACCGAGTCGGTGCTTTTTTTCTAGACCCAGCTTTCTTGTACAAAGTTGGCATTA-3’ ( Mali et al . , 2013 ) ; The poly-NNNNNNNNNNNNNNNNNNN represents the 19nt target sequence . The two 19nt target sequences for excision of the shL3 site ( Δ41 deletion ) were 5’-CCTTGTGATCAATGAAACT-3’ ( gRNA #1 ) and 5’-GTTGTTGCAAGATTGACCC-3’ ( gRNA #2 ) . The two target sequences for the Δ227 deletion of the shR6 site were 5’-GCACTTGGTATTCTGGGTC-3’ and 5’-TGTTTGCTCATTTAAACAC-3’ . The two target sequences for Δ64 deletion of the siL3 site were 5’-TAAAACCGTTTGCTGGGGC-3’ and 5’-TATCCCCAGATCTACTGGG-3’ . Target sequences were identified using the CRISPR gRNA algorithm found at http://crispr . mit . edu/; only gRNAs with scores over 50 were used . These six gene blocks were sub-cloned into the pSC-B-amp/kan plasmid using the StrataClone Blunt PCR Cloning kit ( Agilent Technologies #240207 ) . The target sites of siL3 , shL3 , and shR6 were homozygously deleted from target cells by co-transfecting Cas9 plasmid with each corresponding pair of pSC-B-gRNA plasmids . Briefly , 400 , 000 cells were seeded per well on a 6-well plate the day prior to transfection . Each well was transfected with 940 ng of Cas9-GFP plasmid ( pMJ920 ) ( Jinek et al . , 2013 ) and 450 ng of each pSC-B-gRNA plasmid using Lipofectamine 2000 . Media was replaced next day . One to two days later , cells were sorted for the top 50% population with the highest green fluorescence . Those cells were cultured for an additional week to let them recover . The cells were then sorted by FACS ( BD FACSAria SORP system ) directly into 96-well plates containing a 1:1 ratio of fresh media:conditioned media for single cell cloning . Approximately two to three weeks later , single cell clones were expanded and subjected to genotyping . PCR using both a primer pair that flanked the region to be deleted and another pair containing one flanking primer and one internal primer was used to screen clones for homozygous deletion . For detection of the Δ41 deletion of the shL3 site , the flanking external primers were 5’-TCTGGAATGGGAAGACACCT-3’ ( Fr primer ) and 5’- CCTCCATCATCACCAGATCC-3’ ( Rev primer ) , and the internal Rev primer was 5’-ATATACAAAGTACAGCCCAGT-3’ . For detection of the Δ227 deletion of the shR6 site , the flanking external primers were 5’-GGTGTCATGCTGTGACTGTTG-3’ ( Fr primer ) and 5’-TTTAGCTTAAGTGGCCAGCAA-3’ ( Rev primer ) , and the internal Rev primer was 5’-AAGTTGGTTTACATCTGCAC-3’ . For detection of the Δ64 deletion of the siL3 site , the flanking external primers were 5’-CTTGAGCAGTCAGCAACAGG-3’ ( Fr primer ) and 5’-CAGAGGTTGGACAGGGAAGA-3’ ( Rev primer ) , and the internal Rev primer was 5’-ATATGGGTAATTGAAGGGCTG-3’ . After screening the clones , Sanger sequencing was performed to confirm that the proper deletion had occurred . Three clones were pooled for each si/shRNA target site deletion except for HeyA8 ΔshR6 for which only clone #11 showed homozygous deletion of the shR6 site; clones #1 and 2 were not complete shR6 deletion mutants , but frame-shift mutations did occur in each allele ( as in clone #11 ) making them CD95 knockout clones as depicted in Figure 2—figure supplement 1A and B . Cells were infected with the following pLKO . 2 MISSION Lentiviral Transduction Particles ( Sigma ) : pLKO . 2-puro non-targeting ( scramble ) shRNA particles ( #SHC002V ) , eight non-overlapping shRNAs against human CD95L mRNA ( accession number #NM_000639 ) , TRCN0000058998 ( shL1: GCATCATCTTTGGAGAAGCAA ) , TRCN0000058999 ( shL2: CCCATTTAACAGGCAAGTCCA ) , TRCN0000059000 ( shL3: ACTGGGCTGTACTTTGTATAT ) , TRCN0000059001 ( shL4: GCAGTGTTCAATCTTACCAGT ) , TRCN0000059002 ( shL5: CTGTGTCTCCTTGTGATGTTT ) , TRCN0000372231 ( shL6: TGAGCTCTCTCTGGTCAATTT ) , TRCN0000372232 ( shL2’: TAGCTCCTCAACTC ACCTAAT ) , and TRCN0000372175 ( shL5’: GACTAGAGGCTTGCATAATAA ) , and nine non-overlapping shRNAs against human CD95 mRNA ( accession number NM_000043 ) , TRCN0000218492 ( shR2: CTATCATCCTCAAGGACATTA ) , TRCN00000 38695 ( shR5: GTTGCTAGATTATCGTCCAAA ) , TRCN0000038696 ( shR6: GTGCAGA TGTAAACCAAACTT ) , TRCN0000038697 ( shR7: CCTGAAACAGTGGCAATAAAT ) , TRCN0000038698 ( shR8: GCAAAGAGGAAGGATCCAGAT ) , TRCN0000265627 ( shR27’: TTTTACTGGGTACATTTTATC ) , TRCN0000255406 ( shR6': CCCTTGTGTTT GGAATTATAA ) , TRCN0000255407 ( shR7’: TTAAATTATAATGTTTGACTA ) , and TRCN0000255408 ( shR8’: ATATCTTTGAAAGTTTGTATT ) . Infection was carried out according to the manufacturer’s protocol . In brief , 50 , 000 to 100 , 000 cells seeded the day before in a 6-well plate were infected with each lentivirus at an M . O . I of 3 in the presence of 8 μg/ml polybrene overnight . Media change was done the next day , followed by selection with 3 μg/ml puromycin 24 hrs later . Selection was done for at least 48 hrs until puromycin killed the non-infected control cells . For infection of NB7 cells over-expressing pLenti-CD95L cDNAs with pLKO lentiviral particles as in Figure 1C and D , cells were seeded at 5000 per well on a 24-well plate and infected with an M . O . I . of 20 to ensure complete infection . For infection of MCF-7 cells over-expressing pLNCX2-CD95 cDNAs with pLKO lentiviruses as in Figure 1G , cells were seeded at 7000 per well on a 24-well plate and infected at an M . O . I . of three . 3 μg/ml puromycin was added 48 hrs after infection . For infection of HCT116 , Drosha-/- , and Dicer-/- cells in Figure 3E , cells were seeded at 100 , 000 per well in a 24-well plate and infected at an M . O . I of three . 3 μg/ml puromycin was added 48 hrs after infection . Cells were plated at 50 , 000 to 100 , 000 cells per well in a 6-well plate . Cells were infected with lentivirus generated in 293T cells from the desired pTIP-shRNA vector in the presence of 8 μg/ml Polybrene . Media was replaced 24 hrs later . Selection was done 48 hrs after infection with 3 μg/ml puromycin . Induction of shRNA expression was achieved by adding 100 ng/ml Dox to the media . For infection with the LT3REPIR-shRNA viruses cells were plated and infected as described above for pTIP-shRNA viruses . After selection with 3 μg/ml puromycin was complete , they were plated in 96-well plates and the shRNA expression was induced by adding Dox ( 100 ng/ml ) to the media . The cell confluency over time was measured using Incucyte . siRNAs were either purchased from Dharmacon ( Figures 2I and 4D , Figure 1—figure supplement 1A , Figure 5—figure supplement 2 ) or synthesized by IDT ( Figure 3 and Figure 5—figure supplement 1B ) as sense and antisense RNA ( or DNA for Figure 3B ) oligos and annealed . The sense RNA oligonucleotides had 3' two deoxy-T overhangs . The antisense RNA oligos were phosphorylated at the 5’ end and had 3' two deoxy-A overhangs . siRNAs targeting CD95L ( and controls ) were as follows: siRNA ( Scr , sense: UGGUUUACAUGUUGUGUGA ) , siL1 ( sense: UACCAGUGCUGAUCAUUUA ) , siL2 ( sense: CAACGUAUCUGAGCUCUCU ) , siL3 ( sense: GCCCUUCAAUUACCCAUAU ) , siL4 ( sense: GGAAAGUGGCCCAUUUAAC ) , and siL3MUT ( sense: GGACUUCAACUAGACAUCU ) . The siL3 DNA oligos ( sense: GCCCTTCAATTACCCATAT ) and Scr DNA oligos ( sense: TGGTTTACATGTTGTGTGA ) were used in Figure 3B . Blunt siL3 and siScr RNA oligos without the deoxynucleotide overhangs as well as siL2 and siL3 RNA oligos with Cy5-labelled 5’ or 3’ ends ( IDT ) were used in Figure 3C . DsiRNA used in Figure 1—figure supplement 1 were Dsi13 . X ( sense RNA oligo: CAGGACUGAG AAGAAGUAAAACCdGdT , antisense RNA oligo: ACGGUUUUACUUCUUCUCAGUCCUGUA ) , DsiL3 ( sense RNA oligo: CAGCCCUUCAAUUACCCAUAUCCdCdC , antisense RNA oligo: GGGGAUAUGGGUAAUUGAAGGGCUGCU ) , Dsi-13 . 2 ( sense RNA oligo: AUCUU ACCAGUGCUGAUCAUUUAdTdA , antisense RNA oligo: UAUAAAUGAUCAGCACUGGUAAGAUUG ) , Dsi-13 . 3 ( sense RNA oligo: AAAGUAUACUUCCGGGGUCAAUCdTdT , antisense RNA oligo: AAGAUUGACCCCGGAAGUAUACUUUGG ) , Dsi-13 . 9 ( sense RNA oligo: CUUCCGGGGUCAAUCUUGCAACAdAdC , antisense RNA oligo: GUUGUUGC AAGAUUGACCCCGGAAGUA ) , and a non-targeting DsiRNA control Dsi-NC1 ( Sense:5'-CGUUAAUCGCGUAUAAUACGCGUdAdT , antisense:5’-AUACGCGUAUUAUACGCGAUUAACGAC , IDT #51-01-14-03 ) . Predesigned siRNA SmartPools targeting the 11 downregulated genes were obtained from Dharmacon and used in Figure 4C and Figure 4—figure supplement 2B and C . Each siRNA SmartPool consisted of 4 siRNAs with On-Targetplus modification . The following SmartPools were used: L-014208–02 ( NUCKS1 ) ; L-012212–00 ( CAPZA1 ) ; L-018339–00 ( CCT3 ) ; L-013615–00 ( FSTL1 ) ; L-011548–00 ( FUBP1 ) ; L-017242–00 ( GNB1 ) ; L-014597–01 ( NAA50 ) ; L-020893–01 ( PRELID3B ) ; L-019719–02 ( SNRPE ) ; L-003941–00 ( TFRC ) ; L-006630–00 ( HIST1H1C ) . On-Targetplus non-targeting control pool ( D-001810–10 ) was used as negative control . Transfection efficiency was assessed by transfecting cells with siGLO Red ( Dharmacon ) followed by FACS analysis . HeyA8 cells ( and modified cells derived from parental HeyA8 cells ) were seeded at 750 cells per well on a 96-well plate one day before transfection . Cells were transfected using 0 . 1 μl of Lipofectamine RNAiMAX reagent per well . HCT116 cells ( and modified cells derived from parental HCT116 cells ) were seeded at 4000 cells per well on a 96-well plate one day before transfection . 0 . 2 μl of Lipofectamine RNAiMAX was used for transfection . Media was changed the day after transfection . NB7 cells were seeded at 500 cells per well in a 96-well plate . Next day , cells were infected with the scrambled pLKO lentiviruses or pLKO-shL1 lentiviruses at an M . O . I . of 20 ( to achieve 100% transduction efficiency under conditions omitting the puromycin selection step ) in the presence of 8 μg/ml polybrene and 100 ng/ml of S2 CD95L or LzCD95L for 16 hr . Media was replaced the next day with media containing varying concentrations of recombinant CD95L . Total RNA was extracted and purified using QIAZOL Lysis reagent ( QIAGEN ) and the miRNeasy kit ( QIAGEN ) . 200 ng of total RNA was used to generate cDNA using the high-capacity cDNA reverse Transcription kit ( Applied Biosystems #4368814 ) . cDNA was quantified using Taqman Gene expression master mix ( ThermoFisher Scientific #4369016 ) with specific primers from ThermoFisher Scientific for GAPDH ( Hs00266705_g1 ) , human CD95 for Figure 5—figure supplement 3 ( Hs00163653_m1 ) , human CD95 3’UTR in Figure 2F ( custom probe , Fr primer: GGCTAACCCCACTCTATGAATCAAT , Rev primer: GGCCTGCCTGTTCAGTAACT , Probe: CCTTTTGCTGAAATATC ) , human CD95L ( Hs00181226_g1 and Hs00181225_m1 ) , the shL3 target site in CD95L in Figure 2D ( custom probe , Fr primer: GGTGGCCTTGTGATCAATGAAA , Rev primer: GCAAGATTGACCCCGGAAG TATA , Probe: CTGGGCTGTACTTTGTATATT ) , and downstream of the shL3 site in Figure 2D ( custom probe , Fr primer: CCCCAGGATCTGGTGATGATG , Rev primer: ACTGCCCCCAGGTAGCT , Probe: CCCACATCTGCCCAGTAGT ) . To perform arrayed real-time PCR ( Figure 4—figure supplement 1 ) , total RNA was extracted and used to make cDNA as described for standard real-time PCR . For Taqman Low Density Array ( TLDA ) profiling , custom-designed 384-well TLDA cards ( Applied Biosystems #43422489 ) were used and processed according to the manufacturer’s instructions . Briefly , 50 µl cDNA from each sample ( 200 ng total input RNA ) was combined with 50 µl TaqMan Universal PCR Master Mix ( Applied Biosystems ) and hence a total volume of 100 µl of each sample was loaded into each of the 8 sample loading ports on the TLDA cards that were preloaded with assays from Thermofisher Scientific for human GAPDH control ( Hs99999905_m1 ) and for detection of ATP13A3 ( Hs00225950_m1 ) , CAPZA1 ( Hs00855355_g1 ) , CCT3 ( Hs00195623_m1 ) , FSTL1 ( Hs00907496_m1 ) , FUPB1 ( Hs00900762_m1 ) , GNB1 ( Hs00929799_m1 ) , HISTH1C ( Hs00271185_s1 ) , NAA50 ( Hs00363889_m1 ) , NUCKS1 ( Hs01068059_g1 ) , PRELID3B ( Hs00429845_m1 ) , SNRPE ( Hs01635040_s1 ) , and TFRC ( Hs00951083_m1 ) after the cards reached room temperature . The PCR reactions were performed using Quantstudio 7 ( ThermoFisher Scientific ) . Since each of the port loads each sample in duplicates on the TLDA card and because two biological replicates of each sample were loaded onto two separate ports , quadruplicate Ct values were obtained for each sample . Statistical analysis was performed using Student’s t test . Cells were plated at 600 , 000 per 15 mm dish ( Greiner CELLSTAR , cat#P7237 , Sigma ) after one day of puromycin selection . Total RNA was harvested at 50 hrs after plating for RNAseq analysis . Protein extracts were collected by lysing cells with RIPA lysis buffer ( 1% SDS , 1% Triton X-100 , 1% deoxycholic acid ) . Protein concentration was quantified using the DC Protein Assay kit ( Bio-Rad , Hercules , CA ) . 30 μg of protein were resolved on 8–12% SDS-PAGE gels and transferred to nitrocellulose membranes ( Protran , Whatman ) overnight at 25 mA . Membranes were incubated with blocking buffer ( 5% non-fat milk in 0 . 1% TBS/Tween-20 ) for 1 hr at room temperature . Membranes were then incubated with the primary antibody diluted in blocking buffer over night at 4°C . Membranes were washed 3 times with 0 . 1% TBS/Tween-20 . Secondary antibodies were diluted in blocking buffer and applied to membranes for 1 hr at room temperature . After 3 more additional washes , detection was performed using the ECL reagent ( Amersham Pharmacia Biotech ) and visualized with the chemiluminescence imager G:BOX Chemi XT4 ( Synoptics ) . Cell pellets of about 106 cells were resuspended in about 100 μl of PBS on ice . After resuspension , 4 μl of either anti-CD95 primary antibody ( BD #556640 ) conjugated with fluorescein isothiocyanate ( FitC ) , or the matching Isotype control ( BD #551954 ) , Mouse IgG1 κ conjugated with FitC , were added . Cells were incubated on ice at 4°C , in the dark , for 25 min , washed twice with PBS , and percent green cells were determined by flow cytometry ( Becton , Dickinson ) . A cell pellet ( 500 , 000 cells ) was resuspended in 0 . 1% sodium citrate , pH 7 . 4 , 0 . 05% Triton X-100 , and 50 μg/ml propidium iodide . After resuspension , cells were incubated 2 to 4 hrs in the dark at 4°C . The percent of subG1 nuclei ( fragmented DNA ) was determined by flow cytometry . After treatment/infection , cells were seeded at 500 to 4000 per well in a 96-well plate at least in triplicate . Images were captured at indicated time points using the IncuCyte ZOOM live cell imaging system ( Essen BioScience ) with a 10x objective lens . Percent confluence , red object count , and the green object integrated intensity were calculated using the IncuCyte ZOOM software ( version 2015A ) . The following describes the culture conditions used to produce samples for RNA-Seq in Figure 4 . HeyA8 ΔshR6 clone #11 cells were infected with pLKO-shScr or pLKO-shR6 . A pool of three 293T ΔshL3 clones was infected with either pTIP-shScr or pTIP-shL3 . After selection with puromycin for 2 days , the pTIP-infected 293T cells were plated with Dox in duplicate at 500 , 000 cells per T175 flask . The pLKO-infected HeyA8 cells were plated at 500 , 000 cells per flask . Total RNA was harvested 50 hrs and 100 hrs after plating . In addition , 293T cells were infected with either pLKO-shScr or pLKO-shL1 and RNA was isolated ( 100 hrs after plating ) as described above for the infection with shR6 . Finally , HeyA8 cells were transfected with RNAiMAX in 6-wells with siScr ( NT2 ) or siL3 oligonucleotides ( Dharmacon ) at 25 nM . The transfection mix was removed after 9 hrs and RNA harvested 48 hrs after transfection . Total RNA was isolated using the miRNeasy Mini Kit ( Qiagen , Cat . No . 74004 ) following the manufacturer’s instructions . An on-column digestion step using the RNAse-free DNAse Set ( Qiagen , Cat . No . : 79254 ) was included for all RNA-Seq samples . RNA libraries were generated and sequenced ( Genomics Core facility at the University of Chicago ) . The quality and quantity of the RNA samples were checked using an Agilent bio-analyzer . Paired end RNA-SEQ libraries were generated using Illumina TruSEQ TotalRNA kits using the Illumina provided protocol ( including a RiboZero rRNA removal step ) . Small RNA-SEQ libraries were generated using Illumina small RNA SEQ kits using the Illumina provided protocol . Two types of small RNA-SEQ sub-libraries were generated: one containing library fragments 140–150 bp in size and one containing library fragments 150–200 bp in size ( both including the sequencing adaptor of about 130 bp ) . All three types of libraries ( one RNA-SEQ and two small RNA-SEQ ) were sequenced on an Illumina HiSEQ4000 using Illumina provided reagents and protocols . Adaptor sequences were removed from sequenced reads using TrimGalore ( https://www . bioinformatics . babraham . ac . uk/projects/trim_galore ) , and the trimmed reads were mapped to the hg38 assembly of the human genome with Tophat and bowtie2 . Raw read counts were then assigned to genes using HTSeq . Differential gene expression was analyzed with the R Bioconductor DESeq2 package ( Love et al . , 2014 ) using shrinkage estimation for dispersions and fold changes to improve stability and interpretability of estimates . P values and adjusted P values were calculated using the DESeq2 package . To identify differentially abundant RNAs in cells expressing either shL3 or shR6 , using a method unbiased by genome annotation , we also analyzed the raw 100 bp reads for differential abundance . First , the second end in each paired end read was reverse complemented , so that both reads were on the same strand . Reads were then sorted and counted using the core UNIX utilities sort and uniq . Reads with fewer than 128 counts across all 16 samples were discarded . A table with all of the remaining reads was then compiled , summing counts from each sequence file corresponding to the same sample . This table contained a little over 100 , 000 reads . The R package edgeR ( http://bioinformatics . oxfordjournals . org/content/26/1/139 ) was used to identify differentially abundant reads , and then these reads were mapped to the human genome using blat ( http://genome . cshlp . org/content/12/4/656 . abstract ) to determine chromosomal location whenever possible . Homer ( http://homer . salk . edu/homer/ ) was used to annotate chromosomal locations with overlapping genomic elements ( such as genes ) . Raw read counts in each sequence file were normalized by the total number of unique reads in the file . To identify the most significant changes in expression of RNAs both methods of RNAs-Seq analyses ( alignment and read based ) were used to reach high stringency . All samples were prepared in duplicate and for each RNA the average of the two duplicates was used for further analysis . In the alignment-based analysis , only RNAs that had a base mean of >2000 reads and were significantly deregulated between the groups ( adjusted p-value<0 . 05 ) were considered for further analysis . RNAs were scored as deregulated when they were more than 1 . 5 fold changed in the shL3 expressing cells at both time points and in the shR6 expressing cells at either time points ( each compared to shScr expressing cells ) ( Supplementary file 1 ) . This was done because we found that the pLKO-driven expression of shR6 was a lot lower than the pTIP-driven expression of shL3 ( see the quantification of the two shRNAs in Figure 5—figure supplement 1A ) . This likely was a result of the reduced cellular responses in the shR6 expressing cells . In the read-based analysis , reads were only considered if they had both normalized read numbers of >10 across the samples in each treatment , as well as less than two fold variation between duplicates and >1 . 5 fold change between treatment groups at both time points and both cell lines ( Supplementary file 1 ) . After filtering , reads were mapped to the genome and associated with genes based on chromosomal localization . Finally , all RNAs were counted that showed deregulation in the same direction with both methods . This resulted in the identification of 11 RNAs that were down and 1 that was upregulated in cells exposed to the shRNAs shL3 and shR6 . To determine the number of seed matches in the 3'UTR of downregulated genes , the 3'UTRs of the 11 mRNAs were extracted from the Homo sapiens gene ( GRCh38 . p7 ) dataset of the Ensembl 86 database using the Ensembl Biomart data mining tool . For each gene , only the longest deposited 3'UTR was considered . Seed matches were counted in all 3'UTRs using in-house Perl scripts . GSEA used in Figure 4D was performed using the GSEA v2 . 2 . 4 software from the Broad Institute ( http://software . broadinstitute . org/gsea ) ; 1000 permutations were used . The Sabatini gene lists ( Supplementary file 2 ) were set as custom gene sets to determine enrichment of survival genes versus the nonsurvival control genes in downregulated genes from the RNA seq data; Adjusted p-values below 0 . 05 were considered significantly enriched . The GO enrichment analysis shown in Figure 4F was performed using all genes that after alignment and normalization were found to be at least 1 . 5 fold downregulated with an adjusted p values of < 0 . 05 , using the software available on www . Metascape . org and default running parameters . From the RNA-Seq analysis with HeyA8 ΔshR6 infected with pLKO-shR6 and 293T ΔshL3 clones infected pTIP-shL3 , we analyzed the mature double-stranded RNAs derived from pLKO-shR6 and pTIP-shL3 and found that the most abundant RNA forms were both shifted by one nucleotide . Based on these most abundant species observed after cellular processing , we converted shL3 and shR6 sequences to siRNAs . The genomic target sequence in shL3 ( 21nt ) is 5’-ACUGGGCUGUACUUUGUAUAU-3’ . For the shL3 => siL3 sense strand , one G was added before the A on the 5’ end while the last U on the 3’ end was deleted , and second and third to the last ribonucleotides on the 3’ end ( UA ) were replaced with deoxyribonucleotides for stabilization . For shL3 => siL3 antisense strand , the last three nucleotides on the 5’ end ( AUA ) were deleted and one U and two dTs ( UdTdT ) were added after the last U on the 3’end . The shL3 => siL3 sense strand is 5’- GACUGGGCUGUACUUUGUAdTdA-3’ and antisense strand is 5’-/5Phos/UACAAAGUACAGCCCAGUUdTdT-3’ . The shR6 => siRNA was designed in a similar fashion except that two Gs instead of one G were added to the 5’ end of the sense strand while UUdTdT instead of UdTdT was added to the 3’ end of the antisense strand . The genomic target sequence in shR6 ( 21nt ) is 5’-GUGCAGAUGUAAACCAAACUU-3’ . The shR6 => siR6 sense strand is 5’-GGGUGCAGAUGUAAACCAAAdCdT-3’ and antisense strand is 5’-/5Phos/UUUGGUUUACAUCUGCACUUdTdT-3’ . Both shL3 => siL3 and ShR6 => siR6 siRNA duplexes were purchased from Dharmacon . The pTIP-shRNA libraries were constructed by subcloning libraries of 143nt PCR inserts of the form 5’-XXXXXXXXXXXXXXXXXXXXXXXXXXATAGAGATCGNNNNNNNNN NNNNNNNNNNNNCTCGAGNNNNNNNNNNNNNNNNNNNNNTTTTTGTACCGAGCTCGGATCCACTAGTCCAGTGTGGGCATGCTGCGTTGACATTGATT-3’ into the pTIP-shR6 vector after excising the shR6 insert . The poly-N region represents the 21-mer sense and antisense shRNA hairpin . The intervening CTCGAG is the loop region of the shRNA . The 5 libraries targeting Venus , CD95L ORF , CD95L 3’UTR , CD95 ORF , or CD95 3’UTR were composed of every possible 21-mer shRNA ( i . e . each nearest neighbor shRNA was shifted by 1 nucleotide ) . These libraries were synthesized together on a chip as 143 bp single-stranded DNA oligos ( CustomArray Inc , Custom 12K oligo pool ) . Each shRNA pool had its own unique 5’ end represented by the poly-X region . This allowed selective amplification of a particular pool using 1 of 5 unique Fr primers ( CD95L ORF: 5’-TGGCTTTATATATCTCCCTATCAGTG-3’ , CD95L 3’ UTR: 5’-GGTCGTCCTATCTATTATTATTCACG-3’ , CD95 ORF: 5’-TCTTGTGTCCAGACCAATTTATTTCG-3’ , CD95 3’UTR: 5’-CTCATTGACTATCGTTTTAGCTACTG-3’ , Venus: 5’-TATCATCTTTCATGATGACTTTCCGG-3’ ) and the common reverse primer 5’-AATCAATGTCAACGCAGCAT-3’ . Phusion High Fidelity Polymerase ( NEB #M0530 ) was used to amplify each library pool; standard PCR conditions were used with an annealing temperature of 61°C and 15 cycles . PCR reactions were purified using PCR Cleanup kit ( QIAGEN ) . The pTIP-shR6 vector and each of the amplified libraries were digested with SphI-HF and BsaBI . Digested PCR products were run on either a 2% Agarose gel or a 20% polyacrylamide ( 29:1 ) gel made with 0 . 5 x TBE buffer . PCR products were extracted using either Gel Extraction kit ( QIAGEN ) for extraction from Agarose gels or via electro-elution using D-Tube Dialyzer Mini columns ( Novagen #71504 ) . Purified PCR inserts were then ligated to the linearized pTIP vector with T4 DNA ligase for 24 hr at 16°C . The ligation mixtures were transformed via electroporation in MegaX DH10B T1 cells ( Invitrogen #C6400 ) and plated on 24 cm ampicillin dishes . At least 10 colonies per pool were picked and sequenced to verify successful library construction . After verification , all colonies per library were pooled together and plasmid DNA extracted using the MaxiPrep kit ( QIAGEN ) . The 5 pTIP-shRNA library DNA preps were used to produce virus in 293T cells . NB7 cells were seeded at 1 . 5 × 106 per 145 cm2 dish . Two dishes were infected with each of the 5 libraries with a transduction efficiency of about 10% to 20% . Media was replaced next day . Infected cells were selected with 1 . 5 μg/ml puromycin . Cells infected with the Venus , CD95L ORF , and CD95L 3’UTR-targeting libraries were pooled in a 1:1:1 ratio to make the CD95L cell pool . Likewise , cells infected with the Venus , CD95 ORF , and CD95 3’UTR-targeting libraries were pooled to make the CD95 receptor cell pool . The CD95 and the CD95L cell pools were plated separately each in 2 sets of duplicates seeded at 600 , 000 cells per 145 cm2 dish . One set received 100 ng/ml Dox , and the other one was left untreated ( total of 4 dishes per combined pool; 2 received no treatment and 2 received Dox ) . Cells infected with the different libraries were also plated individually in triplicate with or without Dox on a 96-well plate to assess the overall toxicity of each pool . DNA was collected from each 145 cm2 dish 9 days after Dox addition . The shRNA barcodes were amplified from the harvested DNA template using NEB Phusion Polymerase with 4 different pairs of primers ( referred to as N , N + 1 , N + 2 , and N + 3 ) in separate reactions per DNA sample . The N pair consisted of the primers originally used to amplify the CD95L ORF library ( Fr: 5’-TGGCTTTATATATCTCCCTATCAGTG-3’ and Rev: 5’-AATCAATGTCAACGCAGCAT-3’ ) . The N + 1 primers had a single nucleotide extension at each 5’ end of the N primers corresponding to the pTIP vector sequence ( Fr: 5’-TTGGCTTTATATATCTCCCTATCAGTG-3’ and Rev: 5’-TAATCAATGTCAACGCAGCAT-3’ ) . The N + 2 primers had 2 nucleotide extensions ( Fr: 5’-CTTGGCTTTATATATCTCCCTATCAGTG-3’ and Rev: 5’-ATAATCAATGTCAACGCAGCAT-3’ ) , and the N + 3 primers had 3 nucleotide extensions ( Fr: 5’-TCTTGGCTTTATATATCTCCCTATCAGTG-3’ and Rev: 5’-AATAATCAATGTCAACGCAGCAT-3’ ) . The barcodes from the pTIP-shRNA library plasmid preparations were also amplified using Phusion Polymerase with the N , N + 1 , N + 2 , and N + 3 primer pairs . The shRNA barcode PCR products were purified from a 2% Agarose gel and submitted for 100 bp paired-end deep sequencing ( Genomics Core facility at the University of Chicago ) . DNA was quantitated using the Qubit . The 4 separate PCR products amplified using N , N + 1 , N + 2 , and N + 3 were combined in equimolar amounts for each sample . Libraries were generated using the Illumina TruSeq PCR-free kit using the Illumina provided protocol . The libraries were sequenced using the HiSEQ4000 with Illumina provided reagents and protocols . Raw sequence counts for DNAs were calculated by HTSeq . shRNA sequences in the PCR pieces of genomic DNA were identified by searching all reads for the sense sequence of the mature shRNA plus the loop sequence CTCGAG . To avoid a division by zero problem during the subsequent analyses all counts of zero in the raw data were replaced with 1 . A few sequences with a total read number <10 across all plasmids reads were not further considered . In the CD95L pool this was only one shRNA ( out of 2362 shRNAs ) ( L792' ) and in the CD95 20 shRNAs ( out of 3004 shRNAs ) were not represented ( R88 , R295 , R493 , R494 , R496 , R497 , R498 , R499 , R213' , R215' , R216' , R217' , R220' , R221' , R222' , R223' , R225' , R226' , R258' , R946' , R1197' , R423' ) . While most shRNAs in both pools had a unique sequence two sequences occurred 6 times ( L605' , L607' , L609' , L611' , L613' , L615' , and L604' , L606' , L608' , L610' , L612' , L614' ) . In these cases , read counts were divided by 6 . Two shRNAs could not be evaluated: 1 ) shR6 in the CD95 pool . It had a significant background due to the fact that pTIP-shR6 was used as a starting point to clone all other shRNAs . 2 ) shL3 was found to be a minor but significant contaminant during the infection of some of the samples . For each condition , two technical duplicates and two biological duplicates were available . To normalize reads to determine the change in relative representation of shRNAs between conditions , the counts of each shRNA in a subpool ( all replicates and all conditions ) was divided by the total number of shRNAs in each subpool ( % ) . First , the mean of the technical replicates ( R1 and R2 ) was taken . To analyze the biological replicates and to determine the changes between conditions , two analyses were performed: 1 ) The change in shRNA representation between the cloned plasmid library and cells infected with the library and then cultured for 9 days without Dox ( infection -Dox ) . Fold downregulation was calculated for each subpool as [ ( plasmid %/-Dox1 %+plasmid %/-Dox2 % ) /2] . 2 ) The difference in shRNA composition between the infected cells cultured with ( infection +Dox ) and without Dox . Fold downregulation was calculated for each subpool as [ ( -Dox1 %/+Dox1% ) + ( -Dox1 %/+Dox2% ) + ( -Dox2 %/+Dox1% ) + ( -Dox2 %/+Dox2% ) /4] . Only shRNAs were considered that were at least 5-fold underrepresented in either of the two analyses ( data in Supplementary file 3 ) . The TI in Figure 7A is defined by the sum of the counts of a 6mer or 8mer seed match in the 3'UTRs of critical survival genes divided by the seed match counts in the 3'UTRs of nonsurvival genes . We used the 1882 survival genes recently described in a CRISPR/Cas9 lethality screen by Wang et al . ( Wang et al . , 2015 ) . The survival genes were defined by having a CRISPR score of <−0 . 1 and an adjusted p-value of <0 . 05 . We chose as a control group to these top essential genes the bottom essential genes using inverse criteria ( CRISPR score of >0 . 1 and adjusted p-value of <0 . 05 ) and are referring to them as the ‘nonsurvival genes’ . Both counts were normalized for the numbers of genes in each gene set . 3'UTRs were retrieved as described above . For the survival genes 1846 and for the nonsurvival genes 416 3'UTRs were found . For each gene , only seed matches in the longest 3'UTR were counted . The TI was calculated for each of the 4096 possible 6mer combinations and each of the 65536 possible 8mer combinations ( Supplementary file 4 ) . These numbers were then assigned to the results of the shRNA screen ( Supplementary file 5 ) . An alternative TI was calculated in Figure 7—figure supplement 1B and is based on the top 850 most highly expressed survival genes ( all expressed >1000 average reads ) and 850 expression matched genes not described to be critical for cancer cell survival were selected as controls . For the analyses in Figure 7C and D , the GC content % was calculated for every 6mer in the CD95L ORF shRNA pool . The GC content % was then plotted against the log ( Fold down ) for each shRNA in the CD95L ORF shRNA after infection ( compared to the plasmid composition ) in Figure 7C and after addition of Dox ( compared to cells infected but not treated with Dox ) in Figure 7D . In Figure 7E , the log ( TI ) and GC content % was extracted for every possible 6mer and plotted . Pearson correlation coefficient and associated p-value were calculated in R3 . 3 . 1 . Sylamer is a tool to test for the presence of RNAi-type regulation effects from a list of differentially expressed genes , independently from small RNA measurements ( van Dongen et al . , 2008 ) ( http://www . ebi . ac . uk/research/enright/software/sylamer ) . For short stretches of RNA ( in this case length 6 , 7 , and 8 in length corresponding to the lengths of the determinants of seed region binding in RNAi-type binding events ) , Sylamer tests for all possible motifs of this length whether the motif occurrences are shifted in sequences associated with the list under consideration , typically 3'UTRs when analyzing RNAi-type binding events . A shift or enrichment of such a motif towards the down-regulated end of the gene list is consistent with upregulation of a small RNA that has the motif as the seed region . Sylamer tests in small increments along the list of genes , using a hypergeometric test on the counts of a given word , comparing the leading part of the gene list to the universe of all genes in the list . For full details , refer to ( van Dongen et al . , 2008 ) . Enriched motifs stand out from the back-ground of all motifs tested , as visible in the Sylamer plot . The plot consist of many different lines , each line representing the outcomes of a series of tests for a single word , performed along regularly spaced intervals ( increments of 200 genes ) of the gene list . Each test yields the log-transformed P-value arising from a hypergeometric test as indicated above . If the word is enriched in the leading interval , the log-transformed value has its value plotted on the positive y-axis ( sign changed ) , if the word is depleted the log-transformed value is plotted on the negative y-axis . 3'UTRs were used from Ensembl , version 76 . As required by Sylamer , they were cleaned of low-complexity sequences and repetitive fragments using respectively Dust ( Morgulis et al . , 2006 ) with default parameters and the RSAT interface ( Medina-Rivera et al . , 2015 ) to the Vmatch program , also run with default parameters . Sylamer ( version 12–342 ) was run with the Markov correction parameter set to 4 . Continuous data were summarized as means and standard deviations ( except for all IncuCyte experiments where standard errors are shown ) and dichotomous data as proportions . Continuous data were compared using t-tests for two independent groups and one-way ANOVA for 3 or more groups . For evaluation of continuous outcomes over time , two-way ANOVA was used with one factor for the treatment conditions of primary interest and a second factor for time treated as a categorical variable to allow for non-linearity . Comparisons of single proportions to hypothesized null values were evaluated using binomial tests . Statistical tests of two independent proportions were used to compare dichotomous observations across groups . The effects of treatment on wild-type versus either Dicer-/- or Drosha-/- cells were statistically assessed by fitting regression models that included linear and quadratic terms for value over time , main effects for treatment and cell type , and two- and three-way interactions for treatment , cell-type and time . The three-way interaction on the polynomial terms with treatment and cell type was evaluated for statistical significance since this represents the difference in treatment effects over the course of the experiment for the varying cell types . To test if higher TI is enriched in shRNAs that were highly downregulated , p-values were calculated based on permutated datasets using Mann-Whitney U tests . The ranking of TI was randomly shuffled 10 , 000 times and the W statistic from our dataset was compared to the distribution of the W statistic of the permuted datasets . Test of enrichment was based on the filtered data of at least 5-fold difference , which we define as a biologically meaningful . Fisher Exact Tests were performed to assess enrichment of downregulated genes ( i . e . >1 . 5 downregulated with adjusted p-value<0 . 05 ) amongst genes with at least one si/shRNA seed match . All statistical analyses were conducted in Stata 14 ( RRID:SCR_012763 ) or R 3 . 3 . 1 in Rstudio ( RRID:SCR_000432 ) . | Cells store their genetic code within molecules of DNA . Some of this information will be copied into chemically similar molecules called RNAs , from which the sequence of letters in the genetic code can be translated to build proteins . However , these messenger RNAs are not the only RNA molecules that cells can make . MicroRNAs are other short pieces of RNA that closely match sequences in parts of certain messenger RNAs . The messenger RNAs targeted by microRNAs are broken down inside the cell , which reduces how much protein can be produced from them . Since its discovery , scientists have exploited this process – called RNA interference ( or RNAi for short ) – and designed microRNA-like small interfering RNAs ( siRNAs ) to target particular messenger RNAs and decrease the levels of the corresponding proteins in countless experiments . Two proteins that have been studied in RNAi experiments are CD95 and its interaction partner CD95L . Both of these proteins are important in human cancer cells , and targeting them via RNAi killed cancer cells in an unknown mechanism that the cancer cells were unable to resist . RNAi experiments are designed to be specific , but sometimes they can accidently target other non-target messenger RNAs . Putzbach , Gao , Patel et al . have now analyzed all of the siRNAs that can be made from the messenger RNAs for CD95 and CD95L to mediate RNAi in cancer cells . This revealed that several messenger RNAs , other than those for CD95 and CD95L , were unintentionally being targeted , including many that code for proteins that cells need to survive . Further examination of the messenger RNA for CD95 and CD95L showed that they contain short sequences that are similar to those in the messenger RNAs of the genes that encode these survival proteins . Putzbach et al . were able to study and then predict which siRNA sequences would be toxic to cancer cells . These findings indicate that an RNAi off-target effect may actually be used to kill cancer cells . Future studies will determine whether this effect could be exploited to shrink tumors in animal models of cancer . If successful , this in turn could lead to new treatments for cancer patients . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"cell",
"biology",
"cancer",
"biology"
] | 2017 | Many si/shRNAs can kill cancer cells by targeting multiple survival genes through an off-target mechanism |
The PIWI-interacting RNA ( piRNA ) pathway controls transposon expression in animal germ cells , thereby ensuring genome stability over generations . In Drosophila , piRNAs are intergenerationally inherited through the maternal lineage , and this has demonstrated importance in the specification of piRNA source loci and in silencing of I- and P-elements in the germ cells of daughters . Maternally inherited Piwi protein enters somatic nuclei in early embryos prior to zygotic genome activation and persists therein for roughly half of the time required to complete embryonic development . To investigate the role of the piRNA pathway in the embryonic soma , we created a conditionally unstable Piwi protein . This enabled maternally deposited Piwi to be cleared from newly laid embryos within 30 min and well ahead of the activation of zygotic transcription . Examination of RNA and protein profiles over time , and correlation with patterns of H3K9me3 deposition , suggests a role for maternally deposited Piwi in attenuating zygotic transposon expression in somatic cells of the developing embryo . In particular , robust deposition of piRNAs targeting roo , an element whose expression is mainly restricted to embryonic development , results in the deposition of transient heterochromatic marks at active roo insertions . We hypothesize that roo , an extremely successful mobile element , may have adopted a lifestyle of expression in the embryonic soma to evade silencing in germ cells .
Transposable elements ( TEs ) are mobile genomic parasites that can change their genomic position or increase in copy number , and therefore pose a threat to genome integrity . Many TEs have evolved mechanisms that promote their activity specifically in gonads , thereby introducing new insertions that are inherited by future generations ( Kim et al . , 1994; Leblanc et al . , 2000; Wang et al . , 2018 ) . Accumulation of insertional mutations in germ cells can lead to decreased population fitness and increased risk of disease ( Hancks and Kazazian , 2016; Payer and Burns , 2019 ) . Germ cells , however , harbor protective systems that substantially decrease the likelihood of transposition events . In animal gonads , the main transposon defense mechanism is the PIWI-interacting RNA ( piRNA ) pathway ( reviewed in Czech et al . , 2018; Ozata et al . , 2019 ) . At its core , this system depends on 23- to 30-nt piRNAs to distinguish transposon-derived RNAs from host-encoded transcripts and to direct their associated PIWI proteins to active TE targets . In Drosophila , PIWI-guided repression involves cytoplasmic post-transcriptional mRNA cleavage by Aubergine ( Aub ) and Argonaute-3 ( Ago3 ) and nuclear P-element-induced wimpy testis ( Piwi ) that engages nascent transposon transcripts and instructs co-transcriptional gene silencing ( coTGS ) through heterochromatin formation . coTGS requires additional factors acting downstream of Piwi , including Panoramix ( Panx ) , Nuclear Export Factor 2 ( Nxf2 ) , NTF2-related export protein 1 ( Nxt1 ) , and Cut-up ( Ctp ) , that together form the PICTS complex ( also known as SFiNX ) ( Batki et al . , 2019; Eastwood et al . , 2021; Fabry et al . , 2019; Murano et al . , 2019; Schnabl et al . , 2021; Sienski et al . , 2015; Yu et al . , 2015; Zhao et al . , 2019 ) . PICTS interfaces with general chromatin silencing factors including Su ( var ) 205/HP1a , SETDB1/Eggless ( Egg ) , and its co-factor Windei ( Wde ) , Su ( var ) 3–3/Lsd1 , and its co-factor coRest , Mi-2 , Rpd3 , Ovaries absent , and Su ( var ) 2–10 ( Czech et al . , 2013; Handler et al . , 2013; Koch et al . , 2009; Muerdter et al . , 2013; Mugat et al . , 2020; Ninova et al . , 2020; Osumi et al . , 2019; Rangan et al . , 2011; Sienski et al . , 2015; Yang et al . , 2019; Yu et al . , 2015 ) . While the detailed mechanisms of transcriptional silencing remain to be established , loci targeted by Piwi are decorated in repressive chromatin marks including trimethylated H3K9 ( H3K9me3 ) ( Klenov et al . , 2014; Le Thomas et al . , 2013; Rozhkov et al . , 2013; Sienski et al . , 2012; Wang and Elgin , 2011 ) . Loss of Piwi in Drosophila ovaries results in de-repression of TEs and correlates with a severe reduction in H3K9me3 deposition at their corresponding genomic loci . Piwi and Aub , and to a lesser degree Ago3 , have been detected as maternally deposited proteins in Drosophila embryos ( Brennecke et al . , 2007; Brennecke et al . , 2008; Gunawardane et al . , 2007; Mani et al . , 2014; Megosh et al . , 2006; Rouget et al . , 2010 ) . Considering that pluripotent progenitor cells give rise to multiple cell lineages , including the germline , maintaining genome integrity during the early stages of embryogenesis is potentially critical . Consistent with their adult roles , maternally inherited PIWI proteins have been observed in the pole plasm of syncytial embryos and in pole cells , the germ cell progenitors , after cellularization ( Brennecke et al . , 2008; Dufourt et al . , 2017; Mani et al . , 2014; Megosh et al . , 2006 ) . Though , in adult flies , the piRNA pathway is restricted to the gonad , during the early phases of embryogenesis Piwi is also present in somatic nuclei ( Brennecke et al . , 2008; Mani et al . , 2014; Megosh et al . , 2006 ) . This has long been taken as an indication that the piRNA pathway could play roles also in the developing soma , for example , helping to establish its epigenetic landscape ( Gu and Elgin , 2013; Seller et al . , 2019; Yuan and O'Farrell , 2016 ) . However , probing piRNA pathway function during early embryogenesis has been hampered by a lack of suitable experimental approaches . Disrupting Piwi or other piRNA pathway factors in the female parent either via mutation or RNAi leads to oogenesis defects and often results in sterility or patterning defects that would confound the outcome of analyses ( Cox et al . , 1998; Czech et al . , 2013; Handler et al . , 2013; Khurana et al . , 2010; Klattenhoff et al . , 2007; Klenov et al . , 2011; Li et al . , 2009a; Malone et al . , 2009; Mani et al . , 2014; Muerdter et al . , 2013; Pane et al . , 2007; Park et al . , 2019 ) . RNAi-mediated depletion in embryos or generation of homozygous mutant embryos carrying piRNA pathway defects enables analysis of later developmental stages ( Akkouche et al . , 2017; Gu and Elgin , 2013; Marie et al . , 2017 ) , but not time windows where maternally deposited proteins predominate and generally drive development . Here , we exploit a conditional protein degradation strategy to explore the function of maternally deposited piRNAs during Drosophila embryonic development . We find that Piwi-piRNA complexes present in the embryo are primarily derived from the oocyte , whereas components of the PICTS complex are both maternally deposited and zygotically expressed . An embryonic burst of transposon expression in somatic cells as the zygotic genome becomes active precedes the transient decoration of normally active elements in repressive chromatin marks . Rapid and efficient degradation of maternally deposited Piwi protein in embryos leads to earlier and increased activity of zygotically expressed TEs in concert with loss of repressive marks during the affected developmental stages . Although loss of transposon control in the embryonic soma does not result in an overt morphological phenotype , our results suggest that the piRNA pathway indeed plays a role in regulating the somatic chromatin structure during early embryogenesis . Through these mechanisms , a wave of expression , primarily of the roo transposon , is attenuated , though substantial expression of the TE remains .
The maternal deposition of Piwi , Aub , and Ago3 , noted more than a decade ago ( Brennecke et al . , 2007; Brennecke et al . , 2008; Gunawardane et al . , 2007; Mani et al . , 2014; Megosh et al . , 2006; Rouget et al . , 2010 ) , has long suggested a possible role for the piRNA pathway during embryogenesis . Prior studies have indicated that maternal instructions transmitted via piRNAs are important for defining piRNA clusters in the subsequent generation and/or provide critical information for gaining control over at least some transposons in daughters ( Akkouche et al . , 2013; Akkouche et al . , 2017; Brennecke et al . , 2008; de Vanssay et al . , 2012; Hermant et al . , 2015; Khurana et al . , 2011; Le Thomas et al . , 2014a; Le Thomas et al . , 2014b ) . Both of these functions are relevant in gonadal cells . Yet , prior studies highlighted the presence of maternally deposited Piwi protein in the somatic nuclei of developing embryos ( Brennecke et al . , 2008; Mani et al . , 2014; Megosh et al . , 2006 ) , leading to suggestions that piRNAs might help set the global epigenetic landscape of the embryonic soma ( Gu and Elgin , 2013 ) . To investigate the role of the piRNA pathway during embryogenesis , we first focused on its most well-established role , that of transposon control . Toward this end , we first characterized the expression of transposons throughout Drosophila embryogenesis by RNA-seq and quantitative mass spectrometry ( Figure 1A ) . Transcriptomes of 0–2 hr after egg laying ( AEL ) embryos represent the maternally inherited mRNA pool . Maternal transcripts are cleared and the zygotic genome is activated ( zygotic genome activation [ZGA] ) around nuclear cycle 14 ( NC14; 2–2 . 5 hr AEL ) , and we generated RNA-seq data spanning 1 hr intervals of development from this point up to 10 hr AEL ( stage 13 ) . For comparison , we also included two late-stage embryo time points ( 12–13 hr and 17–18 hr AEL ) as these were times when our prior data indicated that maternal Piwi was no longer detectable in somatic nuclei ( Brennecke et al . , 2008 ) . To take into account different library sizes and facilitate comparability throughout our time-course experiment that only contained two biological replicates per time point , our RNA-seq data was normalized to reads per million ( rpm ) . We benchmarked our dataset by comparing the expression of selected embryonic genes in our RNA-seq to reported transcriptomes in FlyBase ( Graveley et al . , 2011 ) . We found highly similar expression patterns of genes that are dynamically regulated during embryogenesis ( Figure 1—figure supplement 1A , B ) . Furthermore , well-validated maternal ( e . g . , fs ( 1 ) N and gammaTub37C ) and zygotic ( e . g . , Ultrabithorax [Ubx] and wingless [wg] ) genes demonstrated their expected expression patterns in our datasets ( Figure 1—figure supplement 2A ) . We detected only very few transposon transcripts in pre-ZGA embryos ( 0–2 hr AEL ) , as might be expected from their effective silencing by the piRNA pathway in ovaries . TE expression steadily increased following ZGA and peaked between 4 and 6 hr AEL ( Figure 1B ) , similar to what was noted in prior reports ( Batut et al . , 2013 ) . At the peak , transposon RNAs correspond to ~1 . 7% of the total embryonic transcriptome , with levels at the later studied time points decreasing to below 1% of the overall transcriptome . Transposons often show highly dynamic spatio-temporal expression; thus , we analyzed the contribution of individual TE families to the embryonic transcriptome . Interestingly , the majority of transposon expression could be attributed to one single transposon family , roo ( Figure 1C ) . At its peak at 4–6 hr AEL , reads derived from the roo TE accounted for more than 1% of the entire embryonic transcriptome , corresponding to more than 70% of all TE-derived reads overall . From its expression peak , roo mRNA levels declined strongly before leveling off at around 12 hr AEL . While less pronounced , other transposons , such as copia and 297 , also showed dynamic expression changes during embryogenesis ( Figure 1—figure supplement 2F ) . The roo expression peak at 4–6 hr AEL could be due to transcription from germ cell precursors , which become transcriptionally active around 3 . 5 hr AEL ( stage 8 ) ( Van Doren et al . , 1998; Zalokar , 1976 ) . However , the sheer abundance of roo and other transposon transcripts argued strongly that they must emanate at least in part from somatic nuclei as these vastly outnumber the germ cell precursors . To directly test the origin of roo transcripts during embryogenesis , we performed RNA fluorescence in situ hybridization ( RNA-FISH ) . In agreement with our RNA-seq data , roo transcripts were detected as early as stage 6 ( in gastrulating embryos ~3 hr AEL ) and localized predominantly to yolk cell nuclei ( Figure 1D , Figure 1—figure supplement 2B ) . Stage 11 embryos ( ~5 hr AEL ) showed strong roo RNA signal in somatic cells of the mesoderm , similar to earlier reports ( Brönner et al . , 1995; Ding and Lipshitz , 1994 ) . In contrast , roo transcript levels were undetectable by FISH in late-stage embryos ( >10 hr AEL ) . These data indicate a transient somatic burst of roo expression during early Drosophila development . TEs rely on proteins encoded in their open reading frames ( ORFs ) for mobilization . roo is an LTR retrotransposon and , as has been proposed for gypsy in ovarian follicle cells ( Kim et al . , 1994; Leblanc et al . , 2000; Song et al . , 1997 ) , could potentially be packaged into virion-like particles , possibly enabling infection of germ cell precursors as a propagation mechanism . To determine whether roo-encoded proteins are expressed in embryos , we mined quantitative proteomic data from three developmental intervals ( Figure 1A ) . The first , 0–2 hr AEL , represents the time before ZGA when the proteome is derived from maternal protein deposition and zygotic translation of maternal mRNAs . The second , 5–7 hr AEL , represents an interval where zygotic roo expression had become robust , and the third , 10–12 hr AEL , is a time at which roo RNA levels had substantially declined . In transcriptionally silent embryos ( 0–2 hr AEL ) , we detected over 6400 unique proteins . Compared to 0–2 hr embryos , the signal intensity of ~17% or 1114 proteins significantly increased ( p<0 . 01 ) by over 25% in 5–7 hr AEL embryos ( Figure 1—figure supplement 2C ) . We also detected 490 ( or ~8% of ) proteins that significantly decreased ( p<0 . 01 ) by over 25% in 5–7 hr AEL embryos ( Figure 1—figure supplement 2D ) . The majority of proteins ( 4652 or 72% ) , however , did not change by more than 25% between 0–2 hr and 5–7 hr AEL embryos . As with transcriptome analyses , known maternally deposited and zygotically expressed proteins showed their expected patterns of presence in the datasets . Compared to the early time point ( 0–2 hr AEL ) , 5–7 hr AEL embryos showed significant accumulation of roo peptides ( p<0 . 01 ) corresponding to its expression peak . roo encodes a single ORF ( with a predicted protein weight of 272 kDa ) , which contains a group-specific antigen-like protein ( gag ) , a reverse transcriptase ( RT/pol ) , an envelope protein ( env ) , two peptidases-like domains ( Pep ) , and a zinc finger ( Figure 1—figure supplement 2E ) . We detected peptides corresponding to the gag , pol , and env proteins ( Figure 1—figure supplement 2E , bottom ) , indicating potential competence for retrotransposition . We additionally detected proteins derived from other transposons including copia and 297 . Of note , roo ORFs remained detectable at 10–12 hr AEL ( Figure 1—figure supplement 2F ) , possibly suggesting substantial stability , as this was a time at which roo mRNA levels had diminished . The decline in transposon expression from 4 to 6 hr to 10–12 hr intervals of embryogenesis could potentially involve the piRNA pathway . However , piRNA-guided post-transcriptional or co-transcriptional silencing also requires a growing list of additional proteins ( reviewed in Czech et al . , 2018; Ozata et al . , 2019 ) . We therefore probed the expression of known piRNA pathway components during various stages of embryogenesis in our transcriptomic and proteomic datasets . With the exception of Piwi , genes involved in coTGS were both maternally deposited and zygotically expressed during the first ~10 hr of embryogenesis ( Figure 1—figure supplement 2A ) . Components of the PICTS complex , comprising Panx , Nxf2 , Nxt1 , and Ctp , showed abundant protein expression in the 5–7 hr and 10–12 hr AEL time intervals . piRNA-mediated coTGS also depends on several general chromatin modifiers , including Egg and its co-factor Wde ( Osumi et al . , 2019; Rangan et al . , 2011; Sienski et al . , 2015; Yu et al . , 2015 ) . Both of these proteins are required for heterochromatin formation in the embryo , and Egg in particular has previously been implicated in embryonic repeat silencing ( Seller et al . , 2019 ) . Similar to piRNA-specific coTGS factors , proteins involved in general chromatin silencing were both maternally deposited and zygotically expressed and detected at all studied time points , as expected based on their ubiquitous functions ( Figure 1—figure supplement 2A ) . Of note , Piwi mRNA appears to be primarily maternally deposited , with zygotic transcript levels remaining low throughout embryogenesis ( Figure 1—figure supplement 2A , Figure 2A , B ) . In contrast , we noted little or no maternal deposition and low zygotic expression of key components of the piRNA precursor expression and export machinery and of critical piRNA biogenesis factors ( Figure 1—figure supplement 2A ) . Considered together , our expression analyses are consistent with the potential of maternally instructed Piwi protein acting through coTGS during Drosophila embryogenesis . To assess the potential role of the piRNA pathway in regulating the transposon burst during Drosophila embryogenesis , we examined the spatial and temporal expression of coTGS proteins in the developing embryo using light-sheet live fluorescence microscopy . For this purpose , we used two previously published lines carrying a modified BAC expressing either GFP-Piwi or GFP-Panx from its endogenous regulatory region ( Handler et al . , 2013; Sienski et al . , 2015 ) and a GFP-Nxf2 knock-in line that we generated by CRISPR/Cas9 ( Fabry et al . , 2019 ) . As Ctp and Nxt1 have many additional functions , we did not examine their localization in this study . We also crossed in a transgene carrying H2Av-RFP to enable tracking of nuclei . Pre-blastoderm stage embryos ( 0 . 5 hr AEL ) were continuously imaged for >10 hr of embryogenesis . As previously reported ( Brennecke et al . , 2008; Mani et al . , 2014; Megosh et al . , 2006 ) and consistent with its maternal deposition , we detected GFP-Piwi during the pre-blastoderm stage ( NC1–9 , ~0–30 min AEL ) localized to the posterior pole where it formed a crescent-like structure ( Video 1 , Figure 2C ) . As embryogenesis progressed and somatic nuclei migrated to the surface ( NC 9–14 , ~1 . 5–3 hr AEL ) , Piwi localized to somatic nuclei and to the pole plasm surrounding the nuclei of germline progenitor cells , as we and others reported earlier based on immunofluorescence staining of fixed embryos ( Brennecke et al . , 2008; Mani et al . , 2014; Megosh et al . , 2006 ) . In agreement with an earlier report ( Mani et al . , 2014 ) , our dynamic data revealed that nuclear Piwi signal strongly decreased during mitotic cycles , with little fluorescence signal overlapping with H2Av-RFP during nuclear divisions ( Video 1 , Figure 2D ) . We continued to detect Piwi expression in somatic nuclei throughout the first 10 hr of embryogenesis; however , signal intensity decreased over time . This observation was consistent with transcriptomic and proteomic measurements taken over a comparable time course ( Figure 2A , B ) . Similar to Piwi , both Nxf2 and Panx were detected in somatic and pole cell nuclei from the syncytial blastoderm stage ( Videos 2 and 3 , Figure 2—figure supplement 1A–E ) . In contrast to Piwi , Panx and Nxf2 showed strong co-localization with H2Av-RFP during mitotic cycles ( Videos 2 and 3 , Figure 2—figure supplement 1F , G ) , suggesting that while Piwi is predominantly excluded , Nxf2 and Panx are retained in the nucleoplasm during mitosis . Consistent with our RNA-seq and TMT-MS data ( Figure 2—figure supplement 1H–K ) , as embryogenesis progressed , Panx and Nxf2 remain detectable for several hours ( >10 hr AEL ) , closely matching the protein expression of Piwi . Piwi carries epigenetic information in the form of piRNAs ( Brennecke et al . , 2008; Le Thomas et al . , 2014b ) . However , it is unclear if Piwi-piRNA complexes are assembled during oogenesis prior to maternal deposition into the embryo , or whether zygotic piRNA biogenesis and Piwi loading also occurs . We therefore analyzed the expression of GFP-tagged Piwi from reciprocal crosses with control w1118 flies by immunofluorescence staining in early and late-stage embryos . Embryos derived from females expressing GFP-Piwi showed strong maternal deposition of Piwi during early embryogenesis ( Figure 2E , Figure 2—figure supplement 1L ) , with GFP fluorescence in later stage ( >12 hr AEL ) embryos restricted to the germline cells . Consistent with maternal deposition of Piwi , embryos derived from the reciprocal cross showed no GFP signal in the early embryos ( Figure 2F , Figure 2—figure supplement 1M ) . Instead , we only observed GFP-Piwi signal in the developing gonads of late-stage embryos , likely as a result of zygotic expression . Strikingly , Piwi of zygotic origin localized exclusively to the cytoplasm of the germ cell progenitors and was not detected in nuclei , suggesting that zygotically transcribed Piwi is likely not relevant for coTGS until later in development . Piwi proteins are guided by their piRNA co-factors to recognize and co-transcriptionally silence active transposons in the Drosophila ovary ( Le Thomas et al . , 2013; Post et al . , 2014; Sienski et al . , 2015; Sienski et al . , 2012; Yu et al . , 2015 ) . If this pathway were relevant in the embryonic soma , maternally deposited Piwi would require instructions to recognize embryonically expressed elements . To examine this possibility , we immunoprecipitated Piwi from 0 to 8 hr control w1118 embryos as well as from adult ovaries and sequenced the associated small RNAs . Piwi in both tissues existed in complex with 23- to 28-nt piRNAs and showed nearly indistinguishable size profiles that were biased for antisense reads ( Figure 2—figure supplement 1N , O ) . Closer inspection by aligning the reads to transposon consensus sequences revealed similar piRNA levels for the majority of TEs; however , we detected some notable differences ( Figure 2G ) . Piwi in ovaries showed higher levels of antisense piRNAs targeting the TEs mdg1 and 412 , in agreement with the majority of these small RNAs originating from the soma-specific flam piRNA cluster ( Brennecke et al . , 2007; Malone et al . , 2009; Zanni et al . , 2013 ) . Piwi in embryos showed high levels of antisense piRNAs targeting roo ( ~16% or all TE-targeting reads ) and 297 ( ~9% ) , consistent with an ability of maternally deposited Piwi to potentially recognize these TEs when expressed in the embryo . In Drosophila ovaries , coTGS depends on Piwi-mediated recruitment of the PICTS/SFiNX complex and correlates with the deposition of H3K9me3 marks at TE insertions and surrounding genomic regions ( Batki et al . , 2019; Eastwood et al . , 2021; Fabry et al . , 2019; Murano et al . , 2019; Schnabl et al . , 2021; Sienski et al . , 2015; Yu et al . , 2015; Zhao et al . , 2019 ) . Due to the poor conservation of the genomic locations of transposon insertions between different Drosophila strains , we used whole-genome sequencing ( WGS ) to de novo identify the TE insertion sites present in our control w1118 flies ( see Materials and methods ) . This data enabled us to identify over 600 euchromatic transposon insertions that are absent from the dm6 reference genome , and these were used for our chromatin analyses , as most annotated insertions in the dm6 genome assembly were absent from our strain . In order to determine the fate of transposon loci throughout embryogenesis , we performed H3K9me3 chromatin immunoprecipitation followed by sequencing ( ChIP-seq ) on control w1118 embryos at 2 hr intervals covering the period when transposon expression is dynamic ( 0–10 hr AEL ) and a later time point ( 16–18 hr AEL ) well after maternal Piwi protein was no longer detectable in somatic nuclei ( Figures 1A and 2C ) . We included adult ovaries , which show piRNA-guided coTGS , as well as adult heads , a somatic tissue without active piRNA pathway , to compare the changes of this repressive chromatin mark across different stages and tissues of Drosophila development . Early embryos ( 0–2 hr AEL ) showed low levels of H3K9me3 signal at 117 euchromatic , w1118-specific roo insertions ( Figure 3A ) . However , as development progressed , H3K9me3 levels increased with a peak at 6–10 hr AEL ( Figure 3A , Figure 3—figure supplement 1A ) . Thus , deposition of repressive chromatin marks correlated with the RNA expression of roo , yet the maximum of H3K9me3 accumulation lagged behind the RNA expression peak by approximately 2 hr . These data are consistent with a requirement for nascent transcription for efficient conversion of a TE insertion into heterochromatin , as previously reported in yeast ( Bühler et al . , 2006; Shimada et al . , 2016 ) and for the recognition of transposon loci by the piRNA pathway ( Le Thomas et al . , 2013; Post et al . , 2014; Rozhkov et al . , 2013; Sienski et al . , 2015; Sienski et al . , 2012; Yu et al . , 2015 ) . Of note , the deposition of repressive marks trailed the direction of transcription and showed higher signal enrichments in the regions downstream of the transposon insertions , as previously observed for piRNA-dependent silencing in cell culture systems ( Fabry et al . , 2019; Sienski et al . , 2015; Sienski et al . , 2012 ) . Interestingly , H3K9me3 signal at euchromatic roo insertions of 16–18 hr AEL embryos , which lacked maternal Piwi in somatic nuclei and no longer express roo , showed diminished intensities compared to earlier time intervals . Similarly , heads and ovaries , both tissues from adult flies , showed no enrichment of H3K9me3 at euchromatic roo insertions , despite the presence of a functional piRNA pathway in ovaries . Considered together , these data suggest that maternal piRNAs program a response to a burst of roo expression during embryonic development but that the deposition of H3K9me3 marks , likely directed via coTGS , no longer occurs at developmental time points and in tissues where roo is not expressed . This is consistent both with the known requirement for active transcription for targeting by Piwi and with the observed need for continuous engagement of the PICTS/SFiNX complex to maintain H3K9me3 marks on transposon loci ( Batki et al . , 2019; Eastwood et al . , 2021; Fabry et al . , 2019; Le Thomas et al . , 2013; Murano et al . , 2019; Post et al . , 2014; Rozhkov et al . , 2013; Schnabl et al . , 2021; Sienski et al . , 2015; Sienski et al . , 2012; Yu et al . , 2015; Zhao et al . , 2019 ) . To investigate whether this mechanism is specific to roo or more general , we examined the transposon 297 , which is also expressed during embryogenesis ( Figure 1—figure supplement 1F ) and showed high targeting potential by maternally inherited piRNAs ( Figure 2G ) . Genomic loci in close proximity to euchromatic , w1118-specific 297 insertions ( n = 20 ) showed dynamic deposition of H3K9me3 similar to roo ( Figure 3—figure supplement 1B ) . However , while H3K9me3 levels at roo insertions peaked between 6 and 10 hr AEL , 297 insertions showed the maximum H3K9me3 signal intensity between 2 and 8 hr AEL , suggesting that these loci are targeted by coTGS earlier than roo insertions . In contrast , H3K9me3 occupancy at transposons such as mdg1 and 412 that were expressed during embryogenesis but lacked substantial maternal deposition of piRNAs retained low H3K9me3 levels throughout embryogenesis , though they showed a strong enrichment in ovaries ( Figure 3—figure supplement 1C ) . To determine whether the deposition of repressive chromatin marks at euchromatic 297 and roo insertions was specific , rather than reflecting a general trend of H3K9me3 accumulation genome-wide , we analyzed genomic regions not targeted by maternally inherited piRNAs . H3K9me3 signal at constitutive heterochromatin remained stable throughout the sampled time points ( Figure 3—figure supplement 1D ) , while H3K9me3 levels on chromosome 4 increased steadily throughout development ( Figure 3—figure supplement 1E ) . Of note , while ovaries showed no coTGS signature at roo insertions , other transposons , such as Doc , showed a clear accumulation of H3K9me3 marks that was absent in embryos during all assayed time points ( Figure 3B ) . Considered together , these results are consistent with piRNA-guided chromatin modification of a subset of transposons that show activity during Drosophila embryonic development . Though embryonically repressed transposons bore hallmarks of piRNA-guided heterochromatin formation , the reliance of the pathway on maternally deposited Piwi-piRNA complexes prevented a demonstration that silencing depended on the pathway through conventional genetics . Ovaries that lack key piRNA pathway silencing factors show substantial expression changes and produce morphologically altered eggs that largely fail to develop normally ( Cox et al . , 1998; Czech et al . , 2013; Handler et al . , 2013; Khurana et al . , 2010; Klattenhoff et al . , 2007; Klenov et al . , 2011; Li et al . , 2009a; Malone et al . , 2009; Mani et al . , 2014; Muerdter et al . , 2013; Pane et al . , 2007; Park et al . , 2019 ) . To investigate the effect of Piwi depletion on Drosophila embryogenesis without affecting oogenesis , we used the auxin-inducible degron ( AID ) system ( Nishimura et al . , 2009 ) . This protein degradation system comprised an AID-tag , fused to the protein of interest , and the plant-derived F-box protein transport inhibitor response 1 ( TIR1 ) . AID and TIR1 associate with each other in an auxin-dependent manner , with binding of TIR1 to the AID-tagged target leading to the recruitment of the cellular ubiquitination machinery and target protein degradation via the proteasome ( Figure 4A ) . This conditional degradation system has proven effective in several model organisms including Drosophila where it was recently shown to enable degradation of the germ cell-specific protein Vasa ( Bence et al . , 2017 ) . We used CRISPR/Cas9 to insert an amino-terminal GFP-AID tag at the Drosophila piwi locus and crossed these flies to a line expressing the Oryza sativa-derived TIR1 ( OsTIR1 ) protein under the control of the ubiquitin promoter . As a proof of concept , we tested the auxin-induced degradation of Piwi in adult ovaries of flies homozygous for both GFP-AID-Piwi and OsTIR1 . Feeding flies for 24 hr with 5 mM auxin-containing yeast paste was sufficient to induce complete degradation of Piwi in ovaries ( Figure 4B , C , Figure 4—figure supplement 1A ) , and this depletion resulted in the de-repression of transposons ( Figure 4D ) . Notably stronger changes were observed following longer treatments , possibly implying a lag between loss of piRNA pathway function and that of repressive chromatin marks . Following a 1-day treatment , embryos laid by Piwi-depleted females developed without defects and showed similar hatching rates as their control treated siblings ( Figure 4E , F ) . Longer auxin treatments resulted in more frequent deformation of embryos that was accompanied by reduced hatching rates ( Figure 4E , F ) , likely due to patterning defects as a result of Piwi depletion from follicle cells . Drosophila embryos develop within a relatively impermeable chorion , and treatment of embryos directly with auxin showed little impact . However , in dechorionated embryos we observed a near complete degradation of Piwi protein following 30 min auxin treatment of embryos collected for 0–30 min AEL ( Figure 5A , B ) . To investigate the dynamics of auxin-mediated Piwi depletion in embryos , we used light-sheet fluorescence live microscopy . Early blastoderm embryos treated with 5 mM auxin showed rapid degradation of GFP-AID-Piwi signal , which was undetectable after 25 min of treatment ( Figure 5C , Video 4 ) . Of note , the removal of maternal Piwi in this time window did not significantly affect the embryo hatching rate ( Figure 4—figure supplement 1B ) . We next investigated the impact of degrading maternal Piwi from early-stage embryos on transposons . Embryos derived from flies homozygous for GFP-AID-Piwi and OsTIR1 were collected across a 30 min period and treated for an additional 2 . 5 hr with or without 5 mM auxin before RNA extraction , generation of libraries , and differential expression analysis of the sequenced transcriptomes ( Figure 5A ) . These embryos corresponded to 2 . 5–3 hr AEL , the point at which we began to observe zygotic roo transcripts ( Figure 1C ) , and showed minimal differences between control- and auxin-treated embryos for the same set of genes used to benchmark our dataset ( Figure 5—figure supplement 1A ) . The majority of transposons showed no significant expression change upon Piwi depletion; however , roo and 297 were significantly de-repressed ( p<0 . 05 ) by more than twofold ( Figure 5D ) , suggesting that Piwi impacts their expression during embryogenesis . Previous studies suggested that auxin in small concentrations has a negligible impact on Drosophila development ( Bence et al . , 2017; Trost et al . , 2016 ) , but to control for effects of auxin itself on TE regulation , we also evaluated transposon expression in auxin-treated GFP-AID-Piwi embryos that lack OsTIR1 . Without OsTIR1 , 2 . 5–3 hr embryos treated with 5 mM auxin showed no significant changes in transposon expression compared to control siblings treated with PBS ( Figure 5—figure supplement 1B ) . We additionally examined changes in the repressive chromatin mark H3K9me3 to determine whether these were deposited in a piRNA-dependent fashion at euchromatic roo and 297 transposon insertions . We again collected embryos for 30 min and treated with 5 mM auxin ( or PBS as a negative control ) for 6 hr , which yielded embryos 6–6 . 5 hr AEL ( Figure 5A ) and corresponds to the peak in H3K9me3 signal at roo insertions in control w1118 embryos ( Figure 3A , Figure 3—figure supplement 1A ) . Piwi depletion severely impacted H3K9me3 signal over the transposon consensus sequence of roo and 297 , but not that of other TEs ( Figure 5E ) . Additionally , H3K9me3 levels at individual transposon genomic loci ( see Materials and methods for identification of TE insertions in our fly stock ) showed similar patterns . H3K9me3 signal in genome-wide 5 kb bins predominantly changed when roo or 297 insertions were nearby ( Figure 5—figure supplement 1C ) . We identified 154 bins with significantly reduced ( p<0 . 05 ) H3K9me3 occupancy , while only two bins showed an increase . Of the bins with lower H3K9me3 signal , 122 and 10 were within 5 kb of roo or 297 insertions , respectively , thereby illustrating the impact of Piwi on chromatin states at genomic loci specifically targeted by maternal piRNAs . Furthermore , 205 and 63 individual euchromatic , degron strain-specific TE insertions of both roo and 297 , respectively , showed a strong decrease of H3K9me3 levels in Piwi-depleted embryos ( Figure 5F , G , Figure 5—figure supplement 1D–F ) , while H3K9me3 levels at constitutive heterochromatin and on chromosome 4 were not affected ( Figure 5H , I ) . Of note , while roo and 297 TE levels were elevated upon auxin treatment in 2 . 5–3 hr AEL embryos , transposon expression returned to baseline levels comparable to untreated embryos in 6–7 hr and 7–8 hr AEL time intervals ( Figure 5—figure supplement 1G ) . Taken together , these data strongly indicate a relationship between the deposition of repressive H3K9me3 chromatin marks at transposon insertions and maternally deposited Piwi-piRNA complexes ( Figure 5J ) .
Here , we have examined the role of the Piwi-directed coTGS arm of the piRNA pathway during early embryogenesis in Drosophila . By far , most of our insight into the function of piRNAs has derived from studies in germ cells or in the support cells of reproductive tissues . The intriguing observation that piRNAs and their Piwi-family binding partners are maternally deposited has led to speculation regarding potential roles for piRNAs in inter- and transgenerational epigenetic inheritance . Indeed , maternal piRNAs are critical in the suppression of hybrid dysgenesis induced by paternal transmission of I- or P-elements in matings with females that lack these transposons ( Brennecke et al . , 2008; Khurana et al . , 2011 ) . Epigenetic modifications induced by piRNAs appear to aid in piRNA cluster definition in the germline ( Akkouche et al . , 2017 ) . Additionally , maternally deposited Aub-piRNA complexes have been implicated in embryonic gene regulation ( Barckmann et al . , 2015; Dufourt et al . , 2017; Rouget et al . , 2010 ) . Last , maternally inherited piRNAs control transposon expression in interspecies hybrids between Drosophila melanogaster and Drosophila simulans ( Kelleher et al . , 2012 ) and regulate the TE tirant in the somatic compartment of the female gonad in D . simulans ( Akkouche et al . , 2013 ) . Yet , to date , the lack of mechanisms to rapidly deplete maternally deposited PIWI proteins specifically from early embryos has hampered our ability to broadly assess their zygotic roles . By fusing a chemically inducible degron to Piwi , we were able to deplete Piwi-piRNA complexes from dechorionated embryos within less than 30 min of treatment and well before the nuclear accumulation of Piwi that is observed following activation of zygotic transcription . Though nuclear localization of Piwi correlates with the appearance of its potential targets , nascent transcripts of transposons , it is unclear what triggers movement of Piwi into the somatic nuclei . Notably , nuclear translocation of Piwi lags behind in germ cell precursor nuclei , and this correlates with the observation that these nuclei activate transcription of their genomes later in embryogenesis than somatic nuclei do ( Van Doren et al . , 1998; Zalokar , 1976 ) . Our dynamic imaging of Piwi localization also revealed that it shuttles out of nuclei during mitosis , as previously observed ( Mani et al . , 2014 ) . Since other factors of the pathway , namely components of the PICTS/SFiNX complex , remain nuclear , it is likely that Piwi is actively excluded . Several studies have previously shown that nuclear localization of Piwi is conditional upon its binding to a piRNA partners ( Klenov et al . , 2011; Saito et al . , 2009; Yashiro et al . , 2018 ) , but we have no indication that Piwi is unloaded and reloaded during mitotic cycles . Rather , we hypothesize that another mechanism regulates the activity of the Piwi nuclear localization signal , though what purpose this might serve and whether it also occurs in germline and follicle cells or is restricted to embryogenesis remains unclear . It has been suggested that the evolution of the abbreviated piRNA pathway in ovarian follicle cells arose as a consequence of the lifestyle adopted by gypsy family elements . These retrotransposons show their highest expression levels in the support cells surrounding the developing germline . gypsy family elements encode an envelope protein and have been shown to assemble into virus-like particles ( Kim et al . , 1994; Leblanc et al . , 2000; Song et al . , 1997 ) . This has led to the hypothesis that their ancestral propagation strategy combined evasion of TE repression mechanisms present in germ cells with an ability to create particles that could infect the germline , where the element could insert into the genome of the developing oocyte following reverse transcription ( Kim et al . , 1994; Leblanc et al . , 2000; Song et al . , 1997 ) . While this remains speculative , it does provoke questions of whether a similar strategy is adopted by roo in the embryo . roo is a quite successful element , as indicated by it being the element with the highest copy number of individual insertions in our sequenced strains ( 9 . 4% of all identified TE insertions in the w1118 strain and 9 . 9% in our degron line ) . How this is achieved remains mysterious since roo expression is extremely low in the ovary . Moreover , roo does not appear to be a target of the ovarian piRNA pathway since its gonadal expression is not increased nor does its HP1a enrichment and H3K9me3 levels change in piRNA pathway mutant animals ( Figure 5—figure supplement 1H , I; Senti et al . , 2015; Wang and Elgin , 2011 ) . In the embryo , the expression of roo is restricted to somatic cells , especially cell lineages giving rise to the adult mesoderm . Previous studies have suggested that roo expression is activated by twist ( twi ) and snail ( sna ) , which are highly expressed in the embryogenic mesoderm ( Brönner et al . , 1995 ) , and this is consistent with the spatial expression pattern that we also observe . roo expresses the full repertoire of proteins needed to form virus-like particles , and its high expression levels ( exceeding 1% of the transcriptome at its peak ) might enable a strategy of propagation by infection in trans , even if rates of transmission to the germ cell precursors are relatively low . Our data strongly suggests that only maternally deposited piRNAs engage Piwi in the soma of the developing embryo . Since roo is not regulated by the piRNA pathway in the ovary , evolutionary pressures must have driven the development of a set of maternal instructions that are inherited to dampen the burst of roo expression in the developing embryo . Indeed , 16% of maternally deposited piRNAs target roo . Though there are differences between the populations of piRNAs observed in ovaries as compared to embryos , this is mostly driven by the presence of follicle cell piRNAs in in samples taken from the gonad . In our small RNA analyses , we collapse all stages of oogenesis . Thus , it is not clear whether the composition of piRNA populations shifts as the ovariole matures and whether any such shifts enrich for embryonically expressed elements late in oocyte maturation . Irrespective , a set of instructions from maternal piRNA clusters clearly builds a transgenerational ability to recognize roo and other embryonically expressed elements . Consistent with its recognition by Piwi-piRNA complexes and recruitment of the PICTS/SFiNX complex , H3K9me3 marks build at presumably active , euchromatic roo insertions as embryos progress toward stage 13 ( 10 hr AEL ) . The peak of H3K9me3 abundance lags about 2 hr behind the peak of transcription . Since we have little other information on the dynamics of piRNA-mediated silencing , it is not clear whether this is an expected observation or whether there may be mechanisms that antagonize the ability of the piRNA pathway to immediately recognize and direct heterochromatinization of expressed roo insertions . Of note , we see a shorter interval between the embryonic peak of 297 expression and its peak of H3K9me3 accumulation . Imposition of a repressive chromatin state on roo is transient during somatic development . By 13 hr AEL , H3K9me3 peaks over roo insertions have disappeared , but roo expression has not returned . The lack of H3K9me3 also correlates with the absence of critical piRNA pathway proteins in the soma . Overall , this suggests that both the expression of these TEs and the host response via small RNAs is transient . While our data provide compelling evidence of the accumulation of repressive chromatin marks at presumably actively transcribing TE insertions , it does not carry spatial information about the precise cell types affected by H3K9me3 deposition . Though zygotic depletion of maternal Piwi during early embryogenesis does produce a statistically significant change in roo expression ( roughly twofold ) , this transposon remains highly expressed reaching up to 1% of the entire transcriptome in control animals , despite being targeted by the piRNA pathway . This provokes the question of whether targeting of roo by the piRNA pathway is biologically relevant . In favor of this hypothesis are several observations . Roo is expressed in ovaries at very low levels , yet the hallmarks of piRNA-dependent silencing , specifically H3K9me3 , are absent from euchromatic roo insertions . This strongly indicates that roo is not controlled by the piRNA pathway in this tissue . Nonetheless , ovaries produce abundant roo piRNAs , and these are overwhelmingly in the antisense orientation . Additionally , the only uni-strand cluster expressed in germ cells , cluster 20A , has collected roo insertions in the antisense orientation . These piRNAs are abundantly maternally transmitted ( 16% of all piRNAs in embryos ) and persist throughout the time during early embryogenesis when high-level roo expression is proposed to be driven by mesodermal transcription factors . An argument against biological significance is the lack of a clearly observable phenotype in flies following embryonic depletion of maternal Piwi . However , technical limitations enable us to only measure impacts within a single generation . It is entirely possible that the fitness cost of roo occupying 2% of the embryonic transcriptome might be substantial over time or in conditions flies might experience in the wild compared to the controlled rearing conditions in the lab . Perhaps more importantly , our study demonstrates that recognition of a locus by the piRNA pathway does not necessarily impose the creation of a mitotically heritable epigenetic state . This is consistent with observations made by many groups in follicle cells wherein heterochromatin-mediated silencing of somatic transposons requires the continuous presence of the piRNA machinery ( Batki et al . , 2019; Clark et al . , 2017; Dönertas et al . , 2013; Fabry et al . , 2019; Muerdter et al . , 2013; Murano et al . , 2019; Ohtani et al . , 2013; Saito et al . , 2009; Sienski et al . , 2015; Sienski et al . , 2012; Zhao et al . , 2019 ) . These data are at odds with prior observations and speculation that the maintenance of silenced epigenetic states can be primed by Piwi but maintained in a Piwi-independent mechanism throughout adult life ( Gu and Elgin , 2013 ) . The prior study noted these effects after only a 50% reduction in embryonic Piwi protein or RNA , using either of two different strategies . Though our induced proteolytic degradation strategy is unlikely to completely remove all Piwi protein , Piwi was reduced to levels that are undetectable by western blotting ( Figure 5B ) or via the fluorescence of its fused GFP ( Figure 5C ) , which would , if anything , be expected to produce a more profound impact . While it is difficult to reconcile our observations with the interpretation of the prior study , there were substantial differences in what was measured and in how the measurements were made ( i . e . , a different set of genomic loci was studied in different Drosophila strains by different methods ) . The prior work made use of position-effect reporters integrated into pericentromeric heterochromatin and indicated that the expression of these in adults was sensitive to Piwi depletion in the embryo . We did not examine such a reporter , and so it remains possible that H3K9me3 marks deposited in a Piwi-dependent fashion in regions adjacent to large domains of Piwi-independent H3K9 methylation might behave differently than those deposited on active , euchromatic transposons . Considerable consistency between the two studies can be found in the prior observation that HP1a occupancy in embryos did not change substantially on several transposons studied ( maximum of twofold on HeT-A ) ( Gu and Elgin , 2013 ) . The prior study also failed to note large-scale changes in HP1a distribution , as a proxy for methylated H3K9 , and reported only very small changes in HP1a levels on a few transposon families , as assayed in larvae by ChIP-array measurements , which collapse all insertions of a given family into a single data point . The transposons that we do identify as sensitive to Piwi during early embryogenesis do not overlap with those identified in the previous study as being mildly affected by reductions in Piwi at a later developmental stage ( data not shown ) . This is actually consistent with our observation that the effects of profound Piwi depletion on roo and other TEs are transient during embryogenesis . Thus , it seems that the data themselves diverge less between the two studies than do the conclusions drawn . Of note , another recent report found a mild upregulation of transposons in pre-ZGA embryos upon maternal depletion of Piwi; however , this work relied on germ cell-specific knockdown during late stages of oogenesis rather than direct protein depletion in the embryo , thus at least some of the observed effects could stem from TE mobilization during ovary development ( Gonzalez et al . , 2021 ) . A recent detailed and elegant study examined the patterns of H3K9me3 accumulation during early embryogenesis in Drosophila miranda ( Wei et al . , 2021 ) . Though overall , deposition of H3K9me3 did not correlate with the abundance of maternally deposited piRNAs , a set of the earliest heterochromatin nucleating elements were associated with abundant piRNAs . These targeted elements had high copy numbers and showed evidence of recent transposition activity , suggesting that they were under evolutionary pressure for robust silencing both in the ovary and the soma . It should be noted that precise nucleation sites did not necessarily overlap with abundant piRNAs , suggesting that multiple silencing mechanisms might collaborate to repress these transposon families . Considered as a whole , our data are consistent with a role for maternally deposited piRNAs in the recognition of transposon families that have focused their expression and activity during early embryogenesis . However , our data does not support a model wherein the piRNA pathway nucleates heritable patterns of heterochromatin formation that broadly pattern the epigenetic landscape of the adult Drosophila soma , and this is perhaps consistent with our failure to observe consequential developmental abnormalities upon negation of embryonic Piwi function .
All flies were kept at 25°C on standard cornmeal or propionic food . Flies expressing GFP-Nxf2 from the endogenous locus were generated by CRISPR/Cas9 ( Fabry et al . , 2019 ) . Transgenic flies carrying a BAC transgene expressing GFP-Panx and GFP-Piwi were generated by the Brennecke lab ( Handler et al . , 2013; Sienski et al . , 2015 ) and obtained from the Vienna Drosophila Resource Center . Control w1118 flies were a gift from the University of Cambridge Department of Genetics Fly Facility , and flies expressing His2Av-RFP were a gift from the St Johnston lab . Flies between 3 and 14 days after hatching were used for experiments . GFP-AID-Piwi knock-in flies were generated by CRISPR/Cas9 genome engineering . Homology arms of 1 kb flanking the targeting site were cloned into pUC19 by Gibson Assembly and co-injected with pCFD3 ( Addgene # 49410 ) containing a single-guide RNA ( Port et al . , 2014 ) into embryos expressing vas-Cas9 ( Bloomington Drosophila Stock Center # 51323 ) . Flies expressing OsTIR1 under the D . melanogaster Ubiquitin-63E promoter were generated by phiC31 integrase-mediated transgenesis by injection of plasmids containing expression cassettes for proteins into embryos of genotype ‘y w P[y[+t7 . 7]=nos-phiC31\int . NLS]X #12; +; P[y[+t7 . 7]=CaryP]attP2 , ’ resulting in transgene integration on chromosome 3 . Microinjection and fly stock generation was carried out by the University of Cambridge Department of Genetics Fly Facility . Transgenic and knock-in flies were identified by genotyping PCRs and confirmed via Sanger sequencing . Protein concentration was measured using a Direct Detect Infrared Spectrometer ( Merck ) . 20 µg of proteins were separated on a NuPAGE 4–12% Bis-Tris gel ( Thermo Fisher Scientific ) . Proteins were transferred for 2 hr at 100 V , 400 mA , 100 W on an Immun-Blot Low Fluorescent PVDF Membrane ( BioRad ) and blocked for 1 hr in 1× LI-COR TBS Blocking Buffer ( LI-COR ) . Primary antibodies were incubated overnight at 4°C . LI-COR secondary antibodies were incubated for 45 min at room temperature ( RT ) and images acquired with an Odyssey CLx scanner ( LI-COR ) . Fly ovaries were dissected in ice-cold Phosphate-buffered saline ( PBS ) and fixed in 4% PFA diluted in PBS for 15 min at room temperature while rotating . Following three rinses and three 10 min washing steps in PBS-Tr ( 0 . 3% Triton X-100 in PBS ) , ovaries were blocked for 2 hr at RT while rotating in PBS-Tr + 1% BSA . Primary antibody incubation was carried out in blocking buffer overnight at 4°C while rotating , followed by three washing steps for 10 min each in PBS-Tr . All following steps were performed in the dark . Secondary antibodies were diluted in blocking buffer and incubated overnight at 4°C while rotating . Ovaries were washed four times for 10 min in PBS-Tr and stained with 0 . 5 µg/ml DAPI ( Thermo Fisher Scientific ) for 10 min . Following two additional washing steps for 5 min in PBS , ovaries were mounted in ProLong Diamond Antifade Mountant ( Thermo Fisher Scientific ) and imaged on a Leica SP8 confocal microscope using a 40× Oil objective . Embryos were collected and dechorionated in 50% bleach for 1 min . Embryos were transferred into 1 ml fixing solution ( 600 µl 4% PFA in PBS , 400 µl n-heptane ) and fixed for 20 min at RT while rotating . The lower aqueous phase was removed and 600 µl methanol added . The tube was vortexed vigorously for 1 min to remove vitelline membranes . Embryos were allowed to sink to the bottom of the tube and all liquid was removed , followed by two washes with methanol for 1 min each . Embryos were stored at −20°C at least overnight or until further processing . In order to rehydrate embryos , three washes each 5 min with PBST ( 0 . 1% Tween20 in PBS ) were performed and embryos blocked for 1 hr at RT in PBST + 5% BSA . Primary antibodies were incubated overnight at 4°C while rotating in blocking buffer followed by 3 washes for 15 min each with PBST . All following steps were performed in the dark . Secondary antibodies were diluted in blocking buffer and incubated at RT for 2 hr . Embryos were rinsed three times and washed two times for 15 min . Nuclei were stained with 0 . 5 µg/ml DAPI ( Thermo Fisher Scientific ) for 10 min . Following two additional washing steps for 5 min in PBS , embryos were mounted in ProLong Diamond Antifade Mountant ( Thermo Fisher Scientific ) and imaged on a Leica SP8 confocal microscope using a 40× Oil objective . Embryos were collected , dechorionated , and processed as described above until secondary antibody incubation . For all steps containing BSA addition , RNAsin Plus RNase inhibitors were added ( 1:1000 , Promega ) . Following secondary antibody incubation , cells were washed three times for 15 min in PBST at RT while rotating . Embryos were fixed in 4% PFA in PBST solution for 25 min and rinsed three times with PBST for 5 min each . Embryos were pre-hybridized in 100 µl hybridization buffer ( 50% formamide , 5× saline-sodium citrate ( SSC ) , 9 mM citric acid pH 6 . 0 , 0 . 1% Tween20 , 50 µg/ml heparin , 1× Denhardt’s solution [Sigma-Aldrich] , 10% dextran sulfate ) for 2 hr at 65°C . Probes were hybridized in hybridization buffer supplemented with 2 nM of each FISH probe at 45°C overnight . Following washing twice with probe wash buffer ( 50% formamide , 5× SSC , 9 mM citric acid pH 6 . 0 , 0 . 1% Tween20 , 50 µg/ml heparin ) for 5 min and twice for 30 min at 45°C , embryos were incubated in amplification buffer ( 5× SSC , 0 . 1% Tween20 , 10% dextran sulfate ) for 10 min at RT . Hairpins were prepared as described above and embryos incubated in fresh amplification buffer with 120 nM of each probe at RT overnight in the dark . Embryos were washed twice with 5× SSC for 5 min . Nuclei were stained with 0 . 5 µg/ml DAPI diluted in 2× SSC for 15 min . Following washing twice with 2× SSC for 10 min , embryos were mounted in ProLong Diamond Antifade Mountant ( Thermo Fisher Scientific ) and imaged on a Leica SP8 confocal microscope using a 40× Oil objective . Embryos were collected and dechorionated as described above . 1 ml of 1% low melting point ( LMP ) agarose was prepared and embryos transferred into capillaries ( catalog # 100003476381 , Brand ) using a fitting plunger . Embryos were attempted to be positioned vertically in the capillary by twisting until agarose solidified . Capillaries were stored in PBS at RT until imaging . LSFM was performed on a Zeiss Lightsheet Z . 1 ( Carl Zeiss , Germany ) at 25°C with a 20×/1 . 0 Plan-Apochromat water-immersion objective lens . Embryos were lowered carefully out of the capillary into the imaging chamber filled with PBS and positioned directly between the light-sheet illumination objectives ( 10×/0 . 2 , left and right ) . Z-stack images for GFP and RFP ( excitation at 488 and 561 nm , respectively ) were acquired every 2 min for >10 hr with the lowest possible laser intensity ( 2 . 5% for GFP and 10% for RFP ) . Generated data was analyzed in Zeiss ZEN Imaging Software and Fiji ( ImageJ ) . 50 µl of embryos were collected and dechorionated as described above and transferred in 1 ml Crosslinking solution ( 1% formaldehyde in PBS , 50% n-heptane ) and vortexed on high speed for precisely 15 min . 90 µl 2 . 5M glycine solution was added to quench excess formaldehyde and incubated for 5 min at RT while rotating . Embryos were allowed to sink to the bottom of the tube and all liquid was removed . Embryos were washed three times for 4 min with ice-cold buffer A ( 60 mM KCl , 15 mM NaCl , 4 mM MgCl2 , 15mM HEPES pH 7 . 6 , 0 . 5% DTT , 1× PI ) supplemented with 0 . 1% Triton X-100 ( A-Tx buffer ) . All liquid was removed and embryos flash-frozen and stored at −80°C until further processing . Crosslinked embryos were transferred to a 2 ml Dounce homogenizer in 1 ml A-TBP ( Buffer A + 0 . 5% Triton X-100 ) . Following an additional washing step with A-TBP , embryos were lysed in 1 ml A-TBP using 10 strokes with a tight-fitting pestle . Lysate was centrifuged at 3200 g for 10 min at 4°C and supernatant removed . The pellet was resuspended in 1 ml Lysis buffer ( 15 mM HEPES , 140 mM NaCl , 1 mM EDTA , 0 . 5 mM EGTA , 1% Triton , 0 . 5 mM DTT , 10 mM sodium butyrate , 0 . 1% sodium deoxycholate , 1× PI ) and incubated at 4°C for 15 min while rotating . Following centrifugation at 3200 g for 10 min at 4°C , the pellet was washed twice with Lysis buffer and centrifuged again . All liquid was removed , and the pellet resuspended in 300 µl LB3 ( 10 mM Tris-HCl , pH 8 , 100 mM NaCl , 1 mM EDTA , 0 . 5 mM EGTA , 0 . 1% Na-Deoxycholate , 0 . 5% N-lauroylsarcosine , 1× PI ) . Sonication was carried out using the Bioruptor pico ( Diagenode ) for six cycles ( 30 s on , 30 s off settings ) . Debris was removed from the chromatin-containing supernatant by spinning down at full speed for 10 min at 4°C . Prepared chromatin was either frozen down in liquid nitrogen and stored at −80°C or used immediately . 5% of the chromatin fraction was flash-frozen as an input sample . 100 µl magnetic Protein A-coupled Dynabeads ( Thermo Fisher Scientific ) were washed three times in 1 ml blocking solution ( 0 . 2% BSA in PBS ) . The blocking solution was removed using a magnetic rack . 5 µl of anti-H3K9me3 or anti-H3K4me2 polyclonal antibody was diluted in 250 µl blocking solution and incubated with 100 µl washed beads by rotating at 4°C for at least 4 hr up to overnight . The supernatant was removed and beads washed three times in blocking solution as described above . The chromatin solution was added to the beads and incubated at 4°C while rotating overnight . Following four washing steps for 2 min each using ice-cold Lysis Buffer ( 15 mM HEPES , 140 mM NaCl , 1 mM EDTA , 0 . 5 mM EGTA , 1% Triton , 0 . 5 mM DTT , 10 mM sodium butyrate , 0 . 1% sodium deoxycholate , 1× PI , 0 . 05% SDS ) , beads were washed two additional times with ice-cold 1× TE buffer . All liquid was removed and beads resuspended in 200 µl Elution buffer ( 50 mM Tris-HCl , pH 8; 10 mM EDTA; 1% SDS ) . Input samples were thawed and brought up to 200 µl with Elution buffer . Samples were transferred into 200 µl Maxymum Recovery PCR tubes ( Axygen ) and incubated at 65°C for 16–18 hr for reverse crosslinking . RNA contamination was removed by adding 200 µl 1× TE buffer and 8 µl of 1 mg/ml RNase A ( Ambion ) to ChIP and input samples followed by incubation at 37°C for 30 min . Proteins were digested using 4 µl Proteinase K ( 800 U/ml , NEB ) and incubation at 55°C for 2 hr . Reverse crosslinked DNA was recovered using the MinElute PCR purification Kit ( Qiagen ) according to the manufacturer’s recommendation and eluted in 15 µl nuclease-free water . DNA recovery was verified and quantified using 1 µl for Bioanalyzer ( Agilent ) electrophoresis . 50 Drosophila ovaries were dissected in ice-cold PBS . Heads were dislodged by pouring liquid nitrogen over whole flies in a dish followed by shaking and collecting 50 µl broken-off heads in 1 . 5 ml tube . Samples were homogenized in 100 µl Buffer A1 ( 60 mM KCl , 15 mM NaCl , 4 mM MgCl2 , 15 mM HEPES pH 7 . 6 , 0 . 5% DTT , 0 . 5% Triton X-100 , 1× PI ) using a rotating pestle . The volume was brought up to 1 ml with buffer A1 and formaldehyde added to a final concentration of 1 . 8% for crosslinking . Samples were rotated for exactly 15 min at RT and glycine solution added to a final concentration of 225 mM . Samples were allowed to rotate for an additional 5 min and were centrifuged at 4000 g for 5 min at 4°C . The supernatant was removed , the pellet washed twice with buffer A1 and once with buffer A2 ( 140 mM NaCl , 15 mM HEPES pH 7 . 6 , 1 mM EDTA , 0 . 5 mM EGTA , 1% Triton X-100 , 0 . 5 mM DTT , 0 . 1% sodium deoxycholate , 10 mM sodium butyrate , 1× PI ) at 4°C . The pellet was then resuspended in 100 µl A2 buffer supplemented with 1% SDS and 0 . 5% N-laurosylsarcosine and incubated at 4°C for 2 hr while shaking vigorously . Lysate was sonicated using the Bioruptor pico for 16 cycles ( 30 s on , 30 s off ) . The sonicated lysate was spun at full speed for 10 min at 4°C and the supernatant transferred to a new tube . The volume was brought up to 1 ml with A2 buffer supplemented with 0 . 1% SDS . Chromatin used for ChIP was precleared with 15 µl washed Protein A Dynabeads and incubated with antibody coated beads as described above . Further steps were performed as described above for embryo ChIP . Piwi-piRNA complexes were isolated from ovaries or from 0 to 8 hr control w1118 embryos similar to previous reports ( Hayashi et al . , 2016; Mohn et al . , 2015 ) . In short , 100 µl of ovaries were dissected in PBS on ice . 100 µl of embryos were collected on grape juice agar plates and transferred to a mesh strainer . Following dechorionation in 50% bleach , embryos were washed under running tap water for at least 1 min or until bleach smell disappeared . Ovary and embryo samples were washed twice with ice-cold PBS and homogenized in 1 ml lysis buffer ( 10 mM HEPES pH 7 . 3 , 150 mM NaCl , 5 mM MgCl2 , 10% glycerol , 1% Triton x-100 , 1 mM DTT , 1 mM EDTA , 0 . 1 mM PMSF , 1× PI , 1:1000 RNasin [Promega] ) using a 2 ml Dounce homogenizer . Material was lysed with five strokes with a loose pestle and five strokes with a tight pestle on ice . Lysate was incubated for 1 hr at 4°C while rotating and centrifuged at full speed for 10 min to pellet debris . Supernatant was transferred to a new tube and protein concentration determined by Direct Detect ( Millipore ) . 1 mg of lysate per immunoprecipitation was used for the following steps . 50 µl Protein A Dynabeads ( Thermo Fisher Scientific ) were washed with lysis buffer three times for 3 min each . Washed beads were resuspended in 400 µl lysis buffer and 5 µl anti-Piwi ( Hannon Lab ) or rabbit IgG antibodies ( Abcam , ab37415 ) added . Following overnight incubation at 4°C while rotating , beads were washed three times for 5 min in 500 µl lysis buffer . Antibody-coupled beads were added to lysates and volume brought up to 1 ml with lysis buffer . The solution was incubated at 4°C overnight while rotating . Supernatant was removed and saved for quality control western blotting analysis . Beads were washed six times for 10 min with 1 ml wash buffer ( 10 mM HEPES pH 7 . 3 , 150 mM NaCl , MgCl2 , 10% glycerol , 1% Empigen BB Detergent [Merck] , 1× PI ) . For the first wash , 1 µl RNasin was added to the wash buffer and tubes were changed between each wash . 10% of beads were set aside for quality control and 90% resuspended in 1 ml Trizol ( Thermo Fisher Scientific ) and stored at −80°C until further processing . 100 flies were collected in a 1 . 5 ml tube and frozen at −80°C for at least 1 hr . High molecular weight genomic DNA was isolated using the Blood and Cell Culture DNA Mini kit ( Qiagen ) . Flies were homogenized using a rotating pestle on ice for 1 min . 700 µl G2 and 50 µl Proteinase K ( 800 U/ml , NEB ) were added to each tube and incubated at 50°C for 2 hr with occasional tube inversions . Tubes were spun at 5000 g for 10 min at 4°C and supernatant transferred to new tube avoiding debris . A Qiagen Genomic-tip 20/G was equilibrated with 1 ml QBT buffer and allowed to empty by gravity flow . The supernatant containing digested proteins and genomic DNA was added to the column and allowed to flow through . The column was washed three times with 1 ml QC buffer . Elution was carried out with 1 ml QF elution buffer and repeated once . Flow through was transferred to two new tubes ( 1 ml each ) and 700 µl isopropanol added . Tubes were inverted 10 times and centrifuged at full speed for 15 min at 4°C . The pellet was washed with 70% ethanol twice and air-dried for 5 min . 25 µl RNase-free water was added and DNA resuspended by flicking tube gently several times while incubating at 37°C for 2 hr . DNA was stored at 4°C . DNA was sheered using a Covaris S220 ( Covaris ) . 3 µg of genomic DNA was diluted in RNase-Free water and transferred to a AFA Fiber Crimp-Cap ( PN520052 , Covaris ) microtube . Sonication was carried out with the following settings: peak incident power ( W ) 105 , duty factor 5% , cycles per burst 200 , and treatment time 80 s . This resulted in sheared DNA fragments peaking at 500 bp . DNA was recovered using the QIAquick PCR Purification Kit ( Qiagen ) . RNA for RNA-seq and qRT-PCR experiments was isolated using the RNeasy Mini kit ( Qiagen ) with on-column DNA digestion ( RNase-free DNase Set , Qiagen ) according to the manufacturer’s recommendations . RNA for small RNA-seq experiments were isolated using Trizol following the manufacturer’s instructions . 1 µg of total RNA was used as input material for RNA-seq library preparation . The NEBNext Poly ( A ) mRNA magnetic Isolation Module ( NEB ) was used to isolate poly ( A ) RNAs . Libraries were generated with the NEBNext Ultra Directional RNA Library Prep kit for Illumina ( NEB ) according to the manufacturer's instructions . Small RNA libraries were generated as described previously ( Jayaprakash et al . , 2011 ) . In short , 19- to 28-nt-sized small RNAs were purified by PAGE from Piwi-bound RNA isolated from ovaries or embryos . Next , the 3' adapter ( containing four random nucleotides at the 5' end ) was ligated overnight using T4 RNA ligase 2 , truncated KQ ( NEB ) . Following recovery of the products by PAGE purification , the 5' adapter ( containing four random nucleotides at the 3' end ) was ligated to the small RNAs using T4 RNA ligase ( Abcam ) for 1 hr . Small RNAs containing both adapters were recovered by PAGE purification , reverse transcribed , and PCR amplified . WGS libraries were generated using the NEBNext Ultra II DNA Library Prep kit ( NEB ) according to the manufacturer’s recommendation with 1 µg input material . Three PCR amplification cycles were performed . Libraries were quantified using the Library Quantification Kit for Illumina ( Kapa Biosystems ) . Sequencing was performed by the Genomics Core facility at CRUK CI . RNA-seq , ChIP-seq , and small RNA-seq libraries were sequenced on an Illumina HiSeq 4000 according to the manufacturer’s recommendations using single-end 50 bp runs . WGS libraries were sequenced with paired-end 150 bp runs on Illumina HiSeq 4000 or NovaSeq . Reverse transcription was performed using the SuperScript IV reverse transcriptase Kit ( Thermo Fisher Scientific ) with 1 µg of total RNA . qRT-PCR was performed on a QuantStudio Real-Time PCR Light Cycler ( Thermo Fisher Scientific ) in technical triplicates . Expression of targets was quantified using the ddCT method ( Livak and Schmittgen , 2001 ) . Fold-change was calculated as indicated in the figure legends and normalized to rp49 . All primers are listed in Supplementary file 1 . 100 µl of control w1118 embryos for time points 0–2 hr , 5–7 hr , and 10–12 hr AEL were collected in three biological replicates on agar plates and dechorionated . Embryos were then lysed in lysis buffer ( 0 . 1% SDS , 0 . 1 M triethylammonium bicarbonate [TEAB] , 1× Halt Protease and Phosphatase Inhibitor [Thermo Fisher Scientific] ) using a rotating pestle on ice for 2 min or until entirely homogenized . Lysate was heated for 5 min at 90°C and probe sonicated for 20 s ( 20% power with pulse of 1 s ) . Debris was pelleted by centrifugation at full speed for 10 min at 4°C and supernatant transferred to a new tube . Protein concentration was measured using Bradford Assay ( BioRad ) . 100 µg protein was digested with trypsin overnight at 37°C . TMT chemical isobaric labeling was performed as described previously ( Papachristou et al . , 2018 ) . Peptide fractions were analyzed on a Dionex Ultimate 3000 UHPLC system coupled with the nano-ESI Fusion Lumos mass spectrometer ( Thermo Fisher Scientific ) . Embryos were collected for 30 min and dechorionated . Control embryos were transferred into a fine mesh strainer placed in a plastic dish and submerged in PBS . 1 M auxin solution was generated by diluting Indole-3-acetic acid ( IAA ) , a highly permeable small molecule as recently shown for Caenorhabditis elegans embryos ( Zhang et al . , 2015 ) , in water and stored protected from light at −20°C . Auxin-treated embryos were submerged in PBS with indicated auxin concentrations . Embryos were placed at 25°C for appropriate times and harvested for RNA experiments by transferring into 1 ml Trizol followed by RNA extraction . Embryos used for ChIP-seq were processed as described above . Raw fastq files contained 50 bp reads . The first and the last two bases of all reads were trimmed using fastx_trimmer ( http://hannonlab . cshl . edu/fastx_toolkit/ ) . Reads were first aligned to the consensus sequence for all D . melanogaster transposons using STAR ( Dobin et al . , 2013 ) allowing random allocation of multimappers . Unmapped reads were further aligned to D . melanogaster genome release 6 ( dm6 ) keeping uniquely mapping reads . Generated bam files for RNA-seq were further split in reads originating from sense and antisense genomic strands using samtools view options -f 0x10 and -F 0x10 for sense and antisense reads , respectively ( Li et al . , 2009c ) . Indexes were generated using samtools index function . Coverage files were generated using bamCoverage with normalization mode --normalizeUsing CPM ( Ramírez et al . , 2014 ) and applying a scaling factor ( --scaleFactor ) . Scaling factors for individual files were calculated by dividing the sum of mapped reads contained in the file by the sum of all transposon and dm6 mapping reads of the corresponding library . Reads mapping to protein-coding genes were counted with htseq ( Anders et al . , 2015 ) using a feature file downloaded from Ensembl release BDGP6 . 22 . Reads mapping to individual transposons were counted with a custom script using samtools idxstats function to extract reads mapping to individual sequences of the reference genome/transposon consensus sequence . Count files for RNA-seq time-course experiments generated as described above were normalized to rpm to account for differences in library size and allow comparability between time points . Heatmaps displaying expression profiles of genes and transposons during embryogenesis show the mean expression values of the biological replicates , while bar graphs display the individual data points as well as the mean expression and standard deviation . Bar graphs and heatmaps were plotted in R using ggplot2 . RNA-seq experiments comparing auxin- and PBS-treated embryos of the same stage and collection were analyzed using differential expression quantification methods allowing for statistical evaluation of differences between RNA output as a direct result of auxin treatment . Differential expression analysis was performed using DESeq2 ( Love et al . , 2014 ) . MA plots show base mean RNA expression across conditions and were calculated as previously described by Love and colleagues . ChIP-seq reads were normalized by library size and rpm calculated for concatenated replicates using the deepTools2 bamCoverage function ( Ramírez et al . , 2016 ) with bin size 10 bp . MA plots displaying H3K9me3 signal intensity fold-changes between auxin-treated and control samples were calculated using DESeq2 for individual replicates ( n = 3 ) . Metaplots flanking euchromatic transposon insertion sites were calculated using computeMatrix scale-region function from deepTools2 with bin size 10 bp . All scripts used for sequencing analysis are available on GitHub ( https://github . com/mhf27/hannon_roo_fabry2021 , copy archived at swh:1:rev:f088572638701e0ae6f13d9e025642b9476146b5; Fabry , 2021 ) . Reads from small RNA-seq libraries were adapter clipped using fastx_clipper with settings -Q33 -l 15 -a AGATCGGAAGAGCACACGTCT . The first and last four bases of adapter clipped reads were trimmed using seqtk trimfq ( https://github . com/lh3/seqtk; Li , 2021 ) . Only high-quality reads with length between 19 and 31 bp were used for further analysis . Small RNAs were aligned as described above and size profiles plotted in R . Mappability track for dm6 with 50 bp resolution was calculated according to a previously published method ( Derrien et al . , 2012 ) . The de novo transposon insertion calling for the homozygous control w1118 strain and our line carrying both GFP-AID-Piwi and OsTIR1 was performed using the TEMP algorithm ( Zhuang et al . , 2014 ) . In brief , ~500 bp genomic DNA fragments were amplified and sequenced generating 150 bp paired-end reads , which were aligned to dm6 using BWA ( Li and Durbin , 2009b ) . Reads with only one mate aligned to dm6 were extracted and the unmapped mate uniquely aligned to transposon consensus sequences in order to ensure correct directionality calling . Calculated insertion sites were extracted from generated GTF files if they were supported by reads on both sides ( 1p1 ) . Transposon insertion files containing coordinates as well as statistical information have been submitted to GEO ( GSE160778 ) . Euchromatic regions ( chr2R:6460000–25286936 , chr2L:1–22160000 , chr3L:1–23030000 , chr3R:4200000–32079331 , chrX:250000–21500000 ) were defined by measuring H3K9me3 signal genome-wide in sliding windows of 10 kb bins and calculating signal enrichment over input . We identified a total of 632 euchromatic TE insertions in w1118 and 1738 in our degron strain ( GFP-AID-Piwi; OsTIR1 ) . The protein database used to identify peptides from Drosophila genes and transposons was generated by merging an existing database downloaded from FlyBase ( dmel-r6 . 24 . fa ) with translated ORFs of transposons . ORFs were predicted and translated using prodigal ( https://github . com/hyattpd/Prodigal; Hyatt , 2020 ) . ORFs with less than 300 amino acids were removed using seqtk -L 300 and the file was converted to fasta format . Functional analysis of protein sequences was performed using the InterPro web application ( https://www . ebi . ac . uk/interpro/ ) . Protein domains and families for ORF encoded by roo transcripts were predicted using default settings . Raw data files were processed according to previous reports ( Papachristou et al . , 2018 ) . Spectral . raw files were analyzed with the SequestHT search engine on Thermo Scientific Proteome Discoverer 2 . 1 for peptide and protein identification . Data was searched against a modified FlyBase protein database with the following parameters: precursor mass tolerance 20 ppm , fragment mass tolerance 0 . 5 Da . Dynamic modifications were oxidation of methionine residues ( +15 . 995 Da ) and deamidation of asparagine and glutamine ( +0 . 984 Da ) , and static modifications were TMT6plex at any amino-terminus , lysine ( +229 . 163 Da ) , and methylthio at cysteine ( +45 . 988 Da ) . The Reporter Ion Quantifier node included a TMT 6plex ( Thermo Scientific Instruments ) Quantification Method , for MS3 scan events , HCD activation type , integration window tolerance 20 ppm , and integration method Most Confident Centroid . Peptides with an FDR > 1% were removed . The downstream workflow included signal-to-noise ( S/N ) calculation of TMT intensities . Level of confidence for peptide identifications was estimated using the Percolator node with decoy database search . Strict FDR was set at q-value < 0 . 01 . Processed data files were analyzed as described in a previous publication ( Papachristou et al . , 2018 ) using qPLEXanalyzer in R with multimapping peptides included in the analysis . Bar graphs showing protein intensities for Piwi and volcano plots with indicated comparisons were plotted in R using ggplot2 . Statistical tests used for individual experiments are indicated in the figure legends . Statistical analyses applied to hatching rates , qPCR datasets , and ChIP-seq signal intensity were calculated by unpaired ( two-sample ) t-test . Significance of TMT mass spectrometry data was calculated according to Papachristou et al . , 2018 . Differential expression of RNA-seq experiments and differential enrichment of ChIP-seq experiments was calculated using DeSeq2 using adjusted p values as described in Love et al . , 2014 . The number of biological replicates is indicated in the figure legends . | Maintaining the integrity of DNA , which encodes all of the instructions necessary for life , is essential for ensuring the survival of a species , especially when genetic information is transferred across generations . DNA , however , contains selfish , mobile elements , called transposons , that move around the genome , hence their nickname ‘jumping genes’ . Their movement , a process by which these elements also multiply within genomes , can muddle an organism’s DNA if the transposon happens to land in the middle of a gene , creating a mutation which renders the gene inactive . Transposons have also been linked to the development of cancer , which is a group of diseases driven by accumulating genetic mutations . Animals have evolved various ways of protecting their DNA against transposons . These are especially important in developing egg cells and sperm , known collectively as germ cells . These cells can produce small fragments of RNA , a molecule similar to DNA , which are able to identify and disarm transposons . While it is known that these small RNAs effectively protect adult gonads from DNA damage , it has been unclear how germ cells formed during the beginning of life are protected . To find out more , Fabry et al . used a combination of genetic sequencing , protein binding and imaging studies to look at the activity of small RNAs , called piRNAs , which are passed on from the mother to her progeny . By studying the gene expression levels in fruit fly embryos , Fabry et al . showed that certain transposons become highly active in the first few hours of embryo development , posing a potential threat to DNA integrity . The experiments also identified clear signs in the embryos of an active mechanism for controlling transposons that resembles the small RNA system known from adult germ cells . Fabry et al . removed the piRNAs from the embryos and found that without piRNAs , transposons were more active . This indicates a direct role of these small RNAs in controlling transposons in early development and evidence for a maternally inherited defence system in early embryos . This study provides insights into the control of transposons in fly embryos . More research is needed to find out whether these embryonic mechanisms are conserved in other animals , including humans . Studying the intrinsic mechanisms that prevent DNA damage and protect our genome could , in time , help to identify new approaches to possibly treat and prevent diseases involving genetic mutations . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"chromosomes",
"and",
"gene",
"expression",
"developmental",
"biology"
] | 2021 | Maternally inherited piRNAs direct transient heterochromatin formation at active transposons during early Drosophila embryogenesis |
Uterine leiomyomas ( ULs ) are benign tumors that are a major burden to women’s health . A genome-wide association study on 15 , 453 UL cases and 392 , 628 controls was performed , followed by replication of the genomic risk in six cohorts . Effects of the risk alleles were evaluated in view of molecular and clinical characteristics . 22 loci displayed a genome-wide significant association . The likely predisposition genes could be grouped to two biological processes . Genes involved in genome stability were represented by TERT , TERC , OBFC1 - highlighting the role of telomere maintenance - TP53 and ATM . Genes involved in genitourinary development , WNT4 , WT1 , SALL1 , MED12 , ESR1 , GREB1 , FOXO1 , DMRT1 and uterine stem cell marker antigen CD44 , formed another strong subgroup . The combined risk contributed by the 22 loci was associated with MED12 mutation-positive tumors . The findings link genes for uterine development and genetic stability to leiomyomagenesis , and in part explain the more frequent occurrence of UL in women of African origin .
Uterine leiomyomas ( ULs ) , also known as fibroids or myomas , are benign smooth muscle tumors of the uterine wall . They are extremely common; approximately 70% of women develop ULs before menopause ( Stewart et al . , 2017 ) . The symptoms , occurring in one fifth of women , include excessive menstrual bleeding , abdominal pain and pregnancy complications ( Stewart et al . , 2017 ) . In most cases , durable treatment options are invasive ( Stewart , 2015 ) . ULs cause a substantial human and economic burden , and the annual cost of treating these tumors has been approximated to be as high as $34 billion in the United States , higher than the combined cost of treating breast and colon cancer ( Cardozo et al . , 2012 ) . Earlier studies have indicated strong genetic influence in UL susceptibility based on linkage ( Gross , 2000 ) , population disparity ( Wise et al . , 2012 ) and twin studies ( Luoto et al . , 2000 ) . The most striking UL predisposing condition thus far characterized is hereditary leiomyomatosis and renal cell cancer ( HLRCC ) syndrome , caused by high-penetrance germline mutations in the Fumarate hydratase ( FH ) gene ( Multiple Leiomyoma Consortium et al . , 2002; Launonen et al . , 2001 ) . Genome-wide association studies ( GWAS ) have proposed several low-penetrance risk loci but few unambiguous predisposing genes have emerged . Cha et al . reported loci in chromosome regions 10q24 . 33 , 11p15 . 5 and 22q13 . 1 based on a Japanese patient cohort ( Cha et al . , 2011 ) . The 11p15 . 5 locus - near the Bet1 golgi vesicular membrane trafficking protein like ( BET1L ) gene - was later replicated in Caucasian ancestry ( Edwards et al . , 2013a ) . The 22q13 . 1 locus has been replicated in Caucasian , American and Saudi Arabian populations suggesting trinucleotide repeat containing 6B ( TNRC6B ) as a possible target gene ( Edwards et al . , 2013a; Aissani et al . , 2015; Bondagji et al . , 2017 ) . Further UL predisposition loci have been suggested at 1q42 . 2 and 2q32 . 2 by Zhang et al ( Zhang et al . , 2015 ) . and , at 3p21 . 31 , 10p11 . 21 and 17q25 . 3 by Eggert et al ( Eggert et al . , 2012 ) . A recent work reported cytohesin 4 ( CYTH4 ) at 22q13 . 1 as a novel candidate locus in African ancestry ( Hellwege et al . , 2017 ) . While multiple loci and genes have been implicated through these valuable studies it is not straightforward to connect any of them mechanistically to UL development . Most ULs show somatic site-specific mutations at exons 1 and 2 of the mediator complex subunit 12 ( MED12 ) gene ( Mäkinen et al . , 2011; Heinonen et al . , 2014 ) . These observations together with further scrutiny of driver mutations , chromosomal aberrations , gene expression , and clinicopathological characteristics have led to identification of at least three mutually exclusive UL subtypes; MED12 mutant , Fumarate Hydratase ( FH ) deficient , as well as High Mobility Group AT-Hook 2 ( HMGA2 ) overexpressing lesions ( Mehine et al . , 2016 ) . Here we report the most powerful GWAS on uterine leiomyoma to date , and novel genome-wide significant UL susceptibility loci with plausible adjacent predisposition genes . These genes associate UL genesis to two distinct biological mechanisms: Genome stability related processes are implicated by genes Tumor Protein P53 ( TP53 ) and ATM Serine/Threonine Kinase ( ATM ) together with the telomere maintenance genes Telomerase Reverse Transcriptase ( TERT ) , Telomerase RNA Component ( TERC ) and STN1-CST Complex Subunit ( OBFC1 ) . The other prominent group is genes relevant for genitourinary development , specifically Wnt Family Member 4 ( WNT4 ) , Wilms Tumor 1 ( WT1 ) , Spalt Like Transcription Factor 1 ( SALL1 ) , Estrogen Receptor 1 ( ESR1 or ERα ) , Growth Regulation By Estrogen In Breast Cancer 1 ( GREB1 ) , Forkhead Box O1 ( FOXO1 ) , Doublesex and Mab-3 Related Transcription Factor 1 ( DMRT1 ) and CD44 Molecule ( CD44 ) . Our analysis of the X chromosome identifies a risk allele near MED12 that drives UL tumorigenesis towards somatic MED12 mutations . We report altogether 22 genome-wide significant susceptibility loci and compile them into a polygenic risk score . The UL association is then replicated in six independent cohorts of different ethnic origins: individuals of African origin are characterized by the highest risk load . Finally , we investigate the risk alleles’ association to clinical features , molecular UL subtypes , telomere length , gene expression and DNA methylation .
Figure 1 provides an outline of this study . At discovery stage 1 , 428 SNPs emerging from 22 distinct genetic loci passed the genome-wide significance level of 5 × 10−8 . Figure 2 displays a Manhattan plot of these associations ( 15 , 453 UL cases and 392 , 628 controls; linear mixed model ) . Two of the significant loci ( 359/1 , 428 SNPs ) were found on the X chromosome . After linkage disequilibrium ( LD; r2 ≤0 . 3 ) pruning the significant SNPs , a total of 50 LD-independent associations remained: the resulting SNPs are given in Appendix 1—table 2 , and the lead SNPs are summarized in Table 1 . Appendix 1—figure 1 displays the regional structure of each locus together with flanking association values , linkage disequilibrium ( LD ) and genome annotation . Annotation tracks are included for tissue-specific data on open chromatin , topologically associating domains ( TAD ) and other regulatory features ( details in Supplementary Methods ) . A polygenic risk score ( Abraham and Inouye , 2015 ) was compiled based on the discovery stage associations . After LD pruning ( r2 ≤0 . 3 ) the discovery-stage SNPs , 50 SNPs from the 22 distinct loci passed for the initial genomic risk score ( GRS; Appendix 1—table 4 ) . The SNP weights were based on UKBB log-odds . We applied this initial GRS of 50 SNPs to the Helsinki cohort and identified a significant association to the UL phenotype ( p=8 . 3 × 10−10; adjusted p=1 . 1 × 10−8; one-tailed Wilcoxon rank-sum; W = 1 . 69 × 106; 457 cases and 8899 female controls ) . The second stage GWAS combined the UKBB and Helsinki cohorts for a meta-analysis approach . The genome-wide statistics revealed rs117245733 , at 13q14 . 11 , as the only SNP with a suggestive ( p<10−5 ) association in both the UKBB ( OR = 1 . 26; p=4 . 2 × 10−9 ) and Helsinki ( OR = 1 . 82; p=8 . 1 × 10−6 ) cohorts . Figure 3 shows the regional structure and combined association ( fixed effect model p=3 . 1 × 10−12 ) at the locus: the SNP resides on a gene poor region , at a conserved element that displays activity in uterus-specific H3K27ac and DNaseI data ( see ENCODE track details in Supplementary Methods ) . The SNP is independent of the group of associations at FOXO1 ( r2 = 0 . 0; Figure 3 ) . The meta-analysis identified altogether 112 genome-wide significant SNPs not seen in the discovery stage: seven of those were LD-independent ( r2 ≤0 . 3; Appendix 1-table 3 ) and their UKBB log-odds weights were appended to the initial GRS model . The final GRS model of 57 SNPs and their UKBB-based weights is given in Appendix 1—table 4 . Supplementary file 1 gives further details on the meta-analysis results and heterogeneity statistics . The third stage replicated the observations in NFBC and in five different ethnic groups . In NFBC , the SNP identified in the stage two meta-analysis , rs117245733 at 13q14 . 11 , was replicated ( p=0 . 034; linear mixed model; OR = 1 . 50; 95% CI 1 . 03 – 2 . 19 ) . Additional analysis of all 57 SNPs did not reveal other associations: Supplementary file 2 gives further details on the meta-analysis results and heterogeneity statistics . The association between the GRS and UL phenotype was significant ( p=1 . 1 ×10−5; Wilcoxon rank-sum; adjusted p=1 . 1 × 10−4; one-tailed; W = 4 . 7 × 105 ) in NFBC . These case-control distributions of GRS are displayed in Figure 4 . UL susceptibility is known to vary by ancestry ( Wise et al . , 2012 ) . Five different ethnic groups - African , Caribbean , Irish , Indian and ‘other white’ background - were available from the UKBB cohort . A total of 2 , 212 UL cases and 21 , 054 female controls could be utilized for replication ( Appendix 1—table 1 ) . Supplementary file 3 includes all the 57 SNPs and their summary statistics in these five cohorts , together with heterogeneity estimates . Due to the small cohort sizes , none of the single-SNP associations passed genome-wide significance . The GRS model replicated with a significant phenotype association in all five ethnicities ( Appendix 1-Table 6 ) . A summary of test statistics , GRS distributions and the numbers of cases and controls for each population is given in Figure 4 . A more detailed summary of the GRS model and receiver operating characteristic ( ROC ) curve of each cohort are given in Appendix 1—figure 5 . The self-reported ‘Black African’ ( mean GRS 4 . 83 ) had an outstanding risk-load compared to Caucasian ( self-reported ‘White Irish’; mean GRS 4 . 04 ) background ( Figure 4; Wilcoxon rank-sum p<10−15 ) . As expected ( Wise et al . , 2012 ) , the African ethnicity displayed an increased prevalence ( 19% ) compared to the Irish ( 6% ) . Assuming that the observed GRS weights have a linear relationship to the true risk , the GRS difference between African and Irish ancestries explains 9 . 0% of the increased prevalence in the African population . Similar population-specific GRSs could be estimated for the seven populations in the gnomAD database ( Appendix 1—table 7 ) . Appendix 1—figure 6 shows an overview of the GRS for each of the populations . African ancestry has been shown to carry a two-to-three times higher prevalence when compared to Caucasian ancestry ( Wise et al . , 2012 ) . Based on the gnomAD frequencies , the increased GRS of African ancestry explains between 8 – 16% of this population difference . The number of ULs per patient had a significant positive association to GRS ( negative binomial regression p=0 . 001; adjusted p=0 . 0032; rate ratio 1 . 25; 95% CI 1 . 09 – 1 . 43 for one-unit increase in GRS; Appendix 1—figure 7 ) . No association was found between GRS and age at hysterectomy ( Appendix 1—table 6 ) . Testing the 57 GRS SNPs separately did not reveal any associations that pass FDR ( Appendix 1—table 5 ) . Our UL set of 1481 lesions included 1159 ( 78% ) mutation-positive and 322 mutation-negative tumors . The occurrence of mutant tumors did not distribute evenly among the 457 patients . In total 221 ( 48% ) and 123 ( 27% ) patients had all their tumors identified as either MED12-mutation-positive or -negative , respectively , suggesting that genetic or environmental factors contribute to the preferred UL type in affected individuals , as previously observed ( Mäkinen et al . , 2011 ) . Indeed , the 221 mutation positive patients were found to have a significantly higher GRS ( Wilcoxon rank-sum p=7 . 9 × 10−4; adjusted p=0 . 0032; two-sided; W = 1 . 6 × 104 ) . This difference in GRS distributions is visualized in Figure 4 . Comparison against the population controls ( n = 8899 females ) revealed that the above-mentioned patient groups differ by their effect size: the MED12-mutation-positive ( 221 ) subset of patients had an odds ratio of 2 . 28 for one-unit increase in GRS ( 95% CI 1 . 80 – 2 . 88 ) compared to the controls , while the mutation-negative ( 123 ) subset had an odds ratio of 1 . 20 ( 95% CI 0 . 88 – 1 . 66 ) . Thus , the majority of the compiled case-control association signal had arisen from the MED12-mutation-positive subset of the patients . The number of MED12-mutation-positive tumors per patient had a significant positive association to GRS ( p=3 . 2 × 10−4; adjusted p=0 . 002; negative binomial model rate ratio 1 . 43; 95% CI 1 . 13 – 3 . 83 for one-unit increase in GRS; Appendix 1—figure 8 ) . No association between the number of MED12-mutation-negative tumors and GRS was found ( adjusted p=0 . 053; Appendix 1—figure 8 ) . The GWAS signal near MED12 was inspected for any associations to somatic MED12 mutations . Strikingly , the risk allele ( rs5937008 ) did significantly increase the number of MED12-mutation-positive tumors ( p=0 . 0087; negative binomial model rate ratio 1 . 23; 95% CI 1 . 05 – 1 . 44 ) . Among our 457 patients , the median number of MED12-mutation-positive tumors increased from one to two for the risk allele carriers . The risk locus and its effect on the number of MED12-mutation-positive tumors is visualized in Figure 2 . An additional analysis of each of the 57 GRS SNPs did not reveal any further associations ( Appendix 1—table 5 ) . All the genome-wide significant SNPs from UKBB and the meta-analysis stage ( altogether 1 , 540 SNPs ) were tested with a permutation based approach . In total 34 and 24 genes passed the local permutation significance threshold ( p<0 . 05 ) for tumor and matched myometrium data , respectively ( Appendix 1—table 8 ) . Among the hits in tumors were WNT4 ( p=0 . 01; permutation test ) and CDC42 ( p=0 . 03 ) at 1 p , TNRC6B ( p=0 . 02 ) at 22q , FOXO1 ( p=0 . 03 ) at 13q , and DMRT1 ( p=0 . 04 ) at 9 p . None of the local associations passed a genome-wide FDR of 10% . No significant association was observed between the risk allele and MED12 expression ( rs5936989; Appendix 1—figure 11 ) . The full list of eQTL statistics can be found from Supplementary file 4 . Our analysis of the 57 GRS SNPs revealed altogether 17 , 030 ( 9 , 466 in tumors and 7564 in matched myometrium ) cis methylation quantitative trait loci ( cis-meQTL ) with nominal p<0 . 05 . Of these , 145 passed a 10% FDR . Of the plausible predisposition genes , FOXO1 , TERT and WNT4 showed significant meQTL associations ( Appendix 1—table 9 ) . All the cis-meQTLs and annotation for their genomic context are in Supplementary file 5 . The UL predisposition loci at TERT , TERC and OBFC1 were examined for an effect on telomere length . Overall the telomere length was significantly shorter in tumors than in adjacent matched myometrium ( p=0 . 01; Kruskal-Wallis ) , as previously reported ( Rogalla et al . , 1995; Bonatz et al . , 1998 ) . One of the risk alleles at TERT ( rs2736100 ) was significantly associated with shorter telomere length ( p=0 . 01; Kruskal-Wallis ) ( Appendix 1—figure 12 ) . Adjusting for the patient age did not explain away the association . The association was not seen in myometrium . The other two LD-independent SNPs at TERT , rs72709458 and rs2853676 , or the SNPs at TERC ( rs10936600 ) and OBFC1 ( rs1265164 ) did not show association to telomere length ( p=0 . 24 , p=0 . 57 , p=0 . 07 and p=0 . 48 , respectively; Kruskal-Wallis ) . The combined effect of TERT ( rs72709458 , rs2736100 , rs2853676 ) , TERC ( rs10936600 ) and OBFC1 ( rs1265164 ) had a negative trend with telomere length ( p=0 . 055; linear model 95% CI −408 . 5 – 4 . 7 per one risk allele; see Appendix 1—figure 13 ) . In whole genome sequencing data , no association was detected between genotype and the number of somatic structural variants . The DEPICT framework ( Pers et al . , 2015 ) was ran using the genome-wide significant SNPs from the UKBB cohort , in total 1 , 069/1 , 428 autosomal SNPs . The resulting target gene prioritization , pathway enrichment and tissue enrichment results are given in Supplementary file 6 . The analysis did not reveal any significant enrichments with the exception of one pathway related to induced stress . ATM was the highest ranking target gene , and uterus/myometrium were among the highest ranking tissue types . Previous UL association studies ( Cha et al . , 2011; Zhang et al . , 2015; Eggert et al . , 2012; Hellwege et al . , 2017 ) have reported altogether seven genome-wide significant UL susceptibility loci . Two out of the seven loci - that is , 22q13 . 1 ( at TNRC6B ) and 11p15 . 5 ( at BET1L ) - replicated in UKBB using 15 , 453 cases and 392 , 628 controls . Cha et al ( Cha et al . , 2011 ) . highlight OBFC1 ( at 10q24 . 33 ) as a candidate gene and , while the SNP that they reported does not replicate in UKBB , the OBFC1 region is identified in our discovery stage ( rs1265164; Table 1 ) . See Appendix 1—table 10 for a summary of these results .
The UK Biobank genotype-phenotype data revealed 22 novel predisposition loci for UL , most of them in close proximity to highly plausible predisposition genes . The combined UL risk of these loci was replicated in a subsequent analysis of the polygenic risk score ( GRS ) in six independent cohorts from different ethnic backgrounds . Our multi-ethnic replication implies that the discovered loci are indeed involved in UL development , and the early UL association studies have likely been underpowered to detect them . Three previously reported loci , at OBFC1 ( Cha et al . , 2011 ) , TNRC6B ( Cha et al . , 2011; Edwards et al . , 2013a; Aissani et al . , 2015; Bondagji et al . , 2017 ) and BET1L ( Cha et al . , 2011; Edwards et al . , 2013b ) , were also validated , however , the mechanistic connection to UL development remains obscure for the latter two . Though simple association is not sufficient to formally prove causality , 14 out of the 22 risk loci harbor plausible predisposition genes . These genes can be divided into two groups: TERT , TERC , OBFC1 ( all involved in telomere length ) , ATM and TP53 guard stability of the genome . ESR1 , GREB1 , WT1 , MED12 , WNT4 , FOXO1 , DMRT1 , SALL1 , and CD44 play a role in genitourinary development . Estrogen is a well-known inducer of UL growth ( Borahay et al . , 2015 ) . The top association at 6q25 . 2 ( rs58415480 ) resides within intron 107 of Spectrin Repeat Containing Nuclear Envelope Protein 1 ( SYNE1 ) , 130 kb downstream of ESR1 , the latter being the only gene that resides completely within the topologically associating domain ( TAD; Appendix 1—figure 1 ) . While the role of estrogen in leiomyomagenesis has been firmly established , this is the first genetic evidence to this end . The lead SNP at 2 p resides in the third exon of the gene GREB1 . GREB1 is an essential regulatory factor of ESR1 ( Mohammed et al . , 2013 ) . WT1 , WNT4 and FOXO1 are central factors in uterine development and in the preparation for pregnancy ( decidualization ) in endometrium ( Biason-Lauber and Konrad , 2008; Hill , 2018; Kaya Okur et al . , 2016; Tamura et al . , 2017 ) . Perturbations in their function are known to have neoplastic potential . The strongest association at 11p13 ( rs10835889 ) is 40 kb downstream of the closest gene WT1 , at a region with enhancer activity ( Appendix 1—figure 1 ) . WT1 is a transcription factor that acts as both a tumor suppressor and an oncogene ( Yang et al . , 2007 ) . The lead SNP at 1p36 . 12 ( rs2235529 ) resides at the second intron of WNT4 . The risk allele is associated with suggestive upregulation of WNT4 ( Figure 5C ) . WNT4 is known to be overexpressed in uterine leiomyomas with MED12 mutations ( Markowski et al . , 2012 ) , and knock-down of MED12 in UL cells reduces WNT4 expression ( Al-Hendy et al . , 2017 ) . The risk locus in 1p36 . 12 was also associated with several meQTLs suggesting that methylation may have a role in WNT4 regulation ( Figure 5 ) . WNT4 encodes a signaling protein that has a crucial role in sex-determination ( Vainio et al . , 1999 ) , and the WNT signaling pathway has a well-established role in various malignancies such as breast and ovarian cancer ( Peltoketo et al . , 2004 ) . Of note , recent GWAS on gestational duration suggested that binding of the estrogen receptor at WNT4 is altered by rs3820282 ( r2 = 0 . 92 with our lead SNP ) ( Zhang et al . , 2017 ) . Both WNT4 and FOXO1 are decidualization markers regulated by ESR1 ( Kaya Okur et al . , 2016 ) . Though these considerations support WNT4 as a candidate predisposition gene at this locus , the near-by CDC42 has been shown to play a role in uterine pathology , in particular endometriosis ( Powell et al . , 2016 ) , and should not be overlooked in further work . Also MED12 has been implicated in uterine development in a mouse model ( Wang et al . , 2017a ) . DMRT1 is a transcription factor associated with male sex-development ( Lindeman et al . , 2015 ) . CD44 is a plausible fibroid stem cell marker ( Mas et al . , 2015 ) . Mutations in SALL1 and a deletion at the GWAS signal have been associated with Townes-Brocks syndrome , a condition associated with kidney malformations ( Stevens and May , 2016 ) . Thus genes involved in genitourinary development are strikingly associated with UL predisposition . ATM , TP53 , TERT , TERC and OBFC1 could be involved in uterine neoplasia predisposition through genetic instability and telomere maintenance . The lead SNP at 11q ( rs141379009 ) resides in the 22nd intron of ATM , and the SNP at 17 p in the 3′-untranslated region of TP53 . ATM and TP53 are involved in DNA damage response ( Guleria and Chandna , 2016 ) , and they are among the relatively few genes that have been found to be recurrently mutated in leiomyosarcoma ( Lee et al . , 2017 ) . TERT and TERC encode subunits of the telomerase enzyme , which guards chromosomal stability by elongating telomeres ( Blasco , 2005 ) . In addition OBFC1 has been associated with telomere maintenance ( Lee et al . , 2013 ) . TERT is expressed in germ cells as well as in many types of cancers ( Blasco , 2005 ) . The neoplasia predisposing effect of the risk alleles at the TERT locus ( rs72709458; rs2736100; rs2853676 ) has been overwhelmingly documented ( Appendix 1—table 11 ) . Previous studies have reported contradicting observations on the effect of rs2736100 on telomere length ( Liu et al . , 2014; ENGAGE Consortium Telomere Group et al . , 2014; Lan et al . , 2013; Melin et al . , 2012; Choi et al . , 2015 ) . ULs have been shown to display shortened telomeres ( Rogalla et al . , 1995; Bonatz et al . , 1998 ) , potentially provoking chromosomal instability as the lengths of chromosome telomeres are diminished . In our patient cohort , the risk allele at TERT ( rs2736100 ) is significantly associated with shorter telomere length ( Appendix 1—figure 12 ) , whereas the combined effect of SNPs at TERT , TERC and OBFC1 did not reach statistical significance . GRS associated merely with a susceptibility to the most common UL subtype , MED12 mutation positive tumors . Indeed it has been known that MED12-mutation-positive tumors do not distribute randomly among patients ( Mäkinen et al . , 2011 ) , and our data provide at least a partial explanation to this intriguing finding . An outstanding susceptibility locus was identified 250 kb upstream of MED12: our in-house patient cohort - together with a mutation-screening of their 1481 tumors - revealed that the risk allele could facilitate selection of somatic MED12 mutations . It may be that environmental factors contribute more significantly to genesis of MED12 wild-type lesions . In our recent study this tumor type was associated with a history of pelvic inflammatory disease , and thus infectious agents could be one underlying factor ( Heinonen et al . , 2017 ) . Obviously , also the power of GWAS to detect genetic associations to rare UL subtypes – such as the HMGA2 overexpressing or FH deficient subtypes – is reduced . This work highlights several new genetic cornerstones of UL formation , highlights genitourinary development and maintenance of genomic stability as key processes associated with it , and represents another step towards a much-improved understanding of its molecular basis . The proposed risk score can stratify the female population to low and high-risk quartiles that differ by two-fold in their UL risk . The population-specific risk score was inflated towards the African and Caribbean cohorts , which connects the predisposition loci to the excess UL prevalence in these ethnicities . While the increased risk appears minor on an individual level , the population-level burden to women’s health arising from these risk loci is highly significant considering the incidence of the condition . Together with the recent progress in molecular tumor characterization and subclassification , the identification of the genetic components of UL predisposition should pave the way towards more sophisticated prevention and management strategies for these extremely common tumors . The risk SNP with the most immediate potential value is that at estrogen receptor alpha , and our findings should fuel much further work on the interplay between individual germline genetics , endogenous and exogenous hormonal exposure , and occurrence and growth rate of UL .
Figure 1 provides an outline of the four stages that were implemented . The discovery stage was conducted with UK Biobank resources ( UKBB; project #32506; accessed on April 10 , 2018 ) . The resource included pre-imputed genotypes ( version 3; March 2018 ) for a total of 487 , 409 samples ( 486 , 757 samples for the X chromosome ) and 96 million SNPs . The background information on the imputation and data quality control ( QC ) can be found through the UKBB documentation ( www . ukbiobank . ac . uk ) . The UL cases were identified on the basis of both the self-reported uterine leiomyoma ( UL ) phenotype ( UKBB data-field 20002: Non-cancer illness code 1351 ) and International Classification of Diseases ( ICD ) codes ( data-fields 41202 – 41205: Main and secondary diagnosis for ICD10 code D25 and ICD9 code 218 ) . These phenotype data resulted in a total of 20 , 106 UL cases prior to any sample/genotype QC . Sample QC was based on the UKBB annotation as follows . In total 409 , 692 samples passed the initial QC on ethnic grouping ( UKBB data-field 22006 ) : self-identified as ‘White British’ , and similar genetic ancestry based on a principal component analysis ( PCA ) of the genotypes . Further sample QC excluded excess kinship ( field 22021; 408 , 797 samples passed ) , sex-chromosome aneuploidy ( field 22019; 408 , 241 ) and inconsistent gender ( fields 31 and 22001 , and one male with self-reported ULs; 408 , 081 ) . In total 15 , 453 UL cases and 392 , 628 population-matched controls ( 205 , 157 females and 187 , 471 males ) passed all these criteria . Raw genotype calls ( UKBB version 2; Affymetrix UK BiLEVE Axiom , or Affymetrix UKBB Axiom array ) were available for 805 , 426 SNPs: after filtering out low genotyping rate ( <95% ) , Hardy-Weinberg equilibrium ( p<10−10 ) and minor allele frequency ( MAF ) <0 . 001 , the remaining 611 , 887 autosomal genotypes were used to train the mixed model for association testing . Imputed SNPs with MAF <0 . 001 and imputation score ( INFO ) <0 . 3 were excluded . Further SNPs were excluded due to imputation panel differences between cohorts , and the remaining 8 . 3 million SNPs ( Haplotype Reference Consortium , HRC1 . 1 panel ) were tested for case-control association with BoltLMM ( version 2 . 3 . 2 ) ( Loh et al . , 2015 ) . The default linear , infinitesimal mixed model was used to adjust for any underlying population structure . The model included categorical covariates for the 22 UK Biobank assessment centres and two genotyping arrays . The second stage meta-analysis utilized the genome-wide summary statistics from UKBB and the Helsinki cohort of 457 UL cases and 15 , 943 controls . Details on the Helsinki cohort’s imputation , sample and genotype QC are given in the Supplementary Methods . A total of 8 . 3 million SNPs passed imputation QC and were utilized in the meta-analysis with PLINK ( version 1 . 90b3i ) ( Chang et al . , 2015 ) . The SNPs were tested for association in six independent cohorts: Northern Finland Birth Cohort ( NFBC ) and five non-overlapping subsets of UKBB . In addition to the single-SNP association tests , a polygenic risk score ( Abraham and Inouye , 2015Abraham and Inouye , 2015 ) was compiled as follows . The genomic risk score ( GRS ) was computed as a sum over SNP dosages weighted by their observed log-odds: LD pruning ( r2 ≤0 . 3 ) was applied in the order of UKBB association , and the remaining , genome-wide significant SNPs were chosen for the GRS . The log-odds weights were taken from the UKBB statistics ( i . e . logarithm of the dosage-based ORs ) . The resulting GRS model was evaluated using R ( 3 . 3 . 1 ) and the packages PredictABEL ( 1 . 2 – 2 ) and MASS ( 7 . 3 – 45 ) . The Northern Finland Birth Cohort ( NFBC ) had in total 459 UL cases and 4943 controls; details of the imputation , sample and genotype QC are given in the Supplementary Methods . Five non-overlapping , self-reported population-strata were available from UKBB ( data-field 21000 ) and could be utilized as an independent replication: the five self-reported ancestries were ‘Black African’ , ‘Black Caribbean’ , ‘Indian’ , ‘White Irish’ and ‘Other white background’ . Sample QC excluded excess kinship ( field 22021 ) , sex-chromosome aneuploidy ( field 22019 ) and inconsistent gender ( fields 31 and 22001 ) . The numbers of cases and controls that passed the sample QC can be found in Figure 1 . A summary of background variables is given in Appendix 1—table 1 . These five sample subsets did not overlap with the discovery GWAS individuals . A collection of ancestry-informative genotypes was utilized to assess the genetic homogeneity of each of the self-reported ancestry ( details in Supplementary Methods ) . Our in-house patient and tumor data were investigated regarding the identified risk loci . All tumors of ≥1 cm diameter had been harvested and stored fresh-frozen ( details in Supplementary Methods ) . MED12 mutations were screened by Sanger sequencing the MED12 exons 1 and 2 and their flanking sequences ( 60 bp ) from all uterine leiomyoma and matching normal myometrium samples ( Mäkinen et al . , 2011; Heinonen et al . , 2014 ) . The resulting sequence graphs were inspected manually and with Mutation Surveyor software ( Softgenetics , State College , PA ) . Clinical patient data was available for the number of ULs , menopause status , parity , body mass index ( BMI ) and age at hysterectomy ( Appendix 1—figure 2 ) . This study was conducted in accordance with the Declaration of Helsinki and approved by the Finnish National Supervisory Authority for Welfare and Health , National Institute for Health and Welfare ( THL/151/5 . 05 . 00/2017 ) , and the Ethics Committee of the Hospital District of Helsinki and Uusimaa ( HUS/177/13/03/03/2016 ) . For the cis expression quantitative trait loci ( cis-eQTL ) analysis , genes with less than six reads in over 80% of the samples were filtered out . The between sample normalization was done with Relative Log Expression ( RLE ) normalization and each gene was inverse normal transformed . The eQTL analysis was run with FastQTL ( version 2 . 184 ) ( Ongen et al . , 2016 ) separately for 60 tumors and 56 patient-matched unaffected , adjacent myometrium samples using permutation approach . The permutation parameter was set to ‘1000 10000’ . Sequencing batch was used as a covariate . The cis-region was set to be 2 Mb . FDR correction was applied for tumors and matched myometrium separately . DNA methylation was studied in 56 tumors and 36 matched myometrium samples . The methylation calls were analyzed with bsseq ( version 1 . 12 . 2 ) ( Hansen et al . , 2012 ) . Only the methylation in CpG context was considered . Every locus was required to have the coverage of ≥2 in at least 90% of samples . The association between methylation and genotype was studied with MatrixEQTL ( version 2 . 1 . 1 ) using a linear regression model ( Shabalin , 2012 ) . The LD-independent ( r2 ≤0 . 3 ) SNPs from the discovery stage ( Appendix 1—table 2 ) and meta-analysis ( Appendix 1—table 3 ) were considered ( altogether 57 SNPs ) . The SNPs with MAF <0 . 05 in the methylation samples were filtered out . This resulted in 44 SNPs in tumors and 45 SNPs in matched myometrium . Cis methylation quantitative trait loci ( cis-meQTL ) was determined to be within 1 Mb flank from the SNP of interest . To annotate the CpGs with genomic context , the overlap between UCSC’s gene track ( hg19 ) and known CpG islands was studied . As the role of promoter methylation is well known , promoter methylation was studied in addition to gene body methylation . Core promoter was defined as a region −2 kb and +1 kb from the transcription start site . The methylation analysis was performed separately for tumors and matched normal myometrium to study whether the changes in methylation could be observed in both tissues . The whole genome sequenced ( WGS ) samples , in total 71 tumors ( 48 Illumina , 23 Complete Genomics ) and 51 matched myometrium samples ( 28 Illumina and 23 Complete Genomics ) , were prepared following Illumina and Complete Genomics protocols and processed as described previously ( Mehine et al . , 2013 ) . Structural variation was defined as a structural rearrangement ( e . g deletion , inversion or translocation ) not detectable in matched normal myometrium . Structural variation was detected as described in Mehine et al . ( Mehine et al . , 2013 ) The mean telomere length was estimated for Illumina samples using Computel ( version 0 . 3 ) ( Nersisyan and Arakelyan , 2015 ) with the default settings . Clonally related tumors were excluded from the analysis by randomly sampling one tumor to represent each clonally related tumor group . Clonally related tumors had identical changes in driver genes and shared at least a subset of somatic copy-number changes and/or copy neutral loss of heterozygosity ( see Mehine et al . ( Mehine et al . , 2015 ) for further details in identification of the clonally related tumors ) . Kruskal-Wallis test was used to assess the telomere length differences between tumors and matched myometrium as well as between genotypes . Linear model was used to calculate the association between number of risk alleles and telomere length . Pathway enrichment of all genome-wide significant SNPs was tested with DEPICT ( version 1 release 194 ) following the default settings ( Pers et al . , 2015 ) . The tool is designed to integrate multiple GWAS loci for in silico target gene prioritization , pathway enrichment and tissue-specific expression profiling . In short , the DEPICT framework combines phenotype-free co-expression networks , predefined pathways and protein-protein interaction networks in order to reveal functionally connected genes among the multiple risk loci . The tool is restricted to autosomal SNPs . Meta-analysis was implemented with an inverse-variance weighted , fixed effect model . Associations between the risk alleles and other variables were tested assuming an additive genotype model unless otherwise noted . The DHARMa ( 0 . 1 . 5 ) package was applied to evaluate the goodness-of-fit of the binomial and negative binomial models . The contribution of GRS to prevalence was estimated by [ ( Ea/Pi-1 ) / ( Pa/Pi-1 ) ] , where Ea = Pi*GRSa/GRSi assumes a linear relationship between GRS and the true risk , and Px and GRSx are the population-specific prevalence and mean GRS , respectively . All statistical tests were two-tailed unless otherwise noted . Summary statistics were collected from each of the study stages and are available as Appendix 1—table 2 and Supplementary file 1–3 . For each SNP , we report its allele frequency , effect size estimates and association based on the default linear , infinitesimal mixed model . For the meta-analysis stages , we also report the Cochrane’s Q statistic and I2 heterogeneity index in addition to the fixed-effects meta-analysis association and effect size ( random-effects meta-analysis is included for reference ) . BoltLMM reported lambda ( λGC ) 1 . 055 , 1 . 045 and 1 . 016 for UKBB , Helsinki and NFBC , respectively . The X chromosome associations were processed separately and included only the female controls ( λGC 1 . 052 , 1 . 049 and 1 . 005 for UKBB , Helsinki and NFBC , respectively ) . For GWAS , p<5 × 10−8 was reported as significant . The GRS association tests ( Appendix 1—table 6 ) were controlled for family-wise error rate ( FWER ) and reported significant for Holm-Bonferroni adjusted p<0 . 05 . Large families of association tests were controlled for false discovery rate ( FDR; Benjamini-Hochberg method ) and noted significant at FDR < 10% . In the six telomere length association tests and the two structural variation association tests , p<0 . 05 was considered statistically significant . | Fibroids – also known as uterine leiomyomas , or myomas – are a very common form of benign tumor that grows in the muscle wall of the uterus . As many as 70% of women develop fibroids in their lifetime . About a fifth of women report symptoms including severe pain , heavy bleeding during periods and complications in pregnancy . In the United States , the cost of treating fibroids is estimated to be $34 billion each year . Despite the prevalence of fibroids in women , there are few treatments available . Drugs to target them have limited effect and often an invasive procedure such as surgery is needed to remove the tumors . However , a better understanding of the genetics of fibroids could lead to a way to develop better treatment options . Välimäki , Kuisma et al . used a genome-wide association study to seek out DNA variations that are more common in people with fibroids . Using data from the UK Biobank , the genomes of over 15 , 000 women with fibroids were analyzed against a control population of over 392 , 000 individuals . The analysis revealed 22 regions of the genome that were associated with fibroids . These regions included genes that may well contribute to fibroid development , such as the gene TP53 , which influences the stability of the genome , and ESR1 , which codes for a receptor for estrogen – a hormone known to play a role in the growth of fibroids . Variation in a set of genes known to control development of the female reproductive organs was also identified in women with fibroids . The findings are the result of the largest genome-wide association study on fibroids , revealing a set of genes that could influence the development of fibroids . Studying these genes could lead to more effective drug development to treat fibroids . Revealing this group of genes could also help to identify women at high risk of developing fibroids and help to prevent or manage the condition . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"genetics",
"and",
"genomics"
] | 2018 | Genetic predisposition to uterine leiomyoma is determined by loci for genitourinary development and genome stability |
Hydrogen peroxide is the preeminent chemical weapon that organisms use for combat . Individual cells rely on conserved defenses to prevent and repair peroxide-induced damage , but whether similar defenses might be coordinated across cells in animals remains poorly understood . Here , we identify a neuronal circuit in the nematode Caenorhabditis elegans that processes information perceived by two sensory neurons to control the induction of hydrogen peroxide defenses in the organism . We found that catalases produced by Escherichia coli , the nematode’s food source , can deplete hydrogen peroxide from the local environment and thereby protect the nematodes . In the presence of E . coli , the nematode’s neurons signal via TGFβ-insulin/IGF1 relay to target tissues to repress expression of catalases and other hydrogen peroxide defenses . This adaptive strategy is the first example of a multicellular organism modulating its defenses when it expects to freeload from the protection provided by molecularly orthologous defenses from another species .
Bacteria , fungi , plants , and animal cells have long been known to excrete hydrogen peroxide to attack their prey and pathogens ( Avery and Morgan , 1924 ) . Hydrogen peroxide is also a byproduct of aerobic respiration ( Chance et al . , 1979 ) . Cells rely on highly conserved defense mechanisms to degrade hydrogen peroxide and avoid the damage that hydrogen peroxide inflicts on their proteins , nucleic acids , and lipids ( Mishra and Imlay , 2012 ) . The extent to which these protective defenses are coordinated across cells in animals is poorly understood . In the present study , we used C . elegans as a model system to explore whether hydrogen peroxide protective defenses are coordinated across cells . C . elegans is not spared from the threat of hydrogen peroxide . In its natural habitat of rotting fruits and vegetation , C . elegans encounters a wide variety of bacterial taxa ( Samuel et al . , 2016 ) , and this community includes bacteria in many genera known to degrade or produce hydrogen peroxide ( Passardi et al . , 2007 ) . Hydrogen peroxide produced by a bacterium from the C . elegans microbiome , Rhizobium huautlense , causes DNA damage to the nematodes ( Kniazeva and Ruvkun , 2019 ) , and many bacteria—including S . pyogenes , S . pneumoniae , S . oralis , and E . faecium—kill C . elegans by producing millimolar concentrations of hydrogen peroxide ( Bolm et al . , 2004; Jansen et al . , 2002; Moy et al . , 2004 ) . C . elegans may also encounter hydrogen peroxide derived from fruits , leaves , and stems , because plants produce hydrogen peroxide to attack their pathogens ( Arakawa et al . , 2014; Daudi et al . , 2012; Mehdy , 1994 ) . Coordinating hydrogen peroxide cellular defenses could be beneficial because it might enable C . elegans to avoid the energetic cost of unneeded protection . In addition , tight coordination of hydrogen peroxide defenses might be necessary because inducing a protective response at an inappropriate time might cause undesirable side effects . Hydrogen peroxide is an important intracellular signaling molecule , and depletion of hydrogen peroxide by scavenging enzymes may interfere with signal transduction and affect cell behavior and differentiation ( Veal et al . , 2007 ) . Nematodes overexpressing all three catalase genes exhibit a high level of mortality due to internal hatching of larvae , and this phenotype can be suppressed by joint overexpression of the superoxide dismutase SOD-1 ( Doonan et al . , 2008 ) , an enzyme that produces hydrogen peroxide . While catalases can degrade large quantities of hydrogen peroxide , at low hydrogen peroxide concentrations these enzymes accumulate in the ferryl-radical intermediate of their catalytic cycle , which is a dangerous oxidizing agent ( Imlay , 2013 ) . We set out to investigate whether sensory neurons coordinate hydrogen peroxide protective defenses across cells because sensory circuits in the brain collect and integrate information from the environment , enabling animals to respond to environmental change . Specific sensory neurons enable nematodes to smell , taste , touch , and sense temperature and oxygen levels ( Bargmann and Horvitz , 1991a; Chalfie et al . , 1985; Gray et al . , 2004; Mori and Ohshima , 1995; White et al . , 1986 ) . This information is integrated rapidly by interneurons to direct the nematode’s movement towards favorable environmental cues and away from harmful ones ( Kaplan et al . , 2018 ) . Nematodes also use sensory information to modify their development , metabolism , lifespan , and heat defenses ( Apfeld and Kenyon , 1999; Bargmann and Horvitz , 1991b; Mak et al . , 2006; Prahlad et al . , 2008 ) . Understanding how sensory circuits in the brain regulate hydrogen peroxide defenses in C . elegans may provide a template for understanding how complex animals coordinate their cellular defenses in response to the perceived threat of hydrogen peroxide attack . Using a systematic neuron-specific genetic-ablation approach , we identified ten classes of sensory neurons that regulate sensitivity to harmful peroxides in C . elegans . We found that the two ASI sensory neurons of the amphid , the major sensory organ of the nematode , initiate a multistep hormonal relay that decreases the nematode’s hydrogen peroxide defenses: a DAF-7/TGFβ signal from ASI is received by multiple sets of interneurons , which independently process this information and then relay it to target tissues via insulin/IGF1 signals . Interestingly , this neuronal circuit lowers the action of endogenous catalases and other hydrogen peroxide defenses within the worm in response to perception and ingestion of E . coli , the nematode’s primary food source in laboratory experiments . We show that E . coli express orthologous defenses that degrade hydrogen peroxide in the environment and that C . elegans does not need to induce catalases and other hydrogen peroxide defenses when E . coli is abundant . Thus , this neuronal circuit enables the nematodes to lower their own defenses upon sensing bacteria that can provide protection . In the microbial battlefield , nematodes use a sensory-neuronal circuit to determine whether to defend themselves from hydrogen peroxide attack or to freeload off protective defenses from another species .
C . elegans is sensitive to the lethal effects of peroxides . Under standard laboratory conditions , wild-type nematodes have an average lifespan of approximately 15 days ( Kenyon et al . , 1993 ) . In contrast , when grown in the presence of a peroxide ( 6 mM tert-butyl hydroperoxide , tBuOOH ) , the average lifespan of these nematodes is reduced to less than 1 day ( Figure 1A; An et al . , 2005 ) . Previously , we determined the peroxide resistance of nematodes by measuring their lifespan with high temporal resolution in the presence of 6 mM tBuOOH ( Stroustrup et al . , 2013 ) . To investigate whether sensory neurons might regulate the nematode’s peroxide defenses , we measured peroxide resistance in mutant animals with global defects in sensory perception . We first examined osm-5 cilium structure mutants , which lack neuronal sensory perception due to defects in the sensory endings ( cilia ) of most sensory neurons ( Perkins et al . , 1986 ) . These mutants exhibited a 45% increase in peroxide resistance relative to wild-type controls ( Figure 1A and Supplementary file 1 ) . Next , we examined tax-2 and tax-4 cyclic GMP-gated channel mutants , which are defective in the transduction of several sensory processes including smell , taste , oxygen , and temperature sensation ( Coburn and Bargmann , 1996; Komatsu et al . , 1996 ) . These two mutants also exhibited large increases in peroxide resistance compared to wild-type controls ( Figure 1A , Figure 1—figure supplement 1 , and Supplementary file 1 ) . Together , these observations indicate that neuronal sensory perception plays a role in regulating peroxide resistance in nematodes . In C . elegans hermaphrodites , 60 ciliated and 12 non-ciliated neurons perform most sensory functions ( White et al . , 1986 ) . To identify which of these sensory neurons influence the nematode’s peroxide resistance , we systematically measured peroxide resistance in a collection of strains in which specific sensory neurons have been genetically ablated via neuron-specific expression of caspases ( Chelur and Chalfie , 2007 ) or , in one case , via mutation of a neuron-specific fate determinant ( Chang et al . , 2003; Uchida et al . , 2003 ) . Overall , our neuron-ablation collection covered 44 ciliated and 10 non-ciliated neurons , including each of the 12 pairs of ciliated neurons that make up the two amphids ( the major sensory organs ) , 8 of the 13 classes of non-amphid ciliated neurons , and 6 of the 7 classes of non-ciliated sensory neurons ( Supplementary file 10 ) . Individual ablation of ASI , ASG , ASK , AFD , AWC , IL2 and joint ablation of ADE , PDE , and CEP increased the nematode’s peroxide resistance by up to 61% ( Figure 1A–B , Figure 1—figure supplement 1B–C , and Supplementary file 1 ) , whereas individual ablation of ASJ and AWA , and joint ablation of URX , AQR , and PQR reduced peroxide resistance by up to 16% ( Figure 1A–B ) . The remainder of the neurons tested—ADF , ADL , ASE , ASH , AWB , OLL , and joint ablation of ALM , PLM , AVM , PVM , FLP , and PVD—did not affect peroxide resistance ( Figure 1A and C , Figure 1—figure supplement 1D–G , and Supplementary file 1 ) . Altogether , we found that ten classes of sensory neurons can positively or negatively modulate peroxide resistance ( Figure 1B ) . These neurons are known to respond to diverse stimuli , including smell , taste , touch , temperature , and oxygen levels ( Figure 1C ) , suggesting that nematodes might adjust their peroxide resistance in response to multiple types of sensory information . Among all neuronal ablations tested , ablation of ASI , a pair of neurons that sense taste and temperature , caused the largest increase in peroxide resistance ( Figure 1A ) . Thus , we focused on the role of the ASI neuronal pair . ASI neurons secrete many peptide hormones , including DAF-7 ( Meisel et al . , 2014; Ren et al . , 1996 ) , a transforming growth factor β ( TGFβ ) hormone that regulates feeding , development , metabolism , and lifespan ( Dalfó et al . , 2012; Greer et al . , 2008; Ren et al . , 1996; Shaw et al . , 2007 ) . To determine whether DAF-7/TGFβ signaling also regulates peroxide resistance , we examined the effects of mutations in daf-7 . We found that daf-7 ( ok3125 ) null and daf-7 ( e1372 ) loss-of-function mutations increased peroxide resistance two-fold relative to wild-type controls ( Figure 2A and B , and Figure 2—figure supplement 1A–D , and Supplementary file 2 ) . Reintroducing the daf-7 ( + ) gene into daf-7 ( ok3125 ) mutants restored peroxide resistance to wild-type levels ( Figure 2B and Supplementary file 2 ) . Moreover , expression of daf-7 ( + ) only in the ASI neurons was sufficient to reduce the peroxide resistance of daf-7 ( ok3125 ) mutants to wild-type levels ( Figure 2C and Supplementary file 2 ) . daf-7 is also expressed at a low level in ASJ , another pair of chemosensory neurons ( Meisel et al . , 2014 ) , and expression of daf-7 ( + ) only in ASJ rescued the increased peroxide resistance of daf-7 ( ok3125 ) mutants ( Figure 2D and Supplementary file 2 ) . Thus , expression of daf-7 in ASI or ASJ was sufficient to confer normal peroxide resistance . Because ablation of ASI increased peroxide resistance but ablation of ASJ did not ( Figure 1A ) , we reason that ASI neurons are the source of DAF-7 that regulates the nematode’s peroxide resistance . We next asked whether DAF-7/TGFβ from ASI might regulate resistance to additional toxic chemicals from the environment that are not peroxides or directly generate peroxides . We tested sensitivity of daf-7 mutants to arsenite ( a toxic metalloid ) , paraquat ( a redox-cycling herbicide ) , and dithiothreitol/DTT ( a reducing agent ) . We adjusted the concentrations of these compounds to reduce the survival of wild-type nematodes about as much as in the tBuOOH survival assays . Compared with wild-type animals , daf-7 ( ok3125 ) mutants had similar survival in 5 mM arsenite , 25 mM dithiothreitol , and 75 mM paraquat ( Figure 2A and E–G , and Supplementary file 2 ) . Therefore , the DAF-7/TGFβ signal from ASI is a specific regulator of peroxide resistance in the worm . DAF-7/TGFβ signals via the Type 1 TGFβ receptor DAF-1 ( Georgi et al . , 1990 ) to regulate multiple downstream processes ( Dalfó et al . , 2012; Greer et al . , 2008; Ren et al . , 1996; Shaw et al . , 2007 ) . Signaling through the DAF-1 receptor inactivates the transcriptional activity of a complex between the receptor-associated coSMAD , DAF-3 , and the Sno/Ski factor , DAF-5 ( da Graca , 2004; Patterson et al . , 1997; Tewari et al . , 2004 ) . We found that a similar signal-transduction pathway regulates peroxide resistance . daf-1 ( m40 ) loss-of-function mutants showed a two-fold increase in peroxide resistance ( Figure 3A and Supplementary file 3 ) , and the increase in peroxide resistance of daf-7 and daf-1 mutants was almost completely abrogated by null or loss-of-function mutations in either daf-3 or daf-5 ( Figure 3A , Figure 3—figure supplement 1A–C , and Supplementary file 3 ) . The daf-3 ( mgDf90 ) null mutation also suppressed the increase in peroxide resistance of ASI-ablated worms ( Figure 3B and Supplementary file 3 ) . Therefore , the ASI neurons normally function to lower peroxide resistance in the worm using a canonical TGFβ signaling pathway . To determine which cells receive the DAF-7/TGFβ signal from the ASI neurons to regulate peroxide resistance , we restored daf-1 ( + ) gene expression in specific subsets of neurons using cell-type specific promoters in daf-1 ( m40 ) mutants . The cells composing each of these subsets of neurons , as well as the overlap between these subsets are diagramed in Figure 3C . DAF-1/TGFβ receptor is expressed broadly in the nervous system and in the distal-tip cells of the gonad ( Gunther et al . , 2000 ) . Pan-neuronal expression of daf-1 ( + ) with the egl-3 promoter lowered peroxide resistance in daf-1 mutants to the same extent as did expressing daf-1 ( + ) with the endogenous daf-1 promoter ( Figure 3D–E , and Supplementary file 3 ) . Reconstituting daf-1 ( + ) expression in all ciliated neurons ( except BAG and FLP ) using the osm-6 promoter had a minimal effect on peroxide resistance ( Figure 3F and Supplementary file 3 ) , indicating that daf-1 function in ciliated neurons is not sufficient to lower peroxide resistance . In contrast , expression of daf-1 ( + ) in multiple sets of non-ciliated interneurons and pharyngeal neurons using the flp-1 , tdc-1 , glr-1 , or glr-8 promoters lowered peroxide resistance to a similar extent as pan-neuronal expression of daf-1 ( + ) in daf-1 mutants ( Figure 3G–I , Figure 3—figure supplement 1D , and Supplementary file 3 ) , while directed daf-1 ( + ) expression in nine pharyngeal neurons using the glr-7 promoter did not affect peroxide resistance ( Figure 3—figure supplement 1E and Supplementary file 3 ) . The flp-1 promoter is active only in the two AVK interneurons ( Greer et al . , 2008 ) . In addition , the flp-1 , tdc-1 , glr-1 , and glr-8 promoters drive expression in non-overlapping cells , except for the expression overlap in the two RIM interneurons by the tdc-1 and glr-1 promoters ( Greer et al . , 2008; Figure 3C ) . We refer to the sets of neurons where flp-1 , tdc-1 , glr-1 , and glr-8 are expressed as ‘DAF-1-sufficiency sets’ , because expression of daf-1 ( + ) in any one of these sets of neurons is sufficient to lower the peroxide resistance of daf-1 mutant nematodes . We conclude that DAF-1 functions redundantly in AVK interneurons and at least two other separate sets of neurons to lower the nematode’s peroxide resistance . Where does the DAF-3/coSMAD transcription factor function to promote peroxide resistance when the DAF-1/TGFβ-receptor is inactive ? We expected that DAF-3 would function in the same cells as DAF-1 to regulate peroxide resistance , because both of these canonical TGFβ signal-transduction pathway components function in tdc-1 expressing interneurons to regulate feeding , fat storage , egg laying , and dauer-larva formation ( Greer et al . , 2008 ) . In addition , because during signal transduction DAF-1 inhibits DAF-3 , we expected that when DAF-1 is active only in one set of neurons then DAF-3 should be active only in neurons outside that set ( including the neurons of other non-overlapping DAF-1-sufficiency sets ) . This implied that to increase peroxide resistance DAF-3 should be active in all DAF-1-sufficiency sets of neurons . To test that prediction , we examined the effect on peroxide resistance of restoring daf-3 ( + ) expression in just one of the DAF-1-sufficiency sets of neurons in daf-1; daf-3 double mutants . Confirming our prediction , we found that restoring daf-3 ( + ) expression with the tdc-1 promoter was not sufficient to increase peroxide resistance in daf-1; daf-3 double mutants ( Figure 3K , Figure 3—figure supplement 1F , and Supplementary file 3 ) . In contrast , the peroxide resistance of daf-1; daf-3 double mutants increased upon restoring daf-3 ( + ) expression in all four DAF-1-sufficiency sets of neurons with a daf-1 promoter ( Figure 3J and Supplementary file 3 ) . We propose that the combination of the redundant action of DAF-1 in multiple sets of neurons and the repression of DAF-3 by DAF-1 in each of those neurons ensures that the nematode’s peroxide resistance stays low until all DAF-1-sufficiency sets of neurons de-repress DAF-3/coSMAD ( Figure 3L ) . Alternatively , to promote peroxide resistance in animals with reduced DAF-1 activity , DAF-3 function may be necessary only in cells that do not express the tdc-1 promoter . In such a scenario , other signaling molecules would transduce DAF-1 activity in tdc-1-expressing neurons to regulate peroxide resistance . Previous studies have shown that different mechanisms are required downstream of the DAF-3/coSMAD transcription factor to mediate the effects of DAF-7/TGFβ signaling on dauer-larva formation , fat storage , germline size , lifespan , and feeding ( Dalfó et al . , 2012; Greer et al . , 2008; Shaw et al . , 2007 ) . In this section , and later in this manuscript , we used a genetic approach to determine whether DAF-7/TGFβ signaling acts via one or more of these mechanisms to regulate the nematode’s peroxide resistance ( Figure 4A ) . DAF-7 regulates dauer-larva formation via the nuclear hormone receptor DAF-12 , which is the main switch driving the choice of reproductive growth or dauer arrest ( Antebi et al . , 2000 ) . Loss of daf-12 suppresses the constitutive dauer-formation phenotype of daf-7 loss-of-function mutants during development ( Thomas et al . , 1993 ) , but the daf-12 ( rh61rh411 ) null mutation did not suppress the increased peroxide resistance of daf-7 ( ok3125 ) null adults ( Figure 4B and Supplementary file 4 ) . In fact , even though the daf-12 null mutation lowered the peroxide resistance in otherwise wild-type animals , it further increased peroxide resistance in daf-7 mutants . We conclude that daf-12 ( + ) limits the peroxide resistance of daf-7 mutants , and that DAF-7 lowers peroxide resistance and inhibits formation of peroxide-resistant dauer larvae via separate mechanisms . The metabotropic glutamate receptors mgl-1 and mgl-3 are necessary for the increase in fat storage upon DAF-7-pathway inhibition ( Greer et al . , 2008 ) . However , null mutations in either or both of these mgl genes did not affect peroxide resistance in daf-1 mutants ( Figure 4C , Figure 4—figure supplement 1 , and Supplementary file 4 ) . Thus , peroxide resistance and fat storage are also regulated via separate pathways downstream of DAF-1 . Germline size is reduced upon DAF-7-pathway inhibition ( Dalfó et al . , 2012 ) . Mutations in the mes-1 gene cause about 50% of animals to become sterile adults because they fail to form the primordial germ cells during embryogenesis , while the remaining animals develop into fertile adults ( Strome et al . , 1995 ) . Germline-ablated mes-1 ( ok2467 ) mutants showed a 57% increase in peroxide resistance compared to their fertile mes-1 ( ok2467 ) mutant siblings ( Figure 4D and Supplementary file 4 ) , consistent with previous studies ( Steinbaugh et al . , 2015 ) . However , daf-1 ( m40 ) increased peroxide resistance in both germline-ablated and fertile mes-1 mutants ( Figure 4D and Supplementary file 4 ) . In addition , daf-3 ( mgDf90 ) did not affect peroxide resistance in germline-ablated mes-1 mutants ( Figure 4E and Supplementary file 4 ) . Therefore , DAF-1 and the germline regulate peroxide resistance via independent mechanisms . DAF-7-pathway signaling lowers lifespan by promoting insulin/IGF1 receptor signaling ( Shaw et al . , 2007 ) . Previous studies have shown that transcription of at least 11 of the 40 insulin/IGF1 genes in the genome is repressed by the DAF-3/coSMAD in response to lower levels of DAF-7 and DAF-1 signaling ( Liu et al . , 2004; Narasimhan et al . , 2011; Shaw et al . , 2007 ) . We found that deletion of the DAF-3-repressed insulin/IGF1 genes ins-1 , ins-3 , ins-4 , ins-5 , ins-6 , or daf-28 caused increases in peroxide resistance ranging between 11% and 65% ( Figure 5A , Figure 5—figure supplement 1A–B , and Supplementary file 5 ) , suggesting DAF-7 lowers peroxide resistance by promoting signaling by the insulin/IGF1 receptor , DAF-2 . The daf-2 ( e1370 ) strong loss-of-function mutation increased peroxide resistance about three-fold ( Figure 5B and Supplementary file 5 ) , consistent with previous findings ( Tullet et al . , 2008 ) . Double mutants of daf-1 ( m40 ) and daf-2 ( e1370 ) had higher peroxide resistance than the respective single mutants ( Figure 5B and Supplementary file 5 ) . This additive effect suggested that the DAF-1 TGFβ receptor and the DAF-2 insulin/IGF1 receptor regulated peroxide resistance via mechanisms that do not fully overlap , but could also have been due to the receptors acting via fully overlapping mechanisms ( because neither daf-1 ( m40 ) nor daf-2 ( e1370 ) eliminates gene function completely ) . We considered the possibility that a DAF-2-dependent mechanism might mediate some of the effects of DAF-1 on peroxide resistance . If repressing the expression of insulin/IGF1 ligands of DAF-2 mediated part of the increased peroxide resistance of DAF-7-pathway inhibition , then one would expect the FOXO transcription factor DAF-16 to be necessary for those effects . DAF-16 is necessary for the increase in lifespan and most other phenotypes of mutants with reduced signaling by the DAF-2 insulin/IGF1 receptor ( Kenyon et al . , 1993; Lin et al . , 1997; Ogg et al . , 1997 ) . We found that DAF-16 was also necessary for the increase in peroxide resistance of daf-2 ( e1370 ) mutants ( Figure 5C and Supplementary file 5 ) and for the increase in peroxide resistance of an ins-4 ins-5 ins-6; daf-28 quadruple mutant ( Figure 5D and Supplementary file 5 ) . The daf-16 ( mu86 ) null mutation decreased the peroxide resistance of daf-7 ( e1372 ) and daf-1 ( m40 ) mutants by nearly 50% , but caused only a small peroxide resistance reduction in wild-type nematodes ( Figure 5E–F , and Supplementary file 5 ) . Therefore , regulation of peroxide resistance by the DAF-7/TGFβ signaling pathway is , in part , dependent on the DAF-16/FOXO transcription factor . We examined whether other transcription factors might act with DAF-16 to increase peroxide resistance in daf-1 mutants . Like DAF-16 , the NRF orthologue SKN-1 and the TFEB orthologue HLH-30 are activated in response to reduced DAF-2 signaling ( Lin et al . , 2018; Tullet et al . , 2008 ) . The peroxide resistance of daf-1 ( m40 ) hlh-30 ( tm1978 ) double mutants was identical to that of daf-1 single mutants ( Figure 5—figure supplement 1C and Supplementary file 5 ) . Knockdown of skn-1 via RNA interference ( RNAi ) decreased the peroxide resistance of daf-1 ( m40 ) mutants by 30% but did not affect peroxide resistance in wild-type nematodes ( Figure 5G and Supplementary file 5 ) . RNAi of skn-1 also decreased the peroxide resistance of daf-16; daf-1 double mutants , suggesting that DAF-16 and SKN-1 functioned in a non-overlapping manner to promote peroxide resistance in daf-1 ( m40 ) mutants ( Figure 5H and Supplementary file 5 ) . We propose that repression of insulin/IGF1 gene expression by DAF-3/coSMAD leads to a reduction in signaling by the DAF-2/insulin/IGF1 receptor , which subsequently increases the nematode’s peroxide resistance via transcriptional activation by SKN-1/NRF and DAF-16/FOXO ( Figure 5N ) . To identify which target tissues are important for increasing the nematode’s peroxide resistance via DAF-16 in response to reduced DAF-1 signaling , we determined the extent to which restoring daf-16 ( + ) expression in specific tissues using tissue-specific promoters increased the peroxide resistance of daf-16; daf-1 double mutants . As expected , peroxide resistance was increased when we restored daf-16 ( + ) expression with the endogenous daf-16 promoter ( Figure 5I and Supplementary file 5 ) . Restoring daf-16 ( + ) expression only in the intestine increased peroxide resistance , albeit to a lesser extent than did re-expressing daf-16 ( + ) with the endogenous daf-16 promoter ( Figure 5J and Supplementary file 5 ) . Restoring daf-16 ( + ) in neurons slightly increased peroxide resistance ( Figure 5K ) , while restoring daf-16 ( + ) in body-wall muscles had no effect ( Figure 5L and Supplementary file 5 ) . Restoring daf-16 ( + ) expression in the hypodermis decreased peroxide resistance slightly ( Figure 5M and Supplementary file 5 ) ; however , it is difficult to interpret these results because these nematodes looked sickly ( unlike daf-1 and daf-16 single and double mutants ) , consistent with reports that selectively expressing daf-16 ( + ) in the hypodermis is toxic ( Libina et al . , 2003 ) . Therefore , the DAF-16/FOXO transcription factor functions in the intestine and neurons to increase the nematode’s peroxide resistance when DAF-3/coSMAD is active due to reduced DAF-1 function ( Figure 5N ) . To investigate how reduced DAF-7/TGFβ signaling increases peroxide resistance , we used mRNA sequencing ( mRNA-seq ) to identify genes that were differentially regulated between daf-7 ( ok3125 ) mutants and wild-type animals . We extracted mRNA from day two adults and then performed differential expression analyses on the mRNA-seq data for the 9660 genes that had detectable expression . Relative to wild-type animals , daf-7 ( ok3125 ) null mutants decreased the expression of 3641 genes and increased the expression of 3229 genes ( q value < 0 . 001 ) ( Figure 6A and Figure 6—figure supplement 1A ) . These changes in gene expression were consistent with but more extensive than those observed in microarray-based studies with partial loss-of-function TGFβ signaling pathway mutants ( Shaw et al . , 2007; Figure 6—figure supplement 1B and Supplementary file 6 ) . To identify which processes may be influenced by the transcriptomic changes of daf-7 null mutants , we used Gene Ontology ( GO ) term enrichment analysis ( Angeles-Albores et al . , 2016 ) and clustered enriched GO terms based on semantic similarity ( Supek et al . , 2011 ) . We focused the GO analysis on genes upregulated or downregulated more than four-fold . The 1267 genes downregulated more than four-fold in daf-7 ( ok3125 ) mutants were associated with reproduction and with expression in the germline , while the 594 genes upregulated more than four-fold in daf-7 ( ok3125 ) mutants were associated with defense and immune responses and with expression in the intestine ( Figure 6B ) . Because the skn-1 and daf-16 genes were each partially required for the increased peroxide resistance of animals with reduced DAF-7/TGFβ signaling , we expected that the expression of SKN-1 and DAF-16 transcriptional targets would be influenced by daf-7 . The daf-7 ( ok3125 ) mutation increased the expression of genes upregulated by skn-1 ( + ) in wild type animals ( Oliveira et al . , 2009 ) and in daf-2 ( - ) mutants ( Ewald et al . , 2015; Figure 6C , Figure 6—figure supplement 1C , and Supplementary file 6 ) . In addition , the daf-7 ( ok3125 ) mutation increased the expression of genes directly upregulated by DAF-16 ( Kumar et al . , 2015; Figure 6C and Supplementary file 6 ) and , as observed in a previous study ( Shaw et al . , 2007 ) , increased the expression of genes upregulated in a daf-16-dependent manner in daf-2 ( - ) mutants ( Murphy et al . , 2003; Figure 6—figure supplement 1D and Supplementary file 6 ) . Together , these findings suggest that DAF-7/TGFβ represses the induction of direct DAF-16 and SKN-1 target genes . Nematodes can be exposed directly to peroxides through food ingestion , and daf-7 and daf-1 mutants have been shown to exhibit mild feeding defects ( Greer et al . , 2008 ) . Therefore , we considered the possibility that the increase in peroxide resistance of mutants with impaired DAF-7-pathway signaling was due to their reduced feeding . Previous studies have shown that the tyrosine decarboxylase TDC-1 and the tyramine β-hydroxylase TBH-1—biosynthetic enzymes for the neurotransmitters tyramine and octopamine , respectively—are each necessary for the feeding defect of daf-1 mutants as daf-1;tdc-1 and daf-1;tbh-1 double mutants have normal feeding behaviors ( Greer et al . , 2008 ) . Surprisingly , despite restoring normal feeding to daf-1 mutants , tbh-1 and tdc-1 null mutations did not suppress the increased peroxide resistance of daf-1 mutants ( Figure 7A , Figure 7—figure supplement 1A , and Supplementary file 7 ) . In fact , both mutations further increased peroxide resistance in a daf-1 mutant background . Because mutations that restored normal feeding to daf-1 mutants increased the peroxide resistance of daf-1 mutants , these findings suggested that the reduced feeding exhibited by daf-1 mutants in fact reduces the magnitude of their increased peroxide resistance . To investigate whether feeding has a direct effect on peroxide resistance , we first determined whether a wild-type nematode’s feeding history ( before peroxide exposure ) might affect its subsequent peroxide resistance . We transferred nematodes to plates with different concentrations of E . coli for 24 hr prior to the start of the peroxide resistance assay and found that the E . coli concentration before the assay had a dose-dependent effect on peroxide resistance ( Figure 7B and Supplementary file 7 ) . Animals grown on higher concentrations of E . coli had higher peroxide resistance . Strikingly , nematodes grown without E . coli for two days before the assay showed a six-fold decrease in peroxide resistance ( Figure 7C , Figure 7—figure supplement 1B , and Supplementary file 7 ) , even though they had access to plentiful E . coli during the assay . Next , we tested whether reduced ingestion of E . coli was sufficient to mimic the effects of pre-exposure to reduced E . coli levels . Mutants in the pharyngeal-muscle specific nicotinic acetylcholine receptor subunit eat-2 ingest bacteria more slowly due to reduced pharyngeal pumping ( feeding ) ( Avery , 1993; Raizen et al . , 1995 ) . The eat-2 ( ad1116 ) loss-of-function mutation , which causes a strong feeding defect ( Figure 7E ) , decreased peroxide resistance by 25% relative to wild-type animals ( Figure 7D and Supplementary file 7 ) . Therefore , impaired feeding leads to decreased peroxide resistance . Finally , we asked whether feeding and DAF-7 signaling regulate peroxide resistance jointly , or independently . Unlike eat-2 mutants , daf-3 null single mutants did not decrease peroxide resistance compared with wild-type animals ( Figure 3A–B and Figure 6D ) . However , the eat-2 ( ad1116 ) mutation caused a larger decrease in peroxide resistance in daf-3 mutants than in wild-type nematodes ( Figure 7D and Supplementary file 7 ) , suggesting that daf-3 ( + ) promotes peroxide resistance in eat-2 mutants . This effect was not due to an enhancement of the feeding defect of eat-2 mutants by the daf-3 mutation , because eat-2; daf-3 double mutants fed slightly more ( not less ) than eat-2 single mutants ( Figure 7E ) . We propose that DAF-3 is activated in response to reduced feeding , leading to an increase in peroxide resistance . DAF-3 acts as an adaptive mechanism that partially offsets the detrimental effect of reduced feeding on peroxide resistance . Taken together , these findings imply that both feeding on bacteria and DAF-3/coSMAD signaling increase peroxide resistance , but that they attenuate each other’s effects ( Figure 7F ) . This cross-inhibition might enable nematodes to switch between DAF-3-dependent and DAF-3-independent mechanisms of peroxide resistance in response to changes in food ingestion and DAF-7 signal levels . Why does DAF-7 from ASI function to decrease the nematode’s peroxide resistance ? ASI sensory neurons become active in response to perception of water-soluble signals from E . coli ( Gallagher et al . , 2013 ) and induce daf-7 gene expression in a TAX-4-activity-dependent manner ( Chang et al . , 2006 ) . As a result , the ASI neurons upregulate daf-7 expression in response to E . coli ( Gallagher et al . , 2013 ) and lower daf-7 gene expression in response to starvation and low E . coli concentrations ( Entchev et al . , 2015; Ren et al . , 1996 ) . Lowering DAF-7 levels when E . coli is scarce may enable nematodes to prepare for a future of reduced feeding by attenuating the expected reduction in peroxide resistance caused by reduced feeding . But increasing DAF-7 levels when E . coli is abundant may render nematodes more vulnerable to peroxide . We reasoned that perhaps C . elegans decreases its own peroxide self-defenses via DAF-7 signaling from the ASI neurons when E . coli is abundant because C . elegans expects to be safe from peroxide attack in that setting . To test that hypothesis , we first asked whether E . coli can protect nematodes from the lethal effects of peroxides . This required that we re-examine the conditions of the peroxide resistance assays , which were conducted using a lipid hydroperoxide ( tert-butyl hydroperoxide , tBuOOH ) widely used in C . elegans studies due to its stability ( An et al . , 2005 ) . When we used hydrogen peroxide instead of tBuOOH , we could not kill C . elegans even with concentrations as high as 20 mM ( Figure 8A ) which is well above the biologically plausible range of up to 3 mM hydrogen peroxide used by other bacteria to kill C . elegans ( Bolm et al . , 2004; Moy et al . , 2004 ) . This suggested that hydrogen peroxide , but not tBuOOH , was efficiently degraded by E . coli . This bacterium uses several scavenging systems to degrade hydrogen peroxide ( Mishra and Imlay , 2012 ) . The two E . coli catalases , KatG and KatE , are the predominant scavengers of hydrogen peroxide in the environment , and the peroxiredoxin , AhpCF , plays a minor role ( Seaver and Imlay , 2001 ) . E . coli JI377 , a KatG KatE AhpCF triple null mutant strain which cannot scavenge any hydrogen peroxide from the environment ( Seaver and Imlay , 2001 ) , did not protect C . elegans from 1 mM hydrogen peroxide killing ( Figure 8B ) , whereas the E . coli MG1655 parental wild-type strain was protective ( Figure 8B ) . We propose that E . coli protects C . elegans from hydrogen peroxide killing because it expresses catalases that efficiently deplete hydrogen peroxide from the environment , creating a local environment where hydrogen peroxide is not a threat to C . elegans . To determine whether DAF-7 regulates C . elegans hydrogen peroxide resistance , similar to its effects on tert-butyl hydroperoxide resistance , we examined resistance to hydrogen peroxide in daf-7 mutants . In assays with the catalase mutant E . coli JI377 strain , we found that daf-7 ( ok3125 ) increased the nematode’s hydrogen peroxide resistance over two-fold relative to wild-type nematodes ( Figure 8B and Supplementary file 8 ) . ASI-ablation also increased hydrogen peroxide resistance in assays with E . coli JI377 ( Figure 8—figure supplement 1A and Supplementary file 8 ) . We propose that in response to TAX-4-dependent sensory perception of E . coli , the ASI sensory neurons express DAF-7/TGFβ to instruct target tissues to downregulate their hydrogen peroxide defenses . Last , we investigated the possibility that reducing DAF-7-pathway signaling protects C . elegans from hydrogen peroxide killing via a hydrogen peroxide defense mechanism orthologous to the one by which E . coli protects C . elegans . The C . elegans genome contains three catalase genes in tandem—two-newly duplicated cytosolic catalases , ctl-1 and ctl-3 , and a peroxisomal catalase , ctl-2—which are the nematode orthologues of the two E . coli catalases , KatG and KatE ( Petriv and Rachubinski , 2004 ) . In our mRNA-seq analysis we found that ctl-1 and ctl-2 were induced by the daf-7 ( ok3125 ) mutation ( Figure 8—figure supplement 1B ) . In addition , we expected the C . elegans catalase genes to be upregulated in response to reduced DAF-7 signaling , because all three catalase genes have DAF-16 and SKN-1 binding sites in their promoters ( An and Blackwell , 2003; Park et al . , 2009; Petriv and Rachubinski , 2004 ) , and their mRNA and protein expression increase in a DAF-16-dependent manner when DAF-2 signaling is reduced ( Dong et al . , 2007; McElwee et al . , 2003; Murphy et al . , 2003 ) . To determine whether endogenous catalases could protect C . elegans from hydrogen peroxide when E . coli is not able to deplete hydrogen peroxide from the environment , we examined the effects of simultaneously increasing the dosage of all three catalase genes . We found that ctl-1/2/3 overexpression , which increases catalase activity ten-fold ( Doonan et al . , 2008 ) , more than doubled C . elegans hydrogen peroxide resistance in assays with E . coli JI377 ( Figure 8C and Supplementary file 8 ) . To investigate whether one of the endogenous catalases might mediate the increased hydrogen peroxide resistance of nematodes with reduced DAF-7-pathway signaling , we constructed double mutants between daf-1 and individual catalase genes . We found that the cytosolic catalase ctl-1 ( ok1242 ) null mutation abrogated much of the increase in hydrogen peroxide resistance of daf-1 ( m40 ) mutants in assays with E . coli JI377 ( Figure 8D and Supplementary file 8 ) , but the peroxisomal catalase ctl-2 ( ok1137 ) null mutation did not ( Figure 8—figure supplement 1C and Supplementary file 8 ) . Therefore , the increase in hydrogen peroxide resistance of daf-1 mutants is mediated in part by the CTL-1 cytosolic catalase . In line with this functional dependence , ctl-1 mRNA levels were elevated up to two-fold in daf-7 ( ok3125 ) and daf-1 ( m40 ) mutant adults grown on E . coli OP50 ( Figure 8E ) . This upregulation was partially DAF-16-dependent , since the daf-16 ( mu86 ) mutation caused a small but statistically significant reduction in ctl-1 mRNA expression in daf-1 ( m40 ) mutants but not in wild-type animals ( Figure 8E ) . The ctl-1 gene product is expressed only in the intestine ( Hamaguchi et al . , 2019 ) , and this expression was elevated in daf-1 ( m40 ) mutants ( Figure 8F–G , and Figure 8—figure supplement 1D ) . Taken together , these findings suggest that the DAF-7/TGFβ-pathway downregulates catalase gene expression in the intestine , partly via DAF-16 . We propose that DAF-7/TGFβ signaling enables C . elegans to decide whether to induce its own hydrogen peroxide degrading catalases or , instead , freeload on protection provided by molecularly orthologous catalases from E . coli ( Figure 8H ) .
We show here that sensory neurons regulate how long C . elegans nematodes can survive in the presence of environmental peroxides . Peroxide resistance was higher in nematodes with a global impairment in sensory perception ( Figure 1A ) . Using a systematic neuron-specific genetic-ablation approach , we identified ten classes of sensory neurons that influence the nematode’s peroxide resistance , including seven classes of neurons that normally decrease peroxide resistance and three classes of neurons that normally increase it ( Figure 1B ) . Why do so many neurons influence C . elegans peroxide resistance ? One possibility is that these neurons respond to environmental cues correlated with the threat of hydrogen peroxide . Perception of water-soluble attractants by the amphid sensory neurons ASI , ASG , and ASK—neurons that when ablated caused some of the largest increases in peroxide resistance—helps C . elegans navigate towards bacteria ( Bargmann and Horvitz , 1991a ) , its natural food source . We found that E . coli , the nematode’s food in laboratory experiments , influences the nematode’s hydrogen peroxide resistance in three different ways . E . coli perception induces the expression in ASI sensory neurons of DAF-7/TGFβ ( Chang et al . , 2006; Entchev et al . , 2015; Gallagher et al . , 2013; Ren et al . , 1996 ) , a hormone that decreases the nematode’s hydrogen peroxide defenses . In addition , E . coli ingestion increases C . elegans peroxide resistance in a DAF-3/CoSMAD-independent manner ( Figure 7D ) . Last , E . coli expression of scavenging enzymes degrades hydrogen peroxide in the nematode’s environment . We propose that the control of organismic peroxide resistance by neurons that sense bacteria enables nematodes to turn down their peroxide self-defenses when they sense bacteria they deem protective . We show here that C . elegans is safe from hydrogen peroxide attack when E . coli is abundant because hydrogen peroxide degrading enzymes from E . coli protect C . elegans . These E . coli self-defense mechanisms create a public good ( West et al . , 2006 ) , an environment safe from the threat of hydrogen peroxide , that benefits both E . coli and C . elegans . C . elegans freeloads off the hydrogen peroxide self-defense mechanisms from E . coli ( Figure 8H ) , because it uses a public good created by E . coli . E . coli degrades hydrogen peroxide in the environment primarily by expressing two catalases , KatG and KatE , as these enzymes account for over 95% of E . coli’s hydrogen peroxide degrading capacity ( Seaver and Imlay , 2001 ) . Catalase-positive E . coli can protect catalase-deficient E . coli from hydrogen peroxide ( Ma and Eaton , 1992 ) . This facilitative relationship , where one species creates an environment that promotes the survival of another ( Bronstein , 2009 ) , also occurs across bacterial species in diverse environments: in dental plaque in the human mouth , Actinomyces naeslundii protects catalase-deficient Streptococcus gordonii by removing hydrogen peroxide ( Jakubovics et al . , 2008 ) and , in marine environments , catalase-positive bacteria protect the catalase-deficient cyanobacterium Prochlorococcus , the major photosynthetic organism in the open ocean ( Zinser , 2018 ) . Unlike catalase-deficient bacteria receiving hydrogen peroxide protection from surrounding bacteria , C . elegans is not catalase deficient . In C . elegans , TAX-4-dependent sensory perception of E . coli stimulates the expression of DAF-7 in ASI ( Chang et al . , 2006; Entchev et al . , 2015; Gallagher et al . , 2013; Ren et al . , 1996 ) . We found that when DAF-7 signaling is reduced , target tissues induce defense mechanisms that protect C . elegans from hydrogen peroxide . These mechanisms are mediated in part by the DAF-16-dependent expression in the intestine of the cytosolic catalase CTL-1 . Consistent with this regulatory logic , the total catalase specific activity of C . elegans extracts increases with decreasing concentrations of E . coli in the nematode’s surrounding environment ( Houthoofd et al . , 2002 ) . We propose that the TGFβ-insulin/IGF1 signaling hormonal relay that begins with DAF-7 secretion from ASI enables this sensory neuron to communicate to target tissues that they do not need to induce CTL-1 and other hydrogen peroxide protection services because E . coli in the surrounding environment likely provide protection by expressing orthologous hydrogen peroxide degrading enzymes . Thus , this sensory circuit enables nematodes to choose between hydrogen peroxide self-defense and freeloading strategies ( Figure 8H ) . In the complex and variable habitat where C . elegans lives , deciding whether to induce hydrogen peroxide defenses may be challenging . C . elegans cells manage this challenge by relinquishing control of their cellular hydrogen peroxide defenses to a neuronal circuit in the nematode’s brain . This circuit might be able to integrate a wider variety of inputs than individual cells could , enabling a better assessment of the threat of hydrogen peroxide and precise regulation of hydrogen peroxide protective defenses . Is the choice between hydrogen peroxide self-defense and freeloading strategies regulated by DAF-7 limited to inducing hydrogen peroxide protection services in target tissues ? We favor an alternative possibility , that DAF-7 coordinates the induction of a broad phenotypic response to the perceived threat of hydrogen peroxide , because the phenotypic responses to lower DAF-7 signaling follow the expected desirable outcomes for animals that anticipate exposure to hydrogen peroxide: ( i ) re-routing development to form hydrogen peroxide resistant dauer larva ( Riddle and Albert , 1997 ) ; ( ii ) reducing proliferation of germline stem cells ( Dalfó et al . , 2012 ) , to prevent hydrogen peroxide induced damage to their DNA ( Wyatt et al . , 2017; Zong et al . , 2014 ) , ( iii ) reducing oocyte fertilization and egg-laying ( McKnight et al . , 2014; Trent , 1982 ) , to increase the chances of progeny survival; ( iv ) reducing feeding ( Greer et al . , 2008 ) , since many pathogenic bacteria produce hydrogen peroxide; ( v ) avoiding high oxygen concentrations ( Chang et al . , 2006 ) , which are oxidizing; and ( vi ) increasing the nematode’s hydrogen peroxide resistance . These diverse phenotypic responses might be triggered by different DAF-7 levels , reflecting the adaptive benefit of reducing the harm of hydrogen peroxide in each case . Perhaps for this reason , the DAF-7 signal is relayed via different circuits to target tissues mediating some of those responses . The DAF-1 receptor and the DAF-3/DAF-5 complex function in the somatic gonad to regulate germ-cell proliferation ( Dalfó et al . , 2012 ) , and in RIM and RIC interneurons to regulate feeding , fat storage , egg laying , and dauer-larva formation ( Greer et al . , 2008 ) . In contrast , to regulate hydrogen peroxide resistance , DAF-1 functions in at least three different sets of interneurons ( Figure 3L ) . One set includes RIM interneurons , and another comprises only the two AVK interneurons , which are not involved in regulating feeding , egg laying , and dauer-larva formation via DAF-1 signaling ( Greer et al . , 2008 ) . The more complex role of interneuronal DAF-1 signaling in regulating hydrogen peroxide resistance suggests that C . elegans takes great care to avoid inducing hydrogen peroxide protection services in target tissues unless DAF-7 levels are low . Our studies provide a template for understanding how complex animals coordinate cellular hydrogen peroxide defenses . We identify sensory neurons that respond to bacterial cues as important regulators of hydrogen peroxide protection by C . elegans target tissues . Similar regulatory systems may exist in other animals . In mice , sensory neurons involved in pain perception respond to cues from Staphylococcus aureus by releasing neuropeptides that inhibit the activation of hydrogen peroxide producing immune cells ( Chiu et al . , 2013 ) , and some of the neuropeptides secreted by these sensory neurons , including galanin and calcitonin gene-related peptide , also induce hydrogen peroxide protection in target cells ( Cui et al . , 2010; Tullio et al . , 2017 ) . Assigning control of cellular defenses to dedicated sensory circuits may represent a general cellular-coordination tactic used by animals to regulate induction of self-defenses for hydrogen peroxide and perhaps other threats . We show that the two ASI amphid sensory neurons use a multistep signal relay to control the extent to which target tissues protect C . elegans from hydrogen peroxide . In insects and mammals , TGFβ and insulin/IGF1 signaling components regulate cellular antioxidant defenses ( Brunet et al . , 2004; Clancy et al . , 2001; Holzenberger et al . , 2003; Kayanoki et al . , 1994; Liu et al . , 2012; Tatar et al . , 2003 ) , so it will be interesting to determine if a conserved hormonal relay controls hydrogen peroxide defenses in all animals . While a freeloading strategy may provide maximum fitness by inactivating self-defenses in environments where hydrogen peroxide is not a threat , this strategy need not provide maximum health or longevity to the organism . Consistent with this , in addition to lowering peroxide resistance in C . elegans , the ASI , ASG , and AWC amphid sensory neurons also shorten this organism’s lifespan in environments with no hydrogen peroxide ( Alcedo and Kenyon , 2004 ) , and DAF-7/TGFβ signaling from ASI also shortens C . elegans lifespan in those environments ( Shaw et al . , 2007 ) . Because sensory perception and catalases also determine health and longevity in invertebrate and vertebrate animals ( Apfeld and Kenyon , 1999; Libert et al . , 2007; Pérez-Estrada et al . , 2019; Riera et al . , 2014; Shaw et al . , 2007 ) , it is likely that sensory modulation presents a promising approach to induce latent defenses that could increase health and longevity in all animals .
Wild-type C . elegans was Bristol N2 . C . elegans were cultured on NGM agar plates seeded with E . coli OP50 , unless noted otherwise . For a list of all bacterial strains used in this study , see Supplementary file 9 ( Brenner , 1974; Seaver and Imlay , 2001; Kamath et al . , 2001 ) . For a list of all worm strains used in this study , see Supplementary file 10 ( Chang et al . , 2003; Chang et al . , 2006; Wragg et al . , 2007; Greer et al . , 2008; Beverly et al . , 2011; Chang et al . , 2011; Cornils et al . , 2011; Glauser et al . , 2011; Dalfó et al . , 2012; Lee et al . , 2012; Srinivasan et al . , 2012; Yoshida et al . , 2012; Fernandes de Abreu et al . , 2014; Russell et al . , 2014; Kaplan et al . , 2015; Vidal-Gadea et al . , 2015; Krzyzanowski et al . , 2016; Fletcher and Kim , 2017; Horspool and Chang , 2017; Juozaityte et al . , 2017; Hamaguchi et al . , 2019 ) . Double and triple mutant worms were generated by standard genetic methods . For a list of PCR genotyping primers and enzymes , and phenotypes used for strain construction , see Supplementary file 11 . The Ptdc-1::daf-3 ( + ) ::GFP ( pKA533 ) and Pdaf-1::daf-3 ( + ) ::GFP ( pKA534 ) plasmids ( kindly provided by Kaveh Ashrafi ) were injected at 30 ng/µl into daf-1 ( m40 ) IV; daf-3 ( mgDf90 ) X with 20 ng/µl Pmyo-2::RFP and 20 ng/µl Punc-122::DsRed , respectively , as co-injection markers . Automated survival assays were conducted using a C . elegans lifespan machine scanner cluster ( Stroustrup et al . , 2013 ) . This platform enables the acquisition of survival curves with very high temporal resolution and large population sizes . All chemicals were obtained from Sigma . For hydrogen peroxide , tert-butyl hydroperoxide , sodium arsenite , paraquat , and dithiothreitol assays , the compound was added to molten agar immediately before pouring onto 50 mm NGM agar plates . Plates were dried ( Stroustrup et al . , 2013 ) and seeded with 100 µl of concentrated E . coli OP50 resuspended at an OD600 of 20 ( Entchev et al . , 2015 ) . For RNAi experiments , the appropriate E . coli HT115 ( DE3 ) strain was used instead . For hydrogen peroxide assays , E . coli MG1655 or JI377 were used instead ( Seaver and Imlay , 2001 ) . Nematodes were cultured at 20°C until the onset of adulthood , and then cultured at 25°C—to potentially enhance daf-7 mutant phenotypes ( Ren et al . , 1996; Shaw et al . , 2007 ) —in groups of up to 100 , on plates with 10 μg/ml 5-fluoro-2-deoxyuridine ( FUDR ) , to avoid vulval rupture ( Leiser et al . , 2016 ) , prevent matricidal effects of daf-7 pathway mutants ( Shaw et al . , 2007 ) , and eliminate live progeny . As an alternative to FUDR , we inhibited formation of the eggshell of fertilized C . elegans embryos with RNAi of egg-5 ( Entchev et al . , 2015 ) , with identical results ( Figure 2—figure supplements 1A and C-D , and Supplementary file 2 ) . For experiments with daf-1; daf-2 double mutants , which only develop as dauers at 20°C , all strains were grown at 15°C instead of 20°C until the onset of adulthood . For food-conditioning experiments , E . coli OP50 was resuspended in S Basal containing streptomycin ( 50 μg/ml ) and seeded onto plates supplemented with both streptomycin and carbenicillin , each at 50 μg/ml , as described ( Entchev et al . , 2015 ) . For daf-1 , daf-3 , and daf-16 transgenic-rescue experiments , we picked only nematodes exhibiting bright expression of the respective GFP-fusion proteins . Day two adults were transferred to lifespan machine assay plates . A typical experiment consisted of up to four genotypes or conditions , with four assay plates of each genotype or condition , each assay plate containing a maximum of 40 nematodes , and 16 assay plates housed in the same scanner . All experiments were repeated at least once , yielding the same results . Scanner temperature was calibrated to 25°C with a thermocouple ( ThermoWorks USB-REF ) on the bottom of an empty assay plate . Death times were automatically detected by the lifespan machine’s image-analysis pipeline , with manual curation of each death time through visual inspection of all collected image data ( Stroustrup et al . , 2013 ) , without knowledge of genotype or experimental condition . E . coli HT115 ( DE3 ) bacteria with plasmids expressing double stranded RNA targeting specific genes were obtained from the Ahringer and Vidal libraries ( Kamath et al . , 2001; Rual et al . , 2004 ) . Empty vector plasmid pL4440 was used as control . Bacterial cultures were grown in LB broth with 100 μg/ml ampicillin at 37°C , induced with 0 . 1 M isopropyl-thiogalactopyranoside ( IPTG ) at 37°C for 4 hr , concentrated to an OD600 of 20 , and seeded onto NGM agar plates containing 50 μg/ml carbenicillin and 2 mM IPTG . Total RNA was extracted from day two adult animals that were transferred at the L4 stage onto NGM agar plates with 10 μg/ml FUDR seeded with E . coli OP50 and grown at 25°C . RNA extraction and cDNA preparation were performed as described ( Amrit et al . , 2019 ) . Quantitative RT-PCRs were performed using the Biorad CFX Connect machine . PCR reactions were undertaken in 96-well optical reaction plates ( Bio-Rad Hard Shell PCR Plates ) . A 20 µl PCR reaction was set up in each well using the SYBR PowerUp Green Master Mix ( Applied Biosystems , USA ) with 10 ng of the converted cDNA and 0 . 3 M primers . For each gene at least three independent biological samples were tested , each with three technical replicates . Primers used in this study include TTCCATTTCAAGCCTGCTC ( ctl-1 Fwd ) , ATAGTCTGGATCCGAAGAGG ( ctl-1 Rev ) , GGATTTGGACATGCTCCTC ( rpl-32 Fwd ) ( Amrit et al . , 2019 ) , and GATTCCCTTGCGGCTCTT ( rpl-32 Rev ) ( Amrit et al . , 2019 ) . RNA for sequencing was extracted from day 2 adult animals that were transferred at the L4 stage onto NGM agar plates seeded with E . coli OP50 and grown at 25°C . We adapted a nematode lysis protocol ( Ly et al . , 2015 ) for bulk lysis to pool 30 individuals per sample in 240 µL of lysis buffer , which gave a sufficiently low variation among replicates ( Figure 6—figure supplement 1A ) . This enabled us to collect three independent batches of three biological replicates per genotype , one batch per week , yielding a total of nine biological replicates for each genotype . cDNA preparation from mRNA was performed by SmartSeq2 as described ( Picelli et al . , 2014 ) . cDNA was purified using Agencourt AMPure XP magnetic beads . Nextera sequencing libraries were prepared according to the manufacturer’s protocol and purified twice with Agencourt AMPure XP magnetic beads . Paired-end libraries were sequenced on an Illumina HiSeq2500 with a read length of 50 bases and approx . 2 . 0 × 106 reads per sample . RNA-seq reads were aligned to the C . elegans Wormbase reference genome ( release WS265 ) using STAR version 2 . 6 . 0 c ( Dobin et al . , 2013 ) and quantified using featureCounts version 2 . 0 . 0 ( Liao et al . , 2014 ) , both using default settings . The reads count matrix was normalized using scran ( Lun et al . , 2016 ) . Principal component analysis was performed on the log ( normalized counts + 0 . 5 ) matrix with centring and no scaling . Differential analysis was performed using a negative binomial generalized linear model as implemented by DESeq2 ( Love et al . , 2014 ) to compare daf-7 ( ok3125 ) mutants against wildtype . A batch replicate term was added to the regression equation to control for confounding . To access the expression of catalases genes ( ctl-1 , ctl-2 and ctl-3 ) genome coverage of all reads was computed using BEDTools ( Quinlan and Hall , 2010 ) . Moreover , the multi-mapped read counts for all catalases were estimated using featureCounts with the -M option . Resulting counts were then compared using DESseq2 as described above . Gene functional enrichments were determined by using the WormBase Enrichment Suite ( Angeles-Albores et al . , 2016 ) . We clustered and plotted GO terms with q-value < 0 . 001 using REVIGO ( Supek et al . , 2011 ) . Curated gene expression data sets were obtained from WormExp ( Yang et al . , 2016 ) . Transgenic animals expressing a Bxy-CTL-1::GFP fusion under the control of the C . elegans ctl-1 promoter ( Hamaguchi et al . , 2019 ) were scored at the young-adult stage using a fluorescence dissection stereomicroscope ( Zeiss Discovery V12 ) under 100x magnification , following a scheme previously used to score a gcs-1p::GFP reporter with a similar pattern of intestinal expression ( Wang et al . , 2010 ) . Low: only anterior or posterior intestine with patches of GFP . Medium: anterior and posterior intestine with patches GFP , middle of the intestine with dim GFP . High: anterior and posterior intestine with non-patchy GFP expression , middle of the intestine with patchy or dim GFP . Very high: strong and non-patchy GFP expression throughout the intestine . Fluorescence imaging was conducted as previously described ( Romero-Aristizabal et al . , 2014 ) with an Axioskop 2 FS plus microscope ( Zeiss ) equipped with a D470/20x excitation filter , a 500dcxr dichroic mirror , and a HQ535/50 m emission filter ( all from Chroma ) , using a Plan-Apochromat 10 × 0 . 45 NA 2 mm working distance objective lens ( 1063–139 , Zeiss ) . Young adult worms were placed on petri plates with modified Nematode Growth Media ( to minimize background fluorescence ) containing 6 mM levamisole to immobilize the animals ( Romero-Aristizabal et al . , 2014 ) . Images were acquired with a Cool SNAP HQ2 14-bit camera ( Photometrics ) at 4 × 4 binning and 20 ms exposure . We performed background subtraction by removing the mode intensity value of the entire image from each pixel . This procedure removes the background due to the agar and the camera noise , since most pixels in our images were part of the background . All microscopy was performed at 22°C . Pharyngeal pumping was assayed for 30 s on day 2 adults at 25°C using a dissecting microscope under 100x magnification . Statistical analyses were performed in JMP Pro version 14 ( SAS ) . Survival curves were calculated using the Kaplan-Meier method . We used the log-rank test to determine if the survival functions of two or more groups were equal . For pumping-period assays , we used the Tukey HSD post-hoc test to determine which pairs of groups in the sample differ . We used ANOVA to determine whether the fold-change in gene expression of specific gene sets and of all genes were equal . For intestinal GFP expression assays , we used ordinal logistic regression to determine if expression levels were equal between groups . | Cells of all kinds often wage chemical warfare against each other . Hydrogen peroxide is often the weapon of choice on the microscopic battlefield , where it is used to incapacitate opponents or to defend against attackers . For example , some plants produce hydrogen peroxide in response to infection to fight off disease-causing microbes . Individual cells have also evolved defenses to prevent or repair ‘injuries’ caused by hydrogen peroxide . These are similar across many different species . They include enzymes called catalases , which break down hydrogen peroxide , and others to repair damage . However , scientists still do not fully understand how animals and other multicellular organisms might coordinate these defenses across their cells . Caenorhabditis elegans is a microscopic species of worm that lives in rotting fruits . It often encounters the threat of cellular warfare: many types of bacteria in its environment generate hydrogen peroxide , and some can make enough to kill the worms outright . Like other organisms , C . elegans also produces catalases to defend itself against hydrogen peroxide attacks . However , it must activate its defenses at the right time; if it did so when they were not needed , this would result in a detrimental energy ‘cost’ to the worm . Although C . elegans is a small organism containing only a defined number of cells , exactly why and how it switches its chemical defenses on or off remains unknown . Schiffer et al . therefore set out to determine how C . elegans controls these defenses , focusing on the role of the brain in detecting and processing information from its environment . Experiments looking at the brains of genetically manipulated worms revealed a circuit of sensory nerve cells whose job is to tell the rest of the worm’s tissues that they no longer need to produce defense enzymes . Crucially , the circuit became active when the worms sensed E . coli bacteria nearby . Bacteria in the same family as E . coli are normally found in in the same habitat as C . elegans and these bacteria are also known to make enzymes of their own to eliminate hydrogen peroxide around them . These results indicate that C . elegans can effectively decide , based on the activity of its circuit , when to use its own defenses and when to ‘freeload’ off those of neighboring bacteria . This work is an important step towards understanding how sensory circuits in the brain can control hydrogen peroxide defenses in multicellular organisms . In the future , it could help researchers work out how more complex animals , like humans , coordinate their cellular defenses , and therefore potentially yield new strategies for improving health and longevity . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"developmental",
"biology",
"neuroscience"
] | 2020 | Caenorhabditis elegans processes sensory information to choose between freeloading and self-defense strategies |
In mouse hairy skin , lanceolate complexes associated with three types of hair follicles , guard , awl/auchene and zigzag , serve as mechanosensory end organs . These structures are formed by unique combinations of low-threshold mechanoreceptors ( LTMRs ) , Aβ RA-LTMRs , Aδ-LTMRs , and C-LTMRs , and their associated terminal Schwann cells ( TSCs ) . In this study , we investigated the organization , ultrastructure , and maintenance of longitudinal lanceolate complexes at each hair follicle subtype . We found that TSC processes at hair follicles are tiled and that individual TSCs host axonal endings of more than one LTMR subtype . Electron microscopic analyses revealed unique ultrastructural features of lanceolate complexes that are proposed to underlie mechanotransduction . Moreover , Schwann cell ablation leads to loss of LTMR terminals at hair follicles while , in contrast , TSCs remain associated with hair follicles following skin denervation in adult mice and , remarkably , become re-associated with newly formed axons , indicating a TSC-dependence of lanceolate complex maintenance and regeneration in adults .
Our skin is the largest sensory organ of the body and is presented with an array of tactile stimuli , including indentation , stretch , vibration , and hair deflection ( Lumpkin et al . , 2010 ) . To detect , perceive , and respond to such diverse stimuli , morphologically and physiologically distinct classes of low-threshold mechanosensory neurons ( LTMRs ) innervate the skin and associate with cutaneous tactile end organs . These LTMRs , whose cell bodies are located in dorsal root ganglia ( DRG ) and cranial sensory ganglia , carry impulses from their endings in the skin to the central nervous system ( Abraira and Ginty , 2013; Lechner and Lewin , 2013 ) . LTMRs are classified as Aβ , Aδ or C based on their action potential conduction velocities ( Horch et al . , 1977; Rice and Albrecht , 2008 ) . C-LTMRs are unmyelinated and thus have the slowest conduction velocities , whereas Aδ-LTMRs and Aβ-LTMRs are lightly and heavily myelinated , exhibiting intermediate and rapid conduction velocities , respectively . LTMRs are also classified as slowly , intermediately , or rapidly adapting ( SA , IA , and RA-LTMRs ) according to their rates of adaptation to sustained mechanical stimuli ( Burgess et al . , 1968; Johnson and Hsiao , 1992 ) . Genetic labeling of Aβ RA-LTMRs , Aδ-LTMRs , and C-LTMRs has revealed that the cutaneous endings of each LTMR subtype in hairy skin form longitudinal lanceolate endings at one or more hair follicle subtypes ( Li et al . , 2011 ) . More than 90% of the body surface of most mammals is covered by hair follicles , which regulate body temperature , facilitate perspiration , and are involved in the perception of mechanosensory stimuli upon the skin . There are three major hair follicle subtypes in mouse back hairy skin: guard , awl/auchene , and zigzag . The three hair follicle subtypes emerge during different stages of hair follicle morphogenesis and have distinct morphologies and presumably distinct mechanical properties . Guard hairs , which make up ∼1% of the back skin hairs of the mouse , develop during the first wave of hair follicle morphogenesis , beginning E13–14 . 5 . These hairs have the longest hair shafts that have two rows of medulla cells and do not contain kinks . Awl hairs , which form during the second wave of murine hair morphogenesis , beginning E15–16 . 5 , are straight but are shorter than guard hairs and contain three or four rows of medulla cells . Auchene hairs are developmentally and morphologically identical to awl hairs , except that they have one kink in the hair . Awl/auchene hairs together make up ∼20% of back skin hairs and , for the purposes of this study , are considered together . Finally , zigzag hairs form during the third wave of hair morphogenesis beginning at E18 . 5 and are the most abundant type , making up ∼80% of hairs . Zigzag hairs have one row of medulla cells and at least two bends in their hair shaft ( Schlake , 2007; Driskell et al . , 2009; Li et al . , 2011 ) . In addition to their different developmental and morphological properties , zigzag , awl/auchene , and guard hair follicles are each associated with unique combinations of LTMRs . Indeed , guard hair follicles are innervated by Aβ RA-LTMR lanceolate endings and are also associated with Aβ SA1-LTMR Merkel endings . In contrast , awl/auchene hairs are triply innervated by inter-digitated Aβ RA- , Aδ- , and C-LTMR lanceolate endings , whereas zigzag hair follicles are innervated by inter-digitated Aδ- and C-LTMR lanceolate endings ( Li et al . , 2011 ) . Thus , guard , awl/auchene , and zigzag hairs , with their unique combinations of LTMR endings , play neurophysiologically distinct roles in mechanosensory transduction ( Iggo and Andres , 1982; Li et al . , 2011 ) . Light and electron microscopic studies have shown that longitudinal lanceolate endings are arranged parallel to the hair follicle long axis and that each ending is encased in finger-shaped terminal Schwann cell ( TSC ) processes ( Yamamoto , 1966; Halata , 1993; Kaidoh and Inoue , 2000; Takahashi-Iwanaga , 2000 ) . Furthermore , intercellular junctions are observed between axon terminals , TSCs , and hair follicle outer root sheath cells ( Kaidoh and Inoue , 2000 , 2008 ) , suggesting that physical interaction between these different cell types may be essential for mechanotransduction . However , the organization of TSCs and LTMR lanceolate endings , and their functional relationship for maintenance of lanceolate complexes are largely unknown . In this study , we used histologic and genetic approaches to investigate the functional organization , ultrastructural properties , and maintenance of lanceolate complexes associated with the three hair follicle subtypes of the mouse .
We previously reported molecular–genetic strategies that enable visualization of axonal endings of C- , Aδ- , and Aβ RA-LTMRs in the skin and spinal cord of the mouse . We found that C-LTMRs can be visualized in ThCreER; Rosa26LSL-tdTomato mice treated with 4-HT; Aδ-LTMRs are visualized using TrkB ( Nrtk2 ) tauEGFP knockin mice; and Aβ RA-LTMRs are labeled in Npy2r-GFP BAC transgenic mice ( Li et al . , 2011 ) . Remarkably , in back hairy skin of the mouse , C- , Aδ- , and Aβ RA-LTMRs form morphologically similar longitudinal lanceolate endings associated with hair follicles ( Figure 1 ) . Immunostaining with S100 , a glial cell marker , shows that each type of LTMR lanceolate ending is associated with TSCs , with cell bodies of TSCs residing at the base of lanceolate complexes . Moreover , TSCs associated with C- , Aδ- , and Aβ RA-LTMRs exhibit extensive longitudinal processes extending toward the skin surface . These finger-shaped processes are parallel to the long axis of the hair follicle and encase individual longitudinal lanceolate endings of C- , Aδ- , and Aβ RA-LTMRs ( Figure 1 ) . These observations are consistent with previous findings from our group and others ( Iggo and Andres , 1982; Li et al . , 2011; Woo et al . , 2012 ) . 10 . 7554/eLife . 01901 . 003Figure 1 . LTMRs and TSCs form palisade-like lanceolate complexes at hair follicles . ( A ) On back hairy skin sections from ThCreER;Rosa26tdTomato mice , C-LTMRs are visualized using tdTomato fluorescence ( green ) , while TSCs are labeled using S100 immunostaining ( red ) . C-LTMRs form longitudinal lanceolate endings associated with TSCs at zigzag and awl/auchene hair follicles . ( B ) Back hairy skin sections from TrkBtauEGFP animals were stained with anti-GFP to label Aδ-LTMR axonal terminals and anti-S100 ( red ) to label TSCs . Aδ-LTMRs form longitudinal lanceolate endings associated with TSCs at zigzag and awl/auchene hair follicles . ( C ) Back hairy skin sections from Npy2r-GFP animals were stained with anti-GFP to label Aβ RA-LTMR axonal terminals and anti-S100 ( red ) to label TSCs . At a representative awl/auchene hair follicle , Aβ RA-LTMRs form longitudinal lanceolate endings associated with TSCs . Similar patterns can be seen at guard hair follicles ( Li et al . , 2011 ) . Animals around 3 weeks of age were used for these experiments . Scale bar , 20 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 01901 . 003 Whether lanceolate complexes comprised of unique combinations of LTMR subtypes around the three types of hair follicles exhibit unique morphological , ultrastructural , or organizational patterns is an intriguing , unanswered question . Indeed , whole-mount immunostaining of back hairy skin using anti-S100 shows that lanceolate complexes associated with the three different hair follicle subtypes appear to contain different numbers of TSCs ( Figure 2A , B ) . Quantification of TSCs associated with individual hair follicles shows that individual awl/auchene hair follicles have slightly more TSCs compared with zigzag hair follicles , and that guard hair follicles have many more TSCs than both zigzag and awl/auchene hairs ( Figure 2C ) . Since zigzag , awl/auchene , and guard hair follicles have different diameters , we also compared the densities of TSCs at lanceolate complexes by dividing the number of TSCs by the circumference of individual hair follicles . Interestingly , TSC densities for lanceolate complexes associated with zigzag and awl/auchene hair follicles are comparable , while TSCs at guard hair follicles are considerably more compact ( Figure 2D ) . We next asked whether individual TSCs at zigzag , awl/auchene , and guard hair follicles have similar or distinct morphologies . To visualize the morphological properties of individual TSCs at the three hair follicle types , we crossed the Plp1CreER mouse line , which expresses CreER exclusively in glial cells ( Doerflinger et al . , 2003 ) , to a Rosa26YFP reporter mouse line , and achieved sparse labeling of TSCs by treating double transgenic animals with a single , low dose of tamoxifen ( 0 . 01 mg–0 . 03 mg; Figure 2E–G ) . Whole-mount immunostaining with anti-GFP allowed visualization of individual TSCs and measurement of the size of TSC cell bodies , as well as the width and length of their processes ( Figure 2E–G , H–J ) . We found that the morphologies of individual TSCs associated with zigzag and awl/auchene hair follicles are comparable . In contrast , TSCs at guard hair follicles have smaller cell bodies , and fewer and considerably longer processes ( Figure 2H–J ) . The greater numbers and unique morphologies of TSCs associated with guard hair follicles compared to those associated with zigzag and awl/auchene follicles , together with the finding that Aβ RA-LTMRs represent the sole longitudinal lanceolate endings at guard hairs , suggests unique mechanotranduction functions of guard hairs in detecting hair movements . 10 . 7554/eLife . 01901 . 004Figure 2 . Guard , awl/auchene , and zigzag hair follicles have different numbers and morphologies of TSCs . ( A and B ) Whole-mount immunostaining of back hairy skin using anti-S100 shows that awl/auchene ( A ) and zigzag ( Z ) hair follicles ( in panel A ) and guard ( G ) hair follicles ( in panel B ) have different numbers of TSCs . Scale bar , 20 μm . ( C ) Comparisons of numbers of TSCs at individual hair follicles . Awl/auchene hairs ( 5 . 0 ± 0 . 2 TSCs , n = 24 hair follicles ) have slightly more TSCs than zigzag hairs ( 3 . 7 ± 0 . 1 TSCs , n = 64 hair follicles ) ( p<0 . 001 ) . Guard hairs ( 13 . 4 ± 1 . 3 TSCs , n = 5 hair follicles ) have many more TSCs than awl/auchene and zigzag hairs ( p<0 . 001 for both comparisons ) . ( D ) Densities of TSCs at individual hair follicles were calculated by dividing the number of TSCs by the circumference of the hair follicles . Although TSC densities are comparable between zigzag ( 4 . 3 ± 0 . 1 TSCs/100 μm ) and awl/auchene hairs ( 4 . 7 ± 0 . 2 TSCs/100 μm ) ( p=0 . 084 ) , guard hair follicles ( 7 . 7 ± 0 . 9 TSCs/100 μm ) have almost twofold higher densities of TSCs compared to zigzag and awl/auchene hair follicles ( p<0 . 001 for both comparisons with zigzag and awl/auchene hair follicles ) . ( E–G ) . In Plp1CreER;Rosa26YFP animals treated with 0 . 01 mg of tamoxifen , TSCs were sparsely labeled to visualize the morphologies of individual TSC associated with different hair follicle types . Insets in E–G are schematic images of TSCs to summarize the differences observed between TSCs associated with guard hairs vs awl/auchene and zigzag hairs . Scale bar , 20 μm . ( H–J ) Sizes of TSC cell bodies ( H ) , widths of total processes of individual TSCs ( I ) and lengths of TSC processes ( J ) were measured at zigzag , awl/auchene , and guard hair follicles . Although all three parameters are comparable between zigzag and awl/auchene hair follicles , indicating similar morphologies , individual TSCs at guard hair follicles have significantly smaller cell sizes , and narrower and longer processes compared to the other two hair follicle types ( Zigzag hair follicles: cell body size , 64 . 3 ± 4 . 5 μm2; width of processes , 17 . 0 ± 1 . 0 μm; length of processes , 13 . 9 ± 0 . 4 μm; n = 25 hair follicles . Awl/auchene: cell body size , 70 . 1 ± 7 . 6 μm2; width of processes , 16 . 6 ± 0 . 7 μm; length of processes , 13 . 6 ± 0 . 6 μm; n = 10 hair follicles . Guard hair follicles: cell body size , 50 . 0 ± 3 . 5 μm2 , p<0 . 05 compared with zigzag or awl/auchene hairs; width of processes , 13 . 3 ± 0 . 9 μm , p<0 . 05 compared with zigzag or awl/auchene hairs; length of processes , 27 . 3 ± 1 . 1 μm , p<0 . 001 compared with zigzag or awl/auchene hairs; n = 16 hair follicles ) . Animals around 3 weeks of age were used in these whole-mount immunostaining experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 01901 . 004 Three and two distinct LTMR subtypes form lanceolate endings associated with most if not all awl/auchene and zigzag hair follicles , respectively ( Li et al . , 2011 ) . Consistent with this , recent studies indicate that mechanosensory neurons with distinct molecular features can innervate the same hair follicle ( Heidenreich et al . , 2012; Wende et al . , 2012; Woo et al . , 2012 ) . These observations raise the intriguing question of whether physiologically distinct LTMR subtypes that innervate the same hair follicle associate with the same or different TSCs . Based on the interdigitated patterns of LTMR axonal endings at zigzag and awl/auchene hair follicles ( Li et al . , 2011 ) , there are at least two possible organizational patterns of their associated lanceolate complexes: 1 ) Each TSC hosts only one type of LTMR , and the interdigitated axonal terminals of different LTMRs are associated with interdigitated processes of TSCs; 2 ) Processes of TSCs at individual hair follicles are tiled and each TSC hosts interdigitated axonal terminals from two or more LTMR subtypes . To distinguish between these possibilities , we first crossed the Plp1CreER mouse line to the multicolor Cre-reporter Rosa26-Confetti ( Schepers et al . , 2012 ) to label TSCs with multiple reporters . Tamoxifen treatment of Plp1CreER;Rosa26-Confetti mice allows individual TSCs to express only one of the four fluorescent proteins encoded by the Rosa26-Confetti allele . Thus , adjacent TSCs will express different reporters allowing visualization of the organization of TSCs and their associated LTMR endings at individual hair follicles . In trunk hairy skin of Plp1CreER;Rosa26-Confetti animals , adjacent TSCs that are labeled by red and green fluorescence reporters exhibit non-overlapping sets of processes ( Figure 3A , B ) . Thus , the processes of TSCs associated with individual hair follicles are tiled . Moreover , this finding indicates that the organization of individually tiled TSCs and different LTMR endings can be visualized simply by immunostaining with glial cell markers . Interestingly , in back hairy skin sections from ThCreER;Rosa26tdTomato;TrkBtauEGFP mice , in which C-LTMRs , Aδ-LTMRs , and TSCs are visualized with tdTomato , GFP , and S100 immunostaining , respectively , we found that C-LTMRs and Aδ-LTMRs form interdigitated lanceolate endings associated with processes belonging to the same TSC ( Figure 3C , C′ ) . Similar patterns were observed in skin sections from ThCreER;Rosa26tdTomato;Npy2r-GFP mice in which C-LTMRs and Aβ RA-LTMRs were labeled with tdTomato and GFP , respectively ( Figure 3—figure supplement 1 ) . Thus , TSCs are tiled , and the processes of an individual TSC encase the axonal endings of multiple , physiologically distinct LTMR subtypes . 10 . 7554/eLife . 01901 . 005Figure 3 . TSCs are tiled , and a single TSC can host axonal endings from multiple , physiologically distinct LTMR subtypes . ( A and B ) On back hairy skin sections taken from Plp1CreER;Confetti animals , TSCs are randomly labeled with either green or red fluorescence . Neighboring TSCs at both guard ( labeled with ‘G’ in B ) and non-guard ( A ) hair follicles are tiled , exhibiting no overlapping processes . This experiment was done using 109 non-guard hair follicles and 23 guard hair follicles in two animals , all of which exhibited mosaic fluorescence labeling . 100% of these hair follicles exhibited tiled arrangements of TSCs at individual hair follicles . ( C and C′ ) On back hairy skin sections from ThCreER;Rosa26tdTomato;TrkBtauEGFP mice , C-LTMRs were labeled with tdTomato fluorescence ( red ) , Aδ-LTMRs were labeled with anti-GFP ( green ) and TSCs were stained with anti-S100 ( cyan ) . Shown here is an example in which C-LTMR and Aδ-LTMR endings associate with different processes of the same TSC at a zigzag or awl/auchene hair follicle . Thus , a single TSC hosts more than one type of LTMR axonal terminal . Four mice were used for the triple labeling experiment and identical results were observed in each . C′ shows higher magnification of the TSC in the middle of the lanceolate complex shown in C . Scale bars , 20 μm for A and B , 10 μm in C , 5 μm in C′ . Animals around 3 weeks of age were used for these experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 01901 . 00510 . 7554/eLife . 01901 . 006Figure 3—figure supplement 1 . A single TSC hosts axonal endings from Aβ , RA-LTMRs , and C-LTMRs . ( A and B ) On back hairy skin sections from ThCreER;Rosa26tdTomato;Npy2r-GFP mice , C-LTMRs were labeled with tdTomato fluorescence ( red ) , Aβ , RA-LTMRs were labeled with anti-GFP ( green ) and TSCs were stained with anti-S100 ( cyan ) . Shown here are two examples in which C-LTMR ( highlighted by red arrows in ( B ) and Aβ , RA-LTMRs ( highlighted by green arrows in ( B ) endings associate with different processes of the same TSC at awl/auchene hair follicles . Boundaries between adjacent TSCs can be roughly judged by assuming symmetric expansion of processes from their cell bodies . Scale bar , 10 μm . Three mice around 3 weeks of age were used for these experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 01901 . 006 We next investigated the ultrastructural basis of mechanical responsiveness of lanceolate endings , and in particular the ultrastructural relationships between Aβ RA-LTMR , Aδ-LTMR , and C-LTMR axonal terminals , TSCs , and the three hair follicles types . Transverse , ultrathin sections of lanceolate complexes associated with guard , awl/auchene , and zigzag hair follicles were collected , stained , and examined by electron microscopy ( EM ) . Lanceolate complexes at all three hair follicle types appear as arrays of individual units composed of blade-shaped TSC processes and LTMR axonal endings aligned against the basal lamina of the hair follicle ( Figure 4A ) . At guard hair follicles , each Aβ RA-LTMR lanceolate axonal terminal ( pseudo-colored in green ) is encased by TSC processes ( colored in pink ) on three sides ( Figure 4B ) . Thus , cross sections of Aβ RA-LTMR lanceolate complexes appear triangular in shape . There are frequent gaps between the TSC processes at the three edges of the axon . Interestingly , at the gaps or openings facing the hair follicle , the Aβ RA-LTMR axon terminals form small protrusions that are closely aligned , together with adjacent TSC processes , against the basal lamina of the hair follicle epithelial ( Figure 4B , B′ , arrows ) . In comparison , axonal protrusions through the other gaps , on sides facing away from the hair follicle epithelial cell , are less pronounced or absent and , when present , variable in width . Furthermore , the cytoplasm of Aβ RA-LTMR axons is densely packed with mitochondria . At awl/auchene and zigzag hair follicles , axonal terminals are encased on two sides by TSC processes , and thus cross sections of these lanceolate complexes appear spindle-shaped ( Figure 4C , D ) . Compared to lanceolate complexes at guard hair follicles , TSC processes associated with awl/auchene and zigzag follicles tend to be thicker ( widths of EM cross sections of TSC processes at guard hairs: 302 . 7 ± 132 . 7 nm , 17 process sections from two mice; awl/auchene hairs: 562 . 3 ± 76 . 5 nm , 21 process sections from two mice; zigzag hairs: 397 . 7 ± 19 . 4 nm , 32 process sections from three mice; these comparisons are not statistically significant ) , and the axonal processes they encase are thinner ( areas of EM cross sections of axon terminals at guard hairs: 2 . 33 ± 0 . 06 μm2 , 5 axon sections from two mice; awl/auchene hairs: 0 . 76 ± 0 . 05 μm2 , 10 axon sections from two mice , p=0 . 002 compared with guard hairs; zigzag hairs: 0 . 57 ± 0 . 10 μm2 , 19 axon sections from three mice , p=0 . 001 compared with guard hairs ) . In addition , the gaps between TSC processes that expose the axonal endings on the side of hair follicle epithelial cells are larger than those found at guard hairs ( Figure 4C , C′ , D , D′ , arrows ) ( widths of gaps of TSC processes toward guard hair follicles: 82 . 8 ± 13 . 9 nm , 7 axon sections from two mice; awl/auchene hair follicles: 162 . 9 ± 11 . 9 nm , 22 axon sections from two mice , p=0 . 05 compared with guard hairs; zigzag hair follicles: 196 . 6 ± 11 . 9 nm , 35 axon sections from three mice , p=0 . 01 compared with guard hairs ) . Interestingly , at awl/auchene and zigzag follicles , adjacent LTMR axonal endings often share the same TSC process; this was rarely observed for Aβ RA-LTMR endings at guard hairs . Figure 4C shows one such example in which three adjacent axon terminals associated with an awl/auchene hair follicle intervene among four TSC processes and appear bound together within a single unit . Figure 4D shows another example at a zigzag hair follicle , in which two adjacent axon terminals are tightly associated with three TSC processes to form a single unit ( Figure 4D , lower left corner ) . An additional , distinguishing feature of lanceolate complexes at awl/auchene and zigzag hair follicles , compared to those at guard hairs , is that the ultrastructural properties of the lanceolate axons are heterogeneous . Indeed , some axons contain many mitochondria , similar to the Aβ RA-LTMR endings at guard hair follicles , whereas others contain few mitochondria ( Figure 4C , D; Figure 6—figure supplement 2A ) . This ultratructural feature of axon terminals is observed across more than 30 serial EM sections , which spans more than 3 μm along the longitudinal axis of a lanceolate complex ( Figure 5A , B ) . Since lanceolate complexes at awl/auchene hair follicles are triply innervated by Aβ RA- , Aδ- , and C-LTMRs while those at zigzag hair follicles are innervated by Aδ- and C-LTMRs , axons exhibiting different abundances of mitochondria may derive from distinct LTMR subtypes . Thus , the cellular and ultrastructural properties of lanceolate complexes differ among the hair follicle subtypes , and these differences may underlie unique sensitivities or responses of each LTMR subtype following skin indentation or hair deflection . 10 . 7554/eLife . 01901 . 007Figure 4 . The ultrastructural relationships between LTMRs , TSCs , and hair follicle epithelial cells at the three hair follicle subtypes . ( A ) A transmission electron microscopic image of a cross section through a lanceolate complex at a guard hair follicle . Repeating units of axon terminals and TSC processes are regularly arranged in a single layer surrounding the hair follicle ( HF ) . In each unit , a mitochondria-rich axonal terminal ( N ) is encased by TSC processes ( * ) . ( B ) A cross section of the same guard hair follicle shown in A . Axon terminals are pseudo-colored in green; TSC processes are colored in pink; the hair follicle epithelial cell is colored in yellow . Each unit is composed of one axonal terminal encased by two or three TSC processes on three sides . Axon terminals contain a large number of mitochondria . Note that small protrusions of axons ( arrows ) and TSC processes are precisely aligned against the basal lamina of the hair follicle . ( C ) A cross section of a lanceolate complex of an awl/auchene hair follicle . Each axon terminal is encased by two TSC processes on two sides . More than one axonal ending and its associated TSC processes is often packed into a single ‘unit’ . Shown here is one unit composed of three axon terminals intervening among four TSC processes . Compared to guard hair follicles , awl/auchene hair follicles exhibit more area of exposed axon terminal membrane facing the outer root sheath cell of the follicle ( arrow ) . In addition , TSC processes are thicker than those at guard hair follicles , while axon diameters are smaller . Also , the axon terminals appear heterogeneous: some have few mitochondria ( the middle axon ) and some many mitochondria ( the other two axons ) . ( D ) A cross section of a lanceolate complex associated with a zigzag hair follicle . Similar to awl/auchene hair follicles , each axon terminal is encased by two TSC processes on two sides . One of the units in the lower left corner is composed of two axon terminals encased by three TSC processes . As with the awl/auchene complexes , and in comparison to guard hairs , areas of exposed axon terminal membrane adjacent to the hair follicle epithelial cell are large ( arrows ) . In addition , similar to awl/auchenes , the zigzag TSC processes are thicker and axonal sections are smaller compared to guard hair follicles . Axon terminals also display varying amounts of mitochondria . ( B′ , C′ and D′ ) High magnification EM images of cross sections of guard ( B′ ) , awl/auchene ( C′ ) , and zigzag ( D′ ) hair follicles show the gaps between TSC processes and the axon protrusions in between ( arrows ) . Animals around 4 weeks of age were used for these experiments . Scale bars , 2 μm in A; 500 nm for B , C and D; 100 nm for B′ , C′ and D′ . DOI: http://dx . doi . org/10 . 7554/eLife . 01901 . 00710 . 7554/eLife . 01901 . 008Figure 5 . Serial EM cross-sections of C-LTMR and Aδ-LTMR axonal endings at a zigzag hair follicle . Serial EM cross-sections spanning more than 3 µm along two different LTMR axonal terminals of lanceolate complex at a zigzag hair follicle were collected . Representative images that are approximately 0 . 4 μm to 0 . 5 μm apart were shown . ( A ) A1 to A8 are serial EM cross-sections of an axonal terminal that has a relatively small number of mitochondria with low electron density . Further analyses ( Figure 6 , Figure 6—figure supplement 2 ) show that this axonal ending type is likely to be C-LTMR . ( B ) B1 to B8 are serial EM cross-sections of an axonal terminal that is packed with mitochondria with relatively high electron density . Further analyses ( Figure 6 , Figure 6—figure supplement 2 ) show that this axonal ending type is likely to be Aδ-LTMR . Animals around 4 weeks of age were used for these experiments . Scale bar , 500 nm . DOI: http://dx . doi . org/10 . 7554/eLife . 01901 . 008 We next sought to compare the morphological and ultrastructural properties of individually defined Aβ RA-LTMR , Aδ-LTMR , and C-LTMR endings using EM . We found that Wnt1-Cre;TrkAf/f mice , in which the nerve growth factor ( NGF ) receptor gene TrkA ( Nrtk1 ) is selectively ablated in DRG neurons and other cells of neural crest origin , display a complete loss of C-LTMRs ( Figure 6A , A′ ) . Unexpectedly , lanceolate endings of Aβ RA-LTMRs at awl/auchene and guard hair follicles are also missing in Wnt1-Cre;TrkAf/f mice ( Figure 6B , B′ ) , although Aβ RA-LTMR neurons labeled by Npy2r-GFP are still present in DRGs ( Figure 6—figure supplement 1 ) . Thus , C-LTMR and Aβ RA-LTMR endings in the skin are dependent on NGF–TrkA signaling for development . In contrast , the axonal endings of Aδ-LTMRs at both awl/auchene and zigzag hair follicles remain intact in these mutant mice ( Figure 6C , C′ ) . Therefore , by visualizing lanceolate complexes at zigzag hairs by EM in control and Wnt1-Cre;TrkAf/f animals , we can distinguish between the C-LTMR endings that are missing in Wnt1-Cre;TrkAf/f mutants , and Aδ-LTMRs endings that are intact . Interestingly , while axons associated with zigzag hair follicles in control mice have variable numbers of mitochondria , nearly all of the axons at zigzag hair follicles in Wnt1-Cre;TrkAf/f mice exhibit a high density of mitochondria ( Figure 6D , E; Figure 6—figure supplement 2B ) . Assuming that the TrkA mutation has no effect on mitochondria density in Aδ-LTMR axonal terminals , this finding indicates that the lanceolate endings with few mitochondria belong to C-LTMRs while those with abundant mitochondria belong to Aδ-LTMRs or Aβ RA-LTMRs ( Figure 5; Figure 6F–H ) . 10 . 7554/eLife . 01901 . 009Figure 6 . Identification of C-LTMR , Aδ-LTMR , and Aβ RA-LTMR axonal endings using EM . ( A and A′ ) DRG sections from Wnt1Cre;TrkAf/f ( A′ ) and control ( A ) animals were stained with anti-TH . TH+ C-LTMRs are nearly completely lost in TrkA conditional knockout animals compared to control . ( B and B′ ) Whole-mount GFP immunostaining of back skin samples from Wnt1Cre;TrkAf/f;Npy2r-GFP ( B′ ) and control ( B ) animals shows that cutaneous innervation of Npy2r-GFP+ Aβ RA-LTMRs at hair follicles is almost completely lost in TrkA conditional knockout animals compared to control . ( C and C′ ) GFP immunostaining of back skin sections from Wnt1Cre;TrkAf/f;TrkBtauEGFP ( C′ ) and TrkBtauEGFP control ( C ) animals shows that innervation of hair follicles by TrkBtauEGFP+ Aδ-LTMRs remains intact in the TrkA conditional knockout animals compared to control . ( D and E ) Cross sections of lanceolate complexes at zigzag hair follicles from Wnt1Cre;TrkAf/f ( E ) and control ( D ) mice . Axon terminals are pseudo-colored in green; TSC processes are colored in pink; hair follicle epithelial cells are colored in yellow . As shown in Figure 4D , axon terminals at the control zigzag hair follicle have varying numbers of mitochondria . In contrast , all axons at the zigzag hair follicle from Wnt1Cre;TrkAf/f animals exhibit abundant clusters of mitochondria . Thus , axons containing few mitochondria are C-LTMRs , which are lost in TrkA conditional knockout animals , whereas axons containing abundant mitochondria are Aδ-LTMRs , which remain intact in TrkA conditional knockout animals . ( F ) A cross section of a lanceolate complex at a wild-type guard hair follicle . All axons associated with guard hair lanceolate complexes are densely packed with mitochondria and are derived from Aβ RA-LTMRs . ( G ) A cross section of a lanceolate complex at a wild-type awl/auchene hair follicle . Axons with few mitochondria are C-LTMRs; axons with abundant mitochondria are either Aβ RA-LTMRs or Aδ-LTMRs . ( H ) A cross section of a lanceolate complex at a wild-type zigzag hair follicle . Axons with few mitochondria are C-LTMRs , whereas axons with abundant mitochondria are Aδ-LTMRs . Animals around 4 weeks of age were used for these experiments . Scale bars , 50 μm for A and A′; 100 μm for B and B′; 20 μm for C and C′; 500 nm for D–H . DOI: http://dx . doi . org/10 . 7554/eLife . 01901 . 00910 . 7554/eLife . 01901 . 010Figure 6—figure supplement 1 . Aβ RA-LTMR neurons remain intact in Wnt1Cre;TrkAf/f animals . ( A and B ) DRG sections from Wnt1Cre;TrkAf/f;Npy2r-GFP ( B ) and control , TrkAf/f;Npy2r-GFP ( A ) animals were stained with anti-GFP . Aβ RA-LTMR neurons labeled by Npy2r-GFP were still present in DRGs of these TrkA conditional knockout animals ( B ) . Scale bar , 20 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 01901 . 01010 . 7554/eLife . 01901 . 011Figure 6—figure supplement 2 . Quantification of mitochondrial abundance . Ratios of mitochondrial area and axon area in EM cross sections of individual LTMR endings were quantified . ( A ) Histograms showing frequency distributions of mitochondrial area ratios of LTMR endings at guard ( 8 LTMR axons ) , awl/auchene ( 24 LTMR axons ) and zigzag hair follicles ( 24 LTMR axons ) from wild-type animals . Mitochondrial area ratios appear homogenously high in axons associated with guard hair follicles . In contrast , a heterogenous , bimodal distribution of mitochondria area ratios is observed in axons associated with awl/auchene and zigzag hairs . ( B ) Histograms showing frequency distributions of mitochondrial area ratios of LTMR endings at TrkA f/f control ( 35 LTMR axons ) vs Wnt1Cre;TrkAf/f mutant zigzag hair follicles ( 16 LTMR axons ) . A bimodal distribution of mitochondrial area ratios is observed in control LTMR endings . In contrast , the mitochondrial abundance in LTMR endings associated with TrkA mutant hair follicles appears homogenous , exhibiting a unimodal distribution of high mitochondrial area ratios . This analysis indicates that the LTMR axonal endings at TrkA mutant zigzag hair follicles , which exhibit a high mitochondrial density , belong to Aδ-LTMRs , while those that exhibit a low mitochondria density and are absent in TrkA mutants belong to C-LTMRs . Animals around 4 weeks of age were used for these experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 01901 . 011 Several additional interesting ultrastructural features of LTMR lanceolate complexes at all three hair follicle subtypes were observed . In particular , numerous small vesicles are located within the cytoplasm of lanceolate endings of all LTMR subtypes ( Figure 7A–C , arrows ) . Recent studies suggest that these are glutamate-containing vesicles , which may mediate communication with TSCs and thereby regulate lanceolate complex assembly and maintenance ( Woo et al . , 2012; Banks et al . , 2013 ) . Additionally , many pinocytotic vesicles are associated with both the inner and outer surfaces of TSC processes ( Figure 7A–C , arrowheads ) . Similar observations have been made for lanceolate complexes of hair follicles from facial skin or ear hairy skin in mouse , rat , and humans ( Yamamoto , 1966; Cauna , 1969; Hashimoto , 1972; Kaidoh and Inoue , 2000 ) , although the sensory neuron subtype under investigation in those studies was not known . Additionally , a network of longitudinally-oriented collagen fibers can be observed in the extracellular space between lanceolate endings and basement membrane of the follicle and appear as circular structures in EM cross-sections ( Figure 6D , arrow ) . Another network of horizontally oriented collagen fibers appear as stripes around longitudinal lanceolate endings in the outer circle , presumably surrounding circumferential endings ( Figure 6D , arrowhead ) ( Low , 1962; Parakkal , 1969 ) . Such an alignment of collagen fiber networks may provide structural rigidity important for transferring hair follicle deflection into distortion or compression of lanceolate complexes and thereby activation of LTMRs . 10 . 7554/eLife . 01901 . 012Figure 7 . Ultrastructural features of lanceolate complexes revealed by EM using tannic acid-treated specimens . ( A–C ) Cross sections of lanceolate complexes at guard , awl/auchene , and zigzag hair follicles . Small vesicles can be observed within axon terminals ( arrows in A–C ) . TSC processes contain fine filaments that are nearly parallel to the long axis of the follicle and therefore appear as dark spots within the cytoplasm . Numerous pinocytotic vesicles are associated with both the inner and outer surfaces of TSC processes ( arrowheads in A–C ) . ( D–G ) Hemidesmosomes are seen along plasma membranes of hair follicle outer root sheath cells that face LTMR axons and TSC processes ( white arrowheads ) . Fine filament-like structures emanate from the hemidesmosomes , traverse the basal lamina , and form contacts with LTMR axon terminals and TSC processes . Axon terminals are labeled with ‘N’; TSC processes are labeled with ‘S’ . Animals around 4 weeks of age were used in these experiments . Scale bars , 100 nm . DOI: http://dx . doi . org/10 . 7554/eLife . 01901 . 012 To further assess the ultrastructural relationship between physiologically distinct lanceolate complex subtypes and the three hair follicle types , we prepared hair follicle specimens for EM analysis using tannic acid treatment , which preserves morphology and enhances visualization of extracellular matrix components ( Dingemans and van den Bergh Weerman , 1990 ) . Intriguingly , there is an abundance of prominent hemidesmosomes located along the outer membranes of the outer root sheath epithelial cells of the three hair follicle subtypes ( Figure 7D–G , white arrow heads ) . Fine filament-like structures emanate from these hemidesmosomes , pass through the basal lamina , and appear to form direct contacts with the membranes of Aβ RA-LTMR , Aδ-LTMR , and C-LTMR lanceolate axonal endings as well as TSC processes . These tether-like filaments are approximately 100 nm in length and may be similar to anchoring filaments or anchoring fibrils , which are essential components of dermo–epidermal junctions connecting basal keratinocytes with dermal cell types ( Keene et al . , 1987; Regauer et al . , 1990; Burgeson and Christiano , 1997 ) . It is tempting to speculate that these filamentous connections between hair follicle epithelial cells and both LTMR lanceolate endings and TSCs are involved in assembly , maintenance , or mechanotransduction at lanceolate complexes . Consistent with the latter , 100-nm long , tether-like structures were observed in DRG sensory neurons grown in cell culture , where they may be required for mechanosensitive currents of DRG sensory neurons ( Hu et al . , 2010 ) . The intricate structural relationship between LTMRs and TSCs suggests that both cells types are essential for function and maintenance of lanceolate complexes at guard , awl/auchene , and zigzag hair follicles . However , the relative contributions of LTMRs and TSCs for the maintenance of lanceolate complex integrity are unknown . To address this , we first examined the integrity of LTMR axonal terminals following genetic ablation of Schwann cells . We crossed the glial-specific Plp1CreER mouse line to the Rosa26hDTR reporter line , which expresses the human diphtheria toxin receptor ( hDTR ) in a Cre-dependent manner ( Buch et al . , 2005 ) . Plp1CreER;Rosa26hDTR animals were treated with tamoxifen at P16 to activate Cre recombinase and thus glial cell-specific expression of hDTR . These animals were then subjected to injections of DTX at P24 and P27 , which results in selective ablation of hDTR-expressing cells ( Saito et al . , 2001; Cavanaugh et al . , 2009 ) . We observed efficient ablation of TSCs at hair follicles in DTX-treated Plp1CreER;Rosa26hDTR animals , compared to DTX-treated control animals ( Figure 8 ) . In these experiments , TSCs were visualized using tdTomato fluorescence in animals harboring the Rosa26tdTomato allele ( Figure 8 ) . Remarkably , 100% of the hair follicles lacking TSCs exhibited a complete loss of longitudinal lanceolate endings and a partial loss of circumferential endings labeled by Tuj1 and NFH immunostaining ( Figure 8A , A′ , B , B′ ) . A partial loss of CGRP+ circumferential endings was also observed ( Figure 8C , C′ ) . These findings indicate that LTMR axon terminals retract from hair follicles following ablation of TSCs . The disappearance of longitudinal lanceolate endings is coincident with degeneration of TSCs , while circumferential lanceolate endings are more resilient and remain in the absence of TSCs ( Figure 8—figure supplement 1 ) . The partial loss of CGRP+ circumferential axons in the absence of TSCs may be due to instability of sensory–neural complexes at hair follicles that results from the loss of TSCs and LTMR lanceolate endings . The PLP gene is expressed in almost all glial subtypes in the periphery ( Jessen and Mirsky , 2005 ) . Due to limitations of our genetic tools , we cannot further assess the contributions of other glial cell types , such as myelinating and non-myelinating Schwann cells to the anchoring of LTMR endings at hair follicles . However , the fact that LTMR longitudinal lanceolate endings at hair follicles are severely affected regardless of whether they are heavily myelinated ( Aβ RA-LTMR ) , lightly myelinated ( Aδ-LTMR ) or unmyelinated ( C-LTMR ) , whereas unmyelinated free nerve endings remain relatively intact ( Figure 8A , A′ , white squares ) , strongly suggests that TSCs play a major role in maintaining the integrity of lanceolate complexes at hair follicles . 10 . 7554/eLife . 01901 . 013Figure 8 . Genetic ablation of TSCs leads to loss of LTMR innervation at hair follicles . In skin sections from Plp1CreER;Rosa26tdTomato ( A–C ) and Plp1CreER;Rosa26hDTR;Rosa26tdtomato ( A′–C′ ) mice , TSCs were visualized by tdTomato fluorescence . In Plp1CreER;Rosa26hDTR;Rosa26tdtomato animals , treatments with tamoxifen and DTX lead to complete loss of TSCs at hair follicles ( A′–C′ ) . In the absence of TSCs , Tuj1 ( A and A′ , green ) and NFH staining ( B and B′ , green ) shows a complete loss of longitudinal axonal terminals and a partial loss of circumferential axons at the presumptive region for lanceolate complexes; CGRP+ peptidergic nociceptor fibers were also partially lost ( C and C′ , green ) . Small white squares at A and A′ highlight free nerve endings in the epidermis and dermis . Animals around 4 weeks of age were used in these experiments . Scale bars , 50 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 01901 . 01310 . 7554/eLife . 01901 . 014Figure 8—figure supplement 1 . Loss of LTMR endings is coincident with degeneration of TSCs . In skin sections from Plp1CreER;Rosa26tdTomato ( A ) and Plp1CreER;Rosa26hDTR;Rosa26tdtomato ( B ) mice , TSCs were visualized by tdTomato fluorescence . Tuj1 staining shows that at hair follicles with partial loss of TSCs , there is a corresponding loss of longitudinal lanceolate ending while circumferential endings remain . DOI: http://dx . doi . org/10 . 7554/eLife . 01901 . 014 We next asked the reciprocal question; are lanceolate axonal endings required for the maintenance of TSCs ? To denervate hair follicles in back hairy skin , dorsal cutaneous nerves that innervate the right side of the back of adult animals were severed , while nerves innervating the left side remained intact , providing a control . As expected , 17 days following surgery , cutaneous nerves innervating the right , surgically denervated side of the animal had completely degenerated ( Figure 9A′ ) , whereas cutaneous innervation of the left , control side was unaffected ( Figure 9A ) . Strikingly , TSCs , labeled by S100 immunostaining , remained intact and in close association with hair follicles in the absence of LTMR endings ( Figure 9B , B′ ) . Even 44 days following skin denervation , TSCs at denervated hair follicles remained intact ( Figure 9C , C′ , D , D′ , E , E′ ) . In contrast , both Schwann cells associated with circumferential axonal terminals at hair follicles ( Figure 9D , D′; white arrow ) and myelinating Schwann cells in the dermis , visualized by S100 and MBP immunostaining ( Figure 9E , E′ ) , had degenerated in the absence of cutaneous nerves . Similar observations were made when TSCs were genetically labeled using Plp1CreER;Rosa26GCaMP3 mice that had received tamoxifen treatment ‘prior’ to surgical denervation . As in the aforementioned S100 immunostaining experiments , GFP immunostaining of skin sections ( to detect GCaMP3 ) showed that TSCs that were genetically labeled prior to denervation are morphologically intact following surgical denervation of the skin ( Figure 9F , F′ ) . In addition , GFP+ Schwann cells that did not express either S100 or MBP were also observed in the dermis , indicating the presence of PLP-expressing immature Schwann cells that may function to regenerate myelin sheaths ( Figure 9F′ , Figure 9—figure supplement 1 ) . We quantified the number of TSCs at individual guard hair and non-guard hair follicles ( awl/auchene and zigzag hairs combined ) 5 weeks following skin denervation . This quantitative assessment revealed that , despite complete loss of cutaneous LTMR endings on the denervated side of the animal , comparable numbers of TSCs were present at hair follicles on control and denervated sides ( Figure 9G ) . Therefore , TSCs associated with Aβ RA-LTMRs , Aδ-LTMRs , and C-LTMRs , maintain both their location and structural integrity in the absence of axonal processes in adult mice , for at least several weeks . 10 . 7554/eLife . 01901 . 015Figure 9 . TSCs remain intact and associated with hair follicles following dorsal cutaneous nerve axotomy and distal axon degeneration . ( A and A′ ) Whole-mount immunostaining of S100 ( red ) and NFH ( green ) shows that 17 days after dorsal cutaneous nerve axotomy , while NFH+ cutaneous nerve have completely degenerated in the denervated skin ( A′ , right back skin ) compared to the control skin ( A , left back skin ) , TSCs remain intact and associated with hair follicles ( A′ ) . ( B and B′ ) Enlarged views of boxed regions in A and A′ . At control hair follicles ( B ) , TSCs are associated with longitudinal lanceolate endings and circumferential endings ( arrow ) . At denervated hair follicles ( B′ ) , TSCs that were associated with longitudinal lanceolate endings prior to denervation appear normal , while TSCs associated with circumferential axonal terminals ( arrows ) are partially lost ( arrow ) . ( C and C′ ) Whole-mount immunostaining of S100 ( red ) and NFH ( green ) shows that , 44 days after axotomy , TSCs in the denervated skin ( C′ ) remain intact . ( D and D′ ) Enlarged views of boxed regions in C and C′ show that compared to TSCs in the control skin ( D ) , TSCs that were associated with longitudinal lanceolate endings at denervated hair follicles ( D′ ) appear normal , while TSCs associated with circumferential axonal terminals are completely lost ( arrow in D ) . ( E and E′ ) Whole-mount immunostaining of S100 ( red ) and MBP ( green ) shows that , at 44 days after axotomy , while myelinating Schwann cells have completely degenerated , TSCs in the denervated skin remain intact ( E′ ) . ( F and F′ ) In Plp1CreER;Rosa26GCamp3 animals , TSCs were induced to express GCaMP3 by tamoxifen injection before dorsal cutaneous nerve axotomy . Immunostaining for GFP ( green ) and Tuj1 ( red ) shows that , at 14 days following axotomy , while Tuj1+ cuanteous axons have completely degenerated , genetically labeled TSCs remain intact ( F′ ) . ( G ) Quantification of TSC numbers 5 weeks after denervation surgery shows no changes in TSC numbers at both guard hair and non-guard hair follicles in denervated skin compared to those at the control skin . n = 3 . Animals around 8 weeks of age were used for all whole-mount immunostaining experiments in A–E′ and G . Plp1CreER;Rosa26GCamp3 mice around 4 weeks old were used in experiments shown in F and F′ . Scale bars , 50 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 01901 . 01510 . 7554/eLife . 01901 . 016Figure 9—figure supplement 1 . Genetically labeled immature Schwann cells after skin denervation . Plp1CreER;Rosa26LSL-YFP mice were treated with tamoxifen before receiving skin denervation surgeries and examined 4 weeks after denervation . ( A–D ) On the sham-operated control side of the back skin , wholemount immunostaining with GFP ( A ) , MBP ( B ) and S100 ( C ) show that genetically-labeled , YFP+ myelinating Schwann cells are positive for MBP and S100 ( Arrows in A–D ) . ( A′–D′ ) Same staining was performed on the denervated back skin . Genetically labeled , YFP+ TSCs remained at hair follicles after denervation . GFP+ sheaths that are negative for S100 and MBP can be observed ( arrows in A′–D′ ) and they are likely to be genetically labeled immature Schwann cells . Animals around 8 weeks old were used in these experiments . Scale bar , 50 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 01901 . 016 A few weeks following skin denervation , re-innervation of some hair follicles was observed . In contrast to 7 days after denervation , where few fibers were seen in the denervated skin , by 33 days after denervation , approximately 70% of the denervated skin had been innervated ( Figure 10A , B ) . The majority of re-innervation most likely derives from axons that had regenerated from the severed dorsal cutaneous nerves ( Jackson and Diamond , 1984 ) . For the skin regions close to the edges of the denervated territory , the re-innervating fibers may also derive from adjacent intact nerves . Re-innervation in the center of the denervated skin was least robust , which may due to the long distance between that skin region and the cut sites of T9 and T10 dorsal cutaneous nerves ( Figure 10B ) . Nevertheless , re-innervation of most skin regions was prominent by 33 days after denervation ( Figure 10B ) . This finding of re-innervation of denervated hair follicles , together with the observation that TSCs are remarkably stable following skin denervation , prompted us to ask whether TSCs that are associated with LTMR lanceolate complexes prior to denervation , and remain following denervation , become re-innervated by sprouting or regenerating LTMR axons . To test this possibility , TSCs were genetically labeled using Plp1CreER;Rosa26GCaMP3 animals and tamoxifen application prior to denervation . TSCs and axonal processes in denervated back skin were then observed using GFP and Tuj1 double immunostaining at different time points after the surgery . In control hair follicles on the non-denervated side of the back skin , longitudinal lanceolate endings were associated with TSC processes and formed typical , intricate longitudinal lanceolate complexes at guard hair and non-guard hair follicles; circumferential endings were also observed surrounding these longitudinal lanceolate complexes at all hair follicles ( Figure 10C ) . On the denervated side of the back skin , though most longitudinal lanceolate endings were still absent three weeks following skin denervation , prominent re-innervation of guard hair follicles in the form of thick circumferential endings and occasional longitudinal lanceolate endings is readily observed ( Figure 10D , arrow ) . A lesser amount of re-innervation in the form of circumferential endings was also seen at this time point in surrounding non-guard hair follicles ( Figure 10D ) . 1 week later , at week four following skin denervation , prominent circumferential and longitudinal axonal endings were readily observed at the re-innervated guard hair follicles ( Figure 10E ) . Circumferential endings were also observed in the surrounding non-guard hair follicles at this time point . At 5 weeks following denervation , sensory endings associated with re-innervated guard hair follicles were morphologically comparable to those in control skin ( Figure 10F ) . Overall , approximately 90% of hair follicles exhibited at least some re-innervation by 5 weeks after denervation . Lanceolate complexes at guard hairs have considerably more advanced recovery compared to non-guard hair follicles . Most non-guard hair follicles , especially zigzag hair follicles , display only sparse circumferential endings at this stage ( Figure 10F ) . We also addressed the specificity of axonal re-innervation of TSC complexes . For this , skin denervation experiments were done using Npy2r-GFP mice , in which Aβ RA-LTMRs are labeled with GFP ( Li et al . , 2011 ) . In control skin , whole-mount immunostaining using anti-GFP confirmed previous findings that , as expected for Aβ RA-LTMRs ( Li et al . , 2011 ) , Npy2r-GFP+ axon terminals form longitudinal lanceolate endings associated with guard and awl/auchene , but not zigzag hair follicles ( Figure 10G , white arrowheads ) . Remarkably , 83 days following denervation , Npy2r-GFP+ axons had re-innervated awl/auchene and guard hair follicles in the denervated skin region ( Figure 10H , white arrowheads ) . While some of the Aβ RA-LTMR axon terminals exhibited simple circumferential endings around hair follicles at this time point and excess sprouting of GFP+ axons innervating adjacent zigzag hair follicles was occasionally seen , the majority of GFP+ axons had developed typical longitudinal lanceolate endings associated with the denervated guard and awl/auchene hairs . In summary , lanceolate complex TSCs remain intact and tightly associated with hair follicles following skin denervation . Furthermore , these intact TSCs become re-associated with the lanceolate endings of regenerated LTMRs , and the specificity of the LTMR subtype/hair follicle relationship is largely if not completely preserved , at least for Aβ RA-LTMRs , following re-innervation . This observation raises the intriguing possibility that TSCs , which remain intact following nerve injury , serve to attract axonal endings of specific LTMR subtypes to restore proper structure and organization of Aβ RA-LTMRs , Aδ-LTMRs , and C-LTMRs lanceolate complexes at the three hair follicle types . 10 . 7554/eLife . 01901 . 017Figure 10 . Re-innervation of TSCs following dorsal cutaneous nerve axotomy . ( A and B ) Whole-mount immunostaining with NFH shows that axons are missing from denervated skin 7 days after axotomy ( A ) , whereas nerve fibers have begun extending into the denervated area 33 days after dorsal cutaneous nerve axotomy ( B ) . ( C ) In the control skin of Plp1CreER;Rosa26GCamp3 animals , whole-mount immunostaining with Tuj1 ( green ) and GFP ( red ) shows the typical morphology of lanceolate complexes at a guard hair follicle located in the lower right area of the image , as well as surrounding non-guard hair follicles . ( D ) 3 weeks after dorsal cutaneous nerve axotomy , re-innervation of guard hair follicles in the form of circumferential endings can be observed . A few longitudinal lanceolate endings can be occasionally observed ( arrow ) . A small degree of re-innervation can also be seen in the surrounding non-guard hair follicles . ( E ) 4 weeks after dorsal cutaneous nerve axotomy , longitudinal lanceolate endings begin to form at re-innervated guard hair follicles . Circumferential endings can also be observed in the surrounding non-guard hair follicles . ( F ) 5 weeks after axotomy , lanceolate complexes at re-innervated guard hair follicles are comparable to those in the control , uninjured skin . More innervation of the surrounding non-guard hair follicles can also be seen . Guard hairs are labeled with ‘G’ in panels C to F . ( G and H ) Npy2r-GFP animals were subjected to dorsal cutaneous nerve axotomy , as above . G shows whole-mount GFP immunostaining of the control , non-denervated skin ( left back hairy skin ) , while H shows GFP immunostaining of denervated skin ( right back hairy skin ) 83 days after axotomy . In control , uninjured skin , GFP+ Aβ RA-LTMRs form longitudinal lanceolate endings at guard and awl/auchene hair follicles ( G , white arrowheads ) . 83 days after axotomy , GFP+ Aβ RA-LTMRs re-innervate guard and awl/auchene hair follicles in the denervated skin ( H , white arrowheads ) , indicating that the specificity of the hair follicle subtype innervation pattern is conserved during the re-innervation processes . Many GFP+ axonal terminals in the re-innervated skin form longitudinal lanceolate endings that are comparable to those in the control skin . Scale bars , 2 mm for A and B; 20 μm from C to F; 50 μm for G and H . DOI: http://dx . doi . org/10 . 7554/eLife . 01901 . 017
Neurobiologists have long postulated that distinct morphological properties of mechanically sensitive end organs associated with LTMRs underlie the unique physiological properties of LTMR subtypes ( Johnson , 2001; Maricich et al . , 2009 ) . Our findings modify this view and provide new insights into the basis of LTMR subtype properties . Guard hair follicles have the longest hair shafts and are innervated by longitudinal lanceolate endings belonging exclusively to Aβ RA-LTMRs . If , as we believe , LTMR lanceolate endings and their associated TSC processes serve as detectors of hair follicle deflection , then guard hair follicles are armed with many more such detectors , displaying fewer but longer processes , compared to those associated with zigzag and awl/auchene hair follicles . Thus , the unique cellular and morphological properties of guard hair lanceolate complexes may endow guard hair-associated Aβ RA-LTMRs with unique functions or response properties during hair movement . On the other hand , our analysis of awl/achene and zigzag hair follicles indicates that cellular and morphological differences of longitudinal lanceolate complexes are unlikely to account for the unique physiological properties of Aβ RA- , Aδ- , and C-LTMRs at these hair follicle subtypes . Our mosaic genetic labeling of lanceolate complex TSCs shows that these cells are tiled at individual hair follicles and that the processes of adjacent TSCs do not overlap . Importantly , at both awl/auchene and zigzag hair follicles , a single TSC plays host to interdigitated axonal endings of two or more LTMR subtypes . Thus , while Aβ RA- , Aδ- , and C-LTMRs exhibit markedly different rates of adaptation to sustained stimuli , their axons form strikingly similar longitudinal lanceolate endings , they can innervate the same hair follicle and , as shown here for awl/auchene and zigzag hair follicles , they can even share the same TSC . These findings provide strong support for a model , in which the unique physiological properties and adaptation rates of Aβ RA- , Aδ- , and C-LTMRs associated with awl/auchene and zigzag hair follicles are due to intrinsic neuronal differences rather than unique end organ cell types or morphologies . The fine structural relationship between LTMR axonal endings , TSC processes , and hair follicle epithelial cells is likely a key determinant of LTMR mechanical sensitivity . Extensive light and electron microscopic analyses of lanceolate complexes were reported beginning in the 1960s ( Yamamoto , 1966; Cauna , 1969; Halata , 1993 ) . These previous studies , together with findings of the present study , indicate that the general morphology of lanceolate complexes associated with hair follicles from different body regions and between different species is highly conserved . In this study , for the first time , we report EM analysis that defines the ultrastructural properties of Aβ RA-LTMR , Aδ-LTMR , and C-LTMR lanceolate complexes and their associations with the three main hair follicle types; guard , awl/auchene , and zigzag hair follicles . Our analysis revealed that the ultrastructural properties of Aβ RA-LTMRs , Aδ-LTMRs , and C-LTMRs and their TSC processes associated with awl/auchene and zigzag hair follicles are remarkably similar . On the other hand , Aβ RA-LTMR endings associated with guard hair follicles are distinct , having larger caliber lanceolate axonal endings and slightly thinner TSC processes compared to those of Aβ RA- , Aδ- , and C-LTMRs associated with awl/auchene and zigzag hair follicles . In addition , at awl/auchene and zigzag hair follicles , two or more axonal endings , which differ in mitochondrial abundance and belong to different LTMR subtypes , are often observed alternately arranged with three or more TSC processes , forming individual groups or ‘units’ . Thus , axon terminals of different LTMR subtypes can share the same set of TSC processes to form ‘lanceolate complex units’ at awl/auchene and zigzag hair follicles , which together comprise ∼99% of trunk skin hairs . Such an arrangement was never observed at guard hairs; at guard hairs , each Aβ RA-LTMR lanceolate ending associates with three TSC processes to form stereotypical ‘guard hair lanceolate complex units’ . Thus , Aβ RA-LTMR lanceolate complexes at guard hair follicles are morphologically invariant and more precisely arranged into confined units than Aβ RA- , Aδ- , and C-LTMR complexes associated with awl/auchene and zigzag hairs , potentially endowing guard hair lanceolate complexes with higher fidelity or sensitivity to hair deflection . The present study also defines the ultrastructural properties of lanceolate endings of the three physiologically defined LTMR subtypes that innervate hair follicles . A defining feature of Aβ RA-LTMR and Aδ-LTMR lanceolate endings is their highly abundant mitochondria . Neurons critically depend on mitochondrial ATP production for establishing membrane excitability and for calcium buffering ( Kann and Kovacs , 2007 ) , and emerging evidence suggests that mitochondria contribute to neuronal plasticity . An abundance of mitochondria in Aβ RA-LTMRs and Aδ-LTMRs terminals may be needed to support rapid conduction velocities , a distinguishing feature of these LTMR subtypes , compared to the slow-conducting C-LTMRs , whose lanceolate endings contain few mitochondria . We also observed several ultrastructural features of Aβ RA- , Aδ- , and C-LTMR lanceolate complexes that may be crucial to the development and maintenance of lanceolate complexes , and may also underlie their remarkable sensitivity to hair deflection . Abundant small vesicles are found within the cytoplasm of axonal terminals of all LTMR subtypes . Studies by Woo et al . ( 2012 ) and Banks et al . ( 2013 ) suggest that these may be glutamate-containing vesicles that mediate signals from sensory neurons to TSCs . Consistent with this idea , Aβ RA-LTMRs and Aδ-LTMRs express VGluT1 and/or VGluT2 , whereas C-LTMRs express VGluT3 , suggesting that all LTMRs are glutamatergic ( Brumovsky et al . , 2007; Seal et al . , 2009; Woo et al . , 2012; Banks et al . , 2013 ) . Numerous pinocytotic vesicles were observed on both the inner and outer surfaces of TSC processes . These vesicle-like structures were noted previously and suggested to control ion flux around nerve fibers ( Yamamoto , 1966 ) . It is also reasonable to speculate that these invaginations serve as docking sites for axon fingers of longitudinal lanceolate endings ( Takahashi-Iwanaga , 2000 ) . Such an intercalating morphology increases the interaction surface area between TSC processes and LTMRs and may contribute to the mechanical sensitivity of LTMR endings . Most noteworthy , in our opinion , is the presence of intercellular processes that emanate from hemidesmosomes on outer root sheath epithelial cells of hair follicles , forming connections , or tethers , between hair follicle epithelial cells and Aβ RA- , Aδ- , and C-LTMR axon terminals and TSC processes . These filaments resemble anchoring filaments and anchoring fibrils , reported previously in other contexts to mediate dermal–epidermal adhesion ( Keene et al . , 1987; Burgeson and Christiano , 1997 ) . At hair follicles , these filaments may contribute to assembly and maintenance of lanceolate ending-TSC-epithelial cell complexes . Intriguingly , the ∼100-nm long filaments observed in the present study may be functionally analogous to the protein tethers described by Lewin et al . in cell culture experiments and proposed to link mechanosensitive ion channels in sensory neurons to the extracellular matrix ( Hu et al . , 2010 ) . We speculate that the tether-like structures we observed at hair follicles in vivo may serve to facilitate or mediate excitation of lanceolate endings following hair deflection . Thus , we propose that ‘epithelial cell–lanceolate complex tethers’ serve a function that is analogous to the tip links that physically tether adjacent steroecilia of hair cells of the cochlea and vestibular apparatus and serve to open mechanically gated ion channels upon stereocilia movement . In the case of hair follicle epithelial cell–lanceolate complex tethers , we speculate that deflection of hair follicles would place strain on the tethers to activate mechanically gated ion channels situated on the membrane of the LTMR lanceolate ending thus leading to their depolarization . Interestingly , a recent study from Lewin et al . suggests that laminin 332 , a major component of keratinocyte-derived extracellular matrix , prevents formation of protein tethers and inhibits rapidly adapting currents , further supporting the notion that extracellular matrices play a key role in mechanotransduction ( Chiang et al . , 2011 ) . Potential molecular components of these tethers may include type VII collagen , a major component of anchoring fibrils , as well as other collagen proteins , laminins , and integrins , which could assemble into multimeric complexes to form tether structures ( Burgeson and Christiano , 1997 ) . Future studies will be needed to define the molecular components of these epithelial cell–lanceolate complex tethers . Once defined , tissue-specific ablation of key components , coupled with histological and physiological analyses , should reveal the contribution of these tethers for the assembly and maintenance of lanceolate complexes and/or their role in mechanotransduction during hair movement or deflection . The intimate relationships between TSCs and longitudinal lanceolate endings of the three LTMR subtypes at each of the three hair follicle subtypes suggested to us an interdependence of LTMRs and TSCs during maintenance of lanceolate complexes . Indeed , we found that LTMR endings rapidly retract from hair follicles following TSC ablation . And yet , surprisingly , when LTMR endings are lost following a nerve cut , TSCs remain intact and intimately associated with hair follicles . In fact , TSCs of adult mice are remarkably stable in the absence of nerve , for at least several weeks or even months . This effect may be age-dependent , since we found that TSCs are less stable and often degenerate during the 2 weeks following skin denervation of neonates ( data not shown ) . Thus , at neonatal ages , when lanceolate complexes are developing , LTMR axonal endings may be necessary for the maintenance of newly formed TSCs . Consistent with this , it was recently reported that TSCs associated with neonatal hair follicles are disorganized in Wnt1Cre-mediated VGLUT2 mutant mice , suggesting that neuronal glutamate signaling is required for proper organization of TSCs during lanceolate complex development ( Woo et al . , 2012 ) . However , in the same study , blocking glutamatergic signaling through systemic administration of an NMDAR antagonist disrupts TSC structures at hair follicles in adult animals , which is different from our conclusion that sensory nerves are not required for maintenance of TSCs at hair follicles in adult animals . One potential explanation for this discrepancy is that glutamatergic signaling from cells other than LTMRs is needed to maintain TSC processes . Nevertheless , our findings indicate that TSCs serve as anchors or scaffolds that maintain LTMR endings and lanceolate complex structure and integrity , whereas LTMR axonal endings of adult mice are not required for maintenance of TSCs . It is remarkable that , within a few weeks following skin denervation , new axonal processes regenerate from severed dorsal cutaneous nerves and re-assemble with TSCs that had remained associated with denervated hair follicles . Moreover , the pattern of axonal innervation of hair follicles during regeneration exhibits similarities to that which is observed during development . In particular , guard hair follicles are the first to be re-innervated . At individual hair follicles , circumferential endings form first , surrounding TSC processes , and this is followed by gradual formation of longitudinal lanceolate endings during the next several days . This is comparable to the pattern of LTMR lanceolate complex formation observed during development of neonatal animals ( data not shown ) ( Peters et al . , 2002 ) . Furthermore , LTMR-hair follicle subtype specificity appears conserved , at least for Aβ RA-LTMRs , during re-innervation . It is likely that the developmental and regeneration processes that govern LTMR lanceolate ending branching , extension , orientation , size , and TSC association have common mechanisms . Additionally , the precise role played by TSCs during the re-innervation of hair follicles is an intriguing question . During regeneration of the neuromuscular junction ( NMJ ) following axotomy , TSCs proliferate extensively and serve to guide motor axons to their destination by expressing some of the same cues that are used during development ( Reynolds and Woolf , 1992; Sanes and Lichtman , 1999 ) . TSCs at hair follicles may serve a similar role to attract specific subtypes of spouting LTMR axons to re-form lanceolate complexes . Future work will define the significance and molecular mechanisms underlying the intimate relationship between LTMR subtype longitudinal lanceolate endings and TSCs during formation , function , and regeneration of hair follicle-associated LTMR lanceolate complexes .
ThCreER , TrkBtauEGFP , Npy2r-GFP BAC transgenic , Plp1CreER , Rosa26 LSL-tdTomato , Rosa26-Confetti , Rosa26GCaMP3 , and Rosa26LSL-hDTR mouse lines , have been described previously ( Doerflinger et al . , 2003; Gong et al . , 2003; Buch et al . , 2005; Rotolo et al . , 2008; Madisen et al . , 2010; Li et al . , 2011; Schepers et al . , 2012; Zariwala et al . , 2012 ) . Generation of the TrkAf/f ( TrkAF592A ) mouse was described previously ( Chen et al . , 2005 ) . In addition to the point mutation introduced into the ATP binding pocket , two Loxp sites were inserted into introns flanking exon 7 to 12 of the TrkA gene . Wnt1Cre;TrkAf/f animals were generated to achieve ablation of the TrkA gene in cells of neural crest origin . The TrkA conditional knockout animals are viable and fertile , with no obvious behavioral abnormality , though behavioral assays assessing pain , temperature , and touch sensations are likely to reveal sensory defects in these TrkA conditional mutants . Protocols for immunohistochemistry were described previously ( Liu et al . , 2007; Luo et al . , 2009; Li et al . , 2011 ) . Mice were anesthetized using CO2 inhalation and transcardially perfused with PBS ( pH 7 . 4 , 4°C ) followed by 4% paraformaldehyde ( PFA ) in PBS ( pH 7 . 4 , 4°C ) . DRGs and hairy skin were dissected from the perfused mice . DRGs were postfixed in PBS containing 4% PFA at 4°C for 1–2 hr . Hairy skin was postfixed with PBS containing 4% PFA at 4°C overnight . The tissues were cryoprotected in 30% sucrose in PBS at 4°C overnight , embedded in OCT ( Tissue Tek ) and frozen at −20°C . The tissues were sectioned at 20–30 μm using a cryostat . The sections on slides were dried at room temperature for 1 hr , and fixed with 4% PFA in PBS on ice for 15 min . The slides were washed with PBS containing 0 . 1% Triton X-100 ( 0 . 1% PBST ) and blocked with 5% normal serum ( goat or donkey ) in 0 . 1% PBST at room temperature for 1 hr . The tissue sections were incubated with primary antibodies diluted in blocking solution at 4°C overnight . The next day , the sections were washed with 0 . 1% PBST , and incubated with secondary antibodies diluted in blocking solution at room temperature for 1 hr , washed again with 0 . 1% PBST , and mounted with fluoromount-G ( Southern Biotech , Birmingham , AL ) . The primary antibodies used for this study were: rabbit anti-CGRP ( Immunostar , Hudson , WI , 24112 , 1:1000 ) , chicken anti-GFP ( Invitrogen , A10262 , 1:1000 ) , rabbit anti-GFP ( Invitrogen , Carlsbad , CA , A11122 , 1:1000 ) , rabbit anti-NFH ( Millipore , Billerica , MA , AB1982 , 1:1000; Sigma , St . Louis , MO , N4142 , 1:1000 ) , chicken anti-NFH ( Aves Labs , Tigard , OR , NFH , 1:1000 ) , rabbit anti-S100 ( DAKO , Denmark , Z0311 , 1:1000 ) , sheep anti-Tyrosine Hydroxylase ( Millipore , AB1542 , 1:400 ) , rabbit anti-tuj1 ( b-Tubulin ) ( Covance , Princeton , NJ , PRB-435P , 1:1000 ) . The secondary antibodies used were: Alexa 488 , 546 or 647 conjugated goat anti-chicken antibody , Alexa 488 , 546 or 647 conjugated goat anti-rabbit antibodies , Alexa 546 conjugated donkey anti-sheep antibody . All secondary antibodies were purchased from Invitrogen . Protocols for whole-mount skin immunostaining were described previously ( Li et al . , 2011 ) . Back hairy skin from 3- to 10-week-old mice was treated with commercial hair remover , wiped clean with tissue paper and tape stripped until glistening . The skin was then dissected , cut into small pieces and fixed in 4% PFA in PBS at 4°C for 2 hr . The tissue was rinsed in PBS and then washed with PBS containing 0 . 3% Triton X-100 ( 0 . 3% PBST ) every 30 min for 5–8 hr . Then , the skin was incubated with primary antibodies in 0 . 3% PBST containing 5% goat/donkey serum and 20% DMSO at room temperature for 3 to 5 days . After washing with 0 . 3% PBST every 30 min for 5–8 hr , the tissues were transferred to secondary antibodies in 0 . 3% PBST containing 5% goat/donkey serum and 20% DMSO and incubated at room temperature for 2 to 4 days . The tissues were then washed with 0 . 3% PBST every 30 min for 5–8 hr , dehydrated in 50% methanol for 5 min and 100% methanol for 20 min , three times , and lastly cleared in BABB ( Benzyl Alcohol , sigma 402834; Benzyl Benzoate , sigma B-6630; 1:2 ) at room temperature for 20 min . To identify the types of hair follicles innervated by each LTMR class , similar whole-mount preparations of hairy skin were made without removing the hair . Using a confocal microscope , hair follicles were traced to the corresponding hair shafts . The number of rows of medulla cells in the hair shaft was counted to distinguish zigzag ( 1 row ) , awl/auchene ( 3 or 4 rows ) and guard hairs ( 2 rows ) . Adult mice were fixed by cardiac perfusion in 2% paraformaldehyde and 2 . 5% glutaraldehyde solution in 0 . 1 M phosphate buffer ( pH 7 . 2 ) . The back hairy skin was dissected on ice . Trunks of hairy skin with less than 20 hair follicles were carefully examined under dissection microscope to record the types and positions of hair follicles . Tissue samples were immersed in the same fixative overnight , treated with reduced 1% osmium tetroxide for 1 hr . To facilitate visualization of intercellular filaments between hair follicle epithelial cells and lanceolate complexes , some of the tissue samples were treated with 1% tannic acid for 5 min . All tissue samples were then stained with 2% uranyl acetate for 2 hr , dehydrated in graded ethanol and embedded in Epon . Toluidin blue-stained transverse sections of the hair follicles were made to determine the location of lanceolate endings and the positions for collecting ultrathin sections . Transverse thin sections were stained with uranyl acetate and lead citrate followed by examinations with Philips/FEI BioTwin CM120 transmission electron microscope at 80 kV . 4-HT was dissolved in ethanol ( 10 mg/ml ) . For ThCreER animals , 100 μl ( ∼1 mg ) of 4-HT in ethanol was mixed with 200 μl of sunflower seed oil ( Sigma ) , vortexed for 1 min and centrifuged under vacuum for 25 min to remove the ethanol . The 4-HT solution was delivered via oral gavage to ThCreER animals at P13 , P14 , and P15 . For Plp1CreER animals , 2 mg of tamoxifen dissolved in 200 μl of sunflower seed oil was delivered via oral gavage to animals at P13 to P17; 0 . 01 to 0 . 03 mg of tamoxifen were used for sparse labeling . DTX injections were performed as previously described ( Cavanaugh et al . , 2009 ) . 40 ng/g of DTX ( List Laboratories ) was delivered by i . p . injections to mice at P21 . Two treatments , separated by 72 hr , were given . Animals were sacrificed 2 to 3 weeks after the first injection . Animals were anaesthetized with Urethane ( 1500 mg/kg ) . A 2 . 5–3 cm midline incision was made in the dorsal skin under anesthesia . The dorsal cutaneous nerves were exposed under a dissection microscope . Dorsal cutaneous nerves ( T5–L2 ) on the right side were cut with scissors . The skin was then closed with 5-0 sterile sutures . Back skin from the left side of each animal was used as control . Quantifications are presented as mean ± SEM . Statistical analyses were performed using unpaired Student’s t tests . Significance was declared at p<0 . 05 and was indicated by one star in the figures . | Many mammals , such as cats , mice , and sea lions , have long whiskers that are particularly sensitive to touch . However , the hairs that cover the skin of most mammals are also important touch detectors . These hairs grow from follicles that are connected to the ends of the nerve cells that detect and convey touch information to the central nervous system . In mice , three main types of hair follicle—guard hairs , awl hairs , and zigzag hairs—are associated with combinations of three types of nerve endings . Much remains to be understood about how hair follicles and nerve cell endings—which are wrapped by cells called terminal Schwann cells—interact via structures called lanceolate complexes . Now , using a combination of genetics , microscopy and surgical procedures , Li and Ginty have studied these structures in unprecedented detail , and revealed some intriguing structural differences among the three types of hair follicles . Zigzag follicles—which make up the fur undercoat—are associated with fewer terminal Schwann cells than are awl follicles , whilst guard hair follicles have the most . High-resolution analyses revealed that distinct combinations of sensory nerve endings were associated with different types of hair follicle cells—which may underlie the unique responses of the different hair follicle types when the hairs are deflected . Furthermore , an individual terminal Schwann cell can be associated with more than one type of nerve ending , adding another layer of intricacy to the detection of hair movements . Killing the terminal Schwann cells in mice caused a complete loss of sensory nerve endings at hair follicles , which suggests that these cells are essential for maintaining the connection between the hair follicles and nerve cell endings . Interestingly , surgically removing nerve endings from the skin did not lead to a loss of terminal Schwann cells , and the nerve endings eventually grew back and reconnected with the hair follicles . In addition to shedding new light on the structures of lanceolate complexes in different types of hair follicles , the findings of Li and Ginty suggest that terminal Schwann cells maintain the nerve endings at hair follicles and guide their regeneration after damage . Uncovering the molecular mechanisms that control these processes represents an important next step in this research . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"neuroscience"
] | 2014 | The structure and organization of lanceolate mechanosensory complexes at mouse hair follicles |
Cullin-3 ( CUL3 ) -based ubiquitin ligases regulate endosome maturation and trafficking of endocytic cargo to lysosomes in mammalian cells . Here , we report that these functions depend on SPOPL , a substrate-specific CUL3 adaptor . We find that SPOPL associates with endosomes and is required for both the formation of multivesicular bodies ( MVBs ) and the endocytic host cell entry of influenza A virus . In SPOPL-depleted cells , endosomes are enlarged and fail to acquire intraluminal vesicles ( ILVs ) . We identify a critical substrate ubiquitinated by CUL3-SPOPL as EPS15 , an endocytic adaptor that also associates with the ESCRT-0 complex members HRS and STAM on endosomes . Indeed , EPS15 is ubiquitinated in a SPOPL-dependent manner , and accumulates with HRS in cells lacking SPOPL . Together , our data indicates that a CUL3-SPOPL E3 ubiquitin ligase complex regulates endocytic trafficking and MVB formation by ubiquitinating and degrading EPS15 at endosomes , thereby influencing influenza A virus infection as well as degradation of EGFR and other EPS15 targets .
Endocytic trafficking is an essential cellular process for nutrition absorption , signal transduction , cell-cell communication and maintenance of cell homeostasis ( Doherty and McMahon , 2009 ) . Endocytosis is used for the uptake of exogenous and cellular cargo , including many viruses such as influenza A virus ( IAV ) ( Cossart and Helenius , 2014; Edinger et al . , 2014 ) or a cohort of plasma membrane proteins such as the epidermal growth factor receptor ( EGFR ) with their respective ligands ( Tomas et al . , 2014 ) . Cargo entering the endosomal system can be delivered via early ( EE ) and late endosomes ( LE ) to lysosomes ( LY ) for degradation , or alternatively , recycled back to the plasma membrane ( Maxfield , 2014; Wandinger-Ness and Zerial , 2014 ) . Transfer of cargo from EEs to LYs depends on an endosomal maturation process that involves a variety of protein- and lipid-based remodeling events . They include a small GTPase RAB5-to-RAB7 switch , a PtdIns ( 3 ) P to PtdIns ( 3 , 5 ) P2 conversion , and changes in the luminal ion concentrations , most notably a decrease in pH ( Huotari and Helenius , 2011 ) . Together these changes prepare the endosomes for fusion with other LEs and LYs . Endosome maturation also involves sorting of membrane cargo destined for degradation into intraluminal vesicles ( ILVs ) , thereby generating late endosomal vacuoles referred to as multivesicular bodies ( MVBs ) ( Piper and Katzmann , 2007 ) . Ubiquitin has emerged as an important regulator of endocytosis and cargo degradation ( Clague et al . , 2012 ) . It serves as a signal for membrane quality control , endocytic internalization , endosome maturation , and cargo sorting into ILVs by the endosomal sorting complexes required for transport ( ESCRTs ) ( Piper et al . , 2014 ) . Indeed , ubiquitination is indispensable for the interaction of EGFR with the ESCRT-0 complex member HRS and thus for efficient lysosomal targeting ( Eden et al . , 2012 ) . Specific deubiquitination enzymes ( DUBs ) like AMSH and UBPY associate with ESCRT components to cleave and recycle ubiquitin ( Alwan and van Leeuwen , 2007; Meijer et al . , 2012 ) . Ubiquitination also regulates the machinery involved in molecular sorting and the ILV formation , although this regulatory layer is less understood . In a recent study , we observed that Cullin-3 ( CUL3 ) is involved in regulating endosome maturation and endo-lysosomal trafficking in mammalian cells . In CUL3-depleted cells , late endosomes fail to mature properly . They are enlarged and often devoid of ILVs . EGFR degradation is delayed and its ligand EGF accumulates in LE/LYs . Moreover , the cells are resistant to influenza A virus ( IAV ) infection because instead of delivering their capsid components to the cytosol and the nucleus , the endocytosed virus particles remain associated with immature LE vacuoles ( Huotari et al . , 2012 ) . CUL3 is the scaffolding subunit common for a large subfamily of Cullin-RING E3 ubiquitin ligases ( CRLs ) that mainly use the RING-H2 finger protein RBX1 to recruit charged ubiquitin conjugating enzymes ( E2s ) and catalyse the transfer of ubiquitin onto substrates ( Lydeard et al . , 2013 ) . This is stimulated by the conjugation of the ubiquitin-like protein NEDD8 to the cullin subunit ( Duda et al . , 2008; Saha and Deshaies , 2008 ) through an enzymatic cascade analogous to ubiquitination ( Enchev et al . , 2015 ) . CUL3 may be activated at endosomes by DCNL3 , an E3 ligase for NEDD8 , known to associate with membranes by virtue of a covalent lipid modification ( Meyer-Schaller et al . , 2009 ) . CRL activity is further modulated by deneddylation as well as non-catalytically by the COP9 signalosome ( Emberley et al . , 2012; Enchev et al . , 2012; Fischer et al . , 2011 ) . In addition to CUL3 , RBX1 , and NEDD8 , the CRL3 complexes contain a subunit that binds to CUL3 through a Bric-a-brac/Tramtrack/Broad ( BTB ) domain . The BTB domain-containing proteins possess diverse substrate recognition domains and function as CRL3-specific substrate adaptors ( Pintard et al . , 2004 ) . There are about 150 different BTB domain-containing proteins in the human proteome ( Stogios et al . , 2005 ) , indicating that a large cohort of CRL3 substrates exists , most of which are presently unidentified . To gain mechanistic insight into the role of CRL3 in endocytic trafficking , we identified relevant BTB-adaptors by siRNA screening . We found that cells depleted of the BTB-adaptor Speckle-type POZ protein-like ( SPOPL ) were defective for IAV infection and showed aberrant late endosomal vacuoles resembling those observed after CUL3 depletion . Although SPOPL shares 81% sequence identity with the tumor driver Speckle-type POZ protein ( SPOP ) and these two adaptors dimerize in vitro ( Errington et al . , 2012 ) , only SPOPL was found to associate with endosomes . We identified the endocytic adaptor EPS15 as a critical substrate of the CRL3SPOPL complex . Our results indicated that EPS15 ubiquitination by CRL3SPOPL is needed for efficient intraluminal vesicle formation during endosome maturation as well as for uncoating of IAV capsids .
To identify the BTB-domain containing CRL3 substrate-adaptors involved in endosome maturation , we took advantage of the observation that IAV depends on CUL3 activity for cell entry and infection ( Huotari et al . , 2012 ) . We performed siRNA screening in the lung adenocarcinoma cell line A549 depleted for 130 human BTB-containing proteins using expression of the viral nuclear protein ( NP ) as read-out after IAV addition ( Figure 1A and Supplementary file 1 ) . Depletion of 14 BTB proteins caused 50% or greater decrease in the number of IAV infected cells . Previously described assays ( Banerjee et al . , 2013 ) were used to determine which step in the infection pathway was inhibited , including virus binding to cells , endocytic uptake , acid-conversion of viral hemagglutinin ( HA ) , penetration by fusion , uncoating of the viral capsid , and nuclear import of the viral ribonucleoproteins ( vRNPs ) . Out of 14 primary hits , we decided to follow up the BTB proteins SPOP and SPOPL because their depletion gave a phenotype of defective virus uncoating similar to that observed after CUL3 depletion . The other BTB-proteins identified in this screen were required for different steps of efficient IAV infection , and will be studied elsewhere . 10 . 7554/eLife . 13841 . 003Figure 1 . The CRL3 substrate adaptors SPOP and SPOPL are crucial for influenza A virus ( IAV ) infection and uncoating . ( A ) siRNA screen workflow for BTB adaptor proteins with similar IAV infection phenotypes as CUL3 ( left ) . Schematic representation of the CRL3SPOP/L E3 ubiquitin ligase complex ( right ) . CUL3 mediates the formation of ubiquitin chains ( UB ) to its substrates by binding to the ubiquitin charged conjugating enzyme ( E2-UB ) via RBX1 on one side while allowing the interaction with the substrate through the substrate adaptor proteins SPOP or SPOPL on the other side . ( B ) IAV X31 infection assay . Images show A549 cells treated with control siRNA ( siControl ) or siRNA-depleted of CUL3 ( siCUL3 ) , and the BTB-adaptor SPOP ( siSPOP-1 ) or SPOPL ( siSPOPL-1 ) for 72 hr before infection with IAV X31 . IAV infection was quantified by co-staining the cells with NP specific antibodies and DAPI to indicate nuclei . Cells siRNA-depleted for the vATPase subunit ATP6V1B2 ( siATP6V1B2 ) were included for positive control . Scale bar = 100 μm; Data are mean + SD , n > 100 cells per sample , N = 4 . ( C-G ) IAV entry assays . A549 cells were treated with control , SPOP- or SPOPL-specific siRNA , and binding of IAV X31 to the cells was monitored by immunofluorescence staining of the hemagglutinin ( HA ) with anti-H3 antibody ( C ) . The IAV infection was allowed for 0 . 5 hr to follow IAV endocytosis with HA staining ( D ) , for 1 hr to monitor IAV acidification using A1 antibodies ( E ) , for 2 . 5 hr to check IAV uncoating by M1 detection ( F ) and finally for 5 hr to track nuclear import of IAV vRNPs by NP-specific antibodies ( G ) . Nuclei were stained with DAPI and entry steps quantified relative to control . Scale bar = 50 μm; Data are mean + SD , n > 500 cells per sample , N = 3 . **p≤0 . 01 , ***p≤0 . 001; ****p≤0 . 0001 . ( H ) Acid-induced endocytic-bypass entry assay . IAV nuclear import after acid-induced fusion at the PM was monitored in A549 cells using indirect immunofluorescence staining for NP and counterstaining with DAPI for infection quantitation . Note that pH 5 . 4 allows acid-induced endocytic-bypass infection of IAV . Scale bar = 50 μm; Data are mean + SD , n > 500 cells per sample , N = 3 . DOI: http://dx . doi . org/10 . 7554/eLife . 13841 . 00310 . 7554/eLife . 13841 . 004Figure 1—figure supplement 1 . The CRL3 substrate adaptors SPOP and SPOPL are crucial for influenza A virus ( IAV ) infection and uncoating . ( A ) The efficiency of distinct siRNAs targeting SPOP ( siSPOP-1 to siSPOP-3 ) or SPOPL ( siSPOPL-1 to siSPOPL-3 ) were analyzed in HeLa cells by qRT-PCR ( left panel ) . The qRT-PCR measurements were quantified and plotted as mRNA levels relative to GAPDH controls . Data are mean + SD , N = 3 . Immunoblotting of cell extracts using an affinity-purified peptide antibody specifically recognizing SPOPL ( right panel , see Material and methods ) . ( B ) IAV X31 infection assay . HeLa cells treated with siControl , depleted of SPOP ( siSPOP1-2 ) , SPOPL ( siSPOPL1-2 ) , CUL3 ( siCUL3 ) or the vATPase subunit ATP6V1B2 ( siATP6V1B2 ) were infected with IAV X31 , and infected cells were visualized by immunofluorescence staining of the viral protein NP . The assay was quantified as described in the legend to Figure 1 and plotted as percentage ( % ) of NP positive cells compared to control ( siControl ) . Data are mean + SD , n > 500 cells per sample , N = 3 . ( C ) IAV binding assay . A549 cells treated with siControl oligos or depleted as indicated for SPOP ( siSPOP2-3 ) , SPOPL ( siSPOPL2-3 ) , CUL3 ( siCUL3 ) or the vATPase subunit ATP6V1B2 ( siATP6V1B2 ) and incubated with the IAV X31 strain ( MOI = 100 ) for 1 hr at 4°C . IAV binding was monitored by immunofluorescence staining of hemagglutinin ( HA ) with anti-H3 antibody , and quantified by plotting the relative percentage ( % ) of H3 positive cells treated with siSPOP or siSPOPL compared to siControl . Scale bar = 50 μm . Data are mean + SD , n > 100 cells per sample , N = 3 . ( D ) IAV endocytosis assay . Following binding for 1 hr at 4°C , IAV X31 virus was internalized for 0 . 5 hr in control , SPOP- and SPOPL-depleted A549 cells . Intracellular staining of HA was used to monitor IAV endocytosis , and quantified by plotting the relative percentage ( % ) of cells with intracellular H3 treated with siSPOP or siSPOPL compared to siControl . Scale bar = 50μm . Data are mean + SD , n > 100 cells per sample , N = 3 . ( E ) IAV acidification assay . Following binding for 1 hr at 4°C , IAV X31 virus was internalized for 1 hr in control , SPOP- and SPOPL-depleted A549 cells . Acidification of IAV was monitored by A1 immunofluorescence staining detecting the acid-induced conformational switch of HA after IAV entry . The data were quantified by plotting the relative percentage ( % ) of A1 positive cells treated with siSPOP or siSPOPL compared to siControl . Scale bar = 50 μm . Data are mean + SD , n > 100 cells per sample , N = 3 . ( F ) IAV fusion assay . A549 cells treated for 72 hr with siControl oligos or depleted as indicated with three different siRNAs targeting SPOPL ( siSPOPL1-3 ) or CUL3 ( siCUL3 ) were incubated for 1 hr at 4°C with the IAV X31 strain ( MOI = 100 ) , labeled prior to this with R18 dye . The dye allows detecting fusion of the virus with the host membrane by changing its color from red to green . Viral fusion was quantified after 1 hr by FACS analysis and plotted as relative percentage ( % ) IAV fusion/hemifusion compared to siControl . Bafilomycin treatment for 1 hr prior to infection served as a control , since blocked vesicle acidification inhibits HA acidification and thus IAV fusion , but not endocytosis . Data are mean + SD , n > 5000 cells per sample , N = 3 . ( G ) IAV uncoating assay . Following binding for 1 hr at 4°C , IAV X31 virus was internalized for 2 . 5 hr in control , siSPOP and SPOPL-depleted A549 cells in the presence of CHX . Uncoating of IAV particles was detected by immunofluorescence staining against M1 . Dispersed M1 staining in the cytoplasm represents a successful uncoating event , and was quantified by plotting the relative percentage ( % ) of M1-dispersed cells treated with siSPOP or siSPOPL compared to siControl . Scale bar = 50 μm . Data are mean + SD , n > 100 cells per sample , N = 3 . ( H ) IAV nuclear import assay . Following binding for 1 hr at 4°C , IAV X31 virus was internalized for 2 . 5 hr in control , SPOP- and SPOPL-depleted A549 cells in the presence of CHX . Import of NP into the nucleus was detected by indirect immunofluorescence staining with anti-HB64 antibody , and quantified by plotting the relative percentage ( % ) of nuclear NP-positive cells treated with siSPOP or siSPOPL compared to siControl . Scale bar = 50 μm . Data are mean + SD , n > 100 cells per sample , N = 3 . DOI: http://dx . doi . org/10 . 7554/eLife . 13841 . 004 To focus on the involvement of SPOP and it close homologue SPOPL in IAV infection , we used three siRNA oligonucleotides ( siSPOP1-3 and siSPOPL1-3 ) . They reduced the respective target gene expression efficiently as judged by quantitative RT-PCR . Notably , SPOP depletion also influenced mRNA levels of SPOPL , but not vice versa . Immunoblotting using a home-made , affinity-purified antibody specific for SPOPL confirmed downregulation of SPOPL at the protein level ( Figure 1—figure supplement 1A ) . Knockdown of SPOP or SPOPL with the siRNAs resulted in a reduction of viral NP expression in A549 and HeLa cells to levels comparable to CUL3 depletion ( Figure 1B and Figure 1—figure supplement 1B ) , but not as strong as upon depletion of a vATPase subunit ( siATP6V1B2 ) . As expected , cells that were depleted of either SPOP or SPOPL showed a block in IAV uncoating ( Figure 1F , Figure 1—figure supplement 1G ) . As a consequence , subsequent nuclear import of NP was also impaired ( Figure 1G , Figure 1—figure supplement 1H ) . Aside from virus binding to the cell surface , which was elevated in SPOPL-depleted cells , we found that endocytosis , acidification and fusion of the viral particles were comparable to controls ( Figure 1C–E , Figure 1—figure supplement 1C–F ) . The defects in IAV uncoating and nuclear import of vRNPs in SPOP or SPOPL-depleted cells were rescued by a brief acidic incubation of virus bound to the cell surface ( Figure 1H ) . This is known to induce direct fusion of the viral particle at the plasma membrane , thus bypassing to some extent the need for endocytic trafficking ( Stauffer et al . , 2014 ) . Together , these results indicated that SPOP or SPOPL depletion leads to a virus entry defect caused by inhibition of uncoating and release of viral capsid components from endocytic vacuoles . To further characterize the functions of SPOP and SPOPL , we used fluorescence microscopy and cell fractionation to determine their subcellular localization . We found that while GFP-tagged SPOP was predominantly nuclear when expressed in HeLa cells , GFP-tagged SPOPL was cytoplasmic and confined to puncta ( Figure 2—figure supplement 1A ) . Endogenous SPOPL , contrary to SPOP , was mainly in the soluble cytosolic fraction after sedimentation of total cell extracts prepared from HeLa cells ( Figure 2A ) . Moreover , endosome purification revealed that some of the SPOPL and CUL3 co-fractionated with endosomal markers such as EPS15 , HRS and STAM ( Figure 2B ) . We concluded that of the two BTBs , SPOPL was more likely to directly influence endocytic trafficking due to its endosomal localization . 10 . 7554/eLife . 13841 . 005Figure 2 . The BTB adaptor SPOP is nuclear , while SPOPL localizes to endosomes and affects endosome maturation . ( A ) Extracts prepared from HeLa cells were analyzed by differential centrifugation . The indicated proteins were probed in the supernatant ( S ) and pellet ( P ) fractions by immunoblotting after sedimentation of nuclei at 1000 g and after further fractionation of the resulting supernatant at 13 , 000 or 100 , 000 g . The cellular organelles detected by the specific antibodies are marked . ( B ) Endosomal organelles were enriched from HeLa cells by differential centrifugation and then fractionated on a 5 – 20% OptiPrep gradient . After centrifugation , TCA-precipitated gradient fractions ( 1–12 ) were analyzed by immunoblotting with specific antibodies against SPOPL , CUL3 , EPS15 , STAM and HRS . ( C ) HeLa cells expressing GFP-RAB5 and GFP-RAB7 were treated with siControl or siSPOPL and live cell imaging monitored their expression . A set of untransfected cells was treated with 10 μM MLN-4924 to inhibit CRL activity by preventing neddylation . Scale bar = 10 μm . Regions of interest ( squares ) are shown at 5x magnification . Vesicle diameter was quantified by Image J . Data are mean + SD , n = 100 vesicles per sample , N = 3 . ***p≤0 . 001; **p≤0 . 01 . ( D ) HeLa cells were depleted of SPOPL or using siControl , fixed after 72 hr and thin sections analyzed by electron microscopy ( EM ) . MVBs are indicated with a black arrow head . Note the enlarged vacuoles in SPOPL-depleted cells that were found empty , devoid of ILVs . N: nucleus . Scale bar = 2000 nm . ( E ) A549 and HeLa cells were depleted of CUL3 , SPOPL or SPOP for 72 hr , and then cell lysates were prepared and analyzed via immunoblotting with specific antibodies for markers of different cellular compartments . ( CRL3 – Cullin-RING ligase 3 , ER – endoplasmic reticulum ) DOI: http://dx . doi . org/10 . 7554/eLife . 13841 . 00510 . 7554/eLife . 13841 . 006Figure 2—figure supplement 1 . The BTB adaptor SPOP is nuclear , while SPOPL localizes to endosomes and affects endosome maturation . ( A ) HeLa cells were transiently transfected with plasmids expressing GFP-SPOPL or GFP-SPOP respectively , and subcellular localization was analyzed 24 hr later by GFP-live cell microscopy . Maximal projections are shown . Note that GFP-SPOPL is mainly cytoplasmic and forms puncta , while GFP-SPOP accumulates in nuclear speckles . Scale bar = 10 μm . ( B ) HeLa cells stably expressing from the doxycycline ( Dox ) -inducible promoter GFP-RAB7 and either a shRNA control ( shControl ) or a shRNA specifically targeting SPOPL ( shSPOPL-1 ) were induced by the addition of doxycycline ( 1 μg/ml ) for 72 hr . The localization of GFP-RAB7 was analyzed in live-cell microscopy . Scale bar = 10 μm . Regions of interest ( squares ) are shown at 5x magnification . Note the accumulation of highly vacuolated GFP-RAB7-positive vesicles in shSPOPL-expressing cells . ( C ) HeLa cells stably expressing GFP-RAB9 from the Dox-inducible promoter were transfected with siControl or siRNA depleting SPOPL for 72 hr . GFP-RAB9 was monitored by live cell imaging . Scale bar = 10 μm . Regions of interest ( squares ) are shown at 5x magnification . ( D ) HeLa cells were transfected with control siRNA ( siControl ) or siRNA depleting SPOPL ( siSPOPL ) for 72 hr . Total cell extracts were analyzed by immunoblotting directly ( left panel ) or after purification of endosomal organelles by differential centrifugation followed by fractionation on a 5 – 20% OptiPrep gradient ( right panel , fractions 1–12 ) . Cellular organelles detected by the specific antibodies are marked . DOI: http://dx . doi . org/10 . 7554/eLife . 13841 . 006 We next used immunoblotting , fluorescence microscopy and thin sectioning electron microscopy ( EM ) to examine the potential role of SPOPL in the endocytic system . When visualized by the marker GFP-RAB5 using live cell imaging , early endosomes appeared normal in SPOPL-depleted cells compared to control siRNAs ( Figure 2C ) . In contrast , the late endosomal system as monitored by GFP-RAB7 was severely distorted when SPOPL was depleted by a siRNA or a shRNA construct ( Figure 2C and Figure 2—figure supplement 1B ) . The diameter of GFP-RAB7 positive endosomes increased two fold and the swollen vacuoles were clustered in the perinuclear region . A similar defect was observed when CRL activity was pharmacologically inhibited by the addition of MLN-4924 ( Figure 2C ) , which prevents cullin neddylation ( Soucy et al . , 2009 ) . Moreover , the endosome-to-Golgi transport marker GFP-RAB9 revealed strongly enlarged vacuoles upon SPOPL depletion as well , implying a broad effect of SPOPL on the late endosomal transport system ( Figure 2—figure supplement 1C ) . Qualitative EM analysis of SPOPL-depleted cells showed that endosomal vacuoles were empty or contained little dense material and few ILVs were detectable compared to control cells ( Figure 2D ) . The formation of normal MVBs was thus abrogated . To narrow down the role of SPOPL we analyzed the level of early and late endosomal markers as well as markers of recycling endosomes , ER , autophagy and cytosol ( Figure 2E ) . In contrast to SPOP , depletion of SPOPL caused changes in late endocytic markers , in particular the MVB markers EPS15 , HRS , STAM and TSG101 . The level of most early endocytic and recycling markers did not change ( Clathrin , EPSIN1 , Caveolin , Calnexin , RAB11 ) . Notable , out of the tested receptors , EGFR and MET revealed significant level changes , while levels of VEGFR , IGF1R and HER2 - a close relative of EGFR - were not altered . We purified endosomes from whole cell lysates and further fractionated them by density gradient centrifugation . Analysis of the endosome fraction revealed that in contrast to the ESCRT components HRS and STAM and the endocytic adaptor EPS15 , the lysosomal marker LAMP1 and early endocytic marker EEA1 were reduced in endosomal fractions upon SPOPL depletion ( Figure 2—figure supplement 1D ) . Considering both total protein levels and protein levels in the endosome fractions as well as microscopy data , we conclude that the late endosome system , most likely the ILV/MVB formation process , is regulated by SPOPL . Our results indicated that especially EPS15 , but also ESCRT subunits HRS and STAM , accumulate in SPOPL-depleted cells . EPS15 is a scaffolding adaptor protein for clathrin-coated vesicles and a critical regulator of EGFR endocytosis and endosomal sorting ( Gucwa and Brown , 2014; Li et al . , 2014; Roxrud et al . , 2008; van Bergen en Henegouwen , 2009 ) . Interestingly , we found that EGFR was strongly reduced in SPOPL-depleted cells ( Figure 3A ) , while it was elevated in cells depleted of CUL3 ( Huotari et al . , 2012 ) and in cells treated with the CRL-inhibitor MLN-4924 ( Figure 3—figure supplement 1A ) . Conversely , overexpression of HA-tagged SPOPL caused a two fold increase in EGFR and a 50% decrease in EPS15 ( Figure 3B ) . EPS15 also decreased upon SPOPL depletion when a siRNA resistant construct of SPOPL was overexpressed ( Figure 3—figure supplement 1B ) . Quantitative RT-PCR revealed that EPS15 accumulation observed in SPOPL-depleted cells was not associated with increased EPS15 mRNA ( Figure 3—figure supplement 1C ) . Moreover , GFP-tagged EPS15 expressed from the doxycycline ( Dox ) -inducible promoter was elevated at least five fold in SPOPL-depleted cells compared to RNAi-controls ( Figure 3—figure supplement 1D ) . Together , these data suggested that EPS15 is degraded in a SPOPL- and CUL3-dependent manner . 10 . 7554/eLife . 13841 . 007Figure 3 . CRL3SPOPL targets EPS15 for proteasome-dependent degradation . ( A ) Total cell extracts prepared from HeLa cells treated with control siRNA ( siControl ) or RNAi oligos targeting SPOPL , SPOP or CUL3 as indicated were analyzed by immunoblotting for EGFR , EPS15 , SPOP and CUL3 protein levels . GAPDH controls for equal loading . ( B ) Total cell extracts prepared from HeLa cells harboring an empty control plasmid ( empty ) or a plasmid overexpressing HA-tagged SPOPL were analyzed by immunoblotting for EGFR , EPS15 and SPOPL-HA protein levels . GAPDH controls for equal loading . EPS15 and EGFR levels were quantified by Image J . Data are mean + SD , N = 3 . ***p≤0 . 001; *p≤0 . 05 . ( C ) The levels of EPS15-GFP expressed from the doxycycline-inducible promoter were analyzed by immunoblotting of extracts prepared from HeLa cells for 40 hr with either MG132 or chloroquine ( CQ ) . Tubulin ( TUB ) controls for equal loading . ( D ) A549 cells transiently transfected with a plasmid expressing SPOPL-GFP were treated with 10 μM MLN-4924 to stabilize SPOPL-GFP levels . After 6 hr , cells were fixed , stained with specific antibodies and analyzed by confocal immunofluorescence microscopy . Displayed are maximal projections of Z-stack acquisitions , fully covering cell height . Scale bar = 10 μm . Regions of interest ( squares ) are shown at 4x higher magnification . ( E ) Endogenous EPS15 was immunoprecipitated ( IP ) from HEK-293 cells using a specific antibody or unspecific IgG as control , after pretreated with 1 μM MG132 for 30 hr . EPS15 and co-precipitated proteins were eluted and analyzed by immunoblotting using specific antibodies . 40 μg of protein were loaded as input samples . ( F ) In vitro binding of recombinantly purified SPOPL to GST-EPS15 in GST pull-down experiments was analyzed by Coomassie staining ( upper panel ) and immunoblotting ( lower panel ) , respectively . ( G ) In vitro ubiquitination assays . E . coli purified EPS15 and reconstituted CUL3-NEDD8-RBX1 were incubated at 37°C using UBE2R1 ( CDC34 ) as the E2-enzyme and in the presence of SPOPL or without BTB adaptor ( no BTB ) . Aliquots were taken at the indicated time points ( minutes ) and the presence of EPS15 and SPOPL was analyzed by immunoblotting . UB*EPS15 marks the appearance of ubiquitinated EPS15 . DOI: http://dx . doi . org/10 . 7554/eLife . 13841 . 00710 . 7554/eLife . 13841 . 008Figure 3—figure supplement 1 . CRL3SPOPL targets EPS15 for proteasome-dependent degradation . ( A ) Total cell extracts prepared from HeLa cells treated for 24 hr with DMSO or 10 µM MLN-4924 were analyzed by immunoblotting with specific antibodies recognizing the indicated proteins . The activity of the inhibitor is visible by the reduced levels of neddylated CUL3 . ( B ) HeLa cells were simultaneously transfected with siRNA depleting endogenous SPOPL , and either an empty control plasmid or a plasmid encoding siRNA-resistant SPOPL-HA . The levels of endogenous EPS15 and SPOPL-HA were analyzed by immunoblotting . ( C ) mRNA expression of EPS15 and EGFR was analyzed by qRT-PCR in HeLa cells treated with siControl or siSPOPL . The relative mRNA levels compared to GAPDH are shown . Data are mean + SD , N = 3 . ( D ) HeLa cell lines stably expressing EPS15-GFP from the inducible doxycycline-promoter were depleted of SPOPL or treated with control siRNA ( siControl ) . EPS15-GFP was analyzed in live cell imaging . EPS15-GFP total fluorescence was plotted as fold change between control and SPOPL-depleted cells . Scale bar = 20 μm . Data are mean + SD , n = 20 cells per sample , N = 3 . **p≤0 . 01 . ( E ) A549 cells were treated with indicated concentrations of MG132 for 40 hr while being kept in OptiMEM reduced serum medium instead of full medium . Afterwards cell lysates were prepared and analyzed by immunoblotting ( left side ) . Quantification of three independent experiments is shown on the right . Data are mean + SEM , N = 3 ( F ) In vitro ubiquitination assays with E . coli purified EPS15 and reconstituted CUL3-NEDD8-RBX1 in the presence of SPOPL or without BTB adaptor . The reactions were incubated at 37°C using UBE2D1 ( UBCH5 ) as the E2-enzyme . EPS15 and SPOPL were analyzed by immunoblotting . Note the appearance of slower migrating EPS15 forms , representing mono- and di-ubiquitination of EPS15 by the CRL3SPOPL complex in vitro . DOI: http://dx . doi . org/10 . 7554/eLife . 13841 . 008 Ubiquitination can specifically target a protein for degradation either through the 26S proteasome , or through lysosomal pathways ( Doherty and McMahon , 2009; Schreiber and Peter , 2014 ) . To understand which process is responsible for EPS15 degradation , we treated the GFP-tagged EPS15 cell line with either MG132 to block proteasome activity or chloroquine to stop endosome maturation ( Figure 3C ) . GFP-EPS15 was stabilized upon MG132 addition in a concentration-dependent manner , but not by chloroquine . Similarly , endogenous EPS15 levels slightly increased with increasing MG132 concentrations ( Figure 3—figure supplement 1E ) , indicating that EPS15 is degraded by the proteasome . We next examined whether EPS15 is a substrate of the CRL3SPOPL complex . Immunofluorescence microscopy showed SPOPL-GFP in vesicle-like structures that partially overlapped with EPS15 and EGFR ( Figure 3D ) . Moreover , SPOPL-GFP partially co-localized with the early endosomes marker EEA1 , while it was not detected in LAMP1-containing structures . This indicated that SPOPL does not associate with lysosomes . Consistent with this localization data , endogenous immunoprecipitation revealed that EPS15 precipitates with SPOPL but not with SPOP in cell extracts ( Figure 3E ) . Furthermore , recombinant GST-tagged EPS15 was purified from E . coli and tested for its ability to bind to purified SPOPL . SPOPL was eluted specifically with GST-EPS15 in vitro ( Figure 3F ) , demonstrating direct association of the two proteins . With evidence for in vitro and in vivo association between CRL3SPOPL and EPS15 , we tested whether EPS15 could be ubiquitinated by the CRL3SPOPL complex in vitro . We incubated recombinantly purified EPS15 with ubiquitin and neddylated CUL3/RBX1 complexes , with or without the addition of SPOPL . As shown in Figure 3G , EPS15 was readily ubiquitinated in a SPOPL-dependent manner . Depending on the E2 enzyme - UBE2R1 ( CDC34 ) or UBE2D1 ( UBCH5 ) - EPS15 was poly- , mono- and di-ubiquitinated , respectively ( Figure 3G and Figure 3—figure supplement 1F ) . These results demonstrated that EPS15 is ubiquitinated in a CRL3SPOPL-dependent manner in vitro and that ubiquitination leads to its proteasomal degradation . EPS15 contains two ubiquitin-interacting motifs ( UIM ) in its C-terminal domain that serve as a hub for regulation via ubiquitination in vivo . Moreover , the EPS15 amino acid sequence predicts SPOP binding motifs ( Φ-π-S-S/T-S/T , Φ = nonpolar , π = polar ) ( Figure 4A and Figure 4—figure supplement 1A , Zhuang et al . , 2009 ) . 10 . 7554/eLife . 13841 . 009Figure 4 . EPS15 is targeted via a SPOP/SPOPL binding consensus motif . ( A ) Cartoon of human EPS15 domain-organization and the amino-acid sequence . Indicated by color code are the SPOP/SPOPL binding site ( red ) and the lysine residue ( yellow ) , which is ubiquitinated in a CRL3SPOPL–dependent manner in vivo . In addition , the amino-terminal Ca2+-binding EF-hand motifs ( EH ) , the coiled-coil domain involved in dimerization and the two carboxy-terminal ubiquitin-interacting motifs ( UIMs ) involved in ubiquitin-binding are indicated . ( B ) EPS15 ubiquitin-profiling . Peptides containing EPS15 modification sites were quantified with LC-MS/MS after enrichment of the K-ε-GG motif from whole cell digests of HeLa cells treated with siSPOPL or siControl . Normalized precursor mass intensity profiles for EPS15 sites corresponding to K793 , K801 and K693 are shown ( raw data in Figure 4—figure supplement 1B ) . Quantification of the β-Actin K113 and the polyubiquitin K11 linkage peptide control for comparable enrichment . Data are mean ± SD , N = 3 . **p≤0 . 01 . ( C ) Purified SPOPL was incubated as indicated with GST-tagged wild-type EPS15 or GST-EPS15 mutants , where the predicted SPOPL binding motifs have been mutated individually ( GST-EPS15S605-S607A and EPS15S744-S746A , respectively ) , pulled down with glutathione sepharose ( IP [GST] ) and bound proteins were analyzed by Coomassie blue staining ( upper panel ) and immunoblotting ( lower panels ) . Note that SPOPL readily binds to GST-EPS15 and GST-EPS15S605-S607A , but this interaction is strongly reduced with the GST-EPS15S744-S746A mutant . ( D ) HeLa cells stably expressing GFP-tagged wild-type EPS15 , the EPS15S744-S746A or the EPS15K793R mutants from a doxycycline-inducible promoter were transfected as indicated ( + ) with control siRNA or siRNA depleting SPOPL . The levels of EPS15-GFP , EGFR and for control tubulin ( TUB ) were analyzed by immunoblotting with specific antibodies . Experiments were quantified in Fiji and the EPS15 levels plotted as fold-increase compared to controls . Data are mean ± SEM , N = 4 . *p≤0 . 05 . Note that SPOPL depletion does not further increase the levels of both EPS15 mutants . ( E ) Total cell extracts were prepared from HeLa cells expressing either GFP-tagged wild-type , the EPS15S744-S746A mutant or the EPS15K793R mutant in the presence ( + ) or absence ( - ) of HA-tagged SPOPL overexpression . The levels of EPS15-GFP , SPOPL-HA and control GAPDH were analyzed by immunoblotting . Note that overexpression of SPOPL-HA is able to induce degradation of wild-type but not the EPS15S744-S746A-GFP or the EPS15K793R-GFP mutant . DOI: http://dx . doi . org/10 . 7554/eLife . 13841 . 00910 . 7554/eLife . 13841 . 010Figure 4—figure supplement 1 . EPS15 is targeted via a SPOP/SPOPL binding consensus motif . ( A ) Alignments of the carboxy-terminal domains of EPS15 proteins from various species . Conserved SPOPL-binding motifs and putative ubiquitination sites are highlighted by yellow boxes . ( B ) Peptides containing EPS15 modification sites were quantified with LC-MS/MS after enrichment of the K-ε-GG motif from whole cell HeLa digests treated with siSPOPL and siControl . Raw intensities for each of the triplicate LC-MS/MS runs are shown with each of the siControl conditions scaled to 100% intensity . Normalized precursor mass intensity profiles for EPS15 sites corresponding to K793 , K801 and K693 are shown , with only K793 showing significant downregulation in the depletion condition . Quantification of a peptide corresponding to β-Actin K113 and the poly-ubiquitin K11 linkage peptide is also shown to demonstrate that enrichment variations did not influence the quantification of the EPS15 sites . Additionally , the total ion chromatographic intensities for each run are plotted to provide insight into the consistency of each of the separate experiments performed on different days . Data are mean ± SD , N = 3 . ( C ) HeLa cell lines stably expressing wild-type EPS15-GFP , the EPS15S744-746A-GFP mutant or the EPS15K793R-GFP mutant from the inducible doxycycline-promoter were treated with doxycycline for 3 days , and analyzed by live cell imaging . Displayed are maximal projections of Z-stack acquisitions , fully covering cell height . Scale bar = 10 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 13841 . 010 To determine whether EPS15 is also ubiquitinated in a SPOPL-dependent manner in vivo , we prepared cell extracts from SPOPL-depleted and RNAi control cells , and used a monoclonal antibody to enrich for isopeptides containing the K-ɛ-GG remnant motif after trypsin digestion of ubiquitinated substrate proteins ( Kim et al . , 2011 ) . Modified peptides were then eluted and quantified with liquid chromatography coupled to tandem mass-spectrometry ( LC-MS/MS ) ( Figure 4B , Figure 4—figure supplement 1B ) . This analysis identified several ubiquitination sites in EPS15 including K693 and K801 that were ubiquitinated irrespective of the presence or absence of SPOPL . In contrast , ubiquitination of K793 , located in the C-terminal domain of EPS15 close to the ubiquitin-interacting motifs ( UIMs ) ( Figure 4A ) , was significantly reduced in cells lacking SPOPL . The MATH domains of SPOP and SPOPL are very similar and , in the case of SPOP , known to be responsible for substrate recognition ( Errington et al . , 2012 ) . To test whether SPOPL recognizes the same motif in EPS15 , we expressed and purified EPS15 mutant proteins with the three serines in potential SPOPL binding pocket mutated to alanine residues ( S605-607A and S744-746A ) . While binding of SPOPL to the EPS15S605-607A mutant was comparable to wild-type controls , the ability of EPS15S744-746A to interact with SPOPL was severely reduced ( Figure 4C ) . This showed that SPOPL preferentially binds EPS15 through the conserved TSSSV motif . To confirm that the bona fide SPOPL-binding motif and the SPOPL targeted lysine are relevant for turnover of EPS15 in vivo , we compared the levels of C-terminally GFP-tagged wild-type EPS15 , EPS15S744-746A and EPS15K793R stably expressed in HeLa cells from a doxycycline-inducible promoter . Indeed , the steady-state levels of EPS15S744-746A-GFP were increased at least six fold , concomitant with decreased EGFR levels , and no further increase of EPS15S744-746A-GFP levels was detected by simultaneously depleting SPOPL ( Figure 4D ) . Furthermore , we analysed the EPS15K793R mutant , in which in addition to the lysine 793 the neighboring lysine 788 was mutated to an arginine to prevent spurious ubiquitination . Although it showed reduced expression , when compared to wild type , no stabilization was detected after SPOPL depletion , suggesting that the lysine 793 is indeed relevant for EPS15 turnover via CRL3SPOPL in cells . Furthermore , when HA-tagged SPOPL was overexpressed , degradation of the binding mutant EPS15S744-746A-GFP and the SPOPL-ubiquitination-deficient mutant EPS15K793R-GFP were not induced in contrast to wild-type EPS15 ( Figure 4E ) . Like EPS15-GFP , EPS15S744-746A-GFP and EPS15K793R-GFP localized at the cell surface and to punctate intracellular structures ( Figure 4—figure supplement 1C ) , implying that its interaction with SPOPL does not interfere with its subcellular localization . We concluded that CRL3SPOPL binds to a conserved SPOP/SPOPL binding motif in EPS15 . It ubiquitinates EPS15 on lysine K793 , which evidently results in proteasomal degradation of EPS15 . Next , we asked why loss of SPOPL-mediated EPS15 ubiquitination reduced the level of EGFR . We found that EGFR was more rapidly degraded after addition of EGF in SPOPL-depleted cells compared to RNAi-controls , while its internalization judging by EGFR ubiquitination and EGF uptake was similar ( Figure 5A and Figure 5—figure supplement 1A ) . The degradation was blocked by chloroquine , which inhibits the acidification of endosomes and LYs ( Figure 5B ) , confirming that EGFR was degraded in lysosomes . 10 . 7554/eLife . 13841 . 011Figure 5 . Ubiquitination of EPS15 by SPOPL regulates EGFR sorting and degradation . ( A ) HeLa cells transfected with control siRNA ( siControl ) or siRNA depleting SPOPL ( siSPOPL ) were serum-starved for 20 hr and treated with EGF ( 200 ng/μl ) for the indicated times ( minutes ) . EGFR levels and ubiquitination were analyzed in total cell extracts by immunoblotting . TUB controls for equal loading . EGFR levels were plotted as fold-increase compared to basal levels against the time of EGF treatment ( right panel ) . Data are mean ± SEM , N = 5 . *p≤0 . 05 . ( B ) SPOPL-depleted HeLa cells were serum-starved for 20 hr , and pre-treated or not for 30 min with 20 μM chloroquine ( CQ ) . EGF ( 200 ng/μl ) was then added , and EGFR levels analyzed by immunoblotting of total cell extracts prepared at time 0 or after 60 min . TUB controls for equal loading . ( C ) Total cell extracts were prepared from HeLa cells induced ( + ) or not ( - ) to express GFP-tagged EPS15 from the doxycycline ( Dox ) -inducible promoter and treated with control or SPOPL siRNAs . The levels of EGFR , EPS15-GFP and for control GAPDH were analyzed by immunoblotting . EGFR levels were quantified in Image J . Data are mean + SD , N = 3 . ( D ) Total cell extracts prepared from HeLa cells treated for 3 days with control siRNA or RNAi oligos targeting EPS15 and SPOPL , individually and together , were analyzed by immunoblotting for EGFR and EPS15 protein levels . Equal loading was controlled by immunoblotting for GAPDH . ( E ) Cells transfected with control siRNA ( siControl ) or siRNA depleting SPOPL ( siSPOPL ) were analyzed by indirect immunofluorescence for EPS15 ( red ) and HRS ( green ) using super resolution microscopy ( SRM ) in structured illumination mode . Maximal projection is shown ( left panel ) . Scale bar = 5 μM . The squares are shown at 5x higher magnification in the insets . Co-localization of EPS15 and HRS as well as the number and size of EPS15-positive endosomes was quantified using Fiji ( right graphs ) . Data are mean ± SEM , 20 > n < 10 , N = 4; *p≤0 . 05; **p≤0 . 01 . ( F ) Cell extracts were prepared from HeLa cells 72 hr after transfection with control siRNA or siRNA targeting SPOPL ( siSPOPL ) , and incubated with control IgG or antibodies against HRS . Co-precipitated proteins ( IP ) were eluted and analyzed by immunoblotting ( IB ) for the presence of HRS and EPS15 . 40 μg of protein extract was loaded as input samples ( left side ) . The input and IP protein levels were quantified using ImageJ ( right side ) . Data are mean ± SEM , N = 6 . DOI: http://dx . doi . org/10 . 7554/eLife . 13841 . 01110 . 7554/eLife . 13841 . 012Figure 5—figure supplement 1 . Ubiquitination of EPS15 by SPOPL regulates EGFR sorting and degradation . ( A ) A549 cells were treated with siControl or siSPOPL and starved of growth factors . After 1 hr binding at cold and washing , Alexa488 labeled EGF was allowed to be internalized for indicated times ( minutes ) , and the EGF-positive vesicles were counted . Scale bar = 20 μm . Data are mean + SD , n > 500 cells per sample , N = 3 . ( B ) HeLa cells treated with siControl or siSPOPL were stained for EPS15 by indirect immunofluorescence and analyzed by super-resolution microscopy as described in the legend of Figure 5E . Instead of maximal projections , the image shows a single acquisition layer taken at the very bottom of the cell to specifically visualize the size of EPS15-positive patches at the plasma membrane . Scale bar = 5 μm . Regions of interest ( squares ) are shown at 5x magnification . The EPS15 particle area and particle size were quantified by Fiji , n = 10 cells per sample , N = 3 . ( C ) Cell extracts prepared from HeLa cells treated for 72 hr with siControl oligos or siSPOPL were immunoprecipitated with control IgG or antibodies against EPS15 . The immunoprecipitate ( IP ) was analyzed by immunoblotting ( IB ) for the presence of EGFR and EPS15 ( upper panel ) . Appropriate expression of the analyzed proteins was examined by immunoblotting an aliquot of the cell extract ( input ) . The input protein levels and proteins co-immunoprecipitated with EPS15 antibodies were quantified ( lower panels ) . Data are mean ± SEM , N = 3 . *p≤0 . 05; **p≤0 . 01 . ( D ) HeLa cells treated for 72 hr with siControl ( upper row ) or siSPOPL ( lower row ) were stained by indirect immunofluorescence for RAB11 ( green ) and EGFR ( red ) and analyzed by super-resolution microscopy in structured illumination mode . The individual images were overlaid to visualize co-localization ( MERGE ) . Scale bar = 10 μm . Regions of interest ( squares ) are shown at 8x higher magnification on the right . Scale bar = 1 μm . The arrow head marks a recycling vesicle positive for EGFR and RAB11 . Note that the number of recycling vesicles is not significantly altered in SPOPL-depleted cells . DOI: http://dx . doi . org/10 . 7554/eLife . 13841 . 012 To examine whether SPOPL-dependent ubiquitination of EPS15 protects EGFR from lysosomal degradation , we stably overexpressed C-terminal GFP-tagged EPS15 from the Dox-inducible promoter in RNAi-control and SPOPL-depleted cells . Overexpression of EPS15-GFP only decreased the EGFR levels in the absence of SPOPL ( Figure 5C ) , implying that EPS15 accumulation alone may not be sufficient to promote EGFR degradation . Rather SPOPL-dependent ubiquitination directly regulates EPS15 activity . To test whether the presence of EPS15 is required for lysosomal EGFR degradation in SPOPL-depleted cells , we measured EGFR levels in SPOPL-depleted cells that are simultaneously depleted for EPS15 . Indeed , EGFR levels increased in cells lacking both SPOPL and EPS15 , demonstrating that EPS15 is necessary to promote EGFR degradation in the absence of SPOPL ( Figure 5D ) . Finally , we investigated why loss of SPOPL-mediated ubiquitination of EPS15 increased EGFR degradation in lysosomes . EPS15 not only promotes EGFR uptake via clathrin-mediated endocytosis , but it is also localized in places of intracellular sorting and degradation ( Gucwa and Brown , 2014; van Bergen en Henegouwen , 2009 ) . We used super resolution microscopy in structured illumination mode to localize EPS15 in SPOPL-depleted cells and controls . While EPS15 at the plasma membrane was unaffected by SPOPL depletion ( Figure 5—figure supplement 1B ) , EPS15 accumulated in intracellular structures upon SPOPL-depletion and increasingly co-localized with the ESCRT-0 subunit HRS ( Figure 5E ) . The number and size of the EPS15-containing vesicles was unaffected suggesting that SPOPL-dependent ubiquitination of EPS15 is required to remove it from ESCRT-0 complexes . Because HRS is necessary for cargo sorting into the ESCRT pathway for MVB formation ( Bache , 2003 ) , it is plausible that the prolonged co-localization of EPS15 with HRS in the absence of SPOPL results in enhanced EGFR sorting to the lysosome . Indeed , the interaction of EPS15 and EGFR is reduced in SPOPL-depleted cells ( Figure 5—figure supplement 1C ) , while the interaction of EPS15 and HRS is stabilized ( Figure 5F ) , consistent with the idea that EPS15 ubiquitination promotes EGFR stabilization . Additionally , we did not detect enhanced EGFR colocalization with recycling endosomes marked by RAB11 , although RAB11 staining is elevated in siSPOPL-treated cells ( Figure 5—figure supplement 1D ) . Taken together , these data suggest that loss of SPOPL-mediated ubiquitination of EPS15 favors EGFR sorting and trafficking into the lysosomal pathway via HRS ( Figure 6 ) . 10 . 7554/eLife . 13841 . 013Figure 6 . CRL3SPOPL ubiquitinates EPS15 at endosomes EPS15 and thereby regulates EGFR sorting and lysosomal degradation . Schematic model depicting the major EGFR trafficking routes and highlighting possible roles of EPS15 ubiquitination by CRL3SPOPL at endosomes . EGFR is internalized at the plasma membrane by clathrin-mediated and clathrin-independent endocytosis , which involves recognition of the ubiquitinated receptor by EPS15 . EGFR is then either recycled back to the plasma membrane via the endocytic recycling compartment ( ERC ) , or targeted for degradation into late endosomes ( LE ) /lysosomes after its uptake into sorting endosomes ( SE ) / multivesicular bodies ( MVBs ) by the ESCRT machinery . Loss of CRL3SPOPL activity results in enhanced EGFR sorting into the degradative pathway , suggesting that CRL3SPOPL-mediated ubiquitination of EPS15 at endosomes delays EGFR trafficking to lysosomes . For further explanation , see text . DOI: http://dx . doi . org/10 . 7554/eLife . 13841 . 013
EPS15 was originally identified as a substrate for the EGFR kinase that interacts with the clathrin assembly adaptors AP2 and EPSIN1 ( Benmerah et al . , 1995; Chen et al . , 1998; Fazioli et al . , 1993 ) . EPS15 possesses four domains: a N-terminal domain with three EH motifs , a central coiled-coil domain involved in dimerization , and a C-terminal regulatory domain with the AP2-binding site followed by two UIM motifs ( Figure 4A ) ( van Bergen en Henegouwen , 2009 ) . Perturbation of EPS15 and EPSIN1 function blocks endocytosis of EGF and transferrin , demonstrating that they are part of the primary endocytic machinery at the plasma membrane ( Carbone et al . , 1997 ) . In addition , EPS15 promotes vesicular trafficking intracellularly ( Yuan et al . , 2014 ) . Overexpression of shorter isoforms ( EPS15S , EPS15B ) determines cargo recycling or degradation at endosomes ( Chi et al . , 2011; Roxrud et al . , 2008 ) . EPS15 is moreover known to associate with the E3-ligases NEDD4 and Parkin ( PARK2 ) and is mono-ubiquitinated by them ( Fallon et al . , 2006; Polo et al . , 2002; Woelk et al . , 2006 ) . The putative E3 binding sites for these ligases are located in the extreme C-terminal end of the EPS15 regulatory domain close to the UIMs . The binding sites are distinct from the TSSSV motif required for interaction with SPOPL . Several ubiquitination sites have been reported in EPS15 ( Kim et al . , 2011; Savio et al . , 2016; Wagner et al . , 2011 ) and most of them are not affected by downregulation of SPOPL in vivo as observed in our mass-spectrometry analysis ( Figure 4—figure supplement 1B ) . That different E3 ligases appear to ubiquitinate EPS15 in distinct sites , may have to do with its fate in different subcellular locations and/or with response to distinct signals . Our findings support previous reports that ubiquitination inhibits the functions of EPS15 . For example , a single ubiquitin in the C-terminal domain reduces binding of EPS15 to free ubiquitin thus impairing association with ubiquitinated client proteins and co-localization with EGFR on endosomes ( Hoeller et al . , 2006 ) . Our own data shows that ubiquitination by the CRL3SPOPL complex inhibits EPS15 function in endosome maturation ( Figure 6 ) . In this case , ubiquitination may control the timing and duration of EPS15 interaction with ESCRT-0 , which in turn may influence ILV formation and the efficiency of lysosomal sorting and degradation of EGFR and other targets . It may also prevent the remodeling of the endosomal limiting membrane during endosome maturation , leading to loss of components needed for uncoating of incoming IAV capsids . EPS15 forms a trimeric complex with the ESCRT components HRS and STAM and interacts with EGFR and HRS 30 min after EGF uptake ( Bache , 2003; Sigismund et al . , 2005 ) Upon HRS depletion , EPS15 accumulates on endosomes ( Gucwa and Brown , 2014 ) . Our results showed that SPOPL depletion promotes endosomal co-localization of EPS15 with HRS . This implies that EPS15 association with HRS is important for EPS15 turnover and for targeting of the EGFR to lysosomes for degradation . That EPS15 accumulates on ubiquitin-enriched endosomes suggests that in addition to HRS it is recruited by ubiquitinated clients ( Gucwa and Brown , 2014 ) . A change in the phosphorylation state of EPS15 - either due to the mutation EPS15-Y850F or overexpression of the phosphatase PTPN3 – results in faster degradation of EGFR ( Li et al . , 2014 ) . Taken together , these findings indicate that EPS15 promotes sorting of ubiquitinated EGFR at endosomes by interacting with HRS . Because upon SPOPL depletion EPS15 accumulates on HRS-containing vesicles and EGFR degradation is accelerated , we propose that the EPS15-HRS complexes actively promote sorting of EPS15 clients such as the EGFR and MET to lysosomes . Interestingly , recent findings further demonstrate that EPS15 de-ubiquitination by USP9X affects EGFR internalization and its trafficking to lysosomes ( Savio et al . , 2016 ) . In contrast to SPOPL depletion , the depletion of USP9X leads to slower degradation of EGFR . Therefore , it is evident that a cycle of ubiquitination / de-ubiquitination of EPS15 not only regulates the internalization pathway , but is also involved in the sorting of cargo in LEs ( Figure 6 ) . Ubiquitination may increase EPS15 turnover in endosomes . This may first involve EPS15 binding to its own UIM domains followed by poly-ubiquitination to target EPS15 for proteasomal degradation . While SPOPL-depletion promotes the HRS- and ESCRT-dependent fast-forward sorting route for EPS15 client cargo , other sorting processes may be inhibited due to depletion of the ESCRT processing machinery . Indeed , ILV formation is a dynamic and tightly regulated process that involves recycling of the ESCRT components by VPS4 ( Babst et al . , 1998; Sachse , 2004 ) . It is evident that ILV formation depends on SPOPL , and that loss of available ESCRT complexes due to loss of CRL3SPOPL activity prevents the reduction in the surface area of the endosomal limiting membrane . This is reflected in both the lack of ILVs and the apparent size-increase in LEs . The inhibition of IAV entry after SPOPL-depletion was similar to our previous observations with CUL3-depletion ( Huotari et al . , 2012 ) . This late penetrating virus was endocytosed normally and reached a late endosomal compartment acidic enough to induce HA-conversion and membrane fusion/hemifusion . However , the uncoating process that involves dissociation of the matrix protein M1 shell and release of the vRNPs from the endosome did not take place . Recent studies have indicated that uncoating of IAV capsids is a complex process that requires priming of the viral capsid prior to fusion by exposure to low pH and potassium ions in the lumen of endosomes as well as the function of the ubiquitin-vacuolar protein sorting system and interaction with a number of host cell factors in the cytosol ( Banerjee et al . , 2014; Khor et al . , 2003; Martin and Helenius , 1991; Stauffer et al . , 2014 ) . While we found that acidification of IAV occurred in the absence of SPOPL and CUL3 , it is possible that the exposure to an elevated potassium concentration in late endosomes did not take place . It is also possible that cellular factors associated with LE membrane and needed for uncoating , such as the E3 ligase ITCH , did not associate with immature endosomes ( Su et al . , 2013 ) . Moreover , the EGFR level could be crucial for IAV entry , since EGFR signaling is required for efficient IAV infection . Available evidence suggests that virus binding leads to lipid-raft clustering , which activates EGFR and other RTKs and facilitates IAV uptake ( Eierhoff et al . , 2010 ) . Given the strong homology between SPOPL and SPOP particularly in the MATH-domain , it is surprising that their subcellular localizations diverge , with SPOPL localizing to endosomes and preferentially ubiquitinating EPS15 in vivo . SPOP and SPOPL mainly differ by a 18 amino-acid insertion in SPOPL ( Errington et al . , 2012 ) that may be responsible for their distinct subcellular localization , which likely underlies the observed substrate specificity . SPOP and SPOPL have also been shown to form heterodimers , and in vitro experiments have suggested that SPOPL may inhibit SPOP activity by attenuating self assembly in a dose-dependent manner ( Errington et al . , 2012 ) . It will be important therefore to identify additional SPOP and SPOPL substrates , and further test the functional interaction of SPOP and SPOPL in physiological settings . While we could demonstrate that CRL3SPOPL directly ubiquitinates EPS15 , it is likely that SPOPL regulates additional substrates at endosomes . That EPS15 , STAM , and to a lesser extent HRS levels were increased in SPOPL-depleted cells ( Figure 2E ) raises the possibility that CRL3SPOPL directly or indirectly affects these ESCRT components . Finally , it is clear that CUL3 must modify additional targets relevant for endocytic trafficking . Indeed , downregulation of SPOPL decreases EGFR levels , whereas CUL3-depletion results in greatly elevated levels of EGFR , most likely due to a reduced ability of endosomes to fuse with lysosomes . Our RNAi screen identified BTB-adaptors other then SPOPL , and a subset of them showed increased EGFR levels mimicking those defects associated with CUL3 ( MG , AU and MP , unpublished results ) . Detailed analysis of these adaptors may shed light on additional functions of CUL3 at late stages of endocytosis .
HeLa Kyoto were kindly provided by Daniel Gerlich , A549 were obtained from American Type Culture Collection ( ATCC ) , HeLa FRT cells were a kind gift of Stephen Taylor and HEK-293 FRT were bought from Life Technologies . HeLa Kyoto , HeLa FRT and HEK-293 FRT were cultured at 37°C and 5% CO2 in Dulbecco’s modified Eagle Medium ( DMEM , Gibco ) supplemented with 10% fetal calf serum ( FCS ) . A549 were maintained at 37°C and 5% CO2 in DMEM + Glutamax ( Gibco ) + 10% FCS . All cell lines were not passaged longer than 3 months and on a routine basis tested negative for mycoplasma contamination . Site-directed integration into HeLa FRT or HEK-293 FRT was achieved with the plasmid pcDNA5-FRT/TO using the Life Technologies Flp-In System . Stable cell lines were generated by transfecting cells in a 6-well plate setting with 1 . 8 μg pOG44 bearing the Flp recombinase cDNA and 0 . 2 μg of the pcDNA5-FRT/TO by Lipofectamine 2000 or 3000 transfection reagents in Opti-MEM according to the manufacturer’s instructions . Cells were plated the next day into a 15 cm dish , and individual cell clones were picked and expanded after selection with 200 μg/ml Hygromycin B . Stable cell lines were cultured in 4 μg/ml Blasticidin to maintain the Tet-Repressor , and expression from the doxycycline-regulated promoter was induced by addition of 1 μg/ml doxycycline . Generation of a stable cell line expressing a shRNA construct was done similar by using the pSUPERIOR vector and selection in 1 μg/ml Puromycin . Transient transfections of plasmid DNA were performed using Lipofectamine 2000 or 3000 ( Life Technologies ) according to the manufacturers’ instructions with Opti-MEM . For siRNA experiments , a final concentration of 20 nM siRNA was incubated with Lipofectamine RNAimax ( Life Technologies ) in Opti-MEM , and successful RNAi depletion of the target protein was assessed after 72 hr . siRNA screenings were conducted in 96-well optical bottom plates ( Greiner ) , and transfection was performed with 20 nM final siRNA concentration and 0 . 1 μl of Lipofectamine RNAimax per well . Between 1500 and 3000 A540 or HeLa cells were plated and reverse transfected for 72 hr , so that they reached 50–80% confluency on the day of infection . Cells were infected with IAV X31 virus with a concentration resulting in 20–40% infection , fixed with 4% PFA after 10–14 hr , permeabilized and blocked with 0 . 1% Saponin , 1% BSA and 10% FCS in PBS , followed by staining with an antibody against NP ( HB65 , ATCC , unpurified , 1:100 ) in order to detect newly synthesized viral protein . Infection was scored by applying an automated imaging and infection scoring procedure , as previously described ( Banerjee et al . , 2013 ) . Briefly , cells were imaged on a MD2 screening microscope ( Molecular Devices ) equipped with a Photometrics CoolSNAP HQ camera using the 10x objective ( 0 . 3 NA Plan Fluor ) and an automated autofocus . Infection efficiency was quantified using a MATLAB-based protocol , which determines the ratio of NP positive cells compared to the total cell number . The SPOPL and SPOP cDNA were obtained from the Orfeome collection Version 5 , and subcloned using PCR into the KpnI and XhoI sites of pcDNA5-GFP or into pcDNA5-HA-Strep-Strep for C-terminal epitope-tagging . The human EPS15 cDNA was purchased from Sino Biological , while the mouse EPS15 cDNA was a kind gift of Pier Paolo Di Fiore ( Milano , Italy ) . Both were subcloned into pcDNA5-GFP via the KpnI and XhoI restriction sites for C-terminal epitope-tagging . Site directed mutagenesis of the SPOPL and EPS15 coding region was carried out using the Quickchange protocol ( Stratagene ) or the 'Round the Horn' site-directed mutagenesis protocol , and the resulting constructs were verified by sequencing . For the generation of a SPOPL siRNA resistant construct , the following primers were used: 5’ CAG TGT CCA CAA TTC GGG ATA CCT CGG AAA CGG CTA AAA CAG TCC 3’ and 5’ GGA CTG TTT TAG CCG TTT CCG AGG TAT CCC GAA TTG TGG ACA CTG 3’ . For generation of the EPS15 mutants the following primers were used: 5’ CTG CGC TGA CAG GTC CAG TTG CAG 3’ and 5’ CTG CGT CTA CAT TAA ATG GAT CTT CCTC 3’ , 5’ CCG CTG TCA GCA ACG TAG TGA TTA C 3’ and 5’ CCG CTG TGG CTG AAC GAA AAG GAT C 3’ , 5’ AGA TTG GAT TCT CCT GAT CCC 3’ and 5’ GTT GAT GGA TCT TCT CCC 3’ . The RAB5 , RAB7 and RAB9 cDNA were obtained from the Orfeome collection Version 5 and 8 . GFP-RAB5 , GFP-RAB7 and GFP-RAB9 constructs were generated by subcloning the cDNA into pcDNA5-GFP N-terminal tag via the GATEWAY cloning system . Resulting constructs were sequence-verified . To extract mRNA for quantitative real-time PCR analysis , cells were lysed with QIAshredder columns ( Qiagen ) , and RNA isolated with the RNeasy kit ( Qiagen ) following the instructions by the manufacturer . mRNA was then reverse transcribed to cDNA with Superscript II RNase H-Reverse transcriptase ( Life Technologies ) and random primers ( Microsynth ) in the presence of RNAse OUT ( Life Technologies ) . cDNA levels were quantified using SYBR Green PCR Master mix and the Light Cycler 480 SYBR I Master ( Roche ) . Individual samples were normalized to the human housekeeping gene GAPDH . For live cell microscopy , cells were grown in an 8-well chamber slide ( LabTek ) , protein expression induced with 1 μg/ml doxycycline for 8–12 hr and cells were then imaged in imaging medium ( Gibco , no phenol-red , CO2 independent ) + FCS at 37°C using an epifluorescence-based Delta Vision microscope , with 20x , 40x , 60x or 100x objectives ( 20x 0 . 45NA Ph LUCPLFLN ( Long-distance dry ) , 40x 1 . 3NA DIC Oil UApo , 60x 1 . 4NA DIC Oil PlanApo , 100x 1 . 4NA DIC Oil PlanApo ) , a LED illumination source and a Roper CoolSnap HQ camera . GFP-RAB7 vesicle diameter was quantified manually in Image J . EPS15-GFP levels were quantified in Image J by measuring total fluorescence levels – background of a maximal projection . For indirect immunofluorescence , cells were grown on a 20 or 12 mm coverslip , washed with PBS and then fixed in 4% PFA for 5–15 min at room temperature ( RT ) . After extensive washing with PBS , cells were permeablized with 0 . 5% NP-40 in PBS for 2 min , washed 3x with PBS + 0 . 01% Triton-X100 ( PBS-TX100 ) and blocked for unspecific binding by incubation for 1 hr in PBS-TX100 containing 3% BSA . The first antibody was then added for 2 hr at RT in the presence of 3% BSA in PBS-TX100 , followed by secondary antibody incubation with Alexa dye-conjugated anti-rabbit and / or anti-mouse IgGs ( 1:1000 , Life Technologies ) . Coverslips were mounted in Mowiol + DAPI . For super resolution microscopy , cells were grown on 18 mm coverslips ( High Precision 170 +/- 5 μm , Marienfeld ) in 6-well plates ( 10 , 000 cells/6-well ) , transfected with control siRNA and siRNA directed against SPOPL , and prepared for immunofluorescence staining as described above . After the final wash , the cells were shortly dipped into water , mounted onto cover glasses using non-hardening mounting medium VectaShield ( vector laboratories , cat . nr . H-1000 ) , and sealed with nail polish . Acquisitions were taken at the Deltavision OMX ( PlanApoN 60x /1 . 42NA Oil PSF ) in structured illumination mode with a 4 sCMOS OMX V4 ( 15bit range ) camera . After image reconstruction and registration , maximal projections of all acquisitions were generated in Fiji and combined in a montage . Colocalization between two channels was analyzed using the Fiji plugin 'Colocalization Threshold' . To analyze endosome number and size an automatic threshold was applied to maximal projections and analyzed using the Fiji plugin 'Particle analysis' . For thin section electron microscopy , cells were grown on a 12 mm coverslip in 24 well plates and knockdown of target genes was accomplished by siRNA treatment as described previously . Either cells were fixed in 2 . 5% glutaraldehyde with 0 . 05 M sodium cacodylate at pH 7 . 2 , 50 mM KCl , 1 . 25 mM MgCl2 and 1 . 25 mM CaCl2 for 30 min , incubated for 1 hr at RT in 2% OsO4 followed by overnight incubation with 0 . 5% uranyl acetate at 4°C and dehydration was done by stepwise washing of coverslips in EtOH ( range from 50 – 100% ) , followed by washing with 100% Propylenoxid ( PO ) . Cells were then embedded in Epon and heated at 60°C over night for polymerization . Or cells were fixed in 2 . 5% glutaraldehyde in 0 . 1 M Na-Cacodylate buffer ( pH 7 . 4 ) with 0 . 5 mg/ml Ruthenium Red for 2 hr or in 2% formaldehyde , 1 . 5% gutaraldehyde in 0 . 1 Na-Cacodylate buffer ( pH 7 . 4 ) over night at 4°C . After washing , the sample stayed in 2% uranyl acetate over night at 4°C and dehydration was done stepwise in acetone ( range from 50 – 100% ) , embedding stepwise in Epon and samples were heated at 60°C during 48 hr for polymerization . The influenza A virus strain X31 ( A/Aichi/68 , H3N2 ) was purchased from Virapur , CA USA . All virus assays were performed in infection medium , composed of DMEM with 50 mM HEPES and 0 . 2% BSA , pH 6 . 8 . All infection assays were performed in A549 or HeLa cells , using the protocol described above . For drug treatments , cells were pretreated for 1 hr with the drugs , and virus infection was conducted in the presence of the drug . Infected cells were fixed and the infection efficiency quantified by immunofluorescence staining of NP expression as a marker of newly synthesized viral proteins . For bypass nuclear import , influenza virus ( 1 μl /well of 96-well plate ) was pre-bound to cells on ice for 1 hr in infection medium . Cells were then washed once with cold infection medium on ice and replaced by either infection medium ( pH 6 . 8 ) or low pH infection medium ( pH 5 . 4 , buffered with 100 mM citrate buffer ) to induce viral fusion with the plasma membrane . After incubation for 2 . 5 min at 37°C , cells were returned to ice , washed 2 times with cold infection medium to remove traces of acid , and then incubated for 30 min in STOP medium ( DMEM with 50 mM HEPES , pH 7 . 4 supplemented with 20 mM NH4Cl to prevent further endocytic uptake of virus ) . Cells were then fixed , and NP expression was quantified as described above . All IAV entry assays have been carried out essentially as described ( Banerjee et al . , 2013 ) . Imaging was performed with the MD screening microscope ( Molecular Devices ) using the 10x or 20x objectives , and the images quantified using MATLAB-based Cell Profiler modules ( Banerjee et al . , 2013 ) . For binding assays , cells were incubated with the virus inoculum ( 0 . 75 μl/well of 96-well plate ) , on ice for 1 hr in infection medium , washed 3x with cold PBS and fixed with 4% PFA . Bound virus particles were stained with a polyclonal antibody against IAV ( PINDA , polyclonal , rabbit , 1:500 ) , visualized with a fluorescently labeled secondary antibody . To assess IAV endocytosis , virus inoculum ( 0 . 75 μl/well of 96-well plate ) was bound to cells on ice for 1 hr in infection medium , after which inoculums were removed and the cells incubated for 30 min in infection medium at 37°C . Cells were fixed with 4% PFA , and non-internalized virus particles blocked with a polyclonal PINDA antibody ( 1:500 ) against IAV prior to permeabilization . Negative controls were directly fixed after the virus-binding step and termed fixation control . Cells were then permeabilized with 0 . 1% saponin and stained with an antibody against HA ( H3 , monoclonal , mouse 1:100 ) to detect internalized particles , followed by fluorescently labeled secondary antibody detection . Acidification assays were essentially done as the endocytosis assay , except that the cells were incubated longer at 37°C . In brief , virus inoculum ( 0 . 75 μl/well of 96-well plate ) was bound to cells on ice for 1 hr in infection medium , and then incubated at 37°C with warm infection medium for 1 hr . Cells were fixed with 4% PFA , and stained with A1 antibody ( 1:1000 , monoclonal ) specifically recognizing the acidified conformation of HA ( Webster et al . , 1983 ) . Bafilomycin A treatment and siRNA against the vATPase subunit vATP6V1B2 were used as negative controls in these assays . For uncoating assays , the virus inoculum ( 1 . 2 μl/well of 96-well plate ) was bound to cells on ice for 1 hr in infection medium , after which cells were incubated for 2 . 5 hr at 37°C with warm infection medium containing 1 mM cycloheximide ( CHX ) to block viral protein synthesis . Cells were fixed with 4% PFA , and stained with antibody against M1 ( HB64 , ATCC , 1:250 ) , followed by fluorescently-labeled secondary antibodies detection . Bafilomycin A treatment or siRNA against the vATPase subunit vATP6V1B2 were used as negative controls . Cytoplasmic dispersion of M1 and other features characteristic for viral uncoating were detected and quantified using the Advanced Cell Classifier program ( www . cellclassifier . org ) . Nuclear import assays were conducted as the uncoating assay with slight modifications . In brief , after virus binding , cells were incubated for 5 hr at 37°C with warm infection medium containing 1 mM cycloheximide ( CHX ) , which was replaced with fresh CHX after 2 . 5 hr . Bafilomycin A treatment and siRNA against the vATPase subunit vATP6V1B2 were used as negative controls . Cells were fixed with 4% PFA , and incubated with antibody against NP ( HB65 , ATCC , unpurified , 1:10 ) , followed by fluorescently-labeled secondary antibody detection and DAPI staining ( 1:5000 ) . Moreover , cells were stained with WGA ( 1:200 in PBS ) prior to permeablization . Nuclear import of NP was quantified using the MATLAB-based infection scoring program and cells were counted as positive , if a certain signal threshold was achieved in the nucleus . For FACS-based IAV fusion assays , IAV X31 stocks were diluted in PBS to 0 . 1 mg/ml and labeled for 1 hr at RT with R18 and SP-DiOC18 at final concentrations of 0 . 2 mM . The labeled virus particles were filtered through a 0 . 22 μM-pore filter ( Millipore ) and stored at 4°C in the dark . After binding ( 50 μl labeled virus + 150 μl infection medium per well/ 24 well plate setting ) at 4°C for 1 h , internalization for 1 hr and fixation with 4% PFA , cells were washed 3x in FACS buffer ( 20 mM EDTA , 2% FCS in PBS ) and analyzed using a FACS Calibur instrument for red and green signal . The red-to-green ratio was quantified using FlowJo 7 . 6 . Bafilomycin-A-treated cells were used as negative controls showing internalization signal of the virus , but no fusion events , which would be detected as a green signal . Cells were treated with siRNA for 72 hr , followed by serum starvation for at least for 4 hr or overnight by exchanging medium to starvation medium ( DMEM without FCS ) . Then , cells were incubated with a final concentration of 100 ng/ml EGF ( EGF-488 , Life Technologies ) in starvation medium for 1 hr at 4°C to allow binding to the EGF receptors . Internalization was achieved by incubating cells for 0–90 min at 37°C , followed by a short acid wash , fixation with 4% PFA and counterstaining with DAPI and WGA to visualize DNA and the cell shape , respectively . Cells were imaged on a MD screening microscope ( Molecular Devices ) with the 20x objective ( 0 . 75 NA S Fluor ) , and uptake quantified with a MATLAB-based Cell Profiler module measuring the number of EGF-containing particles per cell and the fluorescent intensities of single particles . Cells were washed with PBS , scraped off from the dish , and centrifuged for 5 min at 1200 rpm . Pellets were resolved in extraction buffer ( 20 mM Tris pH 7 . 5 , 100 mM NaCl , 20 mM β-glycero-phosphate , 5 mM MgCl2 , 0 . 2% NP-40 , 10% glycerol , 1 mM NaF , 0 . 5 mM DTT , complete Protease inhibitor mix tablet , 10 mM 1 , 10-Phenantroline ) , incubated 20 min on ice , and centrifuged for 10 min at 7500 rpm at 4°C in order to pellet bulk DNA and RNA , or treated with Nuclease for 20 min . The protein concentration was measured by Bradford assay and samples were equalized . For SDS-PAGE and immunoblotting , 4x LDS sample buffer + 1 mM DTT was added , the extracts boiled and proteins separated by SDS-PAGE using 7–14% gradient polyacrylamide gels . For immunoblotting , proteins were transferred to a PVDF membrane ( Millipore ) using a semi-dry blotting device . Membranes were blocked for 30–60 min with 5% non-fat milk resuspended in PBS supplemented with 0 . 1% Tween 20 ( PBS-T ) , followed by incubation with primary antibodies for 2 hr at RT or over night at 4°C . Membranes were then washed 3 times with PBS-T and incubated for 60 min with HRP-conjugated secondary antibodies resuspended in 5% milk-PBS-T . Membranes were washed three times with PBS-T and incubated with ECL solution ( 100 mM Tris pH 8 . 5 + Luminol and Coumaric acid ) or with SuperSignal West Femto chemiluminescent substrate ( Thermo Fisher Scientific ) . For immunoprecipitation experiments , cell pellets were resolved in IP buffer ( 10 mM Tris pH 7 . 5 , 100 mM KCl , 2 mM MgCl2 , 0 . 5% NP-40 , 300 mM Sucrose , 10 mM β-glycero-phosphate , 0 . 2 mM Na3VO4 , complete Protease Inhibitor tablet , 0 . 5 mM DTT , 1 mM PMSF , 1 μg/ml Leupeptin , 1 μg/ml Pepstatin , 1 mM NaF , 10 mM NEM , 10 mM 1 , 10-Phenantroline ) and broken using a 27G syringe on ice . Cell extracts were cleared by centrifugation for 10 min at 10 , 000 rpm in a table top centrifuge . Supernatants were incubated with Strep-Tactin sepharose ( IBA ) or HA . 7-coupled beads ( Sigma ) for 1 – 2 hr rotating at 4°C . After extensive washing with IP buffer , bound proteins were eluted with 100 mM Glycine pH 2 . 0 , and analyzed by SDS-PAGE and immunoblotting . For immunopreciptations of endogenous EPS15 , extracts adjusted to the same volume and protein concentration were incubated with 10 μg anti-EPS15 antibody ( Santa-Cruz , sc-534 , rabbit ) or anti-rabbit control IgGs for 2 hr , followed by the addition of 40 μl protein G sepharose for an additional hour . Resin was then washed 6 times with IP buffer , bound proteins eluted by boiling in 40 μl of two fold SDS sample buffer and analyzed by SDS-PAGE and immunoblotting . HeLa cells were harvested and resuspended in 6 ml hypotonic swelling buffer ( 20 mM sucrose , 20 mM HEPES NaOH pH 7 . 0 , 10 mM KCl , 5 mM MgCl2 , 10 mM CaCl2 , 5 mM EGTA ) . After 10 min , cells were lysed with 30 strokes in a Dounce homogenizer using a tight pestle , and swelling was stopped by the addition of 6 ml two fold homogenization buffer ( 500 mM sucrose , 20 mM HEPES NaOH pH 7 . 0 , 200 mM KCl , 5 mM MgCl2 , 10 mM CaCl2 , 5 mM EGTA ) . Lysates were centrifuged at 1000 g for 10 min to obtain the post-nuclear supernatant ( PNS ) , which was further centrifuged at 13 , 000 g for 30 min . To purify endosomal structures , the resulting supernatant was centrifuged at 100 , 000 g for 1 hr , and the pellet carefully solubilized in 600 μl homogenization buffer ( 250 mM sucrose , 20 mM HEPES NaOH pH 7 . 0 , 100 mM KCl , 5 mM MgCl2 , 10 mM CaCl2 , 5 mM EGTA ) . Insoluble particles were removed by short centrifugation and the supernatant loaded onto a 5–20% continuous OptiPrep gradient ( PROGEN Biotechnik ) with a total volume of 12 ml , poured according to the manufacturer’s instructions using homogenization buffer for dilution . The gradient was centrifuged at 67 , 000 g for 18 hr , 12 fractions were collected with a pipette and proteins precipitated with 15% TCA for 1 hr . Fractions were centrifuged at 13 , 000 g for 1 hr and protein pellets dissolved in SDS-sample buffer for analysis by SDS-PAGE and immunoblotting . To compare protein content of cytosolic and endosomal fractions HeLa cells ( 10 cell culture dishes Ø 15 cm ) were lysed in 10 ml of isoosmotic homogenization buffer ( 0 . 25 M sucrose , 10 mM triethanolamine , 10 mM acetic acid pH 7 . 8 , 1 mM EDTA , 10 mM β-glycero-phosphate , 0 . 2 mM Na3VO4 , complete Protease Inhibitor tablet , 1 mM NaF , 10 mM 1 , 10-Phenantroline ) and fractionated as described above . A lysate sample was removed , mixed with 4x LDS sample buffer and used as input control for western blotting . After final centrifugation at 100 , 000 g the supernatant ( 10 ml ) was transferred to a new tube and the pellet fraction was resuspended in 10 ml of homogenization buffer . Samples of both fractions were removed , mixed with 4x LDS sample buffer and analyzed via western blotting . For the analysis of SPOP and SPOPL via western blotting supernatant and pellet samples were up-concentrated . The pellet sample was resuspended in only 10% of the original volume and part of the supernatant sample was precipitated with 20% of trichloroacetic acid ( TCA ) for 30 min , centrifuged for 30 min at 13 , 000 g and resuspended in 10% of the original volume . 4x LDS sample buffer was then added to samples for western blot analysis . Ten fold more input sample was loaded for the up-concentrated samples to allow relative comparison with the other markers . Cells treated with siRNAs for 72 hr were incubated in starvation medium from the 2nd day onwards , and stimulated by adding 200 ng/ml unlabeled EGF . After the times indicated , cells were washed with icecold PBS , scraped off the dish , centrifuged for 5 min at 1200 rpm and the pellets frozen in liquid nitrogen . The pellets were then resolved in extraction buffer ( 20 mM Tris pH 7 . 5 , 100 mM NaCl , 20 mM β-glycero-phosphate , 5 mM MgCl2 , 0 . 2% NP-40 , 10% glycerol , 1 mM NaF , 0 . 5 mM DTT , complete Protease inhibitor mix tablet , 10 mM 1 , 10-Phenantroline ) , incubated 20 min on ice , and centrifuged for 10 min at 7500 rpm at 4°C to pellet bulk DNA and RNA , or treated with Nuclease for 20 min . The protein concentration was measured by Bradford and equalized , before adding 4x LDS sample buffer + 1mM DTT and boiling . EGFR degradation was analyzed by immunoblotting . SPOPL was expressed from a modified pET17 plasmid , bearing a PreScission-cleavable His-StrepII2x-SUMO N-terminal solubility tag ( kind gift of Anne Schreiber ) , in the E . coli Rossetta strain growing in autoinduction medium at 16°C over night . Cells were collected by centrifugation and the pellet resolved in lysis buffer ( 500 mM NaCl , 50 mM HEPES pH 8 . 0 , 2 mM DTT , 5% glycerol , 1 mM EDTA and complete Protease Inhibitor cocktail tablet , 1 mM PMSF , 1 μg/ml Leupeptin , 1 μg/ml Pepstatin , 1 mg/ml lysozyme , nuclease , benzamidine ) . After sonication and ultracentrifugation , the supernatant was loaded on a Streptactin Superflow column ( Qiagen ) , bound protein washed and eluted with washing buffer ( 200 mM NaCl , 50 mM HEPES pH 8 . 0 , 2 mM DTT , 5% glycerol ) containing 2 . 5 mM d-Desthiobiotin ( Sigma ) . PreScission protease was added to remove the His-StrepII2x-SUMO-tag over night at 4°C . The eluate was cleaned over a GST- and His- column ( GE Healthcare ) to separate SPOPL from the cleaved tag and the GST-tagged PreScission . The eluate was concentrated to 1 ml using Amicon concentrator tube with a 30 kD cut off , and fractionated on a Superose 6 size exclusion column using an Äkta Pure system ( GE Healthcare ) . SPOPL elution fractions with correct molecular weight were pooled , frozen in liquid nitrogen and stored at -80°C . EPS15 was expressed from a pGEX-6P1 plasmid bearing a GST-tag in the E . coli Rossetta strain growing in autoinduction medium supplemented with 4 μg/ml CaCl2 over night at 16°C . Cells were collected by centrifugation and pellets resolved in lysis buffer as described above . After sonication and ultracentrifugation , the supernatant was loaded on a GST-column ( GE Healthcare ) , bound protein washed and eluted with washing buffer ( 200 mM NaCl , 50 mM HEPES pH 8 . 0 , 2 mM DTT , 5% glycerol ) containing 20 mM reduced glutathione . PreScission protease was added to remove the GST-tag over night at 4°C , and cleaved GST was separated together with GST-tagged PreScission using a GST HiTrap column ( GE Healthcare ) . The eluate was concentrated using an Amicon concentrator with a 100 kD cut off , and fractionated on a Superose 6 size exclusion column equilibrated in buffer ( 150 mM NaCl , 20 mM HEPES pH 7 . 8 , 2 mM DTT , 2% glycerol ) . EPS15 fractions with correct molecular weight were pooled , frozen in liquid nitrogen and stored at -80°C . For binding assays , glutathione sepharose 4B ( GE Healthcare ) aliquots were equilibrated in binding buffer ( 150 mM NaCl , 20 mM HEPES pH 7 . 8 , 2 mM DTT , 2% glycerol ) and incubated with equimolar amounts of GST-EPS15 and untagged SPOPL and incubated for 15 min rotating at RT . Beads were washed 4x at RT with binding buffer , and bound proteins eluted with either 20 mM glutathione in binding buffer or 100 mM Glycine pH 2 . 0 at RT . Eluates were immediately mixed with 4x LDS and DTT , boiled and analyzed by Coomassie Staining or immunoblotting . Cullin-3 and RBX1 used for in vitro ubiquitination reactions were purified together as a complex from insect cells , and neddylated in vitro using purified components as described previously ( Enchev et al . , 2012; Orthwein et al . , 2015 ) . 0 . 5 μM CUL3-NEDD8-RBX1 , 0 . 6 μM SPOPL , 0 . 2 μM UbE1 ( Boston Biochem ) , 0 . 7 μM E2 ( CDC34 or UBCH5B , produced as described in Enchev et al . , 2012 ) , 50 μM Ubiquitin ( Boston Biochem ) and 2 μM EPS15 were incubated in the presence of ATP for various times at 37°C . Reactions were stopped by the addition of 4x LDS with 1 mM DTT , boiled and analyzed by SDS-PAGE and immunoblotting . HeLa cells treated for 72 hr with siControl oligo’s or siRNA depleting SPOPL ( 20 confluent 15 cm dishes per condition ) were scraped in their media and centrifuged at 4°C for 3 min at 300 rcf ( g ) . Cell pellets were washed with ice-cold PBS and snap frozen in liquid nitrogen prior to lysis and digestion . After quickly thawing , pellets were resuspended in Urea lysis buffer ( ULB ) containing 9 M urea , 50 mM Ammonium Bicarbonate , 1 mM sodium orthovanadate , 2 . 5 mM sodium pyrophosphate , and 1 mM β-glycerophosphate such that the final protein concentration would be less than 5 mg/ml . A Branson 250 tip sonicator was used for lysis ( power output of 15 , duty cycle of 70% , 3 rounds sonication , cooling on ice between cycles ) and the lysates were cleared by centrifugation at 20 , 000 rcf ( g ) for 15 min at 15°C . Reduction with 10 mM tris ( 2-carboxyethyl ) phosphine ( TCEP ) was performed at RT for 30 min followed by alkylation with 20 mM iodoacetamide with the pH maintained at 7 . 5 and incubation for 30 min at RT protected from light . Samples were then diluted to 4 M urea with 100 mM ammonium bicarbonate buffer ( pH 8 ) and digested with LysC ( 1:100 ) in a shaking incubator at 37°C for 4 hr . Finally , each digest was diluted to a final concentration of 1 M urea with 100 mM ammonium bicarbonate and sequencing grade trypsin was added at 1:100 dilution for overnight digestion at 37°C in a shaking incubator . The next morning , the trypsin reaction was stopped by formic acid ( FA ) addition to a final concentration of 1% ( pH <3 ) and samples were placed on ice for 15 min for precipitation . Lysates were centrifuged at 2000 rcf ( g ) for 15 min at RT , desalted on c18 columns with 20 mg capacity and bound peptides were washed sequentially with 1 ml , 5 ml , and 6 ml of 0 . 1% FA in dH2O followed by 2 ml of 2% acetonitrile ( ACN ) in 0 . 1% FA in dH2O . Finally , peptides were eluted with 3 x 3 ml 50% ACN in 0 . 1% FA in dH2O and lyophilized for 48 hr to remove all of the FA . Immunoprecipitation of K-ε-GG modified peptides was carried out with an antibody-based system essentially as described in the Cell Signaling Technology , Inc . standard protocol for the PTMScan UbiScan kit . Eluted K-ε-GG peptides were desalted on self-packed StageTips as follows: StageTips were equilibrated with 50 μl 50% ACN in 0 . 1% trifluoroacetic acid ( TFA ) in dH2O ( 1x ) followed by 50 μl dH2O with 0 . 1% TFA ( 2x ) . Samples were loaded in two steps loading 50 μl each time and passing the solution through with a centrifuge at 2000 rcf ( g ) for 30 s . The bound peptides were washed with 55 μl dH2O with 0 . 1% TFA ( 2x ) and eluted with 10 μl 50% acetonitrile ( ACN ) in 0 . 1% TFA in dH2O followed by a second elution pooled with the first of 20 μl 50% acetonitrile ( ACN ) in 0 . 1% TFA in dH2O . Solvents were evaporated in a thermal vacuum centrifuge to ~1 μl and the final peptide product was resuspended in 10 μl 2% ACN in 0 . 1% TFA in dH2O for analysis by LC-MS/MS on a Thermo Fisher Scientific Q-Exactive Plus in data-dependent analysis ( DDA ) mode . For this , peptides were separated on an EASY-nLC 1000 ( Thermo Fisher Scientific ) coupled to a 15 cm fused silica emitter ( 75 μm inside diameter ) . The analytical column was packed with ReproSil-Pur C18-AQ 120 Å and 1 . 9 μm resin ( Dr . Maisch HPLC GmbH ) and peptides were separated with a 120 min gradient that went from 2%–35% acetonitrile in water with 0 . 1% formic acid . A top 12 method was used for DDA acquisition within a mass range of 300–1700 m/z using higher energy collisional dissociation ( HCD ) for peptide fragmentation resulting in high resolution MS/MS data for analysis . Raw files were converted to mzXML using msconvert and searched with Comet as part of the TransProteomicPipeline version 4 . 7 . 0 . Peptides were filtered for a 1% false discovery rate based on their peptide prophet scores and Skyline version 2 . 6 . 0 was used to integrate the area under the MS1 chromatography peaks to generate label-free quantification information . To do this , chromatographic traces were aligned for each quantified peak and integration areas were refined manually for multiple peptides of interest . For this , chromatographic traces were aligned and peak areas selected manually for the peptides of interest . In this study , three modified EPS15 peptides were analyzed with the following sequences: Lys 693 ( K693 ) IDPFGGDPFK*GSDPFASDCFFR , Lys 793 ( K793 ) RSINK*LDSPDPFK , and Lys 801 ( K801 ) LDSPDPFK*LNDPFQPFPGNDSPK . As controls , a peptide containing modified Lys 113 ( K113 ) of β Actin ( VAPEEHPVLLTEAPLNPK*ANR ) and the polyubiquitin linkage peptide corresponding to Lys 11 ( K11 ) of ubiquitin ( TLTGK*TITLEVEPSDTIENVK ) were quantified ( K* = K-ε-GG modification; C = carbamidomethyl-modified cysteine ) . The results reported here represent biological triplicate experiments of RNAi depletion of SPOPL compared to siControl performed on separate days . Each data point within each biological replicate represents the results from technical replicate LC-MS/MS injections . Normalization for MS intensity variations between runs was performed by setting the value of the siControl for each individual biological replicate pair to 1 and scaling the RNAi depletion condition to this set value . Final reported results are the mean ± SD . The following antibodies were used in this study: anti-Actin ( Millipore , MAB1501R , 1:1000 in WB ) , anti-AP2 ( adaptin alpha ) / ( BD Transduction , 610502 , 1:1000 in WB ) , anti-CUL3 ( described in ( Sumara et al . , 2007 ) , 1:1000 in WB ) , anti-Clathrin ( BD Biosciences , 610499 , 1:3000 in WB ) , anti-EEA1 ( BD Transduction 610457 , 1:2500 in WB , 1:200 in IF ) , anti-EGFR ( Millipore , 06–847 , 1:1000 for WB , 1:500 in IF; or Biolegend , 352901 , 1:200 in IF ) , anti-EPS15 mouse ( BD Transduction , 610807 , 1:250 in WB , 1:50 in IF ) , anti-EPS15 rabbit ( Santa Cruz , sc-534 , 1:200 in WB , 1:50 in IF ) , anti-GAPDH ( Sigma G8795 , 1:5000 in WB ) , anti-GFP ( Roche , 11 814 460 001 , 1:1000 in WB ) , anti-H4 ( Abcam , ab16483 , 1:1000 in WB ) , anti-HA . 11 ( Covance , MMS-101R , mouse , 1:1000 in WB and IF ) , anti-HRS ( Sigma , WH0009146M1 , 1:1000 in WB , 1:500 in IF ) , anti-LAMP1 ( Santa Cruz , sc-20011 ( mouse ) , 1:1000 in WB , 1:200 in IF or Abcam ( rabbit ) ab24170 , 1:1000 in WB ) , anti-RAB11 ( Sigma , R5903 , 1:200 in IF ) , anti-SPOPL ( rabbit polyclonal antibody raised against the N-terminal SPOPL peptide MSREPTPPLPGDMST+C and a SPOPL middle region peptide CKDGKNWNSNQATDIM and affinity-purified with the recombinantly expressed and purified SPOPL protein , 1:500 in WB ) , anti-SPOP ( Abcam , ab81163 , 1:500 in WB ) , anti-STAM ( Santa Cruz , SC-133093 , 1:500 in WB ) , anti-Tubulin ( Sigma , T5168 , 1:10000 in WB ) , anti-mouse IgG ( H+L ) CF405S ( Biotum , BI-20080 , 1:1000 in IF ) , anti-rabbit IgG ( H+L ) Alexa Fluor568 ( Thermo Fisher Scientific , A-11036 , 1:1000 in IF ) and anti-K-ε-GG antibody ( Cell Signaling Technology , cat . no . 5562 ) . Furthermore , following antibodies were used in Figure 2E: anti-Calnexin ( kindly provided by Ari Helenius ) , anti-Caveolin 1 ( Santa Cruz , sc-894 , 1:500 in WB ) , anti-CHMP6 ( Sigma , SAB2701297 , 1:1000 in WB ) , anti-EPS15R1 ( Novus Biologicals , NB100-88149 , 1:500 in WB ) , anti-EPSIN1 ( Santa Cruz , sc-8673 , 1:500 in WB ) , anti-HER2 ( Cell Signaling , 2165 , 1:500 in WB ) , anti-IGF1R ( R&D Systems , MAB301 , 1:1000 in WB ) , anti-LC3 ( Novus Biologicals , NB600-1384 , 1:1000 in WB ) , anti-MET ( Santa Cruz , sc-10 , 1:1000 in WB ) , anti-RAB7 ( Cell Signaling , 9367 , 1:1000 in WB ) , anti-SQSTM1 ( American Research Products Inc . , 03-GP62-C , 1:1000 in WB ) , anti-TSG101 ( Novus Biologicals , NB200-112 , 1:500 in WB ) , anti-VEGFR3 ( Millipore , MAB3757 , 1:500 in WB ) The following chemicals were used in this study: MG132 ( Sigma , 1 – 5 μM final conc . ) , Bafilomycin A ( BafA , Sigma , 50 nM final conc . ) , Cyclohexamide ( CHX , Sigma , 10 μM – 1 mM final conc . dependent on assay ) , MLN-4924 ( Active Biochem , 10 μM final conc . ) and Chloroquine ( CQ , Sigma , C6628 , 20 μM final conc . ) . All siRNAs were ordered from Qiagen , and their Allstar-negative and Allstar-death siRNAs were used for controls . The following siRNAs were used to specifically deplete the indicated proteins: NameTarget sequencesiCUL3AACAACTTTCTTCAAACGCTAsiSPOPL_1CAGTTTGGCATTCCACGCAAAsiSPOPL_2GGCCTTAAATTATCTTCAATTsiSPOPL_3GGTGCCTGAGTGTCGTCTATTsiSPOP 1TCAGTTTATCATTTGCTCCsiSPOP_2GGCTCACAAGGCTATCTTATTsiSPOP_3GGAGGAAAUGGGUGAAGUCAUsiEPS15AAACGGAGCTACAGATTATsivATPaseCACGGTTAATGAAGTCTGCTA Data are represented as the means of at least triplicate experiments + standard deviation ( SD ) , except for data in Figure 4D , 5A , E and F , as well as Figure 5—figure supplement 1B and C , for which the results are shown as means + standard error of the mean ( SEM ) . ‘N’ represents the number of replicates , and ‘n’ the number of measured cells . One-tailed Student’s t-tests with unequal variance assumption were performed to compare the datasets for statistical significance when appropriate . | Individual cells can move material , collectively referred to as cargo , from the outside environment into the cell interior via a process known as endocytosis . The cell then has different routes to transport the packages of cargo , called endocytic vesicles , to specific locations within the cell . Protein-based molecular machines move the cargo and control how it is selected and targeted to different destinations . For example , a molecular machine that contains a protein called CUL3 labels other components of the system with a chemical tag to regulate the route cargo takes in mammalian cells . However , it was not clear how CUL3 can selectively attach the chemical labels . Gschweitl , Ulbricht et al . have now found that another protein called SPOPL provides selectivity for the CUL3-based machine during endocytosis in human cells . The experiments show that SPOPL attaches to endocytic vesicles , and that CUL3 and SPOPL work together to label a specific component of these vesicles called EPS15 . The label changes how EPS15 interacts with other proteins . When SPOPL is not present in a cell , EPS15 is unnaturally stable and occupies many of the routes used by endocytic cargos . The cargo directly interacting with EPS15 is then routed on the fast lane to its destination , while other cargo accumulate in a kind of molecular traffic jam . Other proteins like SPOPL are specific for the endocytic system . Exchange of SPOPL with these similar proteins in the CUL3 machine is likely to chemically label a different set of endocytic proteins . Gschweitl , Ulbricht et al . ’s next challenge is to identify the selectivity , targeting and coordination of these exchangeable components in the endocytic system . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"biochemistry",
"and",
"chemical",
"biology",
"cell",
"biology"
] | 2016 | A SPOPL/Cullin-3 ubiquitin ligase complex regulates endocytic trafficking by targeting EPS15 at endosomes |
A large fraction of human cancers contain genetic alterations within the Mitogen Activated Protein Kinase ( MAPK ) signaling network that promote unpredictable phenotypes . Previous studies have shown that the temporal patterns of MAPK activity ( i . e . signaling dynamics ) differentially regulate cell behavior . However , the role of signaling dynamics in mediating the effects of cancer driving mutations has not been systematically explored . Here , we show that oncogene expression leads to either pulsatile or sustained ERK activity that correlate with opposing cellular behaviors ( i . e . proliferation vs . cell cycle arrest , respectively ) . Moreover , sustained–but not pulsatile–ERK activity triggers ERK activity waves in unperturbed neighboring cells that depend on the membrane metalloprotease ADAM17 and EGFR activity . Interestingly , the ADAM17-EGFR signaling axis coordinates neighboring cell migration toward oncogenic cells and is required for oncogenic cell extrusion . Overall , our data suggests that the temporal patterns of MAPK activity differentially regulate cell autonomous and non-cell autonomous effects of oncogene expression .
The Receptor-Tyrosine Kinase ( RTK ) /RAS/ERK signaling axis ( Figure 1A ) is mutated in most human cancers ( Sanchez-Vega et al . , 2018 ) . In normal conditions , the ERK pathway promotes proliferation , differentiation , survival and cell migration ( Johnson and Lapadat , 2002 ) . During oncogenesis , mutations or amplification of ERK pathway components can also promote oncogene-induced senescence ( Hahn and Weinberg , 2002 ) ( OIS ) or oncogenic cell extrusion from epithelial monolayers in the so-called Epithelial Defense Against Cancer response ( EDAC ) ( Hogan et al . , 2009; Kajita et al . , 2010 ) . The mechanisms underlying dose-dependent effects of ERK signaling have been intensely studied using bulk cell population assays . However , the advent of single-cell analysis has shown that single cells often behave qualitatively different than bulk populations . In fact , in vivo and in vitro studies have now shown that pulsatile or sustained ERK activity have different effects on cell behavior ( Albeck et al . , 2013; Aoki et al . , 2013; de la Cova et al . , 2017; Johnson and Toettcher , 2019; Santos et al . , 2007; Bugaj et al . , 2018; Aoki et al . , 2017 ) . Whether different oncogenic perturbations also have different functional outcomes depending on downstream signaling dynamics remains unknown . To address this question , an isogenic single-cell approach with temporal control of oncogene expression is needed . Recent in vivo studies revealed that oncogene expression can trigger tissue level responses involving normal neighboring cells ( Brown et al . , 2017; Ellis et al . , 2019; Clavería et al . , 2013; Sancho et al . , 2013 ) . In specific cases , mosaic oncogene expression leads to either basal extrusion or apical extrusion ( Hogan et al . , 2009; Kajita et al . , 2010 ) ; however , the signaling mechanism responsible for recognition between normal and diseased cells is poorly understood ( Kajita and Fujita , 2015; Clavería and Torres , 2016; Maruyama and Fujita , 2017 ) . Coincidentally , propagating ERK signaling waves requiring the sheddase ADAM17 have been observed in mouse epidermis and intestinal organoids , but the physiological role of these signaling events remains unclear ( Hiratsuka et al . , 2015; Muta et al . , 2018 ) . Observation of interactions between oncogenic and neighboring epithelium with live-cell biosensors could provide insights into the collective signaling preceding oncogenic extrusion . In fact , a recent study using live imaging of calcium biosensors during EDAC of HRASG12V cells showed a calcium signaling wave which propagated through neighboring epithelium to coordinate actin rearrangements and polarized movements during apical extrusion ( Takeuchi et al . , 2020 ) . The mechanistic basis underlying EDAC calcium waves remains unknown . Here , we combine live cell imaging of MAPK activity biosensors with inducible expression of oncogenes to study the effects of oncogene expression on signaling dynamics and how altered MAPK dynamics impact both cell autonomous and non-cell autonomous behaviors in epithelial tissues . Our data shows that pulsatile or sustained ERK signaling resulting from oncogenic perturbations triggers different dynamics-dependent cell behaviors including oncogene-induced paracrine signaling via the ADAM17-AREG-EGFR signaling axis . The resulting signaling gradients are required to coordinate neighboring cell migration and active oncogenic cell extrusion ( EDAC ) . Our study highlights the role of MAPK signaling dynamics in coordinating individual and collective cell behaviors .
To study the effects of oncogene expression on the temporal patterns of MAPK signaling we generated a reporter cell line derived from the chromosomally-normal human breast epithelial line , MCF10A , expressing the ERK Kinase Translocation Reporter ( Regot et al . , 2014 ) ( ERK KTR ) and a fluorescently tagged ERK kinase ( ERK-mRuby2 ) . This combination of biosensors allowed independent measurement of ERK activity and ERK localization in live single cells at high temporal resolution . Then , we introduced 12 different doxycycline-inducible oncogenic perturbations via lentiviral infection and measured ERK signaling dynamics during overexpression ( Figure 1B ) . Our results revealed two qualitatively different responses to oncogene induction: ( i ) increased frequency of ERK activity pulses with no change in ERK kinase localization ( i . e . EGFR , B-RafWT ) , and ( ii ) sustained ERK activity with subsequent nuclear translocation of ERK kinase ( i . e . B-RafV600E , MEK2DD ) ( Figure 1 , and Figure 1—Video 1 ) . We refer to these distinct dynamics as pulsatile or sustained ERK , respectively . Of note , MEK1/2WT expression is capable of exporting ERK into the cytoplasm without changing kinase activity ( Adachi et al . , 2000; Figure 1—figure supplement 1 ) and ERK nuclear accumulation occurs only when activity is sustained , suggesting that ERK activity and ERK localization are not always correlated . Interestingly , expression of B-RafWT or B-RafV600E elicit qualitatively different downstream dynamics even though they differ in a single amino acid and show similar expression levels by immunoblotting ( Figure 1—figure supplement 2 ) . Given that the B-RafV600E is insensitive to negative feedback regulation by ERK ( Yao et al . , 2015 ) , this result suggests that ERK inhibition to B-RafWT is mechanistically involved in the characteristic pulsatile dynamics . Next , we assessed how ERK dynamics affect cell behaviors by measuring cell migration and proliferation . While pulsatile ERK dynamics ( i . e . EGFR or B-RafWT ) consistently correlated with increased cell cycle progression , sustained ERK activity ( i . e . B-RafV600E or MEK2DD ) caused cell cycle arrest and increased migration ( Figure 1D ) . Importantly , observed differences in cell behavior correlated with dynamics independently of the point in the cascade that perturbations were introduced ( EGFR , Raf or MEK ) , suggesting that ERK is responsible for differences in cell behaviors rather than alternate downstream pathways . Moreover , expression of B-RafWT or B-RafV600E , which activate the cascade at the same point , caused different ERK activity dynamics ( i . e . pulsatile or sustained respectively ) and triggered opposing cellular behaviors ( Figure 1D ) . Taken together , these data suggest that ERK activity dynamics can either promote or inhibit proliferation cell autonomously . The sudden increase in migration and the loss of cell-cell contacts observed in cases where ERK activity is sustained ( B-RafV600E and MEK2DD , Figure 1—Video 1 ) are reminiscent of phenotypes described for cells undergoing Epithelial-to-Mesenchymal Transition , or EMT ( Hao et al . , 2019 ) . We sought out to address the role of EMT in oncogene-dependent cell behaviors by immunofluorescent staining of an epithelial marker E-Cadherin ( E-Cad ) and the mesenchymal marker N-Cadherin ( N-Cad ) . While cell migration was clearly increased at 24 hr post-oncogene expression , cells retain E-Cad expression with no clear increase in N-Cad , as was observed in TGFβ-induced EMT ( Figure 1—figure supplement 3 ) . These results indicate that at the time points studied here , altered cell behaviors are either distinct from or precede those resulting from EMT . To examine the non-cell autonomous effects of oncogene expression in epithelial monolayers , we cocultured ‘inducible’ cells ( expressing doxycycline-inducible oncogenes , a constitutively expressed H2B-mClover , and the ERK biosensors ) with ‘neighboring’ reporter cells ( expressing ERK biosensors without inducible oncogenes ) and monitored signaling dynamics upon induction ( Figure 2A ) . Interestingly , expression of B-RafV600E , but not B-RafWT , resulted in waves of ERK activation of neighboring cells ( Figure 2B–C and Figure 2—Video 1 ) . This comparison suggests that oncogenic perturbations that elicit sustained ERK activity propagate ERK activity pulses to neighboring cells . In agreement , other oncogenes that triggered sustained , but not pulsatile , ERK activity also promoted ERK activity waves in the neighboring cells ( Figure 2—figure supplement 1 ) . By using KTRs for p38 and JNK , we observed that neighboring epithelia did not activate other the MAPK pathways ( Figure 2—figure supplement 1 ) . Notably , spontaneous cell death events were also followed by similar ERK signaling waves ( Figure 2—Video 2 ) , indicating that oncogene expression and cell death may be similarly perceived by neighboring cells . We then addressed the mechanistic basis of oncogene-dependent paracrine signaling . Previous studies demonstrated that ERK waves in epithelial monolayers depend on the membrane-tethered sheddase ADAM17 , which releases membrane-bound growth factors that activate EGFR signaling in adjacent cells ( Aoki et al . , 2013; Aoki et al . , 2017; Hiratsuka et al . , 2015 ) . Thus , we hypothesized that oncogenic cell ADAM17 may be decoding ERK signaling dynamics to trigger growth factor release . To test this hypothesis , we generated an ADAM17 knockout ( ADAM17KO ) cell line ( Figure 3A ) and used it as either ‘inducible’ or ‘neighboring’ cells in our coculture assay . Live imaging of WT and ADAM17KO cocultures indicated that ADAM17 is necessary in inducible , but not neighboring cells , to trigger ERK waves in the monolayer ( Figure 3B–C and Figure 2—Video 1 ) . Therefore , ADAM17 decodes ERK activity dynamics in inducible cells to transmit ERK signaling to neighboring cells . Previous work has shown that ADAM17 is weakly phosphorylated compared to other ERK substrates ( Díaz-Rodríguez et al . , 2002 ) , thus the phosphorylation-dephosphorylation kinetics of ADAM17 and the temporal patterns of ERK activity may explain dynamics-specific ADAM17 activation . ADAM17-released growth factors include HB-EGF , TGF-α , Epiregulin , and Amphiregulin ( Zunke and Rose-John , 2017; Rios-Doria et al . , 2015 ) . In order to identify the factors mediating oncogene-induced paracrine signaling we used Tandem-Mass-Tag Mass Spectrometry of supernatant proteins following induction of sustained ERK activity in WT and ADAM17KO cells . A variety of known and unknown ADAM17 substrates were present in the induced cell supernatants , including immune surveillance ( HLA-A/B/C ) , Delta-Notch ( JAG1 ) , and Wnt ( SFRP ) signaling proteins ( Figure 3D and Supplementary file 1 ) . Of note , the EGFR ligand Amphiregulin ( AREG ) was the most upregulated , ADAM17-dependent protein in the supernatant , suggesting that AREG released from inducible cells could act as an oncogene-dependent paracrine signaling molecule . Accordingly , cocultures pre-incubated with AREG function-blocking antibodies or EGFR inhibitors prevented neighboring cell ERK activation without affecting ERK signaling in inducible cells ( Figure 3E–F ) . These results indicate that oncogene-dependent ERK waves are mediated by ADAM17 ( in inducible cells ) , AREG release , and EGFR signaling ( Figure 3G ) . Given that cells surrounding B-RafV600E expressing cells showed pulsatile ERK activity ( Figure 2C ) , we hypothesized that oncogene expression may promote cell proliferation in a non-cell autonomous manner . Accordingly , sustained ERK signaling in inducible cells increased proliferation of neighboring cells up to 10-fold ( Figure 4A–C ) . Together , these data indicate that , depending on ERK dynamics , oncogenic cells can have either cell autonomous or non-cell autonomous contributions to tissue growth . In addition to proliferation , ERK waves have been shown to orient collective cell migration during wound healing ( Aoki et al . , 2017 ) . In cocultures , sustained ERK activity in B-RafV600E-inducible cells correlated with neighboring cell migration towards inducible cells in an ADAM17 and EGFR-dependent manner ( Figure 5A ) . We hypothesized that coordinated migration of neighboring cells could physically contribute to oncogenic cell extrusion ( Hogan et al . , 2009; Leung and Brugge , 2012; Slattum et al . , 2014 ) . To address this hypothesis , we used confocal Z stacks to quantify extrusion of oncogene-expressing cells from monolayers ( Figure 5B and Figure 5—Video 1 ) . Interestingly , while pulsatile ERK activity ( i . e . EGFR and B-Raf ) was not sufficient to extrude cells , sustained ERK activity ( i . e . B-RafV600E and MEK2DD ) led to efficient epithelial cell extrusion apically ( Figure 5C ) . KRASG12V induction did not result in apical extrusion to the extent observed for HRASG12V ( Hogan et al . , 2009 , Figure 5—figure supplement 1 ) . However , since sustained ERK activation in KRASG12V occurs later than B-RafV600E ( Figure 1—figure supplement 1 ) further apical extrusion may also occur at a later time . Taken together , our data suggests that apical extrusion occurs when oncogenic perturbations trigger sustained ERK activity . In mammalian epithelia , apical extrusion eliminates apoptotic cells or crowded cells to maintain homeostasis ( Rosenblatt et al . , 2001; Eisenhoffer et al . , 2012 ) . Similar , but mechanistically different , apical extrusion has been observed for some oncogenic cells during EDAC ( Kajita and Fujita , 2015 ) . We wanted to know whether the extrusion of inducible cells with sustained ERK activity resembled EDAC . To differentiate between pseudostratified or de-laminated ( Grieve and Rabouille , 2014 ) epithelium and extrusion we analyzed confocal images of E-cad membrane staining in induced cocultures . Cells with sustained ERK activity were fully extruded , sitting above WT cells in the plane of the monolayer ( Figure 5—figure supplement 2 ) . These images also demonstrate maintenance of E-Cad at the junctions between WT cells below extruded cells . Quantification of fully-extruded cells at several timepoints showed the majority of oncogenic cells being extruded from 4 to 8 hr after induction , but continuing until 24 hr , when 91% of inducible cells are fully extruded . Both the oncogenic and apoptotic extrusion models involve cytoskeletal rearrangements at the site of extrusion ( Rosenblatt et al . , 2001; Kajita and Fujita , 2015 ) . To observe live actin dynamics in cocultures , we made cell lines stably expressing Utrophin-261-EGFP ( Belin et al . , 2014 ) . Using this tool , we observed transient accumulation of actin at the basal interface of B-RafV600E expressing and neighboring cells that first closed off the basal attachments of inducible cells before they were pushed apically out of the monolayer ( Figure 5—figure supplement 2 and Figure 5—Video 2 ) . These polarized , actin-containing basal protrusions were dependent on EGFR activity as they could be inhibited by EGFR inhibitor . Apoptotic extrusion relies on Sphingosine-1-phosphate ( S1P ) signaling through intrinsic S1P production and juxtracrine activation of the GPCR S1PR2 ( Gu et al . , 2011 ) , yet inhibition of S1P production had only moderate effects on apical extrusion of MEK2DD cells ( Figure 5—figure supplement 2 ) . Together , our results suggest that the apical extrusion of oncogenic cells observed in our experiments are similar to the EDAC mechanism previously described for HRASG12V , V-Src , and other cells ( Hogan et al . , 2009; Kajita et al . , 2010; Kajita and Fujita , 2015 ) . The requirement for paracrine signals in collective migration led to the question of whether paracrine signals were also required for extrusion . To test the role of ADAM17-mediated AREG-EGFR paracrine signals in promoting extrusion , we performed extrusion assays using ADAM17KO cells or in the presence of EGFR inhibitor or AREG function-blocking antibodies . Extrusion of inducible cells was abolished in these conditions ( Figure 5D–G ) , suggesting that ERK signaling waves are required for extrusion . Of note , since ADAM17KO and EGFR inhibition affect ERK activation of neighboring cells without altering ERK dynamics in inducible cells , we hypothesized that that activation of inducible cells alone is not sufficient for extrusion , but that neighboring cell ERK activation may be required . To address this question , ERK-independent ADAM17 activation is needed . Previous studies have shown that the stress MAPK p38 phosphorylates and activates ADAM17 ( Xu and Derynck , 2010 ) . Thus , we used our doxycycline-inducible system to drive the expression of MKK3DD , a constitutively-active MAP2K specific for p38 ( Figure 6—figure supplement 1 ) , to activate ADAM17 independently of ERK . As expected , we found that p38 activation leads to ERK signaling waves ( Figure 6A–B ) , proliferation ( Xu and Derynck , 2010; Figure 6—figure supplement 2 ) , directed migration ( Figure 6C ) and extrusion in an ADAM17 and EGFR dependent manner ( Figure 6D and Figure 6—Video 1 ) . However , B-RafV600E oncogenic signaling , extrusion and proliferation were unaffected by p38 inhibition ( Figure 6—figure supplement 3 ) , suggesting that sustained ERK or p38 activity are each capable of activating ADAM17 paracrine signaling . Using this ERK-independent MKK3DD extrusion system , we found that MEK inhibition decreased directed migration and prevented extrusion , confirming that extrusion requires ERK activity in the neighboring cells ( Figure 6C–D ) . Thus , ERK activity is required for extrusion in both oncogenic and neighboring cells with qualitatively different temporal dynamics . Notably , inhibition of S1P signaling in MKK3DD cocultures also reduced extrusion efficiency despite having unaffected signaling ( Figure 6—figure supplement 4 ) . This result suggests that cell-autonomous ERK or p38 activation in extruded cells may underlie the fundamental differences reported between extrusion of oncogenic and stressed cells . Finally , we asked whether the spatiotemporal properties of paracrine ERK signaling waves are important to coordinate extrusion . We first tested the efficiency of extrusion with altered proportions of B-RafV600E cells in the coculture , as higher proportions will have de-centralized and overlapping signaling events . The proportion of inducible cells was inversely correlated with extrusion efficiency ( Figure 7A ) . Moreover , exogenous addition of AREG , which triggers widespread ERK activation preventing any spatially defined waves , eliminated directed migration of neighboring cells and extrusion ( Figure 7B–C ) . The observation that in cocultures , polarized actin enrichment in neighboring cell basal protrusions is absent with EGFR inhibition , also indicates that growth factor signaling provides directional information during extrusion ( Figure 5—figure supplement 2 ) . Together this data suggests that locally generated paracrine signaling coordinates directed migration of neighboring epithelia to promote extrusion of oncogenic cells ( Figure 8 ) .
A wide variety of ERK pathway alterations occur across human tumors , often resulting in different cancer phenotypes ( Sanchez-Vega et al . , 2018; Bailey et al . , 2018; Hoadley et al . , 2018 ) . To better understand the signaling effects of oncogenic mutations , we used live cell imaging of signaling biosensors upon induction of oncogenes and found that different ERK pathway oncogenes trigger two distinct temporal patterns: pulsatile or sustained ERK activity . While our approach is admittedly different than acquisition of point mutations in vivo , ERK dynamics resulting from oncogene overexpression robustly correlated with the same cellular phenotypes: ( i ) pulsatile ERK activity correlates with increased proliferation and , ( ii ) sustained ERK activity leads to cell cycle arrest similar to OIS ( Hahn and Weinberg , 2002; Courtois-Cox et al . , 2006; Kuilman et al . , 2008 ) . Moreover , we showed that sustained ERK activity in oncogenic cells triggers ERK signaling waves through unperturbed neighboring cells . These signaling waves depend on the ADAM17-EGFR paracrine signaling axis and lead to different non cell-autonomous behaviors such as ( i ) proliferation , ( ii ) directed migration of neighboring cells toward oncogenic cells , and ( iii ) oncogenic cell extrusion ( Figure 8 ) . Our data indicates that cancer mutations can have non-cell autonomous contributions to tissue growth ( Figure 4 ) . Interestingly , studies in mouse epidermis have shown that mosaic oncogene expression promotes proliferation of wild type surrounding cells , which is required to expel mutant outgrowths from the tissue ( Brown et al . , 2017 ) . While the role of ADAM17 in this phenomenon and during early tumorigenesis in vivo is yet unknown , it is tempting to speculate that polypous outgrowths may occur in the presence of non-proliferative oncogenic cells that release growth factors via ADAM17 . The mechanisms that mediate tissue expelling in vivo remain unknown . Previous work in described the process of oncogenic cell extrusion as part of the so called Epithelial Defense Against Cancer ( i . e . EDAC ) ( Hogan et al . , 2009 ) . However , the signals involved in recognition of oncogenic cells , and why only specific oncogenes trigger oncogenic cell extrusion was unclear ( Kajita and Fujita , 2015; Clavería and Torres , 2016; Maruyama and Fujita , 2017 ) . Our data suggests that perturbations that elicit sustained ERK activity ( eg . B-RafV600E , MEK2DD ) , activate ADAM17 , which in turn releases EGFR ligands ( Figure 3 ) . This paracrine signal is critical for oncogenic extrusion ( Figure 5 ) . However , we acknowledge that different cellular states such as apoptosis or overcrowding lead to extrusion by different mechanisms . Of note , our data showed that ERK activation drives extrusion to a higher extent than p38 activation ( Figure 6 ) , which may result from a difference in overall cell autonomous migration in these two cases ( Figure 6—figure supplement 1 ) . Moreover , sphingosine kinase inhibition caused greater defects in extrusion of p38-active cells than ERK-active cells ( Figure 5—figure supplement 2 and Figure 6—figure supplement 4 ) . This finding agrees with work showing that EDAC of transformed HRASG12V cells is less dependent on sphingosine-1-phosphate production than extrusion of crowded or apoptotic cells ( Yamamoto et al . , 2016 ) , where stress signaling may be involved . We and others have identified AREG as one of the key EGFR ligands in mammary epithelial cells ( Sternlicht et al . , 2005; Figure 3 ) ; however , different ligands may be required in other tissues . These ligands , released by ADAM17 , coordinate the migration of neighboring cells by mechanisms that remain unclear . Cultured monolayers are fundamentally different than in vivo tissues; however , the chemo-attractive properties of growth factors for directed migration have been modeled and studied in cell culture ( Devreotes et al . , 2017; Tranquillo et al . , 1988 ) . We propose that local signaling gradients are created by oncogenic cells to coordinate directed migration of neighbors . To support this idea , we show that addition of exogenous AREG or increased fractions of oncogenic cells both prevent directed migration of neighboring cells ( Figure 7 ) , and that during extrusion , polarized actin-containing basal protrusions require growth factor signaling ( Figure 5—figure supplement 2 ) . Localized ERK signaling gradients have also been observed during morphogenesis of Drosophila , avian , and mammalian embryos ( Yang et al . , 2002; Ogura et al . , 2018; Corson et al . , 2003 ) , and in preserving homeostasis of mammalian epidermis and intestinal organoids ( Hiratsuka et al . , 2015; Muta et al . , 2018; Liang et al . , 2017 ) . Thus , in addition to roles in oncogenesis , the ADAM17-EGFR paracrine signaling axis may direct collective behaviors during development . Overall , our results highlight the importance of quantitative live-cell approaches to understand the effects of genetic perturbations and cell-cell communication in tissues . We propose a critical role for ERK signaling dynamics and the ADAM17-EGFR signaling axis in coordinating cell behaviors at the tissue level .
MCF10A human mammary epithelial cells ( ATCC ) were grown at 37° and 5% CO2 in DMEM/F12 ( Gibco ) with 5% horse serum ( HS ) ( Sigma ) , 10 µg/ml Insulin ( Sigma ) , 20 ng/ml EGF ( Peprotech ) , 1x Penicillin-Streptomycin ( P/S ) ( Gibco ) , 0 . 5 mg/ml Hydrocortisone ( Sigma ) , 100 ng/ml Cholera Toxin ( Sigma ) . Cells were passaged every 3 days with 0 . 25% Trypsin-EDTA ( Gibco ) , are mycoplasma free , and were verified by STR-profiling ( ATCC ) . Cell lines were generated with lentivirus produced in HEK293-FTs ( Thermo ) with third-generation packaging plasmids and Lipofectamine 2000 ( Thermo ) . Viral supernatants were collected 48 hr after transfection and incubated in MCF10As with polybrene ( 10 µg/ml , EMD Millipore ) . To create dual-sensor cells , MCF10As were infected with a lentiviral H2B-iRFP vector ( Addgene ) and sorted . We used gateway cloning ( Campeau et al . , 2009 ) to introduce ERK-KTR-mCer3 and ERK1-mRuby2 into PGK pLenti DEST vectors ( Addgene ) , infected and selected the H2B-iRFP MCF10As ( Blasticidin 3 µg/ml and Hygromycin 10 µg/ml Corning ) . We isolated moderately expressing clones using cloning cylinders ( EMD Milipore ) . For inducible cells , a gateway-ready reverse TET trans-activator ( rtTA ) plasmid was created by adding the rtTA with a 2A peptide to the Puromycin resistance gene in a CMV Puro DEST plasmid ( Addgene ) by gibson cloning ( Gibson et al . , 2009 ) . Human coding sequences were acquired from either Addgene or the Thermo Ultimate ORF Collection , sequence verified , and introduced in the rtTA CMV Puro DEST plasmid by gateway cloning ( Campeau et al . , 2009 ) . These plasmids were used for lentivirus , and infected cells were selected with Puromycin ( 1 µg/ml , Sigma ) . Utrophin-261-EGFP cell lines were made by cloning the coding region from pEGFP-C1 Utr261-EGFP ( Addgene ) into a pENTR backbone by Gibson cloning , and then introduced into the pLenti PGK Puro DEST plasmid by gateway cloning . These plasmids were used to generate lentivirus , and infected cells were selected with Puromycin . For inhibitor experiments , small molecules or antibodies and doxycycline were dissolved to a 10X working concentration in imaging media before addition . Final DMSO concentration did not exceed 0 . 15% . Inhibitors used include the MEK inhibitor PD-0325901 , the MMP/ADAM inhibitor Batimastat , the EGFR inhibitor Gefitinib , the p38 inhibitor BIRB-796 , the Sphingosine Kinase inhibitor SKII , and the S1PR2 , inhibitor JTE-013 all from Selleck Chemicals . The p38 inhibitor SB-203580 was obtained from Sigma . Amphiregulin was ordered from Peprotech . Amphiregulin function-blocking antibody is from R and D systems . The ADAM17KO cell lines were created using the CRISPR V2 Neo system ( a gift from Dr . Andrew Holland ) and gRNA oligos targeting R241 of exon 6 . Dual sensor cells were infected with lentivirus carrying this plasmid , selected with Neomycin ( 500 µg/ml , Sigma ) and clonally expanded before western blot validation ( Figure 2B ) . Cells were plated at 3*105 cells/well in fibronectin-treated ( EMD Millipore ) 96-well glass-bottom plates ( Thermo Scientific ) 48 hr before imaging . The following day , monolayers were serum-starved with 0 . 5% HS , phenol-red-free DMEM/F12 containing P/S with 1 mM Na Pyruvate and 10 mM HEPES . For signaling experiments in Figure 1 and Figure 1—figure supplement 1 , media was switched to 0% HS several hours before imaging to limit basal signaling . Monolayers were imaged using a Metamorph-controlled Nikon Eclipse Ti-E epifluorescence microscope with a 20x air objective and a Hamamatsu sCMOS camera . The multi LED light source SpectraX ( Lumencor ) and the multiband dichroic mirrors DAPI/FITC/Cy3/Cy5 and CFP/YPF/mCherry ( Chroma ) where used for illumination and imaging without any spectral overlap . For extrusion and live-actin experiments , a Metamorph-controlled Nikon Eclipse Ti-E spinning-disc confocal ( Yokogawa W1 ) with a 20x or 40X objective , Prime 95-B sCMOS camera ( Photometrics ) and a Multiline laser launch ( Cairn Research ) was used to capture H2B-iRFP and H2B-mClover or Utrophin-261-EGFP images every 1 µm of a 25–30 µm range through monolayers . Temperature ( 37°C ) , humidity and CO2 ( 5% ) were maintained throughout all imaging using OKO Labs control units . Sample sizes were selected by attempting to capture at least 100 cells from each population , with several hundred cells preferred . Key conditions from imaging experiments were performed at least twice , with one replicate presented in figures . Primary time-lapse images were subjected to flat-fielding and registration ( custom software Aikin et al . , 2020 ) before object segmentation and measurements in Cell Profiler . Nuclear positions were used to track individual cells through time-series ( custom software Aikin et al . , 2020 ) and intensity ratios were calculated as previously described ( Regot et al . , 2014 ) . Minimal cleaning of traces excluded cells where tracks switched between two objects , where the KTR ratios were affected by segmentation errors , or where traces represent less than two thirds of the entire time-course . In conditions where cells move rapidly , such as B-RafV600E and MEK2DD , and traces are shorter due tracking errors , track-length restraints were relaxed to include more cells for analysis . Single-cell traces were chosen by random plotting of distinct cells and selection of those that were tracked throughout the whole experiment . Peak counting was performed with software based on findPeaks ( O'Haver , 2014; Mathworks . com ) and modified to detect peaks based on the rate of change between gaussian-fitted minima and maxima from single-cell traces . For directed migration , positions were selected where distinct groups of inducible cells were present in the center of the field of view . Migration was quantified by positional changes over 20 min intervals for specified time windows , from all WT neighboring cells within a 200 µM X 200 µM area centered on the group of inducible cells . The migration angles of neighboring cells are plotted as radial histograms where 0° indicates migration directly towards , and 180° directly away from the center of isolated inducible cell groups . Migration datasets contain many sampled angles from large populations of cells . To overcome issues with high power , we applied subsampling techniques using 1000 iterations of 1000 randomly-selected migration angles each , and presented the median Two-Sample Kolmogorov-Smirnov ( KS ) Test P-values from these iterations ( ‘ns’ , not significant , *p<0 . 05 , **p<0 . 01 , ***p<0 . 001 ) . For extrusion experiments , histograms of mClover and mRuby pixel intensities across each z-stack were fit to gaussian curves using Matlab . The difference in gaussian fitted maxima of inducible cells and neighboring cells for each observation are plotted . Extrusion experiment sample size represents all non-overlapping positions from 2 to 3 independent wells excluding outliers resulting from imaging artifacts . Two-sample T-test significance values compare indicated conditions ( ‘ns’ , not significant , *p<0 . 05 , **p<0 . 01 , ***p<0 . 001 ) . For live-actin imaging experiments , ERK-KTR , H2B-iRFP were infected with the TRE3G::BRAFV600E . These cells were plated in 1% coculture with neighboring ERK-KTR , H2B-iRFP cells containing the Utrophin-261-EGFP construct . Oncogenic cells were identified by lack of green fluorescence and confirmed by images of the KTR , showing activation after induction with doxycycline ( 2 µg/ml ) . Actin enrichment was quantified by manually tracing the border of adjacent Utrophin-261-EGFP cells at the leading edge in contact with oncogenic cells on Fiji . All cells from a single extrusion event are plotted along with their mean . For assessment of protein expression in Figure 1—figure supplement 2 , parental , TRE3G::BRAFWT and TRE3G::BRAFV600E cells were plated in 6-well plastic culture plates , and starved with 0 . 5% HS , DMEM/F12 containing P/S with 1 mM Na Pyruvate and 10 mM HEPES overnight before treatment with media or doxycycline ( 2 µg/ml ) for 24 hr . Samples were lysed with RIPA buffer ( CST ) containing HALT protease and phosphatase inhibitors ( Thermo ) , and reduced in Laemelli SDS buffer ( BioRad ) with BME ( Sigma ) . Samples underwent electrophoresis on 4–15% gradient polyacrylamide gels ( BioRad ) and were immunoblotted with Rabbit anti-BRAF ( CST ) and mouse anti-HSC70 ( Santa Cruz Biotechnology ) , and IRDye donkey anti-rabbit 800 and goat anti-mouse 680 secondary antibodies ( LiCor ) before imaging . For validation of ADAM17 CRISPR-KOs in Figure 3 , suspected clones were grown , lysed , and run on a gel as described above , before immunoblotting with Rabbit anti-ADAM17 ( CST ) and mouse anti-HSC70 ( Santa Cruz Biotechnology ) primary and IRDye donkey anti-rabbit 800 and goat anti-mouse 680 secondary antibodies ( LiCor ) . All images were acquired on an Odyssey Infrared Scanner ( LiCor ) . For mass spectrometry , cells were grown to 90% confluency in T175 flasks and serum starved 24 hr ( see live imaging ) before switching to 15 mL growth factor/serum-free DMEM/F12 +/- Dox for 4 hr . The supernatant was collected and concentrated using 3 kDa cut-off centrifugal filters ( Millipore-Sigma ) . Triplicate samples were quantified by the Pierce Assay ( Thermo Scientific ) , reduced , alkylated , and trypsin digested before labeling with Tandem Mass Tag labels . Peptide fractions were analyzed by LC/MSMS using an Easy-LC 1200 HPLC system interfaced with an Orbitrap Fusion Lumos Tribrid Mass Spectrometer ( Thermo Fisher Scientific ) . Isotopically resolved masses in precursor and fragmentation spectra were processed in Proteome Discoverer software ( v2 . 3 , Thermo Scientific ) . Data were searched using Mascot ( 2 . 6 . 2 , Matrix Science ) against the 2017_Refseq 83 Human database and filtered at a 1% FDR confidence threshold . Monolayers were plated and starved as described above and treated with doxycycline ( Dox , 2 µg/ml ) in the presence of indicated inhibitors for 24 hr . During the final 4 hr , EdU ( 10 µM , Thermo Fischer Scientific ) was added into cultures to label S phase cells then fixed with methanol and washed before Alexa-Fluor Azide 488 click labelling ( Thermo Fischer Scientific ) and DAPI staining ( Thermo Scientific ) . Monolayers were imaged by epifluorescence . Because methanol fixation eliminates fluorescence from fluorescent proteins , cocultures were imaged just before fixation and registered with DAPI and EdU images to determine positions of inducible and neighboring cells . Sample size for population EdU experiments represents all non-overlapping positions from 2 to 3 independent wells , excluding outliers resulting from imaging artifacts . Key conditions were replicated at least twice . Monolayers were plated and starved as described above , and treated with media or doxycycline ( Dox , 2 µg/ml ) in the presence of any indicated inhibitors for 24 hr or timepoints as marked . To induce EMT , parental cells were maintained in full serum supplemented with TGFβ ( 5 ng/ml or 50 ng/ml , R and D Systems ) through splittings over 8 days to induce EMT ( Hao et al . , 2019 ) , then cells were plated and starved as described with consistent TGFβ . Cells were fixed 15 min with 4% PFA in PBS , washed with PBS before incubating 1 . 5 hr in blocking buffer ( PBS + 0 . 3% Triton X-100 + 5% BSA ) , followed by PBS washing and incubation overnight in blocking buffer with added primary antibodies ( Rabbit anti-E-Cadherin , or Rabbit anti-N-Cadherin , both CST ) . The following day , cells were washed in PBS before incubating 2 hr in blocking buffer with secondary antibody ( Donkey anti-Rabbit IgG Alexa Fluor 405 , Abcam ) . Cells were then washed with PBS and stored at 4°C until imaging via spinning disk confocal as described above . All incubations occurred at room temp in the dark , except the overnight primary , which was incubated at 4°C . | In animals , the MAPK pathway is a network of genes that helps a cell to detect and then respond to an external signal by switching on or off a specific genetic program . In particular , cells use this pathway to communicate with each other . In an individual cell , the MAPK pathway shows fluctuations in activity over time . Mutations in the genes belonging to the MAPK pathway are often one of the first events that lead to the emergence of cancers . However , different mutations in the genes of the pathway can have diverse effects on a cell’s behavior: some mutations cause the cell to divide while others make it migrate . Recent research has suggested that these effects may be caused by changes in the pattern of MAPK signaling activity over time . Here , Aikin et al . used fluorescent markers to document how different MAPK mutations influence the behavior of a human breast cell and its healthy neighbors . The experiments showed that cells with different MAPK mutations behaved in one of two ways: the signaling quickly pulsed between high and low levels of activity , or it remained at a sustained high level . In turn , these two signaling patterns altered cell behavior in different ways . Pulsed signaling led to more cell division , while sustained signaling stopped division and increased migration . Aikin et al . then examined the effect of the MAPK mutations on neighboring healthy cells . Sustained signaling from the cancerous cell caused a wave of signaling activity in the surrounding cells . This led the healthy cells to divide and migrate toward the cancerous cell , pushing it out of the tissue layer . It is not clear if these changes protect against or promote cancer progression in living tissue . However , these results explain why specific cancer mutations cause different behaviors in cells . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"cell",
"biology",
"cancer",
"biology"
] | 2020 | MAPK activity dynamics regulate non-cell autonomous effects of oncogene expression |
Recent studies have identified a genetic variant rs641738 near two genes encoding membrane bound O-acyltransferase domain-containing 7 ( MBOAT7 ) and transmembrane channel-like 4 ( TMC4 ) that associate with increased risk of non-alcoholic fatty liver disease ( NAFLD ) , non-alcoholic steatohepatitis ( NASH ) , alcohol-related cirrhosis , and liver fibrosis in those infected with viral hepatitis ( Buch et al . , 2015; Mancina et al . , 2016; Luukkonen et al . , 2016; Thabet et al . , 2016; Viitasalo et al . , 2016; Krawczyk et al . , 2017; Thabet et al . , 2017 ) . Based on hepatic expression quantitative trait loci analysis , it has been suggested that MBOAT7 loss of function promotes liver disease progression ( Buch et al . , 2015; Mancina et al . , 2016; Luukkonen et al . , 2016; Thabet et al . , 2016; Viitasalo et al . , 2016; Krawczyk et al . , 2017; Thabet et al . , 2017 ) , but this has never been formally tested . Here we show that Mboat7 loss , but not Tmc4 , in mice is sufficient to promote the progression of NAFLD in the setting of high fat diet . Mboat7 loss of function is associated with accumulation of its substrate lysophosphatidylinositol ( LPI ) lipids , and direct administration of LPI promotes hepatic inflammatory and fibrotic transcriptional changes in an Mboat7-dependent manner . These studies reveal a novel role for MBOAT7-driven acylation of LPI lipids in suppressing the progression of NAFLD .
Non-alcoholic fatty liver disease ( NAFLD ) is an increasingly common condition that affects roughly one-third of adults in the United States ( Cohen et al . , 2011; Machado and Diehl , 2016; Rinella and Sanyal , 2016; Wree et al . , 2013 ) . The expansion of adipose tissue in obese individuals is strongly associated with the development of NAFLD , yet mechanisms linking obesity to NAFLD and more advanced forms of liver disease such as NASH and cirrhosis are not well understood ( Cohen et al . , 2011; Machado and Diehl , 2016; Rinella and Sanyal , 2016; Wree et al . , 2013 ) . Genome-wide association studies ( GWAS ) provide a powerful unbiased tool to identify new genes and pathways that are involved in human disease , allowing for pinpoint accuracy in identification of new drug targets . This is exemplified by the recent success story of human genetic studies leading to rapid development of monoclonal antibodies targeting proprotein convertase subtilisin/kexin type 9 ( PCSK9 ) for hyperlipidemia and cardiovascular disease ( Hess et al . , 2018 ) . Within the last two years , several independent GWAS studies have identified a liver disease susceptibility locus ( rs641738 ) within a linkage-disequilibrium block that contains genes encoding MBOAT7 and TMC4 ( Buch et al . , 2015; Mancina et al . , 2016; Luukkonen et al . , 2016; International Liver Disease Genetics Consortium et al . , 2016; Viitasalo et al . , 2016; Krawczyk et al . , 2017; Thabet et al . , 2017 ) . Strikingly , the rs641738 variant is associated with all major forms of liver injury including NAFLD , alcoholic-liver disease ( ALD ) , and viral hepatitis-induced fibrosis ( Buch et al . , 2015; Mancina et al . , 2016; Luukkonen et al . , 2016; International Liver Disease Genetics Consortium et al . , 2016; Viitasalo et al . , 2016; Krawczyk et al . , 2017; Thabet et al . , 2017 ) . The rs641738 variant is associated with a C > T missense single nucleotide polymorphism ( SNP ) within the first exon the TMC4 gene , yet available data suggest that TMC4 is not abundantly expressed in human liver ( Mancina et al . , 2016 ) . Based on eQTL studies in the liver it has been suggested that instead reduced expression and activity of MBOAT7 may be mechanistically linked to liver disease progression ( Mancina et al . , 2016; Luukkonen et al . , 2016; International Liver Disease Genetics Consortium et al . , 2016; Viitasalo et al . , 2016; Krawczyk et al . , 2017; Thabet et al . , 2017 ) . The MBOAT7 gene encodes an acyltransferase enzyme that specifically esterifies arachidonyl-CoA to lysophosphatidylinositol ( LPI ) to generate the predominant molecular species of phosphatidylinositol ( PI ) in cell membranes ( 38:4 ) ( Gijón et al . , 2008; Zarini et al . , 2014; Lee et al . , 2012; Anderson et al . , 2013 ) . Given this biochemical activity , MBOAT7 is a unique contributor to the Lands’ cycle , which is a series of phospholipase-driven deacylation and lysophospholipid acyltransferase-driven reacylation reactions that synergize to alter phospholipid fatty acid composition , creating membrane asymmetry and diversity ( Shindou and Shimizu , 2009; Shindou et al . , 2009 ) . It is important to note that MBOAT7 , unlike other lysophospholipid acyltransferases , only diversifies the fatty acid composition of membrane PI species and not phospholipids with other head groups ( Gijón et al . , 2008; Zarini et al . , 2014; Lee et al . , 2012; Anderson et al . , 2013 ) . Despite recent progress in characterizing the selective biochemistry of MBOAT7 ( Gijón et al . , 2008 ) , and the clear genetic links to liver disease ( Buch et al . , 2015; Mancina et al . , 2016; Luukkonen et al . , 2016; International Liver Disease Genetics Consortium et al . , 2016; Viitasalo et al . , 2016; Krawczyk et al . , 2017; Thabet et al . , 2017 ) , there is no information regarding how MBOAT7 activity or its substrate ( LPI ) or product ( PI ) lipids impact liver disease progression . Here , we demonstrate that MBOAT7 expression is suppressed in obese humans and rodents . Furthermore , we show that Mboat7 , but not Tmc4 , loss of function in mice is sufficient to drive NAFLD progression , and show that Mboat7 substrate lipids ( LPIs ) may be critical mediators of obesity-linked liver disease progression .
NAFLD progression is commonly associated with obesity , and it has even been suggested that obesity is a causative factor in the pathogenesis of NAFLD ( Younossi et al . , 2018 ) . Although , there is now strong support that the common MBOAT7 SNP ( rs641738 ) is associated with NAFLD , it is not known whether MBOAT7 expression is significantly altered in obese people . To investigate this , we obtained wedge liver biopsies from sequentially consenting obese bariatric surgery patients and normal weight controls ( Schugar et al . , 2017 ) , and examined the expression levels of MBOAT7 mRNA . We found that hepatic MBOAT7 expression was dramatically reduced in obese people , when compared to normal weight controls ( Figure 1A ) . It is important to note that this striking reduction in MBOAT7 expression was not due to presence of the rs641738 MBOAT7 SNP , as we found the minor allele frequency was quite similar in lean , obese , and severely obese subjects ( Figure 1A ) . To follow up on our findings in human obesity , we also examined hepatic expression levels of Mboat7 in obese leptin-deficient mice and high fat diet-fed rats . Mboat7 expression was reduced in ob/ob mice compared to lean WT controls ( Figure 1B ) . In agreement , we found that high fat diet-induced obesity in Sprague-Dawley rats was also associated with significant reductions in hepatic Mboat7 expression ( Figure 1C ) . Collectively , these results demonstrate that hepatic expression of MBOAT7 is suppressed in obese humans and rodents . Next , we used a systems genetics approach to examine links between Mboat7 expression and adiposity traits in mice by leveraging data generated using the hybrid mouse diversity panel ( HMDP ) ( Ghazalpour et al . , 2012 ) . To induce obesity , all HMDP mouse strains were fed an obesity-promoting high fat and high sucrose diet ( Parks et al . , 2013 ) . Across the different strains in the HMDP , we found that Mboat7 expression in the liver was only modestly correlated ( r = −0 . 244 , p=0 . 01 ) with adiposity ( Figure 1D ) . However , Mboat7 expression in white adipose tissue ( WAT ) was strongly negatively correlated ( r = −0 . 68 , p=2 . 3e-15 ) with both fat pad weight and % body fat in male and female mice ( Figure 1E ) . Given the fact that obesity is commonly associated with insulin resistance and type 2 diabetes mellitus ( Cohen et al . , 2011; Machado and Diehl , 2016; Rinella and Sanyal , 2016; Younossi et al . , 2018 ) , we also examined the relationship between Mboat7 expression and insulin sensitivity in the HMDP . We found that expression of Mboat7 in adipose tissue was negatively ( r = −0 . 38 , p=0 . 0002 ) associated with indices of insulin sensitivity ( Figure 1F ) . Collectively , these data suggest that MBOAT7 may be mechanistically linked to the well-known association between obesity , insulin resistance , and NAFLD progression . To test whether MBOAT7 impacts obesity-linked NAFLD progression we utilized an in vivo knockdown approach in high fat diet-fed C57BL/6 mice . Metabolic phenotyping of global Mboat7-/- mice has been limited due to the fact that these mice are only viable for several weeks after birth due to the critical role Mboat7 plays in cortical lamination and neuronal migration in the brain ( Lee et al . , 2012; Anderson et al . , 2013 ) . To overcome this barrier , we have generated second-generation antisense oligonucleotides ( ASOs ) , which predominately target liver , adipose tissue , and cells within the reticuloendothelial system to selectively knock down Mboat7 expression in adult mice using methods we have previously described ( Schugar et al . , 2017 ) . This ASO approach allows us to circumvent the postnatal lethality of global Mboat7 deletion ( Lee et al . , 2012; Anderson et al . , 2013 ) , and permits investigation into obesity-linked liver disease progression with near complete loss of function of Mboat7 in the liver . Mboat7 ASO treatment resulted in significant reductions of Mboat7 mRNA and protein in the liver ( Figure 2A ) , and white adipose tissue ( Figure 2—figure supplement 1 ) , without altering Mboat7 expression in the brain , spleen , heart , or skeletal muscle ( Figure 2—figure supplement 2 ) . Although Mboat7 ASO treatment significantly reduced adipose Mboat7 expression ( Figure 2 , Figure 2—figure supplement 1 ) , this was not associated with alterations in body weight ( Figure 2B ) , fat mass ( Figure 2—figure supplement 1L ) , adipose gene expression ( Figure 2—figure supplement 1A–K ) , food intake ( Figure 2C ) , energy expenditure ( Figure 2D , and Figure 2—figure supplement 3 ) , or physical activity ( Figure 2—figure supplement 3 ) . Despite this lack of phenotypic differences in adipose tissue , Mboat7 knockdown resulted in large alterations in the liver lipid storage . Mboat7 ASO treatment promoted an increase in liver weight ( Figure 2F ) and striking hepatic steatosis in HFD-fed , but not chow-fed mice ( Figure 2E , I ) . Importantly , Mboat7 ASO-driven hepatic steatosis was consistently seen with two distinct ASOs targeting different regions of the Mboat7 messenger RNA ( Figure 2—figure supplement 4 ) . Mboat7 ASO-driven hepatic steatosis was characterized by accumulation of triglycerides , free cholesterol , and cholesterol esters only in high fat fed cohorts ( Figure 2J–M ) . Despite these significant alterations in hepatic lipids , Mboat7 knockdown did not dramatically alter the levels of triglycerides or cholesterol in circulating lipoproteins ( Figure 2—figure supplement 5 ) . Mboat7 knockdown was also associated with hepatocyte injury as indicated by elevated liver enzyme levels ( AST and ALT ) , but this only occurred in high fat-fed mice ( Figure 2G , H ) . To more comprehensively understand the global effects of Mboat7 knockdown on liver function , we performed unbiased RNA sequencing experiments in control versus Mboat7 ASO-treated mice ( Figure 3 ) . Principal component analysis of RNA expression profiles showed separation by ASO group according to principal component analysis and hierarchical clustering analyses ( Figure 3A , B ) . ASO groups were also partitioned by unsupervised hierarchical clustering ( data not shown ) . Mboat7 knockdown resulted in a number of differentially expressed genes ( DEGs ) , with 124 DEGs being suppressed and 887 DEGs being upregulated by Mboat7 ASO treatment ( top 50 DEGs shown in Figure 3B and overall changes in Figure 3C ) . To understand the major pathways affected by Mboat7 knockdown , we performed Gene Ontology Molecular Function enrichment analysis , and found that many of the pathways that were significantly enriched are mechanistically linked to liver injury such as leukocyte extravasation , monocyte/macrophage activation , and fibrosis/hepatic stellate cell activation ( Figure 3D ) . As predicted Mboat7 mRNA levels were dramatically suppressed by Mboat7 ASO treatment , but quite unexpectedly we also saw that the expression of Tmc4 was reduced in Mboat7 ASO-treated mice ( Figure 3B , E ) . These data shown for the first time that ASO-mediated knockdown of Mboat7 results in coordinated suppression of its neighboring gene Tmc4 ( the rs641738 polymorphism is located in exon 1 of the TMC4 gene ) , indicating some potential cross talk that deserves further exploration ( Figure 3E ) . Quantitative real time PCR ( qPCR ) analysis also confirmed the RNA sequencing results showing that Mboat7 knockdown resulted in elevated expression of gene associated with inflammation ( Tnfa and Il1b ) and early fibrosis ( Col1a2 and Acta2 ) in high fat fed mice ( Figure 3E ) . In addition , Mboat7 knockdown was associated with altered immune cell populations in the liver . Mboat7 knockdown increased Cd8+ T lymphocytes , while decreasing the total number of Cd11b+ macrophages in the liver ( Figure 3—figure supplement 1 ) . Of the hepatic macrophages that were present , more were skewed towards the proinflammatory M1 ( Cd11c+ ) state , and less were skewed towards the alternative M2 ( Cd206+ ) phenotype ( Figure 3—figure supplement 1 ) . Collectively , these data suggest that specifically under high fat feeding conditions , Mboat7 loss of function is associated with dysregulated immune cell homeostasis , inflammation , and gene expression signatures that are consistent with early activation of pro-fibrotic programs in the liver ( Figure 3 ) . Given the fact that the rs641738 polymorphism is located in exon 1 of the TMC4 gene , and we unexpectedly found that Mboat7 ASO treatment also reduces Tmc4 expression ( Figure 3B , E ) , we wanted to examine whether alteration in Tmc4 may also be a key regulator of hepatic steatosis . To understand the role of Tmc4 in hepatic steatosis , we generated global Tmc4 knockout mice using CRISPR-Cas9-mediated gene editing and examined hepatic lipid levels under high fat feeding conditions ( Figure 4 ) . Global Tmc4 knockout mice have marked reductions in hepatic Tmc4 mRNA ( Figure 4A ) and protein ( Figure 4B ) , yet importantly Mboat7 expression is unaltered ( Figure 4A ) . In contrast to the striking hepatic steatosis seen with Mboat7 loss of function ( Figure 2E , Figure 2—figure supplement 4 ) , Tmc4 null mice show similar levels of hepatic lipids when fed a high fat diet ( Figure 4 ) . These data strongly implicate Mboat7 , but not Tmc4 , as the primary mediator of hepatic steatosis seen with the rs641738 variant or in Mboat7 ASO-treated mice . The abnormal accumulation of lipids ( i . e . , lipotoxicity ) in tissues such as the pancreas and liver is frequently associated with the pathogenesis of type 2 diabetes ( Unger and Scherer , 2010; Samuel and Shulman , 2018 ) . Given that Mboat7 knockdown promoted striking hepatic steatosis ( Figure 2E ) , and the fact that adipose tissue expression of Mboat7 is negatively correlated with insulin sensitivity across strains of the HMDP ( Figure 1F ) we examined glucose homeostasis in Mboat7 ASO-treated mice fed a high fat diet ( Figure 5 ) . Mboat7 knockdown was associated with significantly impaired systemic glucose tolerance ( Figure 5A ) . Mboat7 ASO-treated mice also exhibited profound hyperinsulinemia in the fasted state and throughout an intraperitoneal glucose tolerance test ( Figure 5B ) . In parallel , glucose-stimulated C-peptide release was elevated in Mboat7 ASO-treated mice , potentially indicating overproduction of insulin in the pancreatic β-cell ( Figure 5C ) . During an intraperitoneal insulin tolerance test , Mboat7 ASO-treated mice showed blunted plasma glucose lowering ( Figure 5D ) . To more directly measure tissue-specific insulin action in Mboat7 knockdown mice we examined the acute phosphorylation of the insulin receptor ( IR ) and protein kinase B ( Akt ) in response to a single insulin injection ( Figure 5E–G ) . Mboat7 ASO-treated mice had reduced insulin-stimulated phosphorylation of IR and downstream Akt in the liver ( Figure 5E ) . However , Mboat7 ASO treatment did not significantly alter insulin signal transduction in either skeletal muscle or white adipose tissue ( Figure 5F , G ) . These data suggest that ASO-mediated knockdown of Mboat7 promotes hyperinsulinemia , yet impairs insulin action specifically in the liver . Given the fact that Mboat7 is known to catalyze the selective esterification of arachidonyl-CoA to LPI lipids in neutrophils ( Gijón et al . , 2008 ) , we wanted to examine its lysophosphatidylinositol acyltransferase ( LPIAT ) activity in the liver and also examine both substrate LPI and product PI species across a range of tissues . First , we isolated hepatic microsomes from control and Mboat7 ASO-treated mice and assayed LPIAT enzymatic activity using saturated and monounsaturated LPI substrates ( Figure 6A ) . In chow fed mice Mboat7 knockdown only modestly decreased hepatic LPIAT activity toward 18:1 LPI ( Figure 6A ) , yet in high fat diet-challenged mice Mboat7 knockdown resulted in a highly significant ~50% reduction in LPIAT activity using either 16:0 , 18:0 , or 18:1 LPI substrates ( Figure 6A ) . These data suggest that Mboat7 is a significant contributor to total hepatic LPIAT activity , especially under high fat feeding conditions . Aligned with these alterations in enzyme activity , knockdown of Mboat7 in high fat-fed mice resulted in alterations in LPI and PI lipids in a highly tissue-specific manner . Mboat7 ASO treatment did not significantly alter LPI levels in the circulation , but was associated with selective reduction in the 38:3 and 38:4 species of circulating PI lipids ( Figure 6B , Figure 6C , and Figure 6—figure supplements 1–2 ) . In contrast to effects in the circulation , Mboat7 knockdown was associated with significant accumulation of 16:0 and 18:1 LPI species in the liver ( Figure 6D and Figure 6—figure supplement 2 ) . Furthermore , Mboat7 ASO-treatment resulted in reduced 38:3 and 38:4 PI levels , while increasing more saturated ( 34:1 , 34:2 , 36:1 , and 36:2 ) PI species in the liver ( Figure 6E and Figure 6—figure supplement 1 ) . In white adipose tissue , Mboat7 knockdown did not significantly alter LPI levels ( Figure 6—figure supplement 2 ) , but promoted marked accumulation of more saturated PI species both in chow and high fat fed mice ( Figure 6—figure supplement 1 ) . Also , Mboat7 ASO treatment resulted in significant reductions in several PUFA-enriched PI species ( 38:3 , 38:4 , 38:5 , and 38:6 ) in adipose tissue ( Figure 6—figure supplement 1 ) . In contrast to alterations in LPI and PI levels in liver and adipose tissue ( Figure 6 and Figure 6—figure supplements 1–2 ) , Mboat7 ASO treatment did not significantly alter LPI or PI levels in the brain ( Figure 6—figure supplement 3 ) or pancreas ( Figure 6—figure supplement 4 ) . In contrast to marked changes in inositol-containing phospholipids , it is also important to note that Mboat7 knockdown did not significantly alter the hepatic levels of other major phospholipids including phosphatidylcholines , lysophosphatidylcholines , phosphatidylethanolamines , lysophosphatidylethanolamines , phosphatidylserines , phosphatidylglycerols , or phosphatidic acids ( Figure 6—figure supplement 5 ) . Collectively , these data show that ASO-mediated knockdown of Mboat7 primarily alters LPI and PI levels in the liver and white adipose tissue , which creates the potential to induce an imbalance of local lipid mediators that originate from PI metabolism . The major enzymatic product of MBOAT7 ( 38:4 PI ) is a potential reservoir for the generation of arachidonic acid-derived immunomodulatory lipid mediators ( Serhan et al . , 2015; Dennis and Norris , 2015 ) . Therefore , we initially hypothesized that Mboat7 loss of function might limit the pool of arachidonic acid in PI lipids available for arachidonic acid-derived lipid mediator production . However , when we examined arachidonic acid-derived pro-inflammatory ( LTB4 , PGE2 , PGD2 , PGF2α , and TXB2 ) and pro-resolving ( 15R-LXA4 ) lipid mediators there was no apparent alteration in the liver of Mboat7 ASO-treated mice ( Figure 6—figure supplement 6 ) . Collectively , these data do not support a rate-limiting role for MBOAT7 in arachidonic acid-derived lipid mediator production in mice . Given the fact that Mboat7 knockdown is associated with increases in LPI lipids , we next hypothesized that MBOAT7 substrate LPI lipids themselves may be the main drivers of liver disease progression under conditions of HFD-induced obesity and Mboat7 loss of function . In further support of this concept , we found that saturated LPI lipids are significantly elevated in the circulation of humans with advanced fibrosis compared to healthy controls ( Figure 7 ) . To more directly test the hypothesis that LPI lipids may promote liver disease in a Mboat7-dependent manner , we directly treated mice with LPI lipids to transiently increase circulating LPI levels within a physiologic level ( ~2 fold ) ( Figure 8—figure supplement 1 ) . Injection of 16:0 , 18:0 , or 18:1 LPI into control ASO-treated mice did not alter hepatic inflammatory or fibrotic gene expression ( Figure 8 and Figure 8—figure supplement 2 ) . This is not surprising , given that it was previously reported that exogenously provided LPI is rapidly esterified into membrane PI pools ( Darnell and Saltiel , 1991; Darnell et al . , 1991; Jackson and Parton , 2004 ) . In other words , under normal conditions MBOAT7 activity rapidly acylates LPI to divert this signaling lipid into a membrane PI storage pool . We next postulated that under conditions such as obesity where Mboat7 activity is diminished , LPI-driven signaling can be sustained and promote liver injury . In support of this concept , direct injection of exogenous 18 carbon LPIs can rapidly ( 24 hr ) increase the expression of genes characteristic of hepatic inflammation ( Cd11c , Tnfa , IL1b ) and early fibrosis ( Desmin , Col1A1 , Col1A2 , and Acta2 ) in mice with Mboat7 knockdown ( Figure 8 and Figure 8—figure supplement 2 ) . These data suggest that 18 carbon LPI lipids can acutely induce hepatic inflammatory and fibrotic gene expression programs , but only when MBOAT7 function is compromised , as is seen in obesity or with loss of function variants like rs641738 . It is important to note that 18 carbon LPI lipids can only significantly elicit such pro-inflammatory and pro-fibrotic effects in high fat fed mice ( Figure 8 and Figure 8—figure supplement 2 ) , as we did not see this same effect in chow fed cohorts ( not shown ) . Although direct administration of 18 carbon LPI lipids in chow fed mice did not significantly alter the same pro-inflammatory or pro-fibrotic genes as we found with Mboat7 knockdown ( Figure 3 ) , we did find that a small subset of acute phase response genes ( serum amyloid A genes , Saa1 and Saa2 ) and other immunomodulatory genes ( Gdf15 , Ly6d , Lcn2 , Socs2 , etc . ) were altered by 18:0 LPI treatment in a Mboat7-dependent manner ( Figure 8—figure supplement 3 ) . Interestingly , when liver triacylglycerol levels were examined in this LPI injection experiment , we only found significantly elevation in the saline and 18:1 LPI injection groups , but not in the 16:0 LPI injection group ( Figure 8I ) . Collectively , these results suggest that 18 carbon LPI lipids can alter hepatic inflammatory transcriptional programs , but this is highly reliant on both diminished Mboat7 expression/activity and high fat diet feeding . Mboat7 loss of function results in a striking accumulation of neutral lipids including triglycerides and cholesteryl esters ( Figure 2E–I ) , yet the mechanism ( s ) behind this are poorly understood . We therefore examined several potential mechanisms driving the mixed hepatic steatosis in Mboat7 ASO-treated mice . First , we examined the expression of genes involved in lipogenesis , fatty acid oxidation , and cholesterol sensing and export under both fed and fasted conditions ( Figure 9 ) . Unlike many other models of hepatic steatosis , there were no significant alterations in lipogenic gene expression either in the fed or fasted state with Mboat7 knockdown ( Figure 9A ) . Unexpectedly , the expression of carnitine palmitoyl transferase 1 ( Cpt1 ) was significantly elevated with Mboat7 knockdown ( Figure 9A ) , but this would not be expected to promote the accumulation of neutral lipids . The expression of enzymes involved in cholesterol biosynthesis ( Hmgcr and Hmcgs1 ) were modestly reduced in Mboat7 ASO-treated mice , but only in the fasted state ( Figure 9A ) . In addition , expression of the cholesterol efflux regulator Abca1 was elevated in Mboat7 ASO-treated mice , but only in the fasted state . Altogether , these minor differences in hepatic gene expression are unlikely to be driving the lipid accumulation in Mboat7 ASO treated mice . Next , we evaluated whether Mboat7 may influence the export of neutral lipids via packaging on nascent very low density lipoproteins ( VLDL ) . However , Mboat7 ASO-treated mice did not have significant differences in VLDL-TG secretion during a detergent block ( Figure 9B ) . Furthermore , steady state plasma levels of triglycerides and total cholesterol were not significantly altered in Mboat7 ASO-treated mice other than a minor reduction in cholesterol content in both low density lipoproteins ( LDL ) and high density lipoproteins ( HDL ) only in chow-fed cohorts ( Figure 2—figure supplement 5 ) . Another common cause of fatty liver is increased delivery of adipose-derived fatty acids to the liver , as is commonly seen with prolonged fasting and certain types of lipodystrophies . However , Mboat7 knockdown did not alter basal or catecholamine-stimulated adipocyte lipolysis in vivo ( Figure 9C ) , ruling out a role for altered adipose lipolysis as a contributing factor . Finally , we examined the possibility that MBOAT7 may be a determinate of metabolism locally at the surface of cytosolic lipid droplets to regulate hepatic steatosis , given that one recent study reported that MBOAT7 can localize to cytosolic lipid droplets ( Mancina et al . , 2016 ) . We confirmed that MBOAT7 can indeed be found in lipid droplets isolated by sucrose gradient fractionation ( Figure 9F ) , and plays a regulatory role in both the lipidome and proteome of cytosolic lipid droplets . Cytosolic lipid droplets isolated from Mboat7 ASO-treated mice showed marked accumulation of 18:0 LPI and 20:4 LPI , but not 16:0 LPI or 18:1 LPI ( Figure 9D ) . This is in stark contrast to what is seen in the whole liver , where Mboat7 knockdown instead promotes accumulation of 16:0 LPI and 18:1 LPI ( Figure 6D ) . Also , lipid droplets isolated from Mboat7 ASO-treated mice have a reduced level of 38:4 PI but a reciprocal increase in 36:3 and 38:3 PI ( Figure 9E ) , which is quite different when to compared to effects in whole liver ( Figure 6E ) . In addition to alterations in the lipid droplet lipidome , Mboat7 knockdown is associated with accumulation of several proteins involved in lipid synthesis and storage on isolated lipid droplets ( Figure 9F ) . Both perilipin 2 and 3 ( PLIN2 and PLIN3 ) were much more abundant on lipid droplets isolated from Mboat7 ASO-treated mice , as were the critical lipid synthetic enzymes CTP:phosphocholine cytidylyltransferase α ( CCTα ) and glycerol-3-phosphate 4 ( GPAT4 ) . Given that recent reports have shown that lipid droplet targeting of CCTα and GPAT4 are critical regulators of the overall size and triglyceride storage capacity of lipid droplets ( Guo et al . , 2008; Krahmer et al . , 2011; Wilfling et al . , 2013 ) , these proteomic alterations at the lipid droplet surface may to contribute to the hepatic steatosis seen in Mboat7 ASO-treated mice . In order to understand cell autonomous effects of MBOAT7 lipid metabolism we generated MBOAT7-deficient cells via CRISPR-Cas9-mediated genome editing . Huh7 lacking MBOAT7 have increased lipid droplets upon fatty acid loading ( Figure 10A , B ) . This lipid accumulation was due in part to increases in de novo lipogenesis rates ( Figure 10C ) , and reciprocal decreases in fatty acid oxidation rates ( Figure 10D ) . However , MBOAT-deficient Huh7 cells did not have altered turnover of stored triacylglycerol ( Figure 10E ) or cholesterol ester ( Figure 10F ) .
Given that several recent studies have found a strong link between the common rs641738 variant allele and liver disease progression , there is considerable interest in identifying the causative gene within this locus . Although the rs641738 polymorphisms maps to the first exon of the poorly annotated gene TMC4 , this study is the first to demonstrate that genetic deletion of Tmc4 does not result in hepatic steatosis . Instead , selective loss of function of the neighboring gene Mboat7 is sufficient to sensitize mice to high fat diet-driven liver disease progression . The major findings of the current study include the following: ( 1 ) Genetic deletion of Tmc4 does not alter hepatic lipid storage , ( 2 ) Hepatic expression of MBOAT7 is reduced in obese humans and rodents , independent of rs641738 status , ( 3 ) Mboat7 expression in mouse liver and adipose tissue is negatively correlated with obesity and insulin sensitivity across the strains represented in the Hybrid Mouse Diversity Panel , ( 4 ) Mboat7 knockdown promotes hepatic steatosis , hepatocyte death , inflammation , and early gene expression profiles consistent with fibrosis , but this only happens when mice are challenged with a high fat diet , ( 5 ) Mboat7 loss of function promotes striking hyperinsulinemia and hepatic insulin resistance , ( 6 ) Mboat7 knockdown results in tissue-specific reorganization of its substrates ( LPI ) and product ( PI ) lipids , ( 7 ) LPI lipids can rapidly induce hepatic inflammatory and fibrotic gene expression programs in an Mboat7-dependent manner in mice , and ( 8 ) the hepatic steatosis seen with Mboat7 knockdown is not related to differences in lipogenic gene expression , VLDL secretion , or tissue lipolysis , but instead may be related to alterations in hepatic lipid droplet accumulation of lipogenic enzymes ( CCTα and GPAT4 ) that allow for expansion of large lipid droplets ( Guo et al . , 2008; Krahmer et al . , 2011; Wilfling et al . , 2013 ) , and increased rates of de novo lipogenesis and reduced fatty acid oxidation . Collectively , these data support a role for Mboat7-driven acylation of LPI lipids as a key protective mechanism against obesity-linked NAFLD progression . The MBOAT family of enzymes are critical players in determining the composition of fatty acids in cellular membranes , and are emerging as key players in cardiometabolic disease ( Rong et al . , 2013; Harayama et al . , 2014 ) . It is important to note that several other lysophospholipid acyltransferase enzymes have been linked to inflammatory diseases ( Rong et al . , 2013; Harayama et al . , 2014 ) . In particular , lysophosphatidylcholine acyltransferases ( LPCAT1 , LPCAT2 , and LPCAT3 ) are known to regulate tissue inflammation and endoplasmic reticulum stress by altering arachidonic acid availability as well as determining membrane phosphatidylcholine saturation ( Rong et al . , 2013; Harayama et al . , 2014 ) . In fact , the lysophospholipid acyltransferase family sits at a critical signaling nexus , given that they can play key roles in the generation of arachidonic acid-derived lipid mediators as well as regulating the levels of lysophospholipid signaling lipids ( Shindou and Shimizu , 2009; Shindou et al . , 2009 ) . Given the fact that MBOAT7 preferentially generates PUFA-enriched PIs it also has the unique potential to also impact phosphorylated PI species ( PtdInsP , PtdInsP2 , and PtdInsP3 , etc . ) ( Anderson et al . , 2013 ) . In agreement with our findings with PI lipids ( Fil ) in Mboat7 ASO-treated mice , global genetic deletion of Mboat7 selectively lowers the 38:4 molecular species of PIPs , while increasing more saturated species ( Anderson et al . , 2013 ) . Therefore , we cannot rule out that alterations in PIP-dependent signal transduction may also play a role in MBOAT7 loss of function-driven liver disease progression . Unexpectedly , Mboat7 knockdown did not significantly alter hepatic lipid storage or inflammation in chow-fed mice . Instead there is a clear unmasking of Mboat7-dependent fatty liver phenotypes only when mice are fed with a high fat diet . This could potentially indicate that unknown dietary factors play a regulatory role and limit hepatic LPIAT activity via unknown mechanisms ( Figure 6A ) . Our data also support the idea that when mice , rats , and humans become obese MBOAT7 expression is reduced to varying degrees ( Figure 1A–C ) , indicating regulation either at the transcriptional of post-transcriptional level . Future work should focus on understanding how dietary factors and adiposity itself impacts MBOAT7 expression and activity , and under what dietary and BMI conditions are MBOAT7 polymorphisms predicted to be most deleterious . In humans , the rs641738 variant has been clearly linked to end stage fibrotic liver disease ( Buch et al . , 2015; Mancina et al . , 2016; Luukkonen et al . , 2016; International Liver Disease Genetics Consortium et al . , 2016 ) . However , an important limitation of our studies is that high fat diet feeding in mice is not sufficient to drive bridging fibrosis . In future studies , it will be important to examine how Mboat7 loss of function impacts the development of fibrosis in appropriate fibrosis-prone animal models , and whether Mboat7 expression in hepatic stellate cells plays a regulatory role in the progression from NASH to cirrhosis . From this study it is clear that MBOAT7 can diversify the inositol-containing phospholipids and the associated proteome on cytosolic lipid droplets , and this could in part explain the large lipid droplets that accumulate in Mboat7 ASO-treated mice . It is interesting to note that the well-known PNPLA3 variant associated with fatty liver disease ( I148M ) accumulates on lipid droplets , and similarly reorganizes the lipidome and proteome to promote liver disease progression ( BasuRay et al . , 2019; Wang et al . , 2019; Mitsche et al . , 2018; BasuRay et al . , 2017 ) . Our findings here with MBOAT7-driven restructuring of the lipid droplet surface , and those recently published with PNPLA3 ( BasuRay et al . , 2019 ) , suggest that alterations in lipid modifying enzyme access to the surface of cytosolic lipid droplets may be a common mechanism by which human fatty liver develops . Furthermore , we show that genetic deletion of MBOAT7 in Huh7 hepatoma cells can promote cell autonomous increases in cytosolic lipid droplets , and this enhanced lipid storage is associated with augmented lipogenesis rates and reduced fatty acid oxidation rates . Our studies provide new clues into new therapeutic leads for advanced liver diseases . Another particularly striking finding from the current study is that Mboat7 knockdown promotes severe hyperinsulinemia ( Figure 5B ) . Based on our results examining both insulin and C-peptide release during a glucose tolerance test ( Figure 5B , C ) , it is reasonable to assume that Mboat7 knockdown is enhancing glucose-stimulated insulin secretion ( GSIS ) in pancreatic β cells . However , it is important to note that Mboat7 ASO treatment did not alter Mboat7 expression ( data not shown ) or LPI/PI lipids in the pancreas ( Figure 6—figure supplement 4 ) . Therefore , it is tempting to speculate that ASO-mediated knockdown of Mboat7 facilitates the production of an endocrine signaling lipid ( possibly LPI ) that may impact β cell GSIS . Interestingly , there are reports demonstrating that LPI lipids can stimulate β cell GSIS ( Metz , 1986; Metz , 1988 ) . It is interesting to note that expression of Mboat7 in white adipose tissue is modestly negatively correlated with HOMA-IR in the hybrid mouse diversity panel ( Figure 1F ) , yet ASO-mediated knockdown of Mboat7 did not alter insulin signaling in adipose tissue ( Figure 5F ) . These data suggest that additional studies are needed to clarify a potential role for Mboat7 in adipose tissue insulin sensitivity . Collectively , our data suggest that under conditions where MBOAT7 activity is suppressed ( obesity or with the rs641738 variant ) inefficient acylation of LPI substrate allows for these lipids to accumulate and initiate autocrine , paracrine , and potentially endocrine signaling that impact the progression of NAFLD and insulin resistance . Therefore , further investigation into the receptor system ( s ) that sense LPI lipids could hold therapeutic promise in liver disease and other associated metabolic diseases . Collectively , these studies identify MBOAT7-driven acylation of LPI lipids as an important modulator of both liver disease progression and associated type 2 diabetes .
The majority of subjects recruited to examine MBOAT7 expression levels were morbidly obese bariatric surgery patients , but we were able to obtain liver biopsies from 10 subjects with a BMI under 30 as normal weight controls . For recruitment , adult patients undergoing gastric bypass surgery at Wake Forest School of Medicine were consented via written consent and enrolled by a member of the study staff following institutionally approved IRB protocols as previously described ( Shores et al . , 2011 ) . Exclusion criteria included: positive hepatitis C antibody , positive hepatitis B surface antigen , history of liver disease other than NAFLD , Childs A , B , or C cirrhosis , past or present diagnosis/treatment of malignancy other than non-melanocytic skin cancer , INR greater than 1 . 8 at baseline or need for chronic anticoagulation with warfarin or heparin products , use of immunomodulation for or history of inflammatory diseases including but not limited to malignancy , rheumatoid arthritis , psoriasis , lupus , sarcoidosis and inflammatory bowel disease , and greater or equal to seven alcohol drinks per week or three alcoholic drinks in a given day each week . In addition to bariatric surgery patients , a small number of non-obese subjects ( body mass index <30 . 0 ) consented to liver biopsy during elective gall bladder removal surgery ( n = 10 ) . Each subject was assigned a unique identifier which was used throughout the study and did not include any identifiable information about the patient such as name , address , telephone number , social security number , medical record number or any of the identifiers outlined in the HIPAA Privacy Rule regulations . Only the principal investigator had access to the code linking the unique identifier to the study subject . Basic clinical information was obtained via self-reporting and a 15 ml baseline blood sample was obtained at the time of enrollment . A subset of this cohort has been previously described ( Schugar et al . , 2017; Shores et al . , 2011 ) . At the time of surgery , the surgeon collected a roughly 1-gram sample from the lateral left lobe . Wedge biopsies were rinsed with saline and immediately snap frozen in liquid nitrogen in the operating room before subsequent storage at −80°C . For data shown in Figure 7 showing levels of MBOAT7 substrate and product lipids , de-identified patient samples from the Cleveland Clinic hepatology clinic ( IRB # 10–947 ) were analyzed . These patients had biopsy proven Ishak fibrosis scores ( Ishak et al . , 1995 ) of 0 ( normal ) or 4 ( advanced fibrosis ) . For analysis of hepatic MBOAT7 expression , RNA isolated from liver biopsies were used for quantitative real time PCR ( qPCR ) as described below . Sprague Dawley Rats were received at 12 weeks of age and were housed in individual cages , kept at a constant temperature and ambient humidity in a 12-h light/dark cycle . Animals were then randomly assigned to either a standard chow diet or a high-fat diet ( D12492 , 60% fat , Research Diets , New Brunswick , NJ , USA ) ad libitum to establish diet-induced obesity as previously described ( Schugar et al . , 2017 ) . After 6 months of HFD-feeding , livers were excised for standard qPCR analysis of Mboat7 expression . 92 inbred strains of 8-week-old male mice ( 180 individual mice ) were fed a high fat , high sucrose diet ( D12266B , Research Diets , New Brunswick , NJ ) for 8 weeks before tissue collection ( Parks et al . , 2013 ) . Gene expression of Mboat7 in white adipose tissue and liver were measured and correlated with obesity related traits using biweight midcorrelation analysis as previously described ( International Liver Disease Genetics Consortium et al . , 2016 ) . To explore the role of Mboat7 in diet-induced obesity , NAFLD progression , and insulin resistance , we utilized an in vivo knockdown approach in 8 week old adult mice . Selective knockdown of Mboat7 was accomplished using 2’-O-ethyl ( cET ) modified antisense oligonucleotides ( ASO ) . All ASOs used in this work were synthesized , screened , and purified as described previously ( Crooke et al . , 2005 ) by Ionis Pharmaceuticals , Inc ( Carlsbad , CA ) . For Mboat7 knockdown studies , adult ( 8 week old ) male C57BL/6 mice were purchased from Jackson Labs ( Bar Harbor , ME USA ) , and maintained on either a standard rodent chow diet or a high fat diet ( HFD , D12492 from Research Diets Inc ) and injected intraperitoneally biweekly with 12 . 5 mg/kg of either non-targeting control ASO or one of two independent ASOs directed against murine Mboat7 for a period of 20 weeks . Similar results were seen with two independent ASOs targeting different regions of the Mboat7 mRNA , hence key data using one Mboat7 ASO are shown . All mice were maintained in an Association for the Assessment and Accreditation of Laboratory Animal Care , International-approved animal facility , and all experimental protocols were approved by the Institutional Animal Care and use Committee of the Cleveland Clinic ( IACUC protocols # 2015–1519 and # 2018–2053 ) . Global Tmc4 knockout mice ( C57BL/6NJ-Tmc4em1 ( IMPC ) J/Mmjax ) were provided by the Knockout Mouse Phenotyping Program ( KOMP ) at The Jackson Laboratory using CRISPR technology . Briefly , guide RNAs ( GGAACCAGACCTTTTCCCAA and GAGTCAGCGTCAGAAAATGA ) were designed to insert create a 277 bp deletion in exon 3 of the transmembrane channel-like gene family 4 ( Tmc4 ) gene beginning at Chromosome 7 position 3 , 675 , 326 bp and ending after 3 , 675 , 602 bp ( GRCm38/mm10 ) . The mutation is predicted to delete ENSMUSE00001301550 ( exon 3 ) and 131 bp of flanking intronic sequence , including the splice acceptor and donor , and is predicted to cause a change of amino acid sequence after residue 87 and early truncation six amino acids later . Guide RNAs and Cas9 nuclease were introduced into C57BL/6NJ-derived fertilized eggs with well recognized pronuclei . Embryos were transferred to pseudopregnant females . Correctly targeted pups were identified by sequencing and further bred to C57BL/6NJ ( Stock No . 005304 ) to develop the colony . Once the stock mice arrived at the Cleveland Clinic heterozygous mice were intercrossed to generate additional heterozygous ( Tmc4+/- ) and homozygous wild type ( Tmc4+/+ ) and knockout ( Tmc4-/- ) progeny . In addition to wild type mice on the mixed C57BL/6NJ background , we also studied parallel wild type mice on a pure C57BL/6J strain to increase sample size of experimental controls . Experimental mice were fed a high fat diet for 2 weeks prior to necropsy for analysis of hepatic steatosis . To keep results consistent , the vast majority of experimental mice were fasted for 4 hr ( from 9:00 a . m . to 1:00 p . m . ) prior to necropsy . For the fasting versus fed experiments , fed mouse tissue were collected at the beginning of the light cycle ( 7:00 a . m . ) in ad libitum fed mice , whereas the fasted group had food removed at 7:00 p . m . and were necropsy after a 12 hr fast ( 7:00 a . m . ) . At necropsy , all mice were terminally anesthetized with ketamine/xylazine ( 100–160 mg/kg ketamine-20–32 mg/kg xylazine ) , and a midline laparotomy was performed . Blood was collected by heart puncture . Following blood collection , a whole body perfusion was conducted by puncturing the inferior vena cava and slowly delivering 10 ml of saline into the heart to remove blood from tissues . Tissues were collected and immediately snap frozen in liquid nitrogen for subsequent biochemical analysis or fixed for morphological analysis . To determine the level of hepatic injury in mice fed chow and HFD with ASO treatment for 20 weeks , plasma was used to analyze aminotransferase ( AST ) and alanine aminotransferase ( ALT ) levels using enzymatic assays ( Sekisui Diagnostics , Lexington , MA , USA ) . Quantitation of lean and fat mass were done using an EchoMRITM-130 Body Composition Analyzer ( EchoMRI International ) . Hepatic LDs were isolated by sucrose gradient centrifugation as we have previously described ( Ferguson et al . , 2017 ) . Approximately 100 mg of tissue was minced with a razor blade on a cold surface . Minced tissue was transferred to a Potter-Elvehjem homogenizer , and then 200 μl of 60% sucrose was added to the tissue sample and incubated on ice for 10 min . Next , 800 μl of lysis buffer was added and mixed , and then incubated on ice for 10 min . Samples were homogenized with five strokes of a Teflon pestle and transferred to a 2 ml centrifuge tube . Lysis buffer ( 600 μl ) was carefully layered on top of homogenate and centrifuged for 2 hr at 20 , 000 g at 4°C . The tube was then frozen at −80°C and cut at the 1 , 000 μl mark . The bottom piece of the centrifuge tube contained the non-LD fraction , which was allowed to thaw before being transferred to a new tube . The LD fraction was collected by cutting an ∼4–6 mm piece from the top of the ice cylinder and placing it in a new 2 ml tube . To increase the purity of the LD fraction , this process was repeated once more . Briefly , 200 μl of 60% sucrose was added to the LD fraction . Next , 800 μl of lysis buffer was added and mixed followed by careful layering with 600 μl of lysis buffer and then centrifugation for 2 hr at 20 , 000 g at 4°C . After freezing at −80°C , the tube was cut and the LD fraction was collected by cutting an ∼4–6 mm piece from the top of the ice cylinder and placing it in a new tube . Protein analysis was performed using the modified Lowry assay , as previously described ( Ferguson et al . , 2017 ) , and Western blotting was performed as described below . MBOAT7 substrate ( lysophosphatidylinositol ) and product ( phosphatidylinositol ) lipids in the LD fraction were extracted and quantified using the targeted LC-MS/MS methods described below . A targeted lipidomic assay for LPI and PI lipids was developed using HPLC on-line electrospray ionization tandem mass spectrometry ( LC/ESI/MS/MS ) . Plasma and tissues ( liver , brain , pancreas ) from mice fed chow or HFD with ASO injections for 20 weeks were analyzed for precise detection of LPI and PI species ( Figure 6; Figure 6—figure supplements 1–4 ) . Moreover , plasma LPI levels were measured over time after intraperitoneal injections of 18:0 and 18:1 LPI ( Figure 8—figure supplement 1 ) . Standard Solutions: The standards used in this assay were all purchased from Avanti Polar Lipids ( LPI-16:0 , LPI-18:0 , LPI-18:1 , LPI-20:4 , PI-38:4 ) . Internal standards used for these analyses were LPI-17:1 , PI-34:1-d31; all of which were purchased from Avanti Polar Lipids . Standard LPI and PI species at concentrations of 0 , 5 , 20 , 100 , 500 and 2000 ng/ml were prepared in 90% methanol containing two internal standards at the concentration of 500 ng/ml . The volume of 5 μl was injected into the Shimadzu LCMS-8050 for generating the internal standard calibration curves . HPLC Parameters: A silica column ( 2 . 1 × 50 mm , Luna Silica , 5 μm , Phenomenex ) was used for the separation of PI and LPI species . Mobile phases were A ( water containing 10 mM ammonium acetate ) and B ( acetonitrile containing 10 mM ammonium acetate ) . Mobile phase B at 95% was used from 0 to 2 min at the flow rate of 0 . 3 ml/min and then a linear gradient from 95% B to 50% B from 2 to 8 min , kept at 50% B from 8 to 16 min , 50% B to 95% B from 16 to 16 . 1 min , kept 95% B from 16 . 1 to 24 min . Mass Spectrometer Parameters: The HPLC eluent was directly injected into a triple quadrupole mass spectrometer ( Shimadzu LCMS-8050 ) and the analytes were ionized ( ESI negative mode ) . Analytes were quantified using Selected Reaction Monitoring ( SRM ) and the SRM transitions ( m/z ) were 571 → 255 for LPI-16:0 , 599 → 283 for LPI-18:0 , 597 → 281 for LPI-18:1 , 619 → 303 for LPI-20:4 , 885 → 241 for PI-38:4 , 583 → 267 for internal standard LPI-17:1 , and 866 → 281 for internal standard PI-34:1-d31 . Data Analysis: Software Labsolutions LCMS was used to get the peak area for both the internal standards and LPI and PI species . The internal standard calibration curves were used to calculate the concentration of LPI and PI species in the samples . All plasma LPI and PI species were normalized to the PI-34:1-d31 internal standard , while all tissue LPI species were normalized to the 17–1 LPI internal standard and all tissue PI species were normalized to the PI-34:1-d31 internal standard . To more broadly examine the molecular lipid species in Mboat7 knockdown livers we utilized a shotgun lipidomics method for semi-quantitation of multiple lipid species using a method we have previously described ( Gromovsky et al . , 2018 ) . All the internal standards were purchased from Avanti Polar Lipids , Inc ( 700 Industrial Park Drive , Alabaster , Alabama 35007 , USA ) . Ten internal standards ( 12:0 diacylglycerol , 14:1 monoacylglycerol , 17:0 lysophosphatidylcholine , 17:0 phosphatidylcholine , 17:0 phosphatidic acid; 17:0 phosphatidylethanolamine , 17:0 phosphatidylglycerol , 17:0 sphingomyelin , 17:1 lysosphingomyelin , and 17:0 ceramide ) were mixed together with the final concentration of 100 μM each . Total hepatic lipids were extracted using the method of Bligh and Dyer , with minor modifications . In brief , 50 μL of 100 μM internal standards were added to tissue homogenates and lipids were extracted by adding by adding MeOH/CHCl3 ( v/v , 2/1 ) in the presence of dibutylhydroxytoluene ( BHT ) to limit oxidation . The CHCl3 layer was collected and dried under N2 flow . The dried lipid extract was dissolved in 1 ml the MeOH/CHCl3 ( v/v , 2/1 ) containing 5 mM ammonium acetate for injection . The solution containing the lipid extract was pumped into the TripleTOF 5600 mass spectrometer ( AB Sciex LLC , 500 Old Connecticut Path , Framingham , MA 01701 , USA ) at a flow rate of 40 μL/min for 2 min for each ionization mode . Lipid extracts were analyzed in both positive and negative ion modes for complete lipidome coverage using the TripleTOF 5600 System . Infusion MS/MSALL workflow experiments consisted of a TOF MS scan from m/z 200–1200 followed by a sequential acquisition of 1001 MS/MS spectra acquired from m/z 200 to 1200 . The total time required to obtain a comprehensive profile of the lipidome was approximately 10 min per sample . The data was acquired with high resolution ( >30000 ) and high mass accuracy ( ~5 ppm RMS ) . Data processing using LipidView Software identified 150–300 lipid species , covering diverse lipids classes including major glycerophospholipids and sphingolipids . The peak intensities for each identified lipid , across all samples were normalized against an internal standard from same lipid class for the semi-quantitation purpose . In order to quantitate the abundance of lipid mediators generated from arachidonic acid , livers collected from C57BL/6 mice exposed to each treatment condition were subjected to solid phase extraction ( SPE ) and targeted liquid chromatography-tandem mass spectrometry ( LC-MS/MS ) as previously described in detail ( Dalli et al . , 2018 ) Briefly , tissue was minced in ice-cold methanol containing internal deuterated standards ( d4-LTB4 , d4-PGE2 and d5-LXA4 ) used to assess extraction recovery . Samples were loaded on C18 SPE cartridges and eluted methyl formate fractions were concentrated under a gentle stream of N2 gas . Samples were then resuspended in methanol:water ( 50:50 ) and analyzed using a high performance liquid chromatograph ( HPLC , Shimadzu ) coupled to a QTrap5500 mass spectrometer ( AB Sciex ) operating in negative ionization mode . Lipid mediators were identified using specific multiple reaction monitoring transitions , retention time , and diagnostic fragmentation spectra as compared with authentic standards . Quantification of lipid mediators was accomplished using calibration curves constructed with external standards for each mediator . RNA was isolated via RNAeasy lipid tissue mini kit ( Qiagen ) from livers in which mice were fed a high-fat diet and treated with ASOs for 20 weeks . RNA samples were checked for quality and quantity using the Bio-analyzer ( Agilent ) . RNA-SEQ libraries were generated using the Illumina mRNA TruSEQ Directional library kit and sequenced using an Illumina HiSEQ4000 ( both according to the Manufacturer’s instructions ) . RNA sequencing was performed by the University of Chicago Genomics Facility . Raw sequence files will be deposited in the Sequence Read Archive before publication ( SRA ) . Single-ended 50 bp reads were trimmed with Trim Galore ( v . 0 . 3 . 3 , http://www . bioinformatics . babraham . ac . uk/projects/trim_galore ) and controlled for quality with FastQC ( v0 . 11 . 3 , http://www . bioinformatics . bbsrc . ac . uk/projects/fastqc ) before alignment to the Mus musculus genome ( Mm10 using UCSC transcript annotations downloaded July 2016 ) . Reads were aligned using STAR in single-pass mode ( v . 2 . 5 . 2a_modified , https://github . com/alexdobin/STAR ) ( Dobin et al . , 2013 ) with standard parameters but specifying ‘--alignIntronMax 100000 --quantMode GeneCounts’ . Overall alignment ranged from 88–99% with 61–75% mapping uniquely . Transcripts with fewer than one mapped read per million ( MMR ) in all samples were filtered out before differential expression ( DE ) analysis . The filtering step removed 12 , 692/24 , 411 transcripts ( 52% ) . Raw counts were loaded into R ( http://www . R-project . org/ ) ( R Development Core Team , 2015 ) and edgeR ( Robinson et al . , 2010 ) was used to perform upper quantile , between-lane normalization , and DE analysis . Values generated with the cpm function of edgeR , including library size normalization and log2 conversion , were used in figures . Heat maps were generated using pheatmap ( Kolde , 2015 ) . DAVID ( v . 6 . 8 ) ( Huang et al . , 2009 ) was used to identify enriched functional annotations in DE gene ID lists relative to the set of ‘expressed’ genes ( defined as having a median count across samples > 1 read per million mapped ) . Functional annotation to gene ontology was also performed using Ingenuity-IPA software ( Ingenuity Systems , Inc Redwood City , CA . ) as previously described ( Thomas and Bonchev , 2010 ) . Glucose and insulin tolerance tests were conducted as previously described ( Warrier et al . , 2015; Brown et al . , 2010; Brown et al . , 2008a; Brown et al . , 2008b; Izem and Morton , 2009; Helsley et al . , 2016 ) in male mice following 20 weeks of concurrent chow and HFD-feeding with ASO treatment . During the glucose tolerance test plasma insulin and C-peptide levels were measured via ELISA following manufacturer’s instructions ( Millipore ) . Total plasma triacylglycerol levels ( L-Type TG M , Wako Diagnostics ) and total plasma cholesterol levels were quantified enzymatically ( Infinity Cholesterol Reagent , Thermo/Fisher ) . The distribution of cholesterol across lipoprotein classes was performed by fast-protein liquid chromatography ( FPLC ) using tandem superose-6 HR columns coupled with an online enzymatic cholesterol quantification as previously described ( Warrier et al . , 2015; Brown et al . , 2010; Brown et al . , 2008a; Brown et al . , 2008b; Izem and Morton , 2009; Helsley et al . , 2016 ) . WT mice were fed a HFD for 20 weeks while injected with control or Mboat7 targeted ASOs . At 20 weeks , mice were fasted overnight ( 7 p . m . to 7 a . m . ) and were injected with saline or 0 . 35 U/kg insulin into the portal vein for 5 min , as previously described ( Lord et al . , 2016 ) . After 5 min , mice were euthanized and tissues were collected for protein isolation and western blot analysis . Control and Mboat7 ASO-treated mice were fasted overnight , and the following morning rec were injected with 500 mg/kg Triton WR-1339 ( tyloxapol , Sigma ) . and blood was collected every 90 min thereafter . To measure hepatic VLDL secretion , serum TG levels were measured using enzymatic biochemical assays at each time point as described below . Extraction of liver lipids and quantification of total plasma and hepatic triglycerides , cholesterol , and cholesterol esters was conducted using enzymatic assays as described previously ( Warrier et al . , 2015; Brown et al . , 2010; Brown et al . , 2008a; Brown et al . , 2008b; Izem and Morton , 2009 ) . [1-14C]-arachidonyl-CoA was obtained from American Radiolabeled Chemicals . Lysophosphatidylinositol ( LPI ) substrates ( 16:0 LPI , 18:0 LPI , and 18:1 LPI ) and lipid standards used in the enzyme assays were obtained from Avanti Polar Lipids . Liver microsomes isolated from Mboat7 ASO and control ASO treated mice on both chow and high fat diet was used to measure LPIAT activity . The assay buffer contained 50 mM Tris-HCL ( pH 8 . 0 ) , 150 mM NaCl , 50 μM 18:0-LPI , 20 μM [1-14C]arachidonyl-CoA ( 0 . 025 µCi ) , and 15 µg of the microsomes in a total volume of 100 µL . Substrate was prepared in CHAPS ( 0 . 01 mM final concentration ) . The assay mixture was incubated for 30 min at 37°C , and the reaction was stopped by the addition of 1:2:1 ( v/v/v ) chloroform:methanol:2% orthophosphoric acid . The lipids were extracted and separated on a silica–TLC plate using chloroform/acetone/acetic acid/methanol/water ( 50:20:15:10:5 , v/v ) as the solvent system . The individual lipid molecules were identified by migration with respect to standards . Enzymatic products were monitored by autoradiogram , corresponding spots were scraped from the TLC plate , and the radioactivity was quantified with a liquid scintillation counter . Whole tissue homogenates were made from tissues in a modified RIPA buffer as previously described ( Warrier et al . , 2015; Brown et al . , 2010; Brown et al . , 2008a; Brown et al . , 2008b; Izem and Morton , 2009 ) , and protein was quantified using the bicinchoninic ( BCA ) assay ( Pierce ) . Proteins were separated by 4–12% SDS-PAGE , transferred to polyvinylidene difluoride ( PVDF ) membranes , and proteins were detected after incubation with specific antibodies as previously described ( Warrier et al . , 2015; Brown et al . , 2010; Brown et al . , 2008a; Brown et al . , 2008b; Izem and Morton , 2009 ) . Quantification of blots was performed using ImageJ software ( National Institute of Health ) . Information of antibodies is provided in the key resource table . Tissue RNA extraction was performed as previously described for all mRNA analyses ( Warrier et al . , 2015; Brown et al . , 2010; Brown et al . , 2008a; Brown et al . , 2008b; Izem and Morton , 2009 ) . qPCR analyses were conducted as previously described ( Warrier et al . , 2015; Brown et al . , 2010; Brown et al . , 2008a; Brown et al . , 2008b ) , and mRNA expression levels were calculated based on the ΔΔ-CT method . qPCR was conducted using the Applied Biosystems 7500 Real-Time PCR System . Primers used for qPCR are listed in the Key Resource Table . Hematoxylin and eosin ( H and E ) staining of paraffin-embedded liver sections was performed as previously described ( Warrier et al . , 2015; Brown et al . , 2010; Brown et al . , 2008a; Brown et al . , 2008b; Izem and Morton , 2009 ) . Histopathologic evaluation was scored in a blinded fashion by a board-certified pathologist with expertise in gastrointestinal/liver pathology ( Daniela S . Allende – Cleveland Clinic ) . After 20 weeks of HFD-feeding and ASO injections , livers were excised , washed with 1X PBS , and immediately placed into RPMI with Type IV Collagenase ( Sigma Aldrich , St . Louis , Missouri , Lot# 087K8630 ) and DNase I ( Roche , Mannheim , German ) for 45 min at 37° C . Digested clumps of liver were pressed through a 70 um strainer and washed with RPMI with 10% FBS . Cells were centrifuged at 50 g for 10 min; supernatant was then centrifuged at 50 g for 7 min . To pellet the NPC fraction , cells were centrifuged at 300 g for 7 min . Cells were resuspended in BD Pharm Lyse ( BD Biosciences , San Jose , California ) for 5 min on ice . Cells were washed with RPMI with 10% FBS and centrifuged at 300 g for 10 min . ( Gibbons MA , American Journal of Respiratory and Critical Care Medicine , 184 , 2011 ) . For flow cytometry , cells were resuspended in FACS buffer ( 1x PBS , 1% BSA , 0 . 05% sodium azide ) . Cells were aliquoted into 96 well plates at a concentration of ~1×106 cells/mL . Cells were centrifuged at 830 x g for 4 min , resuspended in 50 µl FACS buffer containing 0 . 5 ug of Fcγ Block ( clone 93 , eBioscience , San Diego , California ) , and incubated for 15 min at room temperature . After blocking , cells were stained with a fluorochrome-conjugated antibody panel CD206 , Ly6c , CD3 , CD4 , CD8 , CD11b , and CD11c . – all described in the Key Resource Table ) for 30 min at 4° C in the dark . Cells were washed and centrifuged at 830 x g for 4 min twice with FACS buffer . Stained cells are resuspended in 200 µl of 1% paraformaldehyde and kept in the dark at 4°C overnight . Stained cells were centrifuged at 830 x g for 5 min . Stained cells were resuspended in 300 µl of FACS buffer , and data were collected on a LSRII flow cytometer ( Becton Dickinson Immunocytometry systems , Mountain View , CA ) . Data collected on the LSRII were analyzed using FlowJo software ( Tree Star , Inc , Ashland Oregon ) . Mycoplasma-tested hepatocellular carcinoma cells ( Huh7 ) were cultured under standard conditions in Dulbecco-modified Eagle's minimum essential medium ( D-MEM ) ( GIBCO , Life Technologies , Carlsbad , CA ) supplemented with 10% fetal bovine serum ( FBS , GIBCO ) , 1% l-glutamine , 1% penicillin-streptomycin and 1% nonessential amino acids in a 5% CO2-humidified chamber at 37°C . CRISPR-Cas9 genome editing was accomplished using methods previously described ( Ran et al . , 2013 ) . MBOAT7 sgRNAs were designed by an online tool ( https://www . benchling . com/ ) and cloned into the Lenti-CRISPER v2 vector ( Addgene ( Ran et al . , 2013 ) with D10A nickase version of Cas9 ( Cas9n ) ) . Primers used in this study are listed in Supplementary Table X . MBOAT7 KO cell lines were generated following lentiviral transduction of the Lenti-CRISPER v2-Cas9 D10A- MBOAT7 sgRNA in Huh7 cells . MBOAT7 KO single cell clones were isolated and expanded following FACS sorting . MBOAT7 KO cells were validated by analyzing the expression of MBOAT7 by real-time PCR and Western blot . Primers used for gene editing are shown in the key resource table with the following primer names: MBOAT7-E5-Nick-5F , MBOAT7-E5-Nick-5R , MBOAT7-E5-Nick-3F , MBOAT7-E5-Nick-3R . Complete fatty acid oxidation to CO2 was measured using methods we have previously described ( Brown et al . , 2003 ) . In brief control and MBOAT7Δ HUH7 cells were grown to a confluency of 70–80% in DMEM . Upon reaching confluence , one million cells were seeded for both WT-HUH7 and MBOAT7Δ ( n = 3 ) for 24 hr . After 24 hr the medium was replaced with fresh serum free DMEM medium with 0 . 4 µci of [1-14C]-palmitic acid in 0 . 3% BSA/100 µM cold palmitate to each plate , and the plate was quickly placed in an airtight CO2 collection chamber ( 60 ml Nalgene jar with a fitted rubber stopper , and hanging center-well collection bucket containing What-man filter paper soaked with 50 μl benzothonium hydroxide ) . Substrate incubation was carried out with different time points ( 0 min , 30 min , 60 min and 120 min ) at 37°C . Following each incubation , the reaction was terminated by the addition of 100 µL of 0 . 5 M H2SO4 to the cells using syringe injection . Liberation of [14C]-CO2 was allowed to proceed for an additional 30 min , and then the center-well collection bucket was cut out of the collection chamber and delivered to a liquid scintillation vial . Production of [14C]-CO2 from [14C]-palmitic acid was determined by liquid scintillation counter . Radiation count was normalized to amount of protein , as quantified by BCA assay ( Pierce ) . Measurement of neutral lipid mobilization was accomplished using a pulse-chase approach essentially as described previously ( Brasaemle et al . , 2000 ) . Both control and MBOAT7Δ Huh7 cells were grown till 80% confluent , one million cells were seeded in 6-well plates ( WT-Huh7 and MBOAT7 KO , the next day fresh DMEM medium was added and the cells were incubated with 100 µM cold oleate complexed to bovine serum albumin + 1 µCi [3H]-cholesterol + 0 . 5 µCi [14C]-oleate for 24 hr . After 24 hr of pulse labeling-cells to reach steady state , cells were chased for different time points ( 0 , 1 hr , 2 hr , 4 hr , 8 hr , 24 , 48 hr ) in media containing 6 µM triacsin C ( from 1 mg/ml stock in Me2SO ) without supplemental fatty acids to examine turnover without the re-esterification ( triacsin C blocks acyl-CoA synthetase activity ) . Cells were rinsed with phosphate buffered saline and harvested by scraping at various times; lipids were extracted as previously described and separated by thin layer chromatography using hexane:diethyl ether:acetic acid ( 80:20:1 ) as a solvent system . Triacylglycerol ( TAG ) and cholesterol ester spots were scraped off the plate , and the incorporation of radioactivity into TAG was determined by liquid scintillation counter . Radiation count was normalized to amount of protein , as quantified by BCA assay ( Pierce ) . Measurement of de novo lipogenesis rates was accomplished by tracing [14C]-acetate into [14C]-triacylglycerol essentially as described previously ( Brasaemle et al . , 2000 ) . Confluent cells of both wild type and MBOAT7Δ Huh7 cells were maintained in DMEM medium , from these 1 million cells of both WT and MBOAT7Δ were seeded to 6-well plates for 12 hr , after 12 hr the medium was replaced by serum free DMEM . To measure triacylglycerol synthesis , without the complication of parallel triacylglycerol hydrolysis , cells were treated with lipase inhibitors of triacylglycerol and cholesterol ester such as diethyl-p-nitrophenyl phosphate ( E-600 , 500 µM ) and diethylumelliferyl phosphate ( DEUP , 0 . 35 µM ) for 1 hr . After 1 hr of lipase inhibitor treatment , the cells were incubated with human recombinant insulin ( 100 nM ) as a stimulator for de novo lipogenesis for 30 min . After 30 min of insulin stimulation , the cells were incubated with 1 µCi [14C]acetate and cells were harvested at various time points ( 30 min , 60 min , 120 min and 240 min ) post substrate addition . From each time point cells were rinsed with PBS ( twice ) , lipids were extracted as described above and separated by thin layer chromatography using using hexane:diethyl ether:acetic acid ( 80:20:1 ) as a solvent system . Triacylglycerol ( TAG ) spot was scraped off of the plate , and the incorporation of [14C]-acetate into [14C]-triacylglycerol was determined by liquid scintillation counter . Radiation count was normalized to amount of protein , as quantified by BCA assay ( Pierce ) . In preparation for fluorescence microscopy , 5 × 104 cells of WT and MBOAT7Δ were plated onto 22 mm square glass coverslips in 4-well cell culture chamber slide containing growth media supplemented with 400 µM oleic acid . After 24 hr , medium was removed , washed with PBS ( twice ) and cells were fixed in 4% paraformaldehyde for 30 min . After 30 min of incubation , fixed cells were washed two times with PBS and then BODIPY ( 1 mg/mL ) was added incubated for additional 30 min in dark . After incubation the cells were washed with PBS ( twice ) followed by mounting the slides with ProLong Gold antifade reagent with DAPI ( 4′ , 6- Diamidine-2′-phenylindole dihydrochloride ) . Images were acquired using a Leica DMIRB upright microscope ( Leica Microsystems , GmbH , Wetzlar , Germany ) equipped with a Retiga SRV camera and QCapture Pro software ( QImaging , Surrey , BC , Canada ) . Microsome were isolated from single clones of MBOAT7 Δ and WT-HUH7 cells using microsome isolation kit from Abcam . Proteins were quantified from isolated microsome using BCA method , 30 μg of protein were loaded on SDA-PAGE gels , the expression of MBOAT7 is confirmed by using MBOAT7-rat primary antibody ( 1:1000 ) and anti-rat secondary antibody ( 1:5000 ) . All experiments consisted of a minimum of three replicates and data are presented as mean ± SD . GraphPad Prism 8 . 1 . 1 was used for data analysis . Two‐way ANOVAs with Bonferroni's multiple comparison tests were used to determine significant differences ( p-value<0 . 0002 ( *** ) , <0 . 0001 ( **** ) . | Non-alcoholic fatty liver disease , or NAFLD for short , is a medical condition that develops when the liver accumulates excess fat . It can lead to complications such as diabetes and liver scarring . In humans , mutations that inactivate a protein called MBOAT7 increase the risk of fat accumulating in the liver . Genetic studies suggest that low levels of MBOAT7 in a human’s liver cells increase the severity of NAFLD . Yet the links between MBOAT7 , NAFLD and obesity are not well understood . Helsley et al . used data from humans and from obese mice that had been fed a high-fat diet to investigate the relationship between NAFLD and MBOAT7 . This revealed that people who are obese have lower levels of MBOAT7 in their livers . Next , obese mice were genetically manipulated to produce less MBOAT7 , which led them to develop more severe NAFLD . Helsley et al . then grew human liver cells in the laboratory and lowered their levels of MBOAT7 , which led to excess fat accumulating in the cells . This increase in fat accumulation was , at least in part , due to how these cells metabolize fats when MBOAT7 is reduced: they start making more new fats and consume fewer lipids to produce energy . These findings provide a link between obesity and liver damage in both humans and mice , and show how a decrease in MBOAT7 levels causes changes in fat metabolism that could lead to NAFLD . The results could drive new approaches to treating liver damage in patients with mutations in the gene that codes for MBOAT7 . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"genetics",
"and",
"genomics"
] | 2019 | Obesity-linked suppression of membrane-bound O-acyltransferase 7 (MBOAT7) drives non-alcoholic fatty liver disease |
In embryonic stem cells ( ESCs ) , a core transcription factor ( TF ) network establishes the gene expression program necessary for pluripotency . To address how interactions between four key TFs contribute to cis-regulation in mouse ESCs , we assayed two massively parallel reporter assay ( MPRA ) libraries composed of binding sites for SOX2 , POU5F1 ( OCT4 ) , KLF4 , and ESRRB . Comparisons between synthetic cis-regulatory elements and genomic sequences with comparable binding site configurations revealed some aspects of a regulatory grammar . The expression of synthetic elements is influenced by both the number and arrangement of binding sites . This grammar plays only a small role for genomic sequences , as the relative activities of genomic sequences are best explained by the predicted occupancy of binding sites , regardless of binding site identity and positioning . Our results suggest that the effects of transcription factor binding sites ( TFBS ) are influenced by the order and orientation of sites , but that in the genome the overall occupancy of TFs is the primary determinant of activity .
Enhancers are composed of combinations of transcription factor binding sites ( TFBS ) . An important question is: to what extent do TFBS act independently within enhancers and to what extent do specific interactions between transcription factors ( TF ) underlie enhancer function ? Independence suggests a modular genome in which the effects of multiple binding sites are predictable from their individual effects . Interactions , such as cooperativity between TFs , cause the effect of multiple TFBS to be more ( or less ) than the combination of their individual effects . Constructing models that predict the expression of genes based on the TFBS composition of their surrounding regulatory DNA will require understanding the degree to which sites function independently and how interactions between sites contribute to the activity of regulatory sequences . The extent to which TFs function either independently or through interactions should be reflected in the cis-regulatory grammar of TFBS , defined as the ways that the order , orientation , spacing , and affinity of binding sites impact the activity of enhancers . If TFs function independently then we do not expect strong constraints on the positioning of their binding sites within regulatory elements . If TFs function mostly through interactions with other TFs that require a precise geometry , then we expect strong biases in the positioning of TFBS within regulatory elements . At least three models make predictions of how grammar might influence enhancer activity , the billboard model , the enhanceosome model , and the TF collective model ( Kulkarni and Arnosti , 2003; Spitz and Furlong , 2012 ) . The enhanceosome model posits extensive interactions between bound TFs , resulting in a strict grammar in which only precise positioning of TFBS activate target genes . The enhanceosome model is supported by structural studies of the IFN-β enhancer , where a specific order and spacing of TFBS is required to activate expression ( Panne , 2008; Yie et al . , 1999 ) . In contrast , the billboard model posits a more flexible grammar , where enhancers tolerate changes to the order , spacing , or orientations of TFBS with little change to target gene expression ( Giorgetti et al . , 2010; Kulkarni and Arnosti , 2003 ) . In the billboard model bound TFs function in a largely independent manner . This model was proposed to explain binding site turnover in developmental enhancers and functional conservation of enhancer activity between species despite sequence divergence ( Hare et al . , 2008a; Hare et al . , 2008b; Ludwig et al . , 2000; Visel et al . , 2009 ) . In the TF collective model , specific TFs must be recruited to enhancers but can be recruited either by direct contact with DNA or indirectly through other TFs ( Junion et al . , 2012; Spitz and Furlong , 2012; Uhl et al . , 2016 ) . In the collective model no specific TFBS is required for activity even though the recruitment of individual TFs might be . TFs may function independently in some contexts and may engage in interactions in other contexts . The billboard , enhanceosome , and collective models differ in the importance the precise arrangements of TFBS play in setting the activities of enhancers , and control of gene expression likely incorporates aspects of all three models . Quantifying the extent to which grammar influences activity in different contexts is an important step toward producing more predictive models of gene expression . We and others have used mouse embryonic stem cells ( mESCs ) as a system for studying cis-regulatory grammar and cooperative interactions between the pluripotency factors POU5F1 ( OCT4 ) , SOX2 , ESRRB , and KLF4 ( Dunn et al . , 2014; Fiore and Cohen , 2016; Williams et al . , 2004 ) . The pluripotency factors are a core set of TFs that maintain pluripotency in mESCs and are sufficient to induce pluripotency in terminally differentiated cells ( Feng et al . , 2009; Liu et al . , 2008; Niwa , 2014; Takahashi and Yamanaka , 2006; Zhang et al . , 2008 ) . The pluripotency TFs activate self-renewal genes and repress genes that promote differentiation ( Chambers and Tomlinson , 2009 ) . Based on known physical and genetic interactions , as well as genome-wide binding assays , multiple interacting TFs specify target gene expression in mESCs ( Huang et al . , 2009; Niwa , 2014; Reményi et al . , 2004; Reményi et al . , 2003; Williams et al . , 2004 ) . However , it remains unclear how pluripotency TFs collaborate to drive-specific patterns of gene expression in ESCs , and what role , if any , is played by TFBS grammar in determining specificity in the genome ( Chambers and Tomlinson , 2009; Chen et al . , 2008b ) . Understanding how these factors combine to regulate their target genes is central to understanding the establishment and maintenance of the pluripotent state . We previously addressed these questions by assaying a set of synthetic cis-regulatory elements that represent a small fraction of the possible arrangements of pluripotency TFBS . We identified some evidence for a grammar that is constrained by TFBS arrangement , including OCT4-SOX2 interactions . However , our previous study lacked sufficient power to detect other interactions ( Fiore and Cohen , 2016 ) . Here , we explore the role of grammar for pluripotency TFBS by assaying an exhaustive set of synthetic cis-regulatory elements , composed of TFBS for SOX2 , OCT4 , KLF4 and ESRRB , as well as a limited set of genomic regulatory sequences with comparable configurations of binding sites . The pattern of expression of synthetic regulatory elements is well predicted by a model that incorporates binding site position . However , despite all genomic sequences overlapping ChIP-seq peaks for at least one of the four pluripotency factors , only about a third of sequences drove reporter gene activity above background levels . Additionally , the positional grammar learned from synthetic sequences performed poorly in predicting the activity of genomic sequences . Genomic sequences appear to also include sequence features that recruit additional TFs , either directly through TF-DNA interactions or possibly indirectly through TF-TF interactions . Our results suggest that in the genome the overall occupancy of TFs is the best predictor of binding site activity . Our results with synthetic elements suggest that other aspects of grammar ( order , orientation ) can tune the activity of sites , but these effects are difficult to observe without direct experimental manipulations . In the genome only the number and affinity of sites shows a correlation with activity .
We designed two reporter gene libraries to explore the role of grammar in regulatory elements controlled by the pluripotency TFs . The first library , synthetic ( SYN ) , contains a set of synthetic combinations of consensus TFBS for OCT4 ( O ) , SOX2 ( S ) , KLF4 ( K ) , and ESRRB ( E ) . We did not include sites for NANOG in our libraries as its position weight matrix ( PWM ) has low information content and is not amenable to a synthetic binding site approach . Nanog also appears to be dispensable for reprogramming terminal cells to a pluripotent state ( Wang et al . , 2013; Wang et al . , 2012; Jauch et al . , 2008; Pan and Thomson , 2007; Takahashi and Yamanaka , 2006 ) . We did not incorporate MYC-binding sites in our libraries because MYC often acts independently of the core pluripotency TFs ( Chen et al . , 2012; Chen et al . , 2008c; Liu et al . , 2008 ) . We designed the SYN library to test how interactions between different TFs ( heterotypic interactions ) determine the activities of regulatory elements . If heterotypic interactions depend on the geometry of TF binding , then the order , orientation , and spacing of sites should influence activity . To test this prediction , we designed the SYN library to assay different orders and orientations of the pluripotency binding sites . The SYN library includes all possible 624 unique combinations of two , three , and four TFBS ( 2-mers , 3-mers , and 4-mers , respectively ) , with each TFBS in either the forward or reverse direction ( Supplementary file 1A ) . Each synthetic element in the SYN library contains no more than one copy of a given TFBS . We chose this library design to focus on heterotypic interactions and to avoid the confounding effects of homotypic interactions , which we examined in detail in a previous study ( Fiore and Cohen , 2016 ) . We embedded each TFBS in a constant 20 bp sequence with fixed spacing between sites to ensure that all the sites sit on the same side of the DNA helix . We avoided varying the length of the spacer sequence between sites because increasing the length of spacer sequences risks introducing cryptic binding sites that confound the results . For each TF , we used a consensus binding site based on its position weight matrix ( PWM ) in the JASPAR database ( Sandelin , 2004; Fiore and Cohen , 2016 ) . We did not vary the predicted affinity of the sites in the SYN library because we could not assay a library large enough to vary the affinity of sites while still testing all possible arrangements of sites . Our rationale was to retain the maximum power to detect the effects of the order and orientation of sites , and this required us to compromise on our ability to detect the effects of the spacing and affinity of sites . The highly controlled nature of the SYN library provides maximum power to detect interactions mediated by the order and orientation of sites . The second library includes sequences from the mouse genome that match , as best as possible , members of the SYN library . Using the same PWMs used to design the SYN library , we scanned the mouse genome for combinations of the TFBS for O , S , K , and E within 100 bp of regions bound by any of the four pluripotency TFs in E14 mESCs as measured by ChIP-seq ( Fiore and Cohen , 2016; Bailey et al . , 2009; Chen et al . , 2008c ) . We chose genomic sequences that contain one and only one binding site that scores above the PWM threshold for each factor to mimic the composition of the SYN library . We identified few clusters that included all four binding sites ( <70 ) . We therefore selected 407 genomic sequences with three pluripotency TFBS that could be compared to the exhaustive set of synthetic 3-mer elements . The resulting genomic wild-type library ( gWT ) is composed of 407 unique genomic sequences with combinations of any three of the four TFBS , with each site represented no more than once per sequence ( Materials and methods , Supplementary file 1E-F ) . Although these sequences differ from SYN elements in the individual site affinities , spacings between TFBS , as well as intervening sequence composition , our expectation was that the gWT sequences would test how well interactions learned from the SYN library apply to genomic sequences . To confirm that the activity of the gWT sequences depends on the presence of pluripotency TFBS , we generated matched genomic mutant sequences ( gMUT ) in which all three of the identified pluripotency TFBS were mutated by changing two positions in each TFBS from the highest information content base to the lowest information base according to the PWM ( Figure 1—figure supplement 1 ) . The final gMUT sequences lack detectable TFBS for O , S , K , or E when rescanned with the threshold used to select the gWT sequences . The combined gWT/gMUT library allows us to quantify the contributions of the pluripotency sites to regulatory activity , as well as sample configurations of pluripotency TFBS from the genome that may provide insight into grammar for these sequences . We assayed the cis-regulatory activity of the SYN and gWT/gMUT libraries in mESCs using a plasmid-based Massively Parallel Reporter Assay ( MPRA ) ( Kwasnieski et al . , 2012 ) . Each unique library member described above is present eight times with a different unique sequence barcode ( BC ) in its 3’ UTR ( Fiore and Cohen , 2016 ) . The elements were placed directly upstream of a minimal promoter , mirroring classical tests of enhancer activity . The assay does not , however , test whether elements can function as long-range enhancers . To determine the relative activity of each sequence compared to the minimal promoter included in each construct , we included copies of plasmids with only the minimal promoter paired with over a hundred unique BCs in each library ( Materials and methods ) . Our measurements were highly reproducible between biological replicates , with R2 between 0 . 98 and 0 . 99 for replicates of the SYN library and 0 . 96–0 . 98 for the gWT/gMUT library , and are not driven by abundance biases in the library ( Figure 1—figure supplement 2 ) . After thresholding on DNA and RNA counts , we recovered reads for 100% ( 624/624 ) of our SYN elements and 99% ( 403/407 ) of paired gWT/gMUT sequences . The high concordance between replicates and simultaneous sequencing of the two libraries allowed us to make quantitative comparisons , both within and between libraries . TFBS in synthetic regulatory elements make strong independent contributions to expression . Most synthetic elements drive expression over basal activity regardless of the number , order , or orientation of sites within the element ( Figure 1A ) . Of all SYN elements , 77% ( 6% of 2-mers , 66% of 3-mers , 92% of 4-mers ) were statistically different from basal levels in all three replicates after correcting for multiple hypothesis testing ( Wilcoxon rank-sum test; Bonferroni correction , n = 637; p-values reported in Supplementary file 1C ) . In most cases , three or four consensus binding sites are sufficient to increase expression above basal levels , which suggests strong independent contributions of TFBS to the activity of synthetic elements . Synthetic elements with more binding sites generally drive higher expression than elements with fewer binding sites , supporting the idea that TFBS can contribute to expression in an independent and additive manner . However , the wide range of expression levels observed from different 4-mer elements must be due to the arrangement of the TFBS , as site number , identity , and affinity are fixed . The strong positive effect of adding sites demonstrates an independent effect of TFBS , while the diversity of expression among elements with the same number of sites reveals that grammar can quantitatively modulate activity . In contrast to the synthetic elements , most genomic sequences in the gWT library did not exhibit regulatory activity above basal levels . Only 28% ( 113/403 ) of wild type genomic sequences were statistically different from basal levels in all three replicates ( p<0 . 05 , Wilcoxon rank-sum test; Bonferroni correction , n = 403; p-values reported in Supplementary file 1H ) . This low fraction of active gWT sequences is consistent with observations from functional tests of genomic sequences bound by key TFs in other cell types ( Fisher et al . , 2012; Grossman et al . , 2017; White et al . , 2013 ) . The difference between the SYN and gWT libraries is that the surrounding sequence context in which the pluripotency sites occur in the gWT library varies much more than in the SYN library , and these contextual differences appear to have strong effects on the pluripotency sites . In most cases , the effect of sequence context in the gWT library was strong enough to suppress the independent contributions of the binding sites to activity . For genomic sequences that were statistically different from basal , 99% ( 112/113 ) have a significant difference between matched gWT and gMUT sequences ( Figure 1B; p<0 . 05 , Wilcoxon rank-sum test; Bonferroni correction , n = 403; p-values reported in Supplementary file 1H ) , indicating that the activity of these genomic sequences depends on one or more of the pluripotency TFBS . Our observation that the presence of high-quality pluripotency TFBS is generally insufficient to drive expression demonstrates that binding sites must be presented in the proper surrounding sequence context in order to generate a functional regulatory element . While the overall pattern of expression of SYN elements supports strong independent contributions from binding sites , direct comparisons of different TFBS configurations also support a role for interactions between factors . Pairwise comparisons between 3-mers and their matched 4-mers that include one additional site at either the 5’ or 3’ end , reveal that the position of the extra site can strongly influence expression . For example , the O-K-E 3-mer and the matched O-K-E-S 4-mer drive indistinguishable expression , while the matched S-O-K-E 4-mer drives one of the highest expression levels in the SYN library ( Figure 2A ) . Other examples are consistent with either strong position dependence or both position and orientation dependence ( Figure 2—figure supplement 1A–B ) . Taken together , these results show that when an additional TFBS is added to an existing synthetic element , the position and orientation of the new site can have large effects on activity . Synthetic elements appear to follow a grammar that includes some position specific interactions between TFBS . The ten highest expressing elements in the SYN library all have S and O sites next to each other and in the first two positions ( Figure 2B ) , while the ten lowest expressing 4-mers have a strong bias for O and S in the last two positions ( Figure 2C ) . The 10 highest expressing 4-mers all have K followed by E in the last two positions , while the lowest expressing 4-mers tend to have K and E in the first two positions . The fourth position can have an especially large effect on expression . In the highest 25% of 4-mers S is depleted ( 0/96 ) in the fourth position ( Figure 2D ) , while in the lowest 25% E is virtually depleted ( 1/96 ) in the fourth position ( Figure 2E ) . Conversely , in the fourth position , E is overrepresented in the top 25% ( 64/96 ) while S is overrepresented in the bottom 25% ( 48/96 ) . These patterns also hold for comparisons of the strongest and weakest 3-mer and 2-mer elements ( Figure 2—figure supplement 1C–F ) . These patterns indicate a grammar that includes a bias for S and O sites positioned upstream of K and E sites . This positioning may favor interactions between these factors and the basal transcriptional machinery or TFs recruited by the minimal promoter . As specifying a site at a given position restricts possible sites in neighboring positions , these patterns could also represent favorable interactions between factors . These data show that the precise arrangement of TFBS influences the activities of synthetic elements . While the grammar of O , S , K , and E sites influences the relative activities of the SYN elements , their order and orientation does not appear to contribute to the activity of genomic sequences . We compared the SYN and gWT libraries for elements with configurations of OKE , OSE , OSK , and SKE TFBS . Unlike SYN 3-mer elements , all four classes of gWT sequences span the full range of expression levels observed for the entire library , with only OSK sequences having a higher average expression ( Figure 3—figure supplement 1A ) . Thus , in genomic sequences , the same arrangement of sites embedded in different genomic contexts can either fail to drive detectable activity or drive expression higher than the highest SYN library member . To quantify the divergence in activities between genomic and synthetic elements directly , we matched gWT sequences with pluripotency TFBS-dependent activity to SYN elements with the corresponding order of TFBS . We observed no correlation in regulatory activity between matched site configurations , ( R2 = 0 . 001; Figure 3—figure supplement 1B ) . These data indicate that other variables contribute to the cis-regulatory activity of gWT sequences , such as the spacing and affinities of the sites , or the presence of TFBS for additional factors in flanking sequences that are held constant in the SYN library . To identify additional sequence features that might be contributing to activity , we used a variation of the Random Forest ( RF ) model , an unsupervised machine learning technique . RF models can be applied for either simple classification , assigning observations to group predictions , or classifying individual observations into semi-continuous bins to make quantitative , regression-case predictions . The accuracy of predictions are assessed over a large number of decision trees trained on random subsets of the data , which allows the contribution or ‘variable importance’ of specific features to be measured . As RFs are prone to biases from early random splits in the decision trees for unbalanced data , we used iterative Random Forests ( iRF ) as a tool for feature selection as well as for predicting activity ( Basu et al . , 2018 ) . We first trained a regression-case iRF model on the data from the SYN library . We initialized the models with four features ( Supplementary file 2A ) , representing only the presence or absence of each of the four pluripotency TFBS . This ‘independent’ iRF model had an R2 of 0 . 56 between observed and predicted observations when tested on held-out data for the final iRF iteration ( Figure 3—figure supplement 2 ) . However , the independent iRF model cannot account for the differences in activities between 4-mers , because all 4-mers have identical TFBS composition ( 4-mers R2 = 0 . 00 ) . To identify features that might distinguish between the activities of 4-mers , we trained an additional regression-case iRF model , ‘independent + position’ , initialized with 20 features , representing both the presence and position of the four TFBS in each SYN element ( Supplementary file 2A ) . The 20-term positional model performs well in predicting SYN expression , with an overall R2 of 0 . 87 for the last model iteration on a held-out test set ( Figure 3A ) . The positional iRF model highly weights the presence/absence of the sites , as expected from the performance of the independent iRF model , but also has contributions from the presence of E in the 4th position and S in the first and second positions ( Figure 3B ) . These results reinforce the conclusion that the activity of synthetic sequences depends both on the composition and positioning of TFBS . iRF models trained on the SYN library failed to predict or classify the expression of genomic sequences . While synthetic elements had a range of activities , elements in the gWT library are predominantly inactive , and the small number of active gWT sequences drive expression across an order of magnitude of activity levels ( Figure 3—figure supplement 1A ) . Having such a large number of inactive sequences in the pool makes it difficult to train a model that predicts the relative activities of genomic sequences . Retraining iRF regression models to predict gWT expression fails during the training step and has no correlation with the observed expression data ( independent: R2 = 0 . 03; independent + position: R2 = 0 . 001 ) . In all subsequent analyses of genomic sequences , we limited ourselves to models that attempt to distinguish between active and inactive genomic sequences , without predicting the relative differences in activity among active sequences . However , our first attempt to produce a classifier failed . Training a classification model to distinguish between active and inactive gWT sequences ( top 25% , n = 102; bottom 75% , n = 305 ) using either only independent or independent + position features also fails to perform better than chance ( Independent: Area Under the Receiver Operator Curve ( AUROC ) = 0 . 52 , Area Under the Precision Recall Curve ( AUPRC ) = 0 . 22; Positional: AUROC = 0 . 47 , AUPRC = 0 . 25; Supplementary file 2B ) . Genomic and synthetic elements with the same pattern of sites can drive drastically different expression levels ( Figure 3—figure supplement 1B ) . Other sequence features present in the flanking genomic sequences and absent from the synthetic elements must therefore play a role in setting activity levels , in addition to the identity and position of the individual pluripotency TFBS . Our results with genomic elements suggested that the sequences flanking the pluripotency TFBS play a role in determining cis-regulatory activity . We tested the effect of changing spacer sequences that flank the TFBS in six 4-mer elements from the SYN library . We tested four different spacer sequences , for a total of 30 library members , which includes the original spacer sequence . The new spacers sequences were designed to match the nucleotide content of the original spacers and minimize the creation of new TFBS ( Supplementary file 1J ) . To ensure the dynamic range of the library , we mixed this ‘mini spacer library’ library with a small portion of the SYN library and performed an MPRA . We found that changing the spacer sequences in the SYN library had small , but significant effects on the activities of the 4-mers . The activities of all six 4-mers in the mini spacer library tested with all four spacer sequences remained in the original range of expression for 4-mers ( Figure 3—figure supplement 3A ) . On average , the spacer sequences modified expression by 6% ( 0 . 3–25% , Figure 3—figure supplement 3B ) . Although the overall effects of spacer sequences were small , the rank order of the 4-mers did change for different spacers ( Figure 3—figure supplement 3C ) , supporting the idea that sequence features flanking the binding sites do affect gene expression . These results are consistent with the differences between the SYN and gWT libraries . We attempted to identify other sequence features that might differentiate active and inactive gWT sequences . Sequence-based support vector machines ( kmer-SVMs ) are powerful tools to predict the activity of regulatory elements ( Fletez-Brant et al . , 2013; Chaudhari and Cohen , 2018 ) . To identify sequence features that explain the differences between genomic elements , we trained a gapped kmer SVM ( gkm-SVM ) ( Ghandi et al . , 2016; Ghandi et al . , 2014 ) . The best performing gkm-SVM classified our positive and negative sets with AUROC of 0 . 75 and AUPRC of 0 . 77 ( k = 8 , gap = 2; Figure 4A ) . Although all sequences in the gWT library were selected to contain TFBS for the four pluripotency factors , many of the discriminative 8-mers ( 29/50 ) have motif matches that include at least one pluripotency family member ( Fletez-Brant et al . , 2013; Bailey et al . , 2009; Supplementary file 2D ) . This suggests that the differences between active and inactive genomic sites could be due to the primary pluripotency sites or secondary occurrences of these sites in the intervening sequences that scored below the scanning threshold . Sequences with higher predicted affinity pluripotency TFBS may drive higher expression . To determine if differences in the primary pluripotency sites are part of the signal identified by the SVM , we annotated gWT sequences with PWM-based scores for each TFBS present ( Grant et al . , 2011 ) . For SOX2 , we found no difference in scores between high and low sequences ( Figure 4B; p=0 . 07 , Welch’s t-test ) . For OCT4 , we found a modest difference between the average scores for high and low sequences and a broader but also a significant difference for KLF4 and ESRRB PWM scores ( Figure 4C–E ) . Summing the PWM scores for all of the TFBS further separates high and low sequences ( Figure 4F–G ) . These patterns suggest that the quality of the primary sites contributes to the activity differences observed among gWT sequences . We then asked if secondary sites for the pluripotency TFs might contribute to cis-regulatory activity by calculating predicted occupancy for both gWT sequences and gMUT sequences that lack the primary binding sites ( Materials and methods ) . Predicted occupancy is a metric that includes contributions from any primary , well-scoring TFBS plus contributions from weaker sites that might be missed with traditional motif scanning ( White et al . , 2016; White et al . , 2013; Evans et al . , 2012; Segal et al . , 2008; Zhao et al . , 2009 ) . We found evidence for additional low predicted affinity sites for SOX2 and OCT4 in both high and low sequences , making it unlikely that low-affinity sites strongly contribute to expression differences ( Figure 4—figure supplement 1 ) . Together , these results suggest that the affinities of the primary sites in genomic sequences , which are fixed in synthetic elements , contribute to the regulatory activity of genomic sequences more than the presence of additional sites with low predicted affinity . We also analyzed whether the spacing between binding sites correlated with the activity of cis-regulatory elements . Using the same annotations used to determine the predicted affinities of SOX2 , OCT4 , ESRRB , and KLF4 binding sites , we calculated the edge-to-edge distance between every possible pair of binding sites and plotted the frequency of each spacing for high and low activity sequences ( Figure 4—figure supplement 2 ) . We observed a preference in high activity sequences for closely spaced sites for OCT4 and SOX2 reflecting a known interaction between these TFs . We also observed preferences in high activity genomic sequences for closely spaced KLF4 and OCT4 sites , and for ESRRB and OCT4 sites . Binding site spacing may therefore play a role in setting the relative activities of genomic sequences . A major difference between the synthetic and genomic elements is the presence of sites for TFs besides the pluripotency factors . While the synthetic elements were designed to keep the sequences between pluripotency sites constant , genomic sequences differ in both the length and composition of sequences between the pluripotency sites . The presence of binding sites for additional transcription factors may contribute to the activity of genomic sequences . To identify sites for other factors that could contribute to differences between high and low activity gWT sequences , we examined the top discriminative 8-mers from the gkm-SVM , looking at possible PWM matches for additional TFs ( Supplementary file 2D ) . We then used PWMs for these additional TFs to identify instances of sites for other factors in the genomic sequences ( see Materials and methods ) ( Grant et al . , 2011; Sandelin , 2004 ) . We found significant enrichment for FOXA1 sites ( Figure 4H ) . We also found that FOXA1 and NANOG had higher total PWM scores in the high activity sequences ( Figure 5—figure supplement 1A ) . While FOXA1 is likely not present in mESCs , other family members ( FOXA2 , FOXD1 , FOXP1 ) are expressed in ESCs and have been shown to contribute to the pluripotent regulatory network , and therefore could be acting on the gWT sequences through these binding sites ( Pan and Thomson , 2007; Mulas et al . , 2018; Gabut et al . , 2011 ) . Genomic sequences with higher occupancy by TFs in the genome , as measured by ChIP-seq , have higher average expression in our assay . We annotated the gWT intervals with publicly available ChIP-seq data for additional TFs and with ATAC-seq data from E14 mESCs to determine if differences in accessibility explained the difference between high and low activity sequences ( Supplementary file 2B ) . Both high and low activity gWT sequences were accessible in the genome showing that accessibility does not necessarily correlate with high activity sequences . High activity sequences had a small but significant overlap with NANOG peaks ( Figure 5—figure supplement 1B ) . However , for the 328 genomic sequences with a NANOG ChIP-seq signal , only 16% had an underlying TFBS as determined by motif scanning . Therefore , NANOG might be recruited by other pluripotency TFs to these sequences independent of high-quality TFBS for this factor . If we compare expression levels to the number of overlapping ChIP-seq peaks , including O , S , K , E and these additional TFs , we see that gWT sequences with higher occupancy in the genome have higher average expression in our assay ( Figure 5 ) , which has been previously observed in HepG2 cells ( Ulirsch et al . , 2016 ) . This result supports a model where cumulative occupancy sets activity level . To understand the relative contributions of the sequence features that were enriched individually , we trained iRF models with different subsets of these sequence features and compared their performance on a held-out test set ( Supplementary file 2B ) . None of these models accurately predicted the activity of genomic sequences , likely because most genomic sequences in our collection had no activity above basal levels . Therefore , we attempted to classify active from inactive genomic sequences . We trained an iRF model initialized with 58 features that capture differences between gWT sequences and SYN elements . These features include predicted affinity and preferred spacings between the pluripotency TFBS , the predicted occupancy for the pluripotency TFs , the presence of binding sites for additional TFs , plus chromatin accessibility ( ATAC-seq ) and ChIP-seq peaks for both TFs and histone marks , as well as summary features such as the total primary site affinities for each sequence ( Supplementary file 2B ) . This gWT iRF model classified active from inactive on a held out test set with AUROC = 0 . 67 , and AUPRC = 0 . 46 ( Figure 6A–B , model ‘All’ ) . Models that only included subsets of features — the spacing between elements ( model ‘Spacing’ ) , the strength of the pluripotency sites ( ‘PrimarySites’ ) , or the overlapping ChIP signal ( ‘ChIPSignals’ ) — did not perform as well ( Figure 6A–B ) . The features that best separate active from inactive sequences were related to attributes of the pluripotency sites with the top feature being the summed pluripotency factor predicted affinity per sequence ( ‘OSKE_TotalAffinity’ , Figure 6C ) . Taken together , our data suggest that genomic sequences drive higher expression when they contain strong binding sites with preferred spacing and are embedded in sequences that can mediate the recruitment of other TFs or cofactors .
In this study , we sought to understand how pluripotency factors collaborate to drive specific levels of expression by testing both an exhaustive set of synthetic arrangements of TFBS for OCT4 , SOX2 , KLF4 , and ESRRB and comparable genomic sequences . The experimental design allowed for direct comparisons between the regulatory grammar of synthetic and genomic sequences . The strongest similarity between synthetic and genomic elements is that in both cases activity depends heavily on the number and affinity of binding sites . These results are most consistent with a model in which the overall occupancy of a sequence by its cognate TFs is the primary determinant of that element’s activity . Consistent with this hypothesis , the predictive power of our trained genomic model derived primarily from summing over the number and affinity of binding sites . We also observed correlation between the occupancy of sites as measured by ChIP-seq and their activity in MPRA assays . While there are many steps involved in activating gene expression , the occupancy model posits that the strength of a regulatory element is primarily controlled by its fractional occupancy by TFs . The occupancy model might also explain the surprising result that the activity of genomic elements in our plasmid MPRA experiments do not correlate with experimental measurements of how accessible the chromatin is in their native locations . Plasmid assays might not capture regulation by chromatin , but in many cases plasmid assays do recapitulate the activity of chromosomally integrated elements ( Maricque et al . , 2019; Inoue et al . , 2017 ) . Alternatively , accessible regions may be bound by transcription factors but may not necessarily drive activity , such as in the case of ‘poised’ regulatory elements ( Cruz-Molina et al . , 2017 ) . Nucleosome exclusion is important for regulatory activity ( Khoueiry et al . , 2010 ) and may reflect TF binding , but accessibility itself may not be sufficient for regulatory activity . Another possibility is that open chromatin may not be a direct reflection of the occupancy of an element by its cognate TFs . Other factors besides occupancy by TFs also determine the openness of chromatin , such as chromosome topology , the proximity of origins of replication , and nucleotide composition . This may explain why some genomic sequences with binding sites that reside in open chromatin do not drive high activity in MPRA assays . The prediction is that these regions are open for reasons other than occupancy by cognate TFs . That the activity of genomic elements correlates with TF occupancy as measured by ChIP-seq , but not necessarily open chromatin measurements by ATAC-seq , supports the occupancy model . While TF occupancy was the best predictor of activity , the AUROC and AUPRC analyses show that we are still missing important features that underlie the activity of genomic sequences . Indeed , two-thirds of genomic sequences that contain consensus motifs and reside under a ChIP-seq peak for one of the pluripotency TFs had no activity in our assay . Why don’t all sequences occupied by TFs have strong regulatory activity ? The sequence context in which occupied binding sites occur must contribute heavily to their activity . We attempted to address this issue by examining the regulatory grammar of synthetic elements . Synthetic elements provide a highly controlled system for exploring whether TFBS are constrained by a regulatory grammar . With synthetic elements we found clear evidence that their activity depends on the position and orientation of pluripotency binding sites . Synthetic elements with the same number and affinity of TFBS had different levels of activity depending on the order and orientation of the sites . This result suggests that active regulatory elements in the genome are defined not only by the presence of TF occupied motifs , but also by cues in the surrounding DNA sequences . However , our models that captured the specific regulatory grammar of synthetic elements failed to predict the activity of genomic sequences . Why don’t models that robustly predict the activity of synthetic elements also predict the activity of genomic sequences ? With synthetic elements , each sequence differs from others in the library by only a small number of sequence features . In synthetic libraries , there are many pairs of elements that differ by only a single sequence feature , which provides power to observe experimentally the effect of a single variable . In contrast , libraries of genomic elements are much more diverse , and the analysis of genomic sequences relies on detecting correlations between elements that share sequence features . However , it is difficult to isolate the effect of a single sequence feature because genomic elements that share a certain sequence feature will always be very different in terms of other features . The strength of the synthetic approach is the power it provides to isolate the effects of specific sequence features or pairs of sequence features . The weakness of the synthetic approach is that genomic elements are subject to many context specific constraints , all of which cannot be captured in a single synthetic library . When we changed the spacer sequences in our synthetic library , we found small but reproducible effects on expression . Our interpretation of this result is that changing the spacer sequences did not have large effects on the independent contribution of each TFBS , but did have effects on the interactions between sites ( i . e . the regulatory grammar ) . In the future , we plan to use the regulatory grammar derived from synthetic elements to design experiments that manipulate single features of genomic elements . If the grammar that is learned from synthetic elements reflects real constraints in the cell , then models of synthetic elements should predict the relative effects of single perturbations of genomic elements even if they cannot predict the absolute expression of genomic sequences . A combined approach that leverages both synthetic and genomic sequences should continue to help unravel the rules that govern cis-regulation of expression in cells .
To generate a library that contained both synthetic and genomic elements , we ordered a custom pool of 13 , 000 unique 150 bp oligonucleotides ( oligos ) from Agilent Technologies ( Santa Clara , CA ) through a limited licensing agreement . Each oligo in the SYN pool was 150 bp in length with the following sequence: where [SEQ] is a 40–80 bp synthetic element comprised of concatenated 20 bp building blocks of pluripotency sites , as described previously , with the fifth position of the KLF4 site changed to ‘T’ to facilitate cloning ( Fiore and Cohen , 2016 ) . [FILL] is a random filler sequence of variable length to bring the total length of each sequence to 150 bp , and [BC] is a random 9 bp barcode . The oligonucleotide pool contained all possible combinations of the pluripotency binding sites in both orientations , with no more than one of each site per sequence in lengths of two , three , and four building blocks . The sequence of each of the element is listed in Supplementary file 1B . In total , the SYN library has 624 unique synthetic elements . Each synthetic element is present in the pool eight times , each time with a different unique BC . There are also 112 oligos in the pool for cloning the basal promoter without any upstream element , each with a unique BC . Genomic sequences were represented in the pool by 150 bp oligos with the following sequences: where [SEQ] is either a reference ( gWT ) or mutated ( gMUT ) genomic sequence of 81–82 bps . Reference gWT sequences were selected by choosing regions of the genome within 100 bps of previously identified ChIP-seq peaks for these four pluripotency factors ( Chen et al . , 2008b ) . After excluding poorly sequenced and repetitive regions ( ENCODE Project Consortium , 2012; Waterston et al . , 2002 ) , we scanned the remaining regions using FIMO with the four PWMs used previously to design the synthetic building blocks , with a p-value threshold of 1 × 10−3 ( Grant et al . , 2011; Bailey et al . , 2009; Fiore and Cohen , 2016 ) . Regions that contained more than one overlapping site identified by FIMO were excluded . Binding sites that were located less than 20 bp from each other were then merged into a single genomic element using Bedtools ( Quinlan and Hall , 2010 ) . Elements with no more than one of each site per element were then selected and expanded to 81–82 bp centered on the motifs . Expanded sequences were rescanned to confirm the presence of only three binding sites with the same threshold as used to originally scan the sequences . Sequences that contained restriction sites for were then removed from the library , leaving 407 genomic sequences with combinations of the OCT4 , SOX2 , KLF4 , and/or ESSRB TFBS . We generated matched mutated sequences ( gMUT ) for each of the 407 gWT sequences by changing two positions in each motif from the highest information content base to the lowest information base for that position ( Figure 1—figure supplement 1 ) . The reverse complement position and substitution was made for the reverse orientation of each motif . The mutated sequences were rescanned with all four original PWMs to confirm that no detectable pluripotency TFBS remained , using FIMO with the same p-value threshold ( 1 × 10−3 ) as above . In total , the pool of oligos representing genomic sequences contained 407 wild-type sequences ( gWT ) and the corresponding 407 gMUT sequences . The sequence of each element is listed in Supplementary file 1G . Each of these 814 sequences were associated with eight unique BCs . The primers for gWT and gMUT sequences were identical so all subsequent steps for this library was performed in a single pool . There are also 112 oligos in the pool for cloning the basal promoter without any upstream element , each with a unique BC ( Supplementary file 1F ) . The rest of the array contained sequences not used in this study . For a full list of primers , see Supplementary file 3 . The synthesized oligos were prepared as previously described ( Kwasnieski et al . , 2012; Fiore and Cohen , 2016 ) , except using primers Synthetic_FW-1 and Synthetic_Rev-2 with an annealing temperature of 55°C for the SYN library and primers Genomic_FW-1 and Genomic_Rev-1 with an annealing temperature of 53°C for the gWT/gMUT libraries . PCR products were purified from a polyacrylamide gel as described previously ( White et al . , 2013 ) . Each library was cloned as described previously ( Fiore and Cohen , 2016 ) , with an SYN element ( SYN library ) or either a gWT or gMUT sequence ( gWT/gMUT library ) cloned into the ApaI and SacI sites of plasmid pCF10 . The pou5f1 basal promoter and dsRed reporter gene were amplified from pCF10 using primers CF121 and CF122 , and inserted into the plasmid library pools from the previous step at the XbaI and HindIII sites . Digestion of the libraries with SpeI and subsequent size selection was omitted as the SYN library had less than 2% background and the combined gWT/gMUT library had less than 1% background in the final cloning step . For the mini spacer library , we ordered an oligo pool containing 4-mer elements with different spacer sequences from Integrated DNA Technologies ( Coralville , IA ) . Each oligo in the mini library was 161 bp in length with the following sequence: where [SEQ] represents an oligo sequence described below and [BC] is a random 9 bp barcode . We picked six 4-mer oligos from the original synthetic library to span the 4-mer expression range and swapped out the spacer sequences in the oligos for four other sequences , generating a total of 30 constructs , including the original spacers . Each construct was represented in the pool with five unique barcodes . The sequence of each element is in Supplementary file 1K . The mini spacer library was cloned into the same backbone as the previous libraries . Briefly , pCF10 was digested with ApaI and SacI , and the single-stranded oligo pool was directly assembled into the backbone using HiFi DNA assembly The pou5f1 basal promoter and dsRed reporter gene were amplified from pCF10 using CF121 and CF122 , then ligated into the mini spacer library following the same approach as the SYN , gWT , and gMUT libraries . RW4 mESCs were cultured as described previously ( Xian et al . , 2005; Chen et al . , 2008a ) on 2% gelatin coated plates in standard media ( DMEM , 10% fetal bovine serum , 10% newborn calf serum , nucleoside supplement , 1000 U/ml leukemia inhibitory factor ( LIF ) , and 0 . 1 µM B-mercaptoethanol ) . Approximately 1 million cells at 100% estimated viability were seeded into six-well plates 24 hr prior to transfection . The SYN library and combined gWT/gMUT were transfected in parallel using 10 µL Lipofectamine 2000 ( Life Technologies , Carlsbad , CA ) , 3 µg of plasmid library , and 0 . 3 µg CF128 ( a GFP control plasmid ) per well , as described previously ( Fiore and Cohen , 2016 ) . Four biological replicates of each library pool , the SYN plasmid pool or combined gWT/gMUT plasmid pool , were transfected and the plates were passaged 6 hr post-transfection . For three replicates of each library pool , RNA was extracted 24 hr post-transfection from approximately 9 million cells per replicate , using the PureLink RNA mini kit ( Life Technologies , Carlsbad , CA ) with the fourth transfection replicate reserved for estimating transfection efficiency via fluorescent microscopy and staining for alkaline phosphatase ( AP ) activity , a universal pluripotency marker ( Singh et al . , 2012 ) . Massively parallel reporter gene assays were used to measure the activity of each element as described previously ( Fiore and Cohen , 2016; Mogno et al . , 2013 ) . Briefly , we used Illumina NextSeq ( San Deigo , CA ) sequencing of both the RNA and original plasmid DNA pool , removing excess DNA from the RNA pool using TURBO DNA-free kit ( Life Technologies , Carlsbad , CA ) . cDNA was then prepared using SuperScript RT III ( Life Technologies , Carlsbad , CA ) with oligo dT primers . Both the cDNA and the plasmid DNA pool were amplified using primers CF150 and CF151b , for 13 cycles . The PCR amplification products were digested using XbaI and XhoI ( New England Biolabs , Ipswich , MA ) , ligating the resulting digestion products to custom Illumina adapter sequences , P1_XbaI_X ( where X is 1 through 8 , with in-line multiplexing BC sequences ) to the 5’ overhang and PE2_SIC69_SalI on the 3’ XhoI overhang , each of which is comprised of annealed forward ( F ) and reverse ( R ) strands . An enrichment PCR with primers CF52 and CF53 was then used , and the resulting products were mixed at equal concentration and sequenced on one NextSeq lane . Sequencing reads were filtered to ensure that the BC sequence perfectly matched the expected sequence . For the SYN library , this resulted in 40 million reads combined for the three demultiplexed RNA samples ( P1_XbaI_1 , P1_XbaI_2 , P1_XbaI_3; 12 . 7–13 . 5 million each ) , and 19 . 7 million reads for the DNA library sample ( P1_XbaI_7 ) . For the combined gWT/gMUT libraries , this resulted in approximately 37 million reads combined for the three demultiplexed RNA samples ( P1_XbaI_4 , P1_XbaI_5 , P1_XbaI_6; 9 . 4–16 million each ) , and 19 . 6 million reads for the DNA library sample ( P1_XbaI_8 ) . For each library , BCs that had less than three raw counts in any RNA replicate or less than 10 raw counts in the DNA sample were removed before proceeding with downstream analyses . Expression normalization was performed by first calculating reads per million ( RPM ) per BC for each replicate for both the SYN library and the combined gWT/gMUT library . For each BC , expression was calculated by dividing the RPMs in each RNA replicate by the DNA pool RPMs for that BC . Normalizing by DNA RPMs successfully removed the impact of the representation of the construct in the original pool as the calculated expression has no correlation with the DNA counts for both the SYN library and the combined gWT/gMUT . Within each biological replicate , the BCs corresponding to each synthetic element ( SYN ) or genomic sequence ( gWT/gMUT ) were averaged and then normalized by basal mean expression in that replicate . These normalized expression values were then averaged across biological replicates . All downstream analyses were performed in R version 3 . 3 . 3 and plotted with ggplot2 version 2 . 2 . 1 . Expression summaries per replicate are reported in Supplementary file 1C for the SYN library , Supplementary file 1H for the gWT/gMUT library and Supplementary file 1L for the ‘mini spacer’ library . Custom code , based on Zhao and Stormo’s BEEML algorithm ( Zhao et al . , 2009 ) , was used to compare sequences of interest to a provided Energy Weight Matrix ( EWM ) at a set protein concentration ( mu ) and output a predicted occupancy for that TF as in White et al . ( 2013 ) . Briefly , an energy landscape ( EWM score ) is calculated by comparing all n-mers of each sequence , where n = length of provided motif , to the matrix to generate an array of individual base scores for the forward and reverse orientation of the sequence . Occupancy is then predicted using equation 3 for binding probability at equilibrium , ( 1/ ( 1 + e ( ΔG − μ ) ) ) . Position Frequency Matrices equivalent to the PWMs used for both SYN building block design and for scanning the mouse genome were used to generate EWMs , using the formula RT∗ ln ( Freq ( Base^consensus ) /Freq ( Base^i ) ) to convert the frequency of each base at each position i to a pseudo ΔΔG values for each factor ( White et al . , 2013 ) . Predicted occupancy ( P ( Occ ) ) for the 3-mer SYN elements was calculated for different assumed protein concentrations ( mu = 0 . 5 , 1 , 2 , 4 , 5 , 8 , 10 , 12 ) to determine at what point the SYN elements are predicted to be saturated , where P ( Occ ) ≅ three for each SYN element , that is: approaching one for each TFBS in the sequence . SYN elements were saturated by each of the four pluripotency factors at mu = 8 with the exception of the shorter Oct4 motif , which reached saturation at mu = 10 . Occupancy of gWT and gMUT sequences was predicted for gWT and gMUT at an assumed high protein concentration of mu = 8 for Sox2 , Klf4 , Esrrb , and mu = 10 for Oct4 , consistent with the role of these factors in mESCs . The predicted occupancy of each factor for matched gMUT sequences are reported in Supplementary file 2F as a feature of gWT sequences . iRF models: We built iterative Random Forest ( iRF ) models to classify our data using the R package iRF ( version 2 . 0 . 0 ) ( Basu et al . , 2018 ) . To run the software a model is initialized with 1/p weights for each of p features to be included in fitting the model . In each iteration , p features are reweighted by their Gini Importance ( wk ) , a measure that is calculated by how purely a node , split by feature , separates the classes ( Menze et al . , 2009; Louppe et al . , 2013 ) . Default settings were used for model training , with four iterations of reweighting p features specified for each model as indicated in Supplementary files 2A and 2B . Synthetic data was split into training and test sets by randomly subsetting 50% of the total SYN elements ( total n = 407 ) . Mean normalized expression was the response variable for model fitting for the synthetic models ( see Supplementary file 2E for feature annotations for SYN elements ) . Four iterations of model fitting on training data was used . Genomic data was split into training and test sets by randomly subsetting 50% of the total gWT/gMUT intervals ( total n = 624 ) . Classification as ‘active’ , 1 , if mean normalized gWT expression was greater than or equal to the 3rd quartile and ‘inactive’ , 0 , if mean normalized gWT expression was less than the 3rd quartile ( cutoff value = 1 . 983 ) , was the response variable for model fitting ( see Supplementary file 2F for feature annotations and response values for gWT sequences ) . Four iterations of model fitting on training data was used . gkm-SVM: We used a gapped k-mer Support Vector Machine ( gkm-SVM ) to search for gapped k-mers that distinguish between highly active and inactive genomic sequences ( Ghandi et al . , 2016 ) . We subset sequences from the gWT library into top 25% ( high ) and bottom 25% ( low ) based on expression data for a total of 101 positive and 101 negative intervals for the training set . FASTA sequences were then generated from the mm10 reference genome ( Bioconductor , BioMart ) for each region ( Supplementary file 4 ) . We then used the gkm-SVM R package to classify high vs . low sequences ( Ghandi et al . , 2016 ) . Word length ( L ) values of 6 ( gap = 2 ) , 8 ( gap = 2 ) , and 12 ( gap = 6 ) , were tested with cross validation . Default settings were used for other function options . Three-fold cross validation was chosen due to the the amount of structure in the data , with combinations of OSK binding sites overrepresented in positive training sequences ( Figure 3—figure supplement 1 ) . The best average performance on training data as evaluated by AUCs was the model trained with parameters of L = 8 and gap = 2 ( See Supplementary file 2G for output scores ) . The final gkmer-SVM model includes approximately 1 million unique k-mers ( See Supplementary file 2C for full kmer list and weights ) . All genome coordinates from previous mouse genome builds were converted to mm10 using the UCSC liftover tool ( Kuhn et al . , 2013 ) . Binding matrices for SOX2 , OCT4 , KLF4 , ESRRB were as previously reported ( Fiore and Cohen , 2016 ) . The Bedtools suite ( version 2 . 20 ) was used for manipulations and analysis of bed files ( Quinlan and Hall , 2010 ) . Statistical tests were chosen based on expectations of normalcy , with Wilcoxon rank-sum test used for comparisons of BC expression as these distributions were observed to be skewed for some library members , Welch’s t-test used where sample sizes were equal and roughly normal , and Fisher’s 1-sided tests used for testing for enrichment in small sample sizes . Raw sequencing data for SYN library and gWT/gMUT library can be found under SRA accession number SRR7515851 . Processed sequencing data , specifically demultiplexed barcode counts per replicate , can be found under GEO accession number GSE120240 . Additionally , a table of normalized reads per million ( RPMs ) across replicates for all barcodes are included as Supplementary file 1D for the SYN library , Supplementary file 1I for the gWT/gMUT library , and Supplementary file 1M for the MiniSpacer library . | Transcription factors are proteins that flip genetic switches; their role is to control when and where genes are active . They do this by binding to short stretches of DNA called cis-regulatory sequences . Each sequence can have several binding sites for different transcription factors , but it is largely unclear whether the transcription factors binding to the same regulatory sequence actually work together . It is possible that each transcription factor may work independently and there only needs to be critical mass of transcription factors bound to throw the genetic switch . If this is the case , the most important features of a cis-regulatory sequence should be the number of binding sites it contains , and how tightly the transcription factors bind to those sites . The more transcription factors and the more strongly they bind , the more active the gene should be . An alternative option is that certain transcription factors may work better together , enhancing each other's effects such that the total effect is more than the sum of its parts . If this is true , the order , orientation and spacing of the binding sites within a sequence should matter more than the number . One way to investigate to distinguish between these possibilities is to study mouse embryonic stem cells , which have a core set of four transcription factors . Looking directly at a real genome , however , can be confusing and it is difficult to measure the effects of different cis-regulatory sequences because genes differ in so many other ways . To tackle this problem , King et al . created a synthetic set of cis-regulatory sequences based on the four core transcription factors found in mouse stem cells . The synthetic set had every combination of two , three or four of the binding sites , with each site either facing forwards or backwards along the DNA strand . King et al . attached each of the synthetic cis-regulatory sequences to a reporter gene to find out how well each sequence performed . This revealed that the cis-regulatory sequences with the most binding sites and the tightest binding affinities work best , suggesting that transcription factors mainly work independently . There was evidence of some interaction between some transcription factors , because , of the synthetic sequences with four binding sites , some worked better than others , and there were patterns in the most effective binding site combinations . However , these effects were small and when King et al . went on to test sequences from the real mouse genome , the most important factor by far was the number of binding sites . Synthetic libraries of DNA sequences allow researchers to examine gene regulation more clearly than is possible in real genomes . Yet this approach does have its limitations and it is impossible to capture every type of cis-regulatory sequence in one library . The next step to extend this work is to combine the two approaches , taking sequences from the real genome and manipulating them one by one . This could help to unravel the rules that govern how cis-regulatory sequences work in real cells . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"computational",
"and",
"systems",
"biology"
] | 2020 | Synthetic and genomic regulatory elements reveal aspects of cis-regulatory grammar in mouse embryonic stem cells |
Perceptual abilities of animals , like echolocating bats , are difficult to study because they challenge our understanding of non-visual senses . We used novel acoustic tomography to convert echoes into visual representations and compare these cues to traditional echo measurements . We provide a new hypothesis for the echo-acoustic basis of prey detection on surfaces . We propose that bats perceive a change in depth profile and an ‘acoustic shadow’ cast by prey . The shadow is more salient than prey echoes and particularly strong on smooth surfaces . This may explain why bats look for prey on flat surfaces like leaves using scanning behaviour . We propose that rather than forming search images for prey , whose characteristics are unpredictable , predators may look for disruptions to the resting surface ( acoustic shadows ) . The fact that the acoustic shadow is much fainter on rougher resting surfaces provides the first empirical evidence for ‘acoustic camouflage’ as an anti-predator defence mechanism .
Considerable investigative effort has been focused on how flying bats use echolocation to track and capture moving prey in mid-air . Investigations of this aerial hawking behaviour have led to the recognition of novel prey defenses ( Fullard and Napoleone , 2001; Ratcliffe and Fullard , 2005 ) and methods of prey detection , tracking , and capture ( Goerlitz et al . , 2010; Conner and Corcoran , 2012 ) . Less well understood is how bats detect insects and other prey on surfaces . This foraging behaviour ( gleaning ) is perceptually complex and takes several forms . Some bats exploit sexual advertisement calls of prey to determine their location ( e . g . , Myotis septentrionalis attacks katydids calling for mates on grass tips [ter Hofstede et al . , 2008] , Trachops cirrhosus similarly exploits displaying frogs [Ryan et al . , 1982; Page and Ryan , 2006] ) . Other passive gleaners listen for more subtle prey-generated sounds such as fluttering wings ( e . g . , rhinolophids and hipposiderids , [Siemers and Ivanova , 2004] ) and scorpions' walking noises ( Holderied et al . , 2011 ) . Ground-gleaning bats ( Siemers and Ivanova , 2004 ) use many of the same strategies of hawking including continuous echolocation and shortened duration of pulse intervals to target moving prey . Active gleaning , when a bat detects and removes motionless prey from a surface , appears to be the most demanding gleaning behaviour . There is some variation in call structure in gleaning bats . Gleaning Myotis use short frequency-modulated ( FM ) broadband calls , but behavioural evidence demonstrates these bats are less efficient at this task when background ‘clutter’ obscures target echoes ( Arlettaz et al . , 2001 ) . Some substrate gleaners may beat their wings at potential prey to elicit movement ( Kuc and Kuc , 2012 ) . True active gleaning requires the bat to distinguish a motionless target from its resting surface , which can be a highly structured background . Geipel , Jung , and Kalko ( 2013 ) suggested that some species such as Micronycteris microtis perform this task expertly . M . microtis uses broad-band FM multi-harmonic calls of very short duration ( ≈0 . 2 ms ) emitted as single calls with intervals of ≈31 ms or in groups of two calls with an intragroup pulse of ≈15 ms ( Geipel et al . , 2013 ) . Call start frequencies are ≈143 kHz and end frequencies ≈69 kHz , and there appear to be no terminal feeding buzzes ( Geipel et al . , 2013 ) . A few authors have used behavioural experiments to investigate the role of approach angles or speculate on the use of a search image ( Jones , 2013; Geipel et al . , 2013 ) . Until now , however , no solution to the nature of the underlying acoustic cues has been provided . Passive gleaning bats use broadband , low-intensity calls ( Faure and Barclay , 1994 ) , while active gleaners may use much more variable call structures and intensities . Some call characteristics have obvious advantages ( e . g . , low-intensity calls make bats relatively inaudible to eared prey [Faure and Barclay , 1994] ) , while the importance of other features is unclear , particularly when the surface is highly structured ( Arlettaz et al . , 2001 ) . Previous experiments with gleaning behaviour have either focused on the bats' use of prey-generated sounds or acoustically simple surfaces ( Arlettaz et al . , 2001; Siemers and Ivanova , 2004 ) . To our knowledge , no one has considered the specific echo cues used to find motionless prey . In addition , many insects lack ears and/or the wings of resting prey ( such as moths ) cover the ears , reducing the prey's acoustic sensitivity ( Faure and Barclay , 1994 ) and potentially confounding their auditory defences . It is not clear if this resting posture makes them more difficult for bats to detect either by reducing target size or obscuring their body . We measure conventional single echo cues of root mean square amplitude ( RMS ) and duration ( Figure 1 ) as well as using a novel ‘acoustic tomography’ technique ( Figure 2 , Figure 3 ) to address two predictions about gleaning . First , we test the prediction that acoustic cues used by predators hunting prey are subtle and that novel cues based on image integration are more salient to the predator . Second , we test the prediction that wing position affects prey echoes . In addition , we consider the effect of differently structured surfaces as these may act to conceal resting targets as a form of acoustic camouflage , and that consequently there are perceptual benefits to gleaning from acoustically simple surfaces . We also consider the challenge that prey size and shape may vary greatly and be unpredictable . Thus , we include both multiple surfaces and multiple species of prey in our analysis . 10 . 7554/eLife . 07404 . 003Figure 1 . Echo cue examples for Citheronia regalis on slate with wings in ‘UP’ position . ( A and B ) Envelope of the echo impulse response as a function of measurement angle , ( A ) slate only; ( B ) slate plus moth . Coloured lines indicate start ( top lines ) and end ( bottom lines ) of the echo . ( C ) Echo duration and ( D ) echo root mean square ( RMS ) as a function of measurement angle . Green lines: with moth , blue lines: without moth . DOI: http://dx . doi . org/10 . 7554/eLife . 07404 . 00310 . 7554/eLife . 07404 . 004Figure 2 . Example of tomography analysis for Citheronia regalis on slate with wings in ‘UP’ position . ( A ) Photograph from rear of moth . ( B ) Tomography . Coloured vertical lines indicate example depth measurements ( ‘with moth’ in green; ‘without moth’ in blue ) . ( C ) Depth profile for substrate with ( green ) and without ( blue ) moth as a function of measurement angle . ( D ) Subtraction of tomographies with ( w ) and without ( w/o ) moth . Colour indicates absolute difference . ( E ) Strength of the shadow in percentage ( % ) difference to without moth as a function of measurement angle . Shadow size is measured as all angles where the shadow strength is at least 25% below the substrate without a moth . Overall shadow strength is the mean shadow strength over the entire shadow . DOI: http://dx . doi . org/10 . 7554/eLife . 07404 . 00410 . 7554/eLife . 07404 . 005Figure 3 . Example acoustic tomographies for Citheronia regalis . Specimens were placed on one of four substrates ( A ) bark , ( B ) leaf , ( C ) limestone , and ( D ) slate , and we generated acoustic tomographies with and without the specimen ( left and right in panels E–H , respectively ) . For each species , we generated tomographies with specimens that had their wings ( I ) FLAT and ( J ) UP . DOI: http://dx . doi . org/10 . 7554/eLife . 07404 . 005 Tomographic techniques compile images from a series of sections generated by converting an energy wave into a visual signal . It is used commonly in medical imaging through computed tomography ( CT ) , and acoustic tomography has been applied to some landscape imaging applications ( Duric et al . , 2011 ) . Acoustic tomography in air ( Balleri et al . , 2010 ) uses CT to interpret sound waves rather than the X-rays more normally associated with medical applications ( Duric et al . , 2011 ) but has not been widely applied in ecology ( Balleri et al . , 2010 ) . Acoustic impulse responses represent a range profile of echo reflections that have been used to reveal spatial reflection patterns of bat-pollinated flowers and different leaf shapes and determine their role as nectar guides for echolocating bats ( von Helversen and von Helversen , 1999; Simon et al . , 2011 ) . Our analysis is the first attempt to convert multi-aspect echo-acoustic information into tomographies . Acoustic tomography allows us to transform acoustic information into visual representations . This makes the information-gathering properties of echolocation more conventionally quantifiable giving us access to information not previously accessible . In this study , we use this approach to investigate the perceptual acoustic cues available to active gleaning bats when approaching a motionless prey item on a surface .
Figure 3E–H compares the tomographies of the four surfaces: bark , leaf , limestone , and slate revealing that slate presents a flat smooth surface , leaf has a smooth surface disrupted by venation , limestone is relatively flat but with a coarsely granulated surface , and bark presents a highly structured and rough surface . Depth profiles from acoustic tomographies ( Figure 4 ) were significantly different among substrates ( ANOVA , F3 , 480 = 335 . 8 , p < 0 . 001 , Tukey HSD p < 0 . 001 except limestone and leaf ) from a minimal depth profile on slate to a significant profile on bark , thus , we consider the effect of substrate in all further analyses . 10 . 7554/eLife . 07404 . 006Figure 4 . Tomographies indicate that substrate predicts mean depth profile with no moth present . Slate presents the most acoustically ‘mirror-like’ surface with the most minimal depth profile . Letters indicate significant differences . See also Figure 4—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 07404 . 00610 . 7554/eLife . 07404 . 007Figure 4—source data 1 . Depth measures for each substrate . DOI: http://dx . doi . org/10 . 7554/eLife . 07404 . 007 We found no significant difference in echo-acoustic measurements between wing position using traditional cues ( RMS and duration ) . In the more sensitive tomographies , we compared depth profiles of the substrate generated when a moth was present either with its wings up or down ( Figure 5 ) . Using a linear mixed-effects model , there was no interaction between wing orientation and substrate ( F3 , 105 = 1 . 66 , p = 0 . 18 , Supplementary file 1; Table 1a for model estimates ) . Wing orientation predicted depth profiles with the ‘UP’ position having a greater depth profile on all substrates though the effect here , and with shadow strength , is subtle and due to only a few unusual species ( see ‘Discussion’ ) . The acoustic shadow cast by the moth was affected by both the wing position and substrate giving a significant interaction in a linear mixed-effects model between the orientation of the wings and the substrate on shadow strength , the ‘shadow effect’ ( F3 , 105 = 3 . 19 , p = 0 . 027 , Supplementary file 1; Table 1b for model estimates , Figure 5 ) . 10 . 7554/eLife . 07404 . 008Figure 5 . Measurements from acoustic tomographies . ( A ) Wing orientation predicted mean depth with the UP position ( triangles , solid lines ) having a larger depth profile on all substrates than FLAT position ( crosses , dashed lines ) . The biggest effect was of resting substrate with the smooth surface , slate , causing the largest depth profile for each moth . ( B ) There was a significant interaction between the orientation of the wings and the substrate on the measures of shadow strength ( missing background ) . The largest effect of shadow strength was observed on smooth surfaces ( leaf and slate ) . More negative values indicate more obvious missing substrate with zero being no change from substrate alone . See also Figure 5—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 07404 . 00810 . 7554/eLife . 07404 . 009Figure 5—source data 1 . Shadow and depth measurements . DOI: http://dx . doi . org/10 . 7554/eLife . 07404 . 009 The most dramatic effect was the choice of moth resting substrate ( slate , leaf , limestone , or bark ) . Substrate roughness itself acts to conceal the moth echo and this effect was particularly important using multi-aspect cues from tomographies . There was a significant decrease in the change in depth profiles on acoustically ‘rougher’ surfaces ( i . e . , slate is the most acoustically uniform substrate [Figure 4] and generates the greatest depth profile change when the moth was present [Figure 5A] ) . This was supported by our linear mixed-effects model ( F3 , 108 = 108 . 99 , p < 0001 , Supplementary file 1; Table 1c for model estimates , Figure 5 ) . Substrate ‘roughness’ predicts the change in depth profile when a moth is present . The largest overall acoustic shadow effect ( regardless of wing position ) was seen in the change in shadow strength on leaf and slate substrates where shadow strength was always below 0 ( the point where there was no change in acoustic information ) with moths casting a strong shadow on the substrate ( Figure 5B ) . Using traditional single echo cues ( log transformed echo duration and RMS amplitude ) , we found that angle of approach was a predictor of whether traditional echo cues were significantly different with and without a moth . Duration decreased while RMS increased as the angle of approach reached perpendicular direction to the target surface ( Figure 6—figure supplement 1 ) . We used linear models to assess this effect . The akaike information criterion indicated models with interactions were better fitted ( Figure 6 ) . In a linear model , duration increased with angle on all substrates both without ( F3 , 556 = 40 . 09 , p < 0 . 001 , Supplementary file 1; Table 1d for model estimates ) and with ( F3 , 556 = 60 . 3 , p < 0 . 001 , Supplementary file 1; Table 1e for model estimates ) a moth ( with the exception of slate without a moth ) . Leaf behaved most similarly to slate . In a linear model , RMS amplitude decreased on all substrates as angle increased with ( F3 , 556 = 40 . 48 , p < 0 . 001 , Supplementary file 1; Table 1f for model estimates ) and without ( F3 , 556 = 55 . 7 , p < 0 . 001 , Supplementary file 1; Table 1g for model estimates ) a moth ( Figure 6 ) though the decrease was not steady . There were fluctuations in both duration and RMS caused by the venation creating folding of the leaf . In these cases , echo signals are likely jumping between two points of reflection depending on slight changes in angle . The largest magnitude of change in RMS amplitude with angle was seen on leaf . 10 . 7554/eLife . 07404 . 010Figure 6 . Echo duration and RMS echo amplitude log transformed as a function of angles 0–70° relative to the substrate surface for four substrates ( slate , leaf , limestone , and bark ) with moths present and absent . On all substrates ( except leaf without a moth ) , echo duration decreased towards an angle of incidence that is perpendicular to the substrate surface , with duration higher on substrates that are rougher ( bark ) . Relative RMS echo amplitude increased towards angles perpendicular to the substrate . Smoother , that is , more mirror-like surfaces showed the greatest increase in RMS amplitude for frontal directions . See also Figure 6—figure supplements 1–5 and Figure 6—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 07404 . 01010 . 7554/eLife . 07404 . 011Figure 6—source data 1 . RMS and Duration measurements . DOI: http://dx . doi . org/10 . 7554/eLife . 07404 . 01110 . 7554/eLife . 07404 . 012Figure 6—source data 2 . Data associated with Figure 6—figure supplements 1 , 4 and 5 . DOI: http://dx . doi . org/10 . 7554/eLife . 07404 . 01210 . 7554/eLife . 07404 . 013Figure 6—source data 3 . Data associated with Figure 6—figure supplement 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 07404 . 01310 . 7554/eLife . 07404 . 014Figure 6—source data 4 . Data associated with Figure 6—figure supplement 3 . DOI: http://dx . doi . org/10 . 7554/eLife . 07404 . 01410 . 7554/eLife . 07404 . 015Figure 6—figure supplement 1 . Echo duration ( panels A and B ) and RMS echo amplitude ( panels C and D ) as a function of angle relative to the substrate surface for four substrates ( slate , leaf , limestone , and bark ) . Panels A and C depict mean values of substrate echoes with moths present , and panels B and D show mean values without moths present . On all substrates , echo duration decreases towards an angle of incidence that is perpendicular to the substrate surface , with duration higher on substrates that are rougher ( bark ) . Relative echo intensity ( RMS ) echo amplitude increases towards angles perpendicular to the substrate . Smoother , that is , more mirror-like surfaces show the greatest increase in RMS amplitude for frontal directions . See also Figure 6—source data 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 07404 . 01510 . 7554/eLife . 07404 . 016Figure 6—figure supplement 2 . RMS echo amplitude as a function of 10° angle increments from 1 to 70° to the substrate surface for four substrates ( bark , limestone , leaf , slate ) . Paired boxplots show mean values without and with moths ( Moth ) present . There were no significant differences between pairs . Letters indicate significant values between angle increments . See also Figure 6—source data 3 . DOI: http://dx . doi . org/10 . 7554/eLife . 07404 . 01610 . 7554/eLife . 07404 . 017Figure 6—figure supplement 3 . Duration as a function of 10° angle increments from 1 to 70° to the substrate surface for four substrates ( bark , limestone , leaf , slate ) . Paired boxplots show mean values without and with moths ( Moth ) present . Significant differences between duration pairs are indicated with an asterisk . See also Figure 6—source data 4 . DOI: http://dx . doi . org/10 . 7554/eLife . 07404 . 01710 . 7554/eLife . 07404 . 018Figure 6—figure supplement 4 . Correlations between mean RMS echo amplitude from 1 to 70° when moths were present vs when moths were absent . RMS comparisons are significant ( bark R = 1 . 0 p < 0 . 001 , limestone R = 1 . 0 , p < 0 . 001 , leaf R = 0 . 98 , p < 0 . 001 , slate R = 0 . 99 , p < 0 . 001 ) . See also Figure 6—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 07404 . 01810 . 7554/eLife . 07404 . 019Figure 6—figure supplement 5 . Correlations between mean duration from 1 to 70° when moths were present vs when moths were absent . Duration comparisons are significant except for slate ( bark R = 0 . 99 p < 0 . 001 , limestone R = 0 . 96 , p < 0 . 001 , leaf R = 0 . 76 , p < 0 . 001 , slate R = 0 . 06 , p = 0 . 50 ) . See also Figure 6—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 07404 . 019 We used a two factor ANOVA on angles from 0 to 70° averaged in 10° increments to compare the effect of angle of ensonification and moth presence on echo duration and RMS on each substrate . We considered the presence of a moth and angle of ensonification as main effects and explored the interaction to test the hypothesis that the effect of a moth is only significant from certain directions . There was no significant effect of the presence of the moth on RMS at any angle though in almost all cases RMS showed a trend towards higher values when the moth was present ( Figure 6—figure supplement 2 ) . Angle of ensonification did have a significant main effect on RMS on all substrates ( p < 0 . 001 in all models ) . A Tukey HSD test between angles of ensonification showed specific pairwise differences ( Figure 6—figure supplement 2 ) . Considering duration , there was a significant interaction between moth and angle on all substrates except limestone , so we treated the comparisons between moth and no-moth separately for each substrate on each angle of ensonification with p-values adjusted for multiple comparisons . At almost all angles of ensonification on all four substrates , duration was longer when a moth was present; however , the effect was only significant in nine of the 28 cases , five of which were on slate . On this substrate , the effect was seen from 20 to 70° , and the magnitude of the effect was particularly large from 50 to 70° ( Figure 6—figure supplement 3 ) .
The low intensity and broadband calls of many gleaning bats differ strongly from the high amplitude and narrowband search calls of aerial hawking species , and it has been demonstrated that one problem with gleaning is an excess of echoes from objects other than the target , referred to as ‘echo clutter’ ( Arlettaz et al . , 2001 ) . This has led to two alternative explanations for behavioural guilds within gleaning . In the first case , echolocation itself is rarely used because acoustic masking disrupts target echoes making prey-generated sounds more important and most bats functionally passive gleaners ( Arlettaz et al . , 2001 ) . In the second case , broadband calls provide detailed information on texture using spectral cues , which allows for active gleaning in bats to distinguish prey amongst clutter ( Schmidt , 1988 ) . There has been considerable debate regarding these alternative mechanisms with many authors concluding that alternative signals ( prey-generated sounds but also vision and olfaction ) are key to prey localization ( Kalko and Condon , 1998; Arlettaz et al . , 2001; Eklöf et al . , 2002 ) ; however , some taxa do appear to rely on echolocation alone . The best example of gleaning in insectivores by echolocation alone was provided in behavioural experiments with the neotropical bat M . microtis ( Geipel et al . , 2013 ) . In this case , M . microtis was shown to use acoustic cues as the sole sensory modality for prey perception with individuals performing a three-dimensional hovering ( scanning ) flight in front of prey on leaves while emitting short , multi-harmonic broadband calls . The authors point out that given the short distance between target and substrate , the bats should experience considerable backward masking but they speculate that the hovering in combination with the continual scanning was a key component allowing for ensonification of the target and prey perception by altering the angle between target and background . This might have the effect of reducing clutter and thus the problematic masking ( Geipel et al . , 2013 ) . They further speculate that at obtuse angles , the smooth leaf surface contributes to the ‘mirror effect’ ( Siemers et al . , 2005 ) , which increases the relative strength of the target echo over the substrate clutter echo , as the latter is mirrored away at oblique angles . The power of tomographic imaging is that it provides a representation of the echo source distribution , which lies at the basis of the image perceived by the bat giving us insight into which cues may be most important to the bat rather than an exact measure of the bat's perception . Tomographic imaging is not simply a new tool , but from an information theoretic point of view it provides multi-echo interpretation , which is more powerful in detecting subtle acoustic phenomena . In our analysis , the most salient cue we measured was the disruption of a surface by the presence of a target prey item revealed in tomographic imaging . This effect , which we dub the ‘acoustic shadow’ , may present fundamental mechanisms allowing bats to find a target on a surface . Interestingly , we find some evidence for acoustic shadowing even in single echo RMS amplitude measurements ( Figure 1D ) . Such an acoustic shadow is particularly salient if bats were forming search images ( Jones , 2013 ) . Our data suggest that scanning behaviour combined with constant patch exploitation and patch fidelity may indicate that true gleaning taxa like Micronycteris form a perceptual construct of a predictable hunting environment and then search for disruptions of that construct . This represents a particularly powerful cue in that the bat needs only to perceive a difference in an expected surface , analogous to detecting a prey item resting on a lighter background by a local drop in brightness , rather than the full features of the prey target whose characteristics depend on size , shape , orientation , and taxon and are therefore unpredictable . One important consideration here is the degree of image integration between our experimental system and the biology of the bat . Imaging by sound as we have here gives an indication of which objects create echoes and how strong their overall contribution is . This represents a novel way to measure and interpret the spatial reflection patterns of echo-acoustic energy and is an important new tool in hypothesis generation ( finding potentially new cues ) and measuring echo-acoustic information in greater detail . However , a bat's image from a single call will be far less detailed ( though this cannot currently be quantified ) . To compensate , bats may use scanning behaviour to change the echo direction slightly between calls and combine multi-aspect range profiles ( impulse responses ) into a 2D or 3D image through biological tomography analogues . The remaining question is whether a tomographic analogue is biologically plausible , and under realistic conditions , what levels of detail bats are able to achieve . While no behavioural experimentation has yet been conducted specific to our hypothesis , one significant clue to the biological plausibility of a tomographic analogue is the very broadband scanning behaviour of Micronycteris ( Geipel et al . , 2013 ) . This behaviour should provide discrimination of depth changes down to 1 mm and combined with the change in approach angle and very short pulse intervals should lead to continuous information during scanning ( see an extensive discussion by Geipel et al . ( 2013 ) ) . In our analysis , the resting surface significantly affects the potential for successful bat gleaning . Mirror-like surfaces ( e . g . , leaf or slate ) provide more target detection potential across all four acoustic cues we considered . This was subtle for conventional single echo cues and varied with angle but was particularly apparent in multi-aspect integrated image interpretation from tomographies , which provided a far more sensitive measure of acoustic information . This suggests a new phenomenon of acoustic camouflage . Camouflage is most commonly associated with visual concealment with a long history of study and evolutionary interest ( e . g . , cited as an example in The Origin of Species [Darwin , 1859] ) . Camouflage devices include homochromy , countershading , and disruptive colouration ( Robinson , 1981 ) leading to the failure of a predator to detect prey . However , non-visual perceptual forms of camouflage are more rarely considered and are thus harder to define . In our biosonar case , ‘rough’ surfaces may create echo camouflage for prey making it virtually impossible for bats to differentiate the prey from the unpredictable signals of the surface . This may also constitute ‘mimicry’ in some cases if the bat detects the prey but dismisses the signal as part of the general variability of the surface ( see a discussion of camouflage vs . mimicry in Robinson ( 1981 ) ) . Subsequent experimentation with multiple surface gradients is clearly indicated to determine the nature of this effect . In contrast , a moth's resting position had a very small effect on acoustic cues , mainly due to three specimens . We did not treat these as statistical outliers for biological reasons . Two of these species were hawkmoths ( Sphingidae ) with very unusual resting positions when wings were up , and the third was a very small butterfly with extremely thin wings . The effect of resting position may be principally hard to detect in our data due to differences in size and this is the main reason for using multiple moth species in our data set . These species were selected to cover a range of potential target sizes and densities ( larger moths often have more dense wings and stronger wing venation ) . We suggest that in most cases , resting position does not provide a meaningful acoustic adaptation . However in some cases , such as hawkmoths , where the body is unusually large compared to the wings , which are folded very tightly when at rest , resting position may be important in detectability . Our data suggest that acoustically smooth surfaces like leaf and slate will act as mirrors such that among traditional cues at perpendicular directions near zero degrees , RMS will show a peak while duration will be shortest . Across angles from 0 to 70° , these cues ( Figure 1; Figure 6—figure supplements 2–5 ) showed strong correlation with and without moths present ( except duration on slate ) . At an increasing angle of sound incidence , the echo from the closest parts of the surface arrives earlier and the most distant parts' echo later , so the overall echo duration increased as the angle moves away to either side ( Figure 1A–C ) . The rougher the surface , the more reflectors contribute to the echo and the greater the magnitude of increase in duration ( Figure 6—figure supplement 1 ) . This suggests that the effect of the moth's presence is subtle and very difficult to detect but that mirror-like surfaces make the task easier by reflecting more information at specific angles yielding more salient cues to the bat . While we did not assess the variability of different surface types ( e . g . , multiple bark samples or taxa ) , we suggest that it is the properties of the surface rather than the actual item which matters—thus , smooth surfaces like beech bark may be more acoustically similar to leaves and slate than the bark used here . Further analysis of surface properties is necessary . Using the method of acoustic tomography , we have detected a novel acoustic effect , the acoustic shadow , which we present as potentially the most salient cue in gleaning behaviour . Our data also suggest a new effect of acoustic camouflage and the importance of choice in hunting surfaces to overcome this effect for gleaners . Our analysis thus provides a new set of hypotheses regarding the echo-acoustic mechanism of gleaning behaviour and an adaptive explanation for the patch fidelity and leaf scanning behaviour of M . microtis ( Geipel et al . , 2013 ) . If the bat is unsure if a prey object is present , scanning maximizes the chance that the prey is ensonified from an angle that returns the most informative single echo cues ( Siemers et al . , 2005 ) making prey detection more likely , but this scanning also would allow multi-aspect integration potentially leading to biological tomography analogues . We provide clear evidence for the role that search images may play and thus a strong hypothesis that these are key to gleaning but , contrary to previous speculation ( Jones , 2013 ) , we suggest the search image is not necessarily for prey , but for surfaces . These effects on the surface image are most salient on mirror-like surfaces . Rough surfaces provide little information even in the most salient acoustic shadow cue . Thus , insects resting on these surfaces are effectively camouflaged by the acoustic roughness of the substrate .
To accommodate potential target variability , we selected two pinned moth specimens from each of 16 species , which varied strongly in size and shape ( Sphinx kalmiae , Sphecodina abbottii , Paonias myops , Dryocampa rubicunda , Citheronia regalis , Erebia pawloskii , Boloria chariclea , Macrurocampa marthesia , Catocala ilia , Ennomos magnaria , Campaea perlata , Lophocampa maculata , Haploa confusa , Grammia virguncula , Grammia virgo , Panopoda rufimargo ) with their wings positioned upwards ( UP ) in a V in line with their bodies ( like many diurnal butterflies ) or flat/downwards ( FLAT ) roughly in the plane of the substrate surface ( like many moths ) ( Figure 3 ) . We placed each specimen on four substrates ( slate , artificial leaf ( leaf ) , limestone , and rough bark; Figure 3 ) for tomography ( 16 × 2 × 4 = 128 images ) . For each combination of moth , wing position , and substrate , echoes were taken ( details see below ) from 281 directions , and from the echo impulse responses , we measured four acoustic cues . Two are traditional single echo measures ( duration and RMS amplitude ) , while two are derived from tomographies , which integrate multiple echoes ( depth and shadow strength ) . For each echo , we calculated echo duration ( time from the start of the echo until the end of the echo as determined from the envelope of the impulse response with a common constant threshold manually set to be above the noise floor; compare Figure 1A , B ) , and—as measure of acoustic energy—RMS amplitude ( calculated over the time period from 1 . 64 ms to 3 . 0 ms echo delay; compare Figure 1A , B ) with and without moths present to quantify the effect of measurement angle on acoustic information . We turned each set of 281 impulse responses into an acoustic tomography using inverse radon transformation ( Balleri et al . , 2010 ) . A tomography is a scaled 2-dimensional acoustic representation of spatial echo origin and strength in a cross section through the target object in the plane echo measurements were taken from . Using tomographies , we can investigate changes to the acoustic profiles of the substrates by the presence of the prey item . We considered two variables , the mean depth profile and the acoustic shadow strength cast by prey on the substrate ( compare Figure 2; details see below ) . We used a custom-made acoustic tomography system ( Balleri et al . , 2010 ) , consisting of a 1/4″ ultrasound microphone ( type 40BF ) , pre-amplifier ( type 26AB; both G . R . A . S Sound & Vibration , Holte , Denmark ) , 2200C amplifier ( Larson Davis Inc . , Depew , NY; gain +40 dB ) , and a custom-made ferro-electret foil loudspeaker ( Emfit Ltd . , Vaajakoski , Finland ) driven by a PZD350 M/S high-voltage amplifier ( TREK Inc . , Lockport , NY ) . Microphone and speaker pointed in the same direction and were positioned at a distance of 20 mm from centre to centre to simulate the arrangement of a bat's mouth and ear . Loudspeaker and microphone were mounted on an adjustable lever arm moved by a LT360 turntable ( LinearX systems Inc . , Battle Ground , WA ) . Microphone , loudspeaker , and turntable were connected to a NI-DAQ BNC-2110 card operated through LabView v . 8 . 0 ( both National Instruments , Austin , TX ) with custom-programmes MisureBinauralaverage . vi ( rev 337 ) and FinalProgramBinaural . vi ( rev 185 ) ( Balleri et al . , 2010 ) . Acoustic measurements were taken in a 2 × 2 . 3 × 4 m semi-anechoic room . We played linear frequency-modulated sweeps from 250 to 10 kHz of 10-ms duration at 500 kHz sampling rate and 16-bit resolution and recorded sample-synchronously at the same rate and resolution . The echo impulse response was calculated by pulse-forming through deconvolution with the echo recorded perpendicularly from a 60 × 60 cm metal plate . Envelopes were calculated using the absolute value of the Hilbert transform of the impulse response . Echoes of each target were taken from a distance of 100 cm from the target from 281 radial positions ( 0 . 5° steps over ±70° relative to perpendicular to the substrate surface ) . We used MatLab ( v7 . 5 , MathWorks , Natick , MA ) to turn sets of impulse responses into acoustic tomographies ( Balleri et al . , 2010 ) . We imposed a colour scale in each tomographic image scaled to amplitude where pixel colour is an indicator of acoustic energy ( Figure 2 ) . In these cross section-like tomographic images , we then defined areas of interest , one including the entire image of the moth including the image of the substrate below , and the other just including the substrate below the moth . From these areas , we then derived two measures of echo-acoustic salience . First , we calculated the change in depth profile from the tomography as the range of pixels starting with the first pixel above background noise ( either moth or substrate echo ) to the last such pixel of substrate echo in a perpendicular direction relative to the substrate surface ( see Figure 2B ) , to derive a detailed depth profile , which builds the informational basis of the bats' perception of an object projecting off the surface ( see Figure 2C ) . Second , we measured ‘shadow size’ and ‘shadow strength’ . Shadow size and strength were determined by first subtracting the tomography without a moth from the equivalent tomography with the moth ( see Figure 2D ) . For each column of pixels ( horizontal position ) , we then measured how much darker ( in percent of pixel brightness , i . e . , acoustic energy ) the tomographic image of the substrate got through the shadow cast by the moth ( shadow strength profile; Figure 2E ) . From this shadow strength profile , we then defined the shadow area as all horizontal positions when the substrate echo had fallen by at least 25% and measured ‘shadow size’ from the first to the last such pixel . ‘Shadow strength’ was then measured as the mean percentage value of the shadow strength profile over the entire shadow area ( see Figure 2E ) . For each specimen , we scaled the measurements' area of interest to specimen size by calculating the respective shadow size . To avoid bias introduced by different echo clarity on different substrates , we measured shadow size cast by each specimen from slate , because this substrate gave the most homogeneous substrate echoes and thus the most clearly defined shadows ( Figure 2 ) . Subsequently , we applied this same shadow size to all other substrates but centred the defined area by eye with the visible moth to account for slight variations ( <2 mm ) in moth pin position . | While bats are far from blind , they are famed for their use of sound waves to home in on their prey . As they fly , bats send out a series of high-frequency calls that bounce off nearby objects , including insects . By listening to the echoes , the bats are able to build up an auditory image of their environment and thus pinpoint the location of their prey . Although echolocation is effective for localizing flying insects , it is less suited to detecting those that are resting on surfaces . This is due to the difficulty of distinguishing sound waves that bounce off the insect from those that are reflected by the surrounding surface . While some species of bats get around this problem by listening for faint sounds made by prey , such as mating calls or the fluttering of insect wings , a number of bat species have found a way to detect entirely motionless prey using echolocation . Clare and Holderied have now worked out how bats might do this . A method was devised to convert the sound waves that bounce off an object into visual signals , and thus , open them up for analysis by human observers . This technique was used to scan moths of various shapes and sizes resting on different surfaces: smooth slate , leaves , coarse limestone , and rough bark . These experiments showed that the difference in the strength of the echoes from the moth and its surroundings varied depending on the texture of the surface . Specifically , the difference was greatest when the insect was resting on smooth slate and smallest when it was on rough bark , suggesting that choice of surface affects how easy it is to spot an insect . Unexpectedly , however , the data also indicate that bats may search for insects by seeking out interruptions in the echoes from the surface , rather than trying to detect echoes from the prey itself . This makes the bats' task a little easier as it means that they do not have to make adjustments for the differing sizes and shapes of insects . Instead , they can use an acoustic search image for what the surface is like and look for missing parts covered up by the prey . Clare and Holderied's findings thus generate a number of predictions about the behaviour of bats and insects in their natural environment . Bats should prefer searching for insects on smooth surfaces rather than rough ones; but insects might attempt ‘acoustic camouflage’ by choosing to rest on rough surfaces rather than smooth . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"ecology"
] | 2015 | Acoustic shadows help gleaning bats find prey, but may be defeated by prey acoustic camouflage on rough surfaces |
T cell receptor ( TCR ) activation leads to a dramatic reorganisation of both membranes and receptors as the immunological synapse forms . Using a genetic model to rapidly inhibit Zap70 catalytic activity we examined synapse formation between cytotoxic T lymphocytes and their targets . In the absence of Zap70 catalytic activity Vav-1 activation occurs and synapse formation is arrested at a stage with actin and integrin rich interdigitations forming the interface between the two cells . The membranes at the synapse are unable to flatten to provide extended contact , and Lck does not cluster to form the central supramolecular activation cluster ( cSMAC ) . Centrosome polarisation is initiated but aborts before reaching the synapse and the granules do not polarise . Our findings reveal distinct roles for Zap70 as a structural protein regulating integrin-mediated control of actin vs its catalytic activity that regulates TCR-mediated control of actin and membrane remodelling during formation of the immunological synapse .
T cell receptor ( TCR ) activation induces a highly structured reorganisation of the receptors at the immunological synapse . The mature synapse forms over several minutes and is segregated into three concentric regions , called the central , peripheral and distal supramolecular activation complexes ( SMACs ) ( reviewed in Jenkins and Griffiths ( 2010 ) ) . The regions can be distinguished by the differential localization of molecules . For instance , Lck and PKC-θ cluster with the TCR within the central SMAC ( cSMAC ) , while integrins and associated talin are excluded centrally but accumulate at the peripheral SMAC ( pSMAC ) ( Monks et al . , 1998 ) and actin clusters within the distal SMAC ( dSMAC ) . In secretory synapses , formed by cytotoxic T lymphocytes ( CTL ) , secretion occurs into a small secretory cleft next to the cSMAC and within the pSMAC ( Stinchcombe et al . , 2001b ) . The membranes around the secretory cleft are very tightly opposed around the area where cytotoxic proteins are secreted . Exactly how this organisation of membranes is set up as CTL and target cells interact is not known . Zap70 is a key kinase in the TCR signalling pathway . Via its tandem SH2 domains , it associates with tyrosine phosphorylated CD3 and zeta chains and is itself phosphorylated by Src family kinases in response to TCR stimulation ( Au-Yeung et al . , 2009; van Oers et al . , 1996 ) . Zap70 acts as a critical effector of downstream signalling after initial engagement of TCR . Loss of Zap70 in humans leads to Severe Combined Immunodeficiency ( SCID ) characterized by the absence of CD8 T cells and the presence of non-functional CD4 T cells ( Arpaia et al . , 1994; Chan et al . , 1994; Elder et al . , 1994 ) . Defects in thymic development are revealed in mice deficient in Zap70 where no mature T cells develop due to a block in positive selection ( Negishi et al . , 1995; Kadlecek et al . , 1998 ) . Due to developmental abnormalities , studies on the role of Zap70 in CTL-mediated killing have been limited . The derivation of mice expressing an engineered Zap70 mutant , the catalytic activity of which can be blocked by the use of a small molecule inhibitor ( Levin et al . , 2008; Au-Yeung et al . , 2010 ) has changed this . This analog-sensitive Zap70 protein [Zap70 ( AS ) ] has a methionine to alanine substitution in its catalytic site which allows it to accommodate the bulky ATP-competitive inhibitor , 3-MB-PP1 , which impairs Zap70 ( AS ) catalytic function but has little effect on wild-type Zap70 . This model , with Zap70 ( AS ) controlled by the addition of a rapidly acting small molecule inhibitor that is genetically selective , has opened the way to studying the role of Zap70 in functional mature T cells . Importantly , this system is able to distinguish the roles played by the catalytic activity of Zap70 as opposed to its structural contributions since the inhibited kinase is present , associates with the TCR , is tyrosine phosphorylated by Lck and has the capacity to recruit other signalling molecules . Initial studies with this system have shown that , when added to CD4 T cells containing the Zap70 ( AS ) allele , 3-MB-PP1 inhibits Zap70 ( AS ) catalytic activity within 30 s , thereby leading to loss of LAT and ERK phosphorylation and ablation of the calcium increase in response to TCR cross-linking . Inhibition of Zap70 ( AS ) does not affect its phosphorylation by the upstream kinase , Lck , nor does it affect the catalytic activity of T cells expressing wild-type Zap70 . Intriguingly , studies of CD4 T cells revealed that Zap70 catalytic activity was not required for integrin activation , but rather that Zap70 plays a catalytic-independent role in integrin activation ( Au-Yeung et al . , 2010 ) . The Zap70 ( AS ) model therefore provides an unprecedented opportunity for examining the role of integrin vs TCR mediated events during the formation of the immunological synapse as TCR activation is required for integrin activation ( Burbach et al . , 2007 ) . This is particularly interesting since integrin activation alone has been shown to trigger polarisation of both centrosome and secretory granules to the immunological synapse in Natural Killer ( NK ) cells; although both activating receptor and integrin activation are required for degranulation ( Bryceson et al . , 2005; March and Long , 2011; Bryceson et al . , 2009 ) as well as remodelling of synaptic actin required for cytokine secretion ( Brown et al . , 2012 ) . Studies on cytokine secretion from CD4 cells suggest that cdc42 may be involved in actin remodelling at the site of secretion ( Chemin et al . , 2012 ) . Centrosomal docking at the plasma membrane is a key step for polarised secretion from CTL , delivering secretory granules along microtubules to the point of centrosomal contact at the synapse where cortical actin is reduced at the site of secretion ( Stinchcombe et al . , 2006; Stinchcombe and Griffiths , 2007 ) . The importance of centrosomal docking at the plasma membrane is clear in CTL in which Lck expression is conditionally deleted . In Lck-deficient CTL the centrosome only partially migrates towards the synapse and does not reach the plasma membrane; the secretory granules are not delivered to the synapse , and targets are not killed ( Tsun et al . , 2011 ) . Previous studies have revealed roles for Lck , LAT , SLP76 and Zap70 ( Lowin-Kropf et al . , 1998; Kuhne et al . , 2003; Tsun et al . , 2011 ) in TCR-induced polarisation of the microtubule organising centre ( MTOC ) towards the immunological synapse . A study using the Jurkat variant , P116 , which lacks Zap70 , showed only 50% of P116 cells were able to polarise their MTOC towards the synapse after superantigen stimulation ( which activates via a LAT-independent pathway [Bueno et al . , 2006] ) while this number increased to 75% if Zap70 was re-expressed in these cells ( Blanchard et al . , 2002 ) . This same study showed that cSMAC recruitment of PKCθ and LAT , was impaired in Zap70-deficient P116 cells , although CD3ζ clustering was not . The link between these events was not clear . In this study , we analysed the requirements of Zap70 catalytic activity , distinct from its structural role , in the formation of the immunological synapse , polarisation of the centrosome and granules as well as subsequent cytotoxic functions by effector CD8 CTL . We have taken advantage of the Zap70 ( AS ) system in which the structural protein , Zap70 , is present but whose catalytic activity can be rapidly and selectively inhibited . This provided us with a unique system in which to ask whether , as in NK cells , integrin activation is sufficient for centrosome and granule polarisation and whether there are distinct roles for the structural functions vs catalytic activity of Zap70 in formation of the immunological synapse .
C57BL/6 Zap70 ( AS ) CTL were generated by stimulation with irradiated allogeneic Balb/c splenocytes in vitro , each week for 2 weeks , before using the activated CTL for assays . The level of cytotoxicity was determined by lactate dehydrogenase ( LDH ) release from P815 target cells . When we examined the ability of Zap70 ( AS ) CTL to induce target cell death in the presence of the 3-MB-PP1 inhibitor , we saw a complete abrogation of killing ( Figure 1A ) . Given that the inhibition of Zap70 has been shown to impair CD4 T cell activation and cytokine production ( Au-Yeung et al . , 2010 ) , we examined the ability of CTL to produce cytokines after a 5 hr in vitro stimulation with anti-CD3ε . CTL with an inactive Zap70 demonstrated a loss of IFN-γ , TNF-α and IL-2 cytokine production ( Figure 1B ) . Therefore , despite being previously activated , CTL still rely on Zap70 signalling for production of cytokines . 10 . 7554/eLife . 01310 . 003Figure 1 . T cell killing and cytokine production is dependent on the catalytic activity of Zap70 . ( A ) Target cell lysis of P815 targets by Zap70 ( AS ) CTL in the presence ( squares ) or absence ( circles ) of 10 µM 3-MB-PP1 . Graphs show the mean percentage cytotoxicity of triplicates ± SD for effector to target ( E:T ) ratios shown; representative of three independent experiments . ( B ) FACS analysis of intracellular staining for IFN-γ ( y-axes ) TNF-α and IL2 ( x-axes ) production by Zap70 ( AS ) CTL stimulated with anti-CD3 ±10 µM 3-MB-PP1 . DOI: http://dx . doi . org/10 . 7554/eLife . 01310 . 003 We examined the ability of CTL lacking Zap70 catalytic activity , to form immunological synapses . We determined whether they formed cSMAC by looking for the clustering of Lck and PKC-θ at the synapse and whether a pSMAC was formed by assaying their ability to clear the integrin-associated protein , talin , into a concentric ring around the cSMAC . Zap70-inactive CTL were able to bind and form conjugates with target cells almost as well as Zap70-active CTL , with 60% of Zap70 ( AS ) CTL ( n = 70 ) forming conjugates in the presence of 3-MB-PP1 compared with 67% ( n = 60 ) without inhibitor . When activated Zap70 ( AS ) CTL were conjugated to P815 target cells in the presence of 3-MB-PP1 , their ability to clear talin into a ring at the pSMAC was impaired ( Figure 2A ) . Instead , accumulation of talin labelling was seen across the synapse , when viewed in the z plane ( Figure 2A , inset ) . cSMAC formation was also impaired , because the same conjugates displayed a drastic reduction in the accumulation of Lck and PKC-θ at the cSMAC ( Figure 2B , C ) . These results indicate that Zap70 activity is important in the redistribution of talin and signalling proteins during the formation of a stable synapse . 10 . 7554/eLife . 01310 . 004Figure 2 . Inhibition of Zap70 activity impairs formation of both the cSMAC and pSMAC . ( A–C ) Confocal projections of Zap70 ( AS ) CTL conjugated to P815 targets ±10 µM 3-MB-PP1 . Cells are labelled with Hoechst ( nuclei , blue ) and antibodies against γ-tubulin ( AlexaFluor 546; red ) and either talin ( A ) , Lck ( B ) or PKC-θ ( C ) ( AlexaFluor-488; green ) in the xy plane ( scale bar , 5 µm ) or as 1 µm reconstructions en face across the synapse in the xz plane ( insets , scale bar , 3 µm ) . Graphs show the quantitation of conjugates with the percentages of conjugates displaying talin rings ( A ) , Lck cSMACs ( B ) and PKC-θ cSMACs ( C ) at the synapse . DOI: http://dx . doi . org/10 . 7554/eLife . 01310 . 004 Previous data have demonstrated that talin is required for F-actin polarisation to the synapse ( Wernimont et al . , 2011 ) . Therefore , given the loss of talin clearance , immunofluorescence microscopy was used to examine the ability of CTL-P815 conjugates to clear actin and form the distal SMAC ( dSMAC ) in the absence of ZAP activity ( Figure 3A ) . Conjugates were scored based on the phenotypes of total actin clearance , partial , or no clearance from the synapse as described in ‘Materials and methods’ ( Figure 3A ) . The majority of conjugates formed with Zap70-active CTL accumulate actin at the contact site , which clears into a ring of actin at the dSMAC of the synapse ( 76 . 5% , n = 65 ) ( Figure 3B , black ) consistent with other studies of CTL ( Tsun et al . , 2011; Zhao et al . , 2012 ) . In contrast , when CTL lack Zap70 catalytic activity , they lose this ability , and the proportion of conjugates displaying clearance drops to 26% ( n = 83 ) ( Figure 3B ) . When actin clearance is examined in conjugates which do not display a cSMAC ( 75% , n = 83 ) , the effect is even more pronounced , with 91 . 5% of conjugates failing to clear actin . Interestingly , accumulation of actin at the synapse was still observed in the absence of Zap70 activity ( Figure 3C ) . These results reveal that actin accumulates across the synapse in the absence of Zap70 catalytic activity , but fails to clear to form the outer ring or dSMAC . Moreover , cSMAC formation is impaired in the absence of actin clearance . 10 . 7554/eLife . 01310 . 005Figure 3 . Actin clearance from synapses formed by CTL requires Zap70 activity . ( A ) Confocal projections of Zap70 ( AS ) CTL conjugated to P815 targets showing actin organisation at the synapse ±10 µM 3-MB-PP1 , in the xy plane ( scale bar , 5 µm ) or as 1 µm reconstructions en face across the synapse in the xz plane ( insets , scale bar , 3 µm ) . Labelling with Hoechst ( blue ) and antibodies against actin ( AlexaFluor-546; red ) and Lck ( AlexaFluor-488; green ) . ( B ) Quantitation of actin organisation at the synapse in conjugates in the presence ( n = 83 ) or absence ( n = 64 ) of 10 µM 3-MB-PP1 and ( C ) for conjugates in the presence of 10 µM 3-MB-PP1 ( n = 83 ) , in which a cSMAC , identified by Lck clustering , is present ( cSMAC ) or absent ( no cSMAC ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01310 . 005 Previous studies have shown that phosphorylated ERK may play a role in actin reorganisation as ERK co-localises with actin at the immune synapse , is implicated in granule and MTOC polarisation , and is required for CTL degranulation ( Robertson et al . , 2005; Jenkins et al . , 2009 ) . To determine how the inhibition of Zap70 catalytic activity affects downstream signalling including ERK phosphorylation in CTL , we performed western blot analysis of Zap70+/− and Zap70 ( AS ) CTL lysates ( Figure 4A ) . Inhibition of Zap70 catalytic activity does not impair its capacity to be phosphorylated on tyrosine 319 , which is inducibly phosphorylated upon CD3 crosslinking , and remains similarly phosphorylated in the presence of a high concentration of 3-MB-PP1 ( 10 μM ) . This result is consistent with Zap70 as a substrate of the upstream Src kinase Lck . However , the phosphorylation of LAT , a substrate of Zap70 , as well as PLCγ and ERK all depend on Zap70 catalytic activity . These results suggest a signalling pathway from Zap70 via the LAT signalosome and PLCγ is required for activation of ERK in CTL . It has also been reported that a PI3K-dependent pathway is required for activating ERK and CTL degranulation ( Robertson et al . , 2005 ) . To determine whether this pathway may also be dependent on Zap70 activity we probed for phosphorylation of AKT , which is activated downstream of PI3K . Indeed , Akt phosphorylation was sensitive to Zap70 inhibition , suggesting that ERK is activated downstream of Zap70 through a LAT and PLCγ pathway or alternatively through a pathway that also includes AKT . 10 . 7554/eLife . 01310 . 006Figure 4 . TCR signalling downstream of Zap70 is impaired in the absence of Zap70 catalytic activity . ( A ) In vitro generated Zap70+/− and Zap70 ( AS ) CTLs were left unstimulated or were stimulated for 2 min by soluble anti-CD3 ( 10 μg/ml ) and cross-linking secondary antibodies , in the presence of vehicle alone ( DMSO ) or 5 or 10 μM 3-MB-PP1 . The phosphorylation status of the indicated TCR signalling molecules was determined by Western blot analysis . Molecular weights: ZAP-70 , 70kD; LAT , 38kD; PLCγ 150kD; ERK 42-44kD; Akt , 60kD; Vav-1 , 100kD . ( B ) Vav-1 immunoprecipitated from Zap70+/− OT-I and Zap70 ( AS ) OT-I CTL treated with vehicle or 3-MB-PP1 as shown . Immunoprecipitates were probed for phosphorylation on tyrosine 160 or total tyrosine phosphorylation of Vav-1 . DOI: http://dx . doi . org/10 . 7554/eLife . 01310 . 00610 . 7554/eLife . 01310 . 007Figure 4—figure supplement 1 . ICAM adhesion assays . Zap70+/− and Zap-70 ( AS ) CTL were stimulated with soluble anti-CD3 in the presence of 10 μM 3-MB-PP1 or DMSO alone . The graphs show the percentage of cells adherent to ICAM-1 coated plates after 10 min of stimulation . Data shown are the mean ± SEM from triplicate samples from one of three independent experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 01310 . 007 Since our previous studies had shown that integrin activation occurs in the absence of Zap70 catalytic activity in naïve CD4 T cells ( Au-Yeung et al . , 2010 ) , we asked whether the integrin-mediated activation of Vav-1 ( Gao et al . , 2005 ) was seen in CTL in which Zap70 catalytic activity was inhibited . This was of particular interest because Vav-1 has been shown to play an important role in both TCR- and integrin-mediated activation of actin reorganisation in T cells ( del Pozo et al . , 2003 , Garcia-Bernal et al . , 2009; Tybulewicz , 2005 ) . We found that TCR activation triggers Vav-1 phosphorylation independently of Zap70 kinase activity , with Vav-1 phosphorylation occurring in both Zap70+/− and Zap70 ( AS ) CTL lysates from cells treated with 3-MB-PP1 , both for overall Vav-1 phosphorylation as well as at Y160 , a site that is selectively phosphorylated upon TCR stimulation or αvβ3 integrin-mediated activation ( Gao et al . , 2005; Miletic et al . , 2006 ) . Only a slight decrease in Vav1 phosphorylation is evident in Zap70 ( AS ) CTL with 10 μM 3MB-PP1 suggesting Zap70 catalytic function makes only a minimal contribution to Vav-1 phosphorylation ( Figure 4B ) . These results suggest that Vav-1 phosphorylation , and presumably its GEF activation , occurs independently of Zap70 catalytic activity , while the pathway leading to ERK and PI3K activation are dependent on Zap70 catalytic activity . This supports the idea that integrin activation occurs in CTL in which Zap70 is catalytically inhibited , as previously observed in Treg ( Au-Yeung et al . , 2010 ) . We examined integrin activation further by asking whether there were differences in adhesion or in the speed of movement of Zap-inhibited CTL . Adherence to an ICAM-1 coated plate was measured before and after TCR activation using anti-CD3ε for 10 min . TCR activation increased the percentage of CTL binding to ICAM-1 from 20% to 30% for Zap70+/− and from 16% to 34% for Zap70 ( AS ) CTL . 3-MB-PP1 treatment resulted in 32% binding for Zap70+/− and 25% for Zap70 ( AS ) CTL ( Figure 4—figure supplement 1 ) . While there does appear to be some reduction in CTL adhesion upon 3-MB-PP1 treatment , these data also suggest there is residual Zap70 catalytic-independent integrin function . We also examined the level of integrin activation by determining the speed of CTL movement using live cell imaging . We found no difference in the speed of movement on an ICAM-1 surface with CTL moving with an average speed of 9 μm/min when Zap70 was catalytically active or inactive ( Videos 1 and 2; n>84 each ) . These results support the idea that Zap70 catalytic activity is not required for integrin activation in CTL . Our finding that conjugate formation is not impaired in Zap70-inactive CTL also supports this model . 10 . 7554/eLife . 01310 . 011Video 1 . Live cell imaging of Zap70 ( AS ) OT-I CTL transfected with Lifeact-EGFP ( green ) moving on a glass coverslip coated with 0 . 5 µg/ml ICAM−1 + 0 . 1% DMSO . DOI: http://dx . doi . org/10 . 7554/eLife . 01310 . 01110 . 7554/eLife . 01310 . 012Video 2 . Live cell imaging of Zap70 ( AS ) OT-I CTL transfected with Lifeact-EGFP ( green ) moving on a glass coverslip coated with 0 . 5 µg/ml ICAM−1 +10 µM 3-MB-PP1 . DOI: http://dx . doi . org/10 . 7554/eLife . 01310 . 012 Thus , distinct TCR-dependent signalling events are influenced by Zap70 scaffolding and kinase activity . Importantly , catalytically inhibited Zap70 can still participate in signal transduction without active catalysis , by interacting with other signalling proteins via its phosphorylated tyrosines . These results in activated CTL are consistent with previous studies in naïve CD4 T cells , which demonstrated a Zap70 kinase-independent scaffold function , where Zap70 forms a complex with the adapter protein Crk , which activates the GTPase Rap1 , which subsequently regulates integrin-mediated adhesion ( Au-Yeung et al . , 2010 ) . Centrosome polarisation to the synapse is an important step in CTL killing , because centrosome docking at the plasma membrane is responsible for directing the cytolytic granules to the secretory cleft at the cSMAC . Given that NK cells can polarise both centrosome and granules to the synapse in response to integrin activation alone ( Bryceson et al . , 2005 ) , it was of interest to ask whether the centrosome could polarise in CTL in which Zap70 was catalytically inhibited but integrin activation could still occur . Our previous studies have shown that Lck is essential for docking of the centrosome at the plasma membrane since when Lck expression is inducibly turned off in mature CTL , the centrosome polarises towards the immune synapse , but does not dock at the plasma membrane . In the absence of centrosome docking at the plasma membrane , the granules cannot be delivered to the immunological synapse and CTL killing is abolished ( Tsun et al . , 2011 ) . Zap70 is a substrate of Lck and its phosphorylation by Lck activates Zap70’s kinase activity . Therefore we asked whether centrosome polarisation occurs in the absence of Zap70 catalytic activity by using immunofluorescence microscopy to examine the ability of Zap70 ( AS ) CTL-P815 conjugates to polarise their centrosomes ( γ-tubulin labelling ) ( Figure 5A ) to the synapse identified by Lck labelling ( green ) in the presence or absence of inhibitor . Conjugates were classified according to the location of the centrosome relative to the synapse and nucleus ( as described in ‘Materials and methods’ and illustrated in Figure 5A ) , and by the distance between the centrosome and the synapse ( Figure 5B ) . The majority of conjugates ( 65% ) with catalytically active Zap70 , displayed centrosomes tightly polarised to the cSMAC ( Figure 5B ) ( n = 118 ) , in accordance with previously published work ( Jenkins et al . , 2009 ) . In contrast , Zap70 inhibition disrupted centrosome polarisation , with 54% of 3-MB-PP1 treated conjugates showing a centrosome location >5 μm from the synapse ( Figure 5B ) ( n = 85 ) . 10 . 7554/eLife . 01310 . 008Figure 5 . Centrosome and granule polarisation to the synapse is impaired in Zap70 inactive CTL . ( A and C ) Confocal projections of Zap70 ( AS ) CTL conjugated to P815 targets , labelled with Hoechst ( blue ) and antibodies against Lck ( AlexaFluor-488; green ) , γ-tubulin ( AlexaFluor-546; red ) and ( C ) LAMP-1 ( AlexaFluor-633; white ) ( scale bars , 5 µm ) illustrating centrosome ( A ) and granule ( C ) polarisation in CTL . Quantitation of conjugates in the presence ( n = 325 ) or absence ( n = 163 ) of 10 µM 3-MB-PP1 , showing distance of centrosome from the synapse ( B ) or granule polarisation phenotypes ( D ) , illustrated in ( A ) and ( C ) , as a percentage of total conjugates formed . ( NB LAMP-1 stains both CTL and target lysosomes . ) DOI: http://dx . doi . org/10 . 7554/eLife . 01310 . 008 To determine whether inhibition of Zap70 catalytic activity affected granule polarisation , the Zap70 ( AS ) ±3-MB-PP1 conjugates were labelled with an antibody to LAMP-1 ( CD107a ) ( Figure 5C , white ) , a lysosomal membrane protein used as a marker of the secretory granules in T cells ( and lysosomes in the target cells ) , and scored for the location of the granules , relative to the synapse ( as described in ‘Materials and methods’ ) . Immunofluorescence microscopy was used to identify granule location relative to the synapse identified by Lck labelling ( green ) ( Figure 5C ) . Although the majority ( 44% ) of control ( Zap70 ( AS ) +DMSO ) conjugates ( n = 117 ) show granules tightly clustered to the synapse , conjugates treated with 3-MB-PP1 showed a loss of granule polarisation to the synapse ( Figure 5D ) ( n = 92 ) . The loss of centrosome and granule polarisation in cells lacking Zap70 catalytic activity is consistent with the loss of cytotoxicity . These results suggest that , in contrast to NK cells , integrin activation is not sufficient for centrosome and granule polarisation to the synapse . We also examined centrosome polarisation using live cell microscopy . In order to examine conjugates formed by identical TCR interactions we used CTL derived from lines crossed onto the TCR transgenic OT-I background . Zap70 ( AS ) OT-I CTL were transfected with Lifeact-EGFP ( to label polymerised actin ) and PACT-mRFP ( to label the centrosome ) and treated with 3-MB-PP1 . Target cell killing by Zap70 ( AS ) OT-I CTL show complete inhibition of killing when treated with 3-MB-PP1 ( Figure 6—figure supplement 1 ) . Live cell imaging of Zap70-active OT-I CTL show the centrosome polarising right up to the contact site within 6 min of initial interaction with the target ( Figure 6; Video 3; n = 34 ) . However although the centrosome begins to polarise towards the synapse in Zap70 ( AS ) OT-I CTL treated with 3-MB-PP1 , the centrosome fails to reach the contact site with the closest point of contact ∼3 μm at 6 min after initial contact ( Figure 6; Video 4; n = 50 ) . In each video examined from Zap-deficient CTL , centrosome polarisation began , but aborted before the centrosome docked at the contact site formed by the synapse . 10 . 7554/eLife . 01310 . 009Figure 6 . CTL lacking Zap70 catalytic activity show abortive centrosome polarisation . Single frames from Videos 3 and 4 showing 3 min intervals of Zap70 ( AS ) OT-I CTL ±10 µM 3-MB-PP1 . CTL were transfected with Lifeact-EGFP ( green ) and mPACT-RFP ( centrosome marker , red ) , and target cells expressing farnesylated mTagBFP2 ( blue ) . Scale bar , 5 µm; n = 50 for inhibitor and 34 for control treatments . DOI: http://dx . doi . org/10 . 7554/eLife . 01310 . 00910 . 7554/eLife . 01310 . 010Figure 6—figure supplement 1 . Lysis of targets by Zap70 ( AS ) OT-I CTL is specifically inhibited by 3-MB-PP1 . Lysis of EL4 target cells by Zap70 ( AS ) OT-I CTL ±10 µM 3-MB-PP1 with percentage target cell death ( y-axis ) and E:T ratio ( x-axis ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01310 . 01010 . 7554/eLife . 01310 . 013Video 3 . Live cell imaging of Zap70 ( AS ) OT-I CTL transfected with Lifeact-EGFP ( green ) and mPACT-RFP ( centrosome marker , red ) , with EL4 target cells expressing farnesylated mTagBFP2 ( blue ) +0 . 1% DMSO . DOI: http://dx . doi . org/10 . 7554/eLife . 01310 . 01310 . 7554/eLife . 01310 . 014Video 4 . Live cell imaging of Zap70 ( AS ) OT-I CTL transfected with Lifeact-EGFP ( green ) and mPACT-RFP ( centrosome marker , red ) , with target cells labelled with EL4 target cells expressing farnesylated mTagBFP2 ( blue ) +10 µM 3-MB-PP1 . DOI: http://dx . doi . org/10 . 7554/eLife . 01310 . 014 In order to determine the defect in synapse formation and centrosome polarisation and docking at higher resolution we examined the ultrastructure of the contact site formed between CTL and targets when Zap70 catalytic activity was inhibited . Conjugates formed between CTL and targets were fixed and processed for electron microscopy at 25 , 40 and 60 min after mixing . At 25 ( Figure 7A ) and 40 ( Figure 7C ) minutes , the contact sites formed between CTL with active Zap70 and target cells show typical secretory synapses ( Stinchcombe et al . , 2001b ) with flat stretches of tight membrane interactions between the cells surrounding a central intercellular gap between the two cell membranes , termed the secretory cleft . The cleft forms an extracellular space containing heterogeneous membranous and granular material . Centrioles , lytic granules and Golgi elements are polarised up to the CTL plasma membrane in the middle of the contact site , opposite the secretory cleft . At these stages of interaction , the target cell appears relatively intact although the endoplasmic reticulum at 40 min is slightly swollen and vacuolated , characteristic of target cell death ( Figure 7C ) . By 60 min , there is marked evidence of target cell death with extensive endoplasmic reticular vacuolation and swollen mitochondria , with the CTL appearing to retract from an etiolated , dying target ( Figure 7E ) . 10 . 7554/eLife . 01310 . 015Figure 7 . The structure of the immunological synapse is severely impaired in CTL upon Zap70 inhibition . Electron micrographs of single ( A–F , left panel G ) or non-sequential serial ( right panels , G ) thin ( 50–70 nm , A–E , G ) or semi-thin ( 70–100 nm , F ) lead-stained sections through the contact site formed between Zap70 ( AS ) CTL ( CTL ) and P815 target cells ( target ) , conjugated for 25 ( A and B ) , 40 ( C and D ) or 60 ( E–G ) min at 37C in the absence ( A , C , E , G ) or presence ( B , D , F ) of 10 µM 3-MB-PP1 . Secretory cleft ( SC ) ; interdigitations , between CTL and targets ( black arrowheads ) ; centrioles ( white arrowheads ) , lytic granules ( asterisks ) ; Golgi elements ( G ) and nuclei ( N ) in CTL , and endoplasmic reticular ( ER ) and mitochondria ( m ) in target cells , are indicated in lower power images for ( A–F ) and all images for ( G ) . Scale bars: low power images , 2 μm; high power images bar , 1 μm . Only the ends of the mother centriole appendages are visible in ( B ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01310 . 01510 . 7554/eLife . 01310 . 016Figure 7—figure supplement 1 . EM quantitation . Quantitation of images of Zap70 ( AS ) CTL-target conjugates formed ±3-MB-PP1 at 25 , 40 , or 60 min after conjugation in EM images in which CTL showed both contact with a target and at least one centriole within the section; ( n = 24 @ 25 min , 51@ 40 min and 22 @ 60min +10 µM 3-MB-PP1 ; n = 26 @ 25 min , 30 @ 40 min and 60 @ 60 min −10 µM 3-MB-PP1 ) . Graphs show percentage conjugates with ( A ) a secretory cleft at the contact site in the absence ( black bars ) or presence ( grey bars ) of 10 µM 3-MB-PP1; ( B ) centriole distances from the synapse; ( C ) 0 , 1–2 , 3–4 or 5–6 membrane projections >1000 nm/cell across the contact site ( y-axis ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01310 . 016 In the absence of Zap70 kinase activity the ultrastructure of the synapse is very different ( Figure 7B , D , F ) . There is no evidence of the tight , flat membrane–membrane associations between CTL and target . Secretory clefts are absent , and neither centrioles , Golgi cisternae nor lytic granules are polarised towards the target cells . Instead , the contact site is highly interdigitated , formed of long projections from the CTL surface ( black arrowheads ) . Conjugates formed between CTL lacking Zap70 activity and targets after 25 , 40 and 60 min of interaction show similar phenotypes . There is no evidence of target cell death at any of these time points in these samples . Notably points of contact between CTL and target are restricted to the tips of projections , providing limited sites for receptor interactions between the cells . Quantitation of the EM data reveal that with Zap70 ( AS ) CTL + 3-MB-PP1 the secretory cleft is absent in >85% of conjugates analysed at any time point ( Figure 7—figure supplement 1A ) ; the centrosome is not docked and instead is found at distances greater than 1 μm from the synapse in >70% of conjugates ( Figure 7—figure supplement 1B ) and projections are present between CTL and target ( Figure 7—figure supplement 1C ) . In contrast , in the absence of inhibitor Zap70 ( AS ) CTL showed secretory clefts in >50% of conjugates observed at 25 and 40 min , reduced to 27% only at 60 min , when many of the target cells were dead . Tightly polarised centrosomes are seen most frequently at the 25 min time-point when 69% of conjugates have at least one centriole <500 nm from the plasma membrane and 85% of conjugates show no projections between CTL and target . Interestingly up to 60% of CTL lacking Zap70 catalytic activity show the centrosome within 3 μm of the synapse in conjugates formed at 25 min ( Figure 7—figure supplement 1B ) , consistent with the initial polarisation up to this point observed in live imaging ( Figure 6 ) , decreasing to 35% in conjugates formed at 60 min . The unusual ultrastructural morphology of the synapse formed by CTL lacking Zap70 activity most closely resembles that seen in wild-type CTL at very early stages of synapse formation , when the target cell is intact and before the centrosome has polarised to the synapse ( Figure 7G ) . At this stage of synapse formation there are many interdigitations between CTL and target and neither the flattening of the membranes between the two cells , nor the secretory cleft are seen . Here too the contact points between CTL and target are only found at the tips of projections between CTL and target . These observations support the idea that loss of Zap70 catalytic activity arrests the reorganisation of membranes at the synapse at an early stage prior to formation of the secretory cleft . Importantly these images reveal the ultrastructure that underlies the accumulation of actin seen across the synapses formed by Zap70 catalytically inhibited CTL and at the early stages of synapse formation in wild-type CTL , revealing the first link between the clearance of actin and the flattening of the membranes to form an extended area of contact during synapse formation .
In this paper we make use of the Zap70 ( AS ) model to study the role of Zap70 catalytic activity in CTL . While earlier studies revealed that target cell killing was inhibited in the absence of Zap70 activity ( Au-Yeung et al . , 2010 ) , the underlying mechanisms for the loss of cytotoxicity were not investigated . We find that in the absence of Zap70 catalytic activity formation of the immunological synapse is arrested at a stage where actin-rich interdigitations dominate the interface between the two cells , with actin accumulated across the synapse and TCR unable to coalesce to form a cSMAC . The membranes between CTL and target are unable to flatten to provide an extended area of contact and the secretory cleft does not form . Signalling downstream of Zap70 is disrupted and neither centrosome nor granules polarise . Our studies provide the first insights into the membrane reorganisations that accompany actin remodelling to form the mature immunological synapse . The Zap70 ( AS ) mouse offers a unique opportunity to examine the role of the catalytic activity of Zap70 in CTL , which can be inhibited rapidly and specifically by addition of the inhibitor 3-MB-PP1 . Previous studies using Zap70 ( AS ) CD4 cells demonstrated that although addition of 3-MB-PP1 inhibits phosphorylation of LAT and ERK , Zap70 itself can still be phosphorylated and take part in signalling . We see the same picture in CD8 CTL . Furthermore we show that phosphorylation of PLCγ , ERK and AKT are all dependent on Zap70 catalytic activity , suggesting that the LAT signalosome and PLCγ are required for ERK activation in CTL , providing a rationale for previous observations implicating both ERK ( Robertson et al . , 2005 ) and PLCγ ( Le Floc’h et al . , 2011 ) in CTL-mediated killing . We also find that catalytically inactive Zap70 can itself be phosphorylated in CTL , supporting a kinase-independent scaffolding role for Zap70 in TCR regulation of integrin-mediated adhesion as reported in Treg cells ( Au-Yeung et al . , 2010 ) . These results support the idea that the initial stages of synapse formation , during which there is an accumulation of actin across the synapse , may well be integrin mediated . Consistent with this is our finding that Vav-1 phosphorylation mediated by the TCR , and possibly involving integrins ( Riteau et al . , 2003; Garcia-Bernal et al . , 2005 , 2009 ) , appears to be independent of Zap70 catalytic activity , with Y160 phosphorylation occurring in Zap70 catalytically inactive CTL . Although earlier studies have implicated Zap70 in the activation of Vav-1 both in vivo ( Deckert et al . , 1996; Kadlecek et al . , 1998; Michel et al . , 1998 ) and in vitro ( Brunati et al . , 1995; Han et al . , 1998 ) , our present study shows phosphorylation of Vav-1 in the absence of Zap-70 catalytic activity , likely mediated by Lck . Our data showing that speed of movement on an ICAM-1 substrate is the same for both Zap70 catalytically active and inactive CTL supports the idea that integrin activation is independent of Zap70 catalytic activity . These results differ from those of an earlier study ( Evans et al . , 2011 ) using the less specific inhibitor , piceatannol , where the reduced speed of T cell migration may have resulted from inhibition of target kinases other than Zap70 ( http://www . kinase-screen . mrc . ac . uk/screening-compounds/345911 ) . The Zap70 ( AS ) model provides a unique opportunity to ask whether , as in NK , integrin activation alone might stimulate centrosome and granule polarisation to the synapse ( Bryceson et al . , 2005 ) . Our results suggest that CTL differ from NK cells in this respect . Although centrosome polarisation is initiated in CTL lacking Zap70 activity , it is aborted before reaching the synapse , and granule polarisation does not take place . These results suggest an underlying difference in the mechanism of polarisation that may reflect the different roles played by NK and CTL in recognising and destroying virally infected and tumour cells . In the absence of Zap70 catalytic activity synapse formation is disrupted at both the level of protein reorganisation and ultrastructure . We find that both the integrin-associated talin , as well as actin , accumulate across the immunological synapse , but do not clear to form the concentric pSMAC and dSMAC actin rings characteristic of the synapse . Lck , the kinase that phosphorylates and activates Zap70 and is a marker of the cSMAC , fails to cluster to form a cSMAC . This observation fits with previous observations on P116 Jurkat cells lacking Zap70 , which failed to cluster PKCθ and LAT at the synapse ( Blanchard et al . , 2002 ) . Our results suggest that the loss of cSMAC formation arises from a failure of actin to clear from the synapse . Actin has been proposed to act as a ‘picket fence’ ( Morone et al . , 2006 ) , impeding the movement of proteins laterally within the membrane and an increased density of actin across the synapse might therefore disrupt cSMAC formation by impeding the movement of TCR . In keeping with this model inhibition of actin reorganisation with jasplakinolide on planar lipid bilayers prevents cSMAC formation ( Beemiller et al . , 2012 ) as well as with earlier observations that actin dynamics are required for effective TCR signalling stemming either from the use of actin inhibitors ( Valitutti et al . , 1995; Delon et al . , 1998; Tskvitaria-Fuller et al . , 2003 ) or depletion of actin regulatory proteins such as dynamin 2 ( Gomez et al . , 2005 ) , or ezrin ( Roumier et al . , 2001 ) , all of which inhibit T cell activation . However our results reveal a much more profound disruption of the whole contact site with actin rich protrusions providing very few points of contact for receptor interactions between the two cells which would impair TCR coalescence to form the cSMAC . Although it has been known for many years that extensive regions of interdigitations ( Kalina and Berke , 1976; Sanderson and Glauert , 1977 , 1979 ) , flattened areas and secretory clefts ( Bykovskaja et al . , 1978; Carpen et al . , 1982; Stinchcombe et al . , 2001b ) can form between killer cells and targets , the functional significance of these differences in membrane organisation has not been clear . Our studies reveal temporal and structural links between the accumulated actin and the extensive interdigitations between CTL and target . Comparing the light and EM images reveals that actin-rich interdigitations seen by EM correspond to confocal images in which actin has accumulated across the synapse , while actin ‘clearance’ in confocal images corresponds to EM images in which the membranes have flattened between the two cells and in which actin-rich protrusions are only apparent at the edges of the synapse reaching out around the target ( Figure 7A , C ) . Consistent with our observations , Ueda et al noted ‘invasive pseudopodia’ forming interdigitations between CD4 T cells and B cells at early stages of interaction ( Ueda et al . , 2011 ) . Our data reveal that , in the absence of Zap70 catalytic activity , the reorganisation of the membranes between CTL and target does not take place . The secretory cleft is also notably absent from synapses formed by Zap70 catalytically-inactive CTL . Although one possible explanation for loss of the secretory cleft , into which granules secrete their content , could be that this cleft results as a loss of secretion from these cells , this seems highly unlikely as the cleft forms properly in synapses made by secretion-deficient CTL lacking Rab27a or Munc13-4 ( Stinchcombe et al . , 2001a , 2004 ) . The very striking loss of the cleft structure in synapses formed by Zap70 inactive CTL demonstrates that cleft formation requires Zap70 activity . In the absence of Zap70 activity , downstream TCR signalling is severely impaired , Lck does not cluster to form a cSMAC and although the centrosome begins to polarise it does not reach the synapse . With the loss of centrosome migration , the lytic granules remain dispersed within Zap70 inactive CTL , do not polarise to the synapse and the target cell is not killed . Previous studies with CTL lacking Lck showed that the centrosome was able to polarise around the nucleus towards the synapse , but was unable to dock at the plasma membrane ( Tsun et al . , 2011 ) . Since live cell imaging was not used to examine centrosome polarisation in Lck-deficient CTL , a direct comparison is not possible . Interestingly actin and talin clearance from the synapse were also impaired in Lck-deficient CTL . Overall our data support a two-stage model in which the initial accumulation of actin at the synapse is mediated by TCR-facilitated integrin activation ( inside out activation of integrins ) , in which Zap70 plays a scaffolding role ( Au-Yeung et al . , 2010 ) , but the subsequent reorganisation of actin requires Zap70 catalytic activity . Integrin-mediated actin dynamics proceed normally in Zap-inactive CTL . Not only is Vav-1 phosphorylated , but actin accumulates at the synapse and our live cell studies show that actin reorganisation required for integrin-mediated motility is unimpaired in the absence of Zap70 activity . The loss of Zap70 catalytic activity does not affect the integrin driven accumulation of actin at the synapse , but results in loss of actin reorganisation once the synapse has formed . Our studies reveal a link between the accumulation of actin and the extensive actin-rich interdigitations formed between CTL and target . It is interesting to note that the only points of contact between CTL and target are at the tips of the interdigitations , and TCR proteins localised to these tips would appear as microclusters . When Zap70 is catalytically active actin clears centrally , the membranes flatten into an extended area of contact and TCR microclusters coalesce to form the cSMAC . Our data are consistent with studies showing that actin clearance is required for the coalescence of microclusters to form the cSMAC ( Campi et al . , 2005; Babich et al . , 2012; Beemiller et al . , 2012; Yi et al . , 2012 ) and we propose that this marks a transition from a highly interdigitated to a flattened interface between the two cells . In this study we find that inhibition of Zap70 catalytic activity arrests the progress of synapse formation at an early stage . We find that , unlike NK cells ( Bryceson et al . , 2005 ) , CTL are unable to polarise their centrosome and granules in response to integrin activation alone . Our findings also reveal a surprising new role for Zap70 in controlling the reorganisation of membranes to form the interface at the immunological synapse , including the cSMAC and secretory cleft . Without these rearrangements , the synapse is not functional . Our findings point to distinct roles played by Zap70 as a structural protein regulating integrin-mediated control of actin vs the role of it’s catalytic subunit controlling TCR mediated control of actin and membrane remodelling during formation of the immunological synapse .
Single-cell suspensions of naive splenocytes were generated using a 70 µM nylon strainer ( Becton Dickinson ) . Equal numbers of Zap70 ( AS ) or Zap70+/− responders were stimulated with BALB/c-derived stimulator splenocytes ( irradiated at 3000 Rad ) in RPMI 1640 , 10% FCS , L-glutamine , sodium pyruvate , 50 U/ml penicillin/streptomycin ( Gibco ) , 50 µM β-2-mercaptoethanol with 100 U/ml human recombinant IL-2 ( Roche , UK ) ( c-RPMI ) , and cultured at 37°C and 5% CO2 . After 4 days , CTL were purified by separation over Ficoll Histopaque 1083-1 ( Sigma–Aldrich , UK ) , washed three times and resuspended in c-RPMI , at 1 × 106 cells/ml . CTL were stimulated every 7 days , up to four times . P815 and EL4 mouse target cells were maintained in RPMI , 10% FCS and L-glutamine . Zap70 ( AS ) OT-I or Zap70+/− OT-I were activated for 4 days with SIINFEKL as previously described ( Jenkins et al . , 2009 ) . 3-methylbenzyl-pyrazolopyrimidine ( 3-MB-PP1 ) was synthesised as described ( Levin et al . , 2008 ) and a stock solution at 10 mM ( 1000x ) was dissolved in DMSO . Antibodies used for immunofluorescence studies were: mouse anti-actin ( AC-40 ) , rabbit anti-actin and rabbit anti-γ-tubulin ( Sigma–Aldrich , United Kingdom ) ; Mouse anti-mouse Lck ( 3A5 ) ( Millipore , United Kingdom ) ; rat anti-mouse CD8 ( YTS192 ) ( gift from H Waldmann , Oxford University ) ; rat anti-mouse CD107a ( LAMP-1 , 1BD4 ) ( Developmental Studies Hybridoma Bank , University of Iowa , Iowa City , IA ) and all secondary Alexa Fluor antibodies ( 405 , 488 , 546 and 633 ) were obtained from Invitrogen . Western blot analysis of CTL lysates was performed as previously described ( Au-Yeung et al . , 2010 ) with antibodies obtained: rabbit antibodies against phospho-ERK T202/Y204 , Zap70 Y319 , LAT Y171 , AKT S473 , total LAT and total AKT ( Cell Signaling , Danvers , MA ) ; rabbit anti-PLCγ Y783 ( Biosource , San Diego , CA ) ; anti-phosphotyrosine ( 4G10 ) , and anti-PLCγ ( total ) ( Millipore ) ; anti-ERK 1 , ERK 2 , and Vav-1 ( Santa Cruz , Santa Cruz , CA ) , Vav Y160 ( R&D Systems , Cambridge , MA ) and anti-β actin ( Sigma , Ronkonkoma , NY ) . Immunoprecipitations were performed by stimulation of CTL with soluble anti-CD3ε and goat anti-Armenian hamster ( Jackson Immunoresearch , West Grove , PA ) crosslinking antibodies for 2 min , followed by centrifugation and resuspension of cells in 1% NP-40 alternative lysis buffer with protease inhibitors . Vav-1 was immunoprecipitated with anti-Vav-1 ( Santa Cruz ) coated protein G sepharose beads ( GE Healthcare , Pittsburgh , PA ) and eluted from the beads with SDS sample buffer containing 1% dithiothreitol . Zap70 ( AS ) CTL ( 5–8 days after the 3rd or 4th stimulation ) and P815 target cells were washed in RPMI , resuspended at 4 × 106 cells/ml and mixed 1:1 , with 10 µM 3-MB-PP1 or DMSO-only control and incubated in suspension for 5 min , before aliquoting onto glass multi-well slides ( Hendley ) , and incubated for a further 15 min at 37°C . Samples were then placed on ice and fixed for 5 min with −20°C methanol , washed six times in PBS and blocked in blocking buffer ( PBS +1% BSA [Sigma] ) . Primary antibodies were resuspended in PBS +0 . 2% BSA and incubated for 1 hr at room temperature , or overnight at 4°C . Samples were washed extensively in PBS +0 . 2% BSA before adding secondary antibodies for 40 min at room temperature . Nuclei were stained with Hoechst ( 1:10 , 000 ) in PBS for 5 min before mounting with 1 . 5 coverglass and Mowiol . Samples were examined using the Andor Revolution Spinning Disk microscope ( with an Olympus microscope , 100x objective ) and lasers exciting at 405 , 488 , 543 and 633 nm . CTL derived from Zap70 ( AS ) OT-I and Zap70+/− OT-I were nucleofected with PACT-mRFP ( Gillingham and Munro , 2000 ) and Lifeact-EGFP using 5 million cells/nucleofection and the mouse T cell nucleofector kit ( Lonza , Germany ) with 1 ml mouse T cell nucleofector medium with Component B ( Lonza ) added post-nuclefection . CTL were topped up with c-RPMI , split between three wells of a 12-well plate and used for imaging 24 hr post nucleofection . EL4 target cells expressing farnesylated mTagBFP2 were pulsed with 1 µM SIINFEKL peptide for 40 min and washed in DMEM to remove peptide and serum , then resuspended in DMEM at 6 . 5 × 105 cells/ml . 250 µl of these target cells were aliquoted onto the glass coverslip of 35 mm glass bottom culture dishes ( MatTek , Ashland , MA ) pre-coated overnight with 0 . 5 µg/ml ICAM-1/FC ( R&D Systems , Minneapolis , MN ) at 4°C . Targets were allowed to settle and adhere for 5 min , after which 1 . 5 ml of RPMI lacking phenol red ( Gibco , United Kingdom ) supplemented with 10% FCS , 25 mM HEPES , and 50 U/ml penicillin and streptomycin ( Imaging medium ) were added to each chamber with 10 µM 3-MB-PP1 or 0 . 1% DMSO . Approximately 2 million CTL were centrifuged at 1200 rpm for 4 min and resuspended in 280 µl of imaging medium with 10 µM 3-MB-PP1 or 0 . 1% DMSO . CTL were added drop-wise over target cells before imaging was started . CTL-target interactions were imaged at 37°C using an Andor ( United Kingdom ) Revolution Spinning Disk microscope with 20x or 100x objective , 1 . 2x camera adapter and environmental chamber ( Oko-lab , Japan ) for temperature and CO2 regulation . Serial confocal 0 . 8 µm Z-stacks were taken at 20 s intervals with excitation of 405 nm , 488 nm , and 561 nm at each Z plane . Videos were processed and analysed using Andor iQ2 software ( Andor Technologies , United Kingdom ) , Imaris x64 ( Bitplane , Switzerland ) and ImageJ ( NIH , USA ) . Conjugate formation was assessed by counting the percentage of CD8 labelled cells conjugated to targets , with all cell nuclei labelled with Hoechst . cSMAC formation was determined by Lck or PKC-θ clustering at the synapse . Clusters were scored when >80% of Lck or PKC-θ in the cell was clustered in the centre of the synapse . Actin rings were classified as ‘cleared’ when a contiguous ring of actin formed the dSMAC; any interruption in this ring was scored ‘partial’ and when a solid wall of actin was observed across the synapse this was scored as ‘not cleared’ as illustrated in Figure 3 . Centrosome position was determined by measuring the distance between the γ-tubulin labelled centrosome and the Lck or PKC-θ labelled cSMAC or contact site using Imaris software ( Bitplane ) . Centrosomes were categorised as ‘docked’ when the centrosome was <1 μm from the synapse; proximal when on the synapse side of the nucleus within 1–3 μm and distal when >3 μm from the synapse . Granules were scored as polarised when >75% of granules were clustered within 5 μm of the synapse; partial when >50% of granules were within 5 μm of the synapse and distal when >75% of granules were >6 μm from the synapse . CTL were taken 5–6 days after the 2nd or 3rd stimulation and incubated overnight in the presence of 1 mg/ml horseradish peroxidase , ( HRP ) ( Serva , Germany ) added directly to the growth medium , to load the secretory lysosomes via the endocytic pathway . Cells were washed extensively in RPMI to remove serum and residual HRP from the medium , resuspended in RPMI to a final concentration of 1–2 × 106 cells/ml and mixed 1:1 with P815 targets ( pre-washed and resuspended to 1–2 × 106 cells/ml as above ) ±10 µM 3-MB-PP1 . Cells were left in suspension at room temperature for 5 min , after which they were mixed gently ( by pipetting up and down ) , plated in 4-well tissue culture dishes ( Nunc ) at 0 . 5 ml/well and incubated at 37°C for a further 20 , 35 , or 55 min before fixation with 2% paraformaldehyde and either 1 . 5% or 3% gluteraldehyde ( Stinchcombe et al . , 2001b ) . Samples were further processed for DAB cytochemistry , osmium fixation and urynal acetate staining and EPON embedding as previously described ( Stinchcombe et al . , 2001a , 2001b , 2011; Jenkins et al . , 2009 ) . Thin ( 50–70 nm ) and semi-thin ( 100–150 nm ) sections were stained with lead citrate and viewed using a Phillips C100 TEM ( FEI ) . Images were captured using Kodak photographic negative film ( Kodak , United Kingdom ) and digital electron micrographs produced using a Flextight X5 scanner ( Hasselblad , United Kingdom ) . Thin sections prepared from samples of Zap70 ( AS ) CTL conjugated to targets for 25 , 40 or 60 min ±10 µM 3-MB-PP1 , were stained with lead citrate and imaged using a Phillips C100 TEM microscope . Images of CTL-target conjugates in which at least one centriole was present were analysed , with 22–60 conjugates for each time point ±10 µM 3-MB-PP1 . The secretory cleft was defined as a clear gap within the centre of the contact site between the CTL and target , bounded by an area of tight , flat , membrane–membrane contact on each side . Centrosome distance was measured as the shortest distance from the point of the centriole closest to the plasma membrane and the synapse membrane itself . Measurements were classified as < 500 nm ( i . e . , one centriole-barrel length , tightly associated ) , 500–1000 nm ( polarised ) , or >1000 nm ( unpolarised ) from the plasma membrane . The contact site was defined as the distance between the furthest points of contact between CTL and target at the synapse . The number of membrane extensions projecting from the surface of each CTL with a visible centriole within the contact site was recorded for each CTL condition . Only extensions >1000 nm were included in this analysis . Cytotoxicity was examined using the CytoTox 96 Non-Radioactive Cytotoxicity Assay ( Promega , United Kingdom ) . P815 target cells were resuspended in phenol-free RPMI , 2% FCS at 105 cells/ml in a round bottom 96-well plate . CTL ±10 µM 3-MB-PP1 , were added at effector:target ( E:T ) ratios shown , and plates were incubated at 37°C for 4 hr . The absorbance of the supernatants at 490 nm determined the release of lactate dehydrogenase and % target cell lysis . Zap70 ( AS ) CTL were cultured with or without hamster anti-CD3ε ( 145-2C11 , Becton Dickinson Bioscience , United Kingdom ) for 5 hr at 37°C in 96-well round bottom plates at approx 0 . 5–2 × 106 cells/well in c-RPMI medium containing 5 μg/ml GolgiPlug ( Becton Dickinson ) ±10 μM 3-MB-PP1 . The cells were then washed with PBS ( containing 0 . 1% BSA and 0 . 02% sodium azide ) , stained with anti-mouse CD8α-PerCPCy5 . 5 ( Pharmingen ) for 30 min on ice , permeabilised by paraformaldehyde fixation using the BD Cytofix/Cytoperm Kit ( Becton Dickinson ) and stained for intracellular cytokine production using anti-mouse IFNγ-FITC ( clone XMG1 . 2 ) , anti-mouse TNFα ( clone MP6-XT22 ) , and anti-mouse IL-2 ( clone JES6-5H4 ) ( Pharmingen , United Kingdom ) . Lymphocytes were washed and analysed on a FACScalibur and analysed using CellQuestPro software ( Becton Dickinson ) . In each assay , any cytokine positive cells isolated from wells with no were subtracted from the % cytokine positive cells incubated with peptide to yield the final value . CTL ( 5 × 105 per well ) were stimulated in triplicate wells for 10 min with 5 μg/ml soluble anti-CD3e ( clone 145-2C11 ) and 50 μg /ml crosslinking goat anti-Armenian hamster ( Jackson Immunoresearch ) Abs in 96 well plates coated with 10 μg/ml recombinant ICAM-1-Fc ( R&D Systems ) . Cells were washed three times with pre-warmed media , removed with cell dissociation buffer ( Gibco ) and plate-bound cells were counted by FACS . The percentage of cells bound was calculated as ( [number of live CD4 cells bound to the plate ) ÷ ( input number of cells per well] ) × 100 . Zap70 ( AS ) OT-I and Zap70+/− OT-I CTL were nucleofected with LifeAct-eGFP and imaged as described in ‘Live cell imaging’ above using the 1 . 2x camera adapter x20 objective lens . Serial confocal 1 . 6 µm 10 planes of Z-stacks were taken at 20 s intervals for 15 min ( = 45 time points ) with excitation of 488 nm . Videos were analysed using the ‘Spots module’ of Imaris software ( 7 . 6 . 0 Bitplane ) to detect each cell as a spot . Only cells with track durations >140 s and track displacement lengths >13 μm were analysed in order to exclude immobile cells . The mean speed for each set of conditions was calculated from 40–50 cells per condition . | White blood cells are responsible for defending the body against infection and disease . Cytotoxic T-lymphocytes , or cytotoxic T cells , are white blood cells that recognise and kill cells that are infected , cancerous or otherwise damaged . Receptors on the surface of these T cells recognise ‘foreign’ molecules on the surface of diseased or damaged cells: this activates the T cells , which then release cytotoxic proteins that destroy the target cells . During this process the T cell and the target cell are brought into close contact with each other , and their membranes undergo a dramatic rearrangement to form an ‘immunological synapse’ . Although the structure of the immunological synapse is understood in detail , the mechanisms controlling the membrane reorganisation are not well understood . Previous studies have shown that an enzyme called Zap70 needs to be present to activate receptors involved in cell adhesion , termed integrins . Now , Jenkins , Stinchcombe et al . have shown a dual role for Zap70 in the formation of the immunological synapse . Jenkins , Stinchcombe et al . used mice that had been engineered to produce a modified version of Zap70 that worked as normal until its activity was ‘switched off’ by the addition of a specific drug . When Zap70 was switched off , integrins were still activated but the formation of the immunological synapse was halted with only finger-tip-like contacts between the T cell and the target cell . These contacts were formed by projections from the T cell made of a protein called actin , which forms a kind of scaffolding within cells . Without active Zap70 , the T cell receptors could not trigger the dynamic rearrangement of the actin proteins and the membrane remodelling required to form a tight contact between the two cells . This disrupted the delivery of the cytotoxic proteins to their target . These results clearly show that Zap70 has at least two distinct roles that it must carry out for an immunological synapse to form . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"cell",
"biology",
"immunology",
"and",
"inflammation"
] | 2014 | Distinct structural and catalytic roles for Zap70 in formation of the immunological synapse in CTL |
The spindle assembly checkpoint ( SAC ) monitors and promotes kinetochore–microtubule attachment during mitosis . Bub1 and BubR1 , SAC components , originated from duplication of an ancestor gene . Subsequent sub-functionalization established subordination: Bub1 , recruited first to kinetochores , promotes successive BubR1 recruitment . Because both Bub1 and BubR1 hetero-dimerize with Bub3 , a targeting adaptor for phosphorylated kinetochores , the molecular basis for such sub-functionalization is unclear . We demonstrate that Bub1 , but not BubR1 , enhances binding of Bub3 to phosphorylated kinetochores . Grafting a short motif of Bub1 onto BubR1 promotes Bub1-independent kinetochore recruitment of BubR1 . This gain-of-function BubR1 mutant cannot sustain a functional checkpoint . We demonstrate that kinetochore localization of BubR1 relies on direct hetero-dimerization with Bub1 at a pseudo-symmetric interface . This pseudo-symmetric interaction underpins a template–copy relationship crucial for kinetochore–microtubule attachment and SAC signaling . Our results illustrate how gene duplication and sub-functionalization shape the workings of an essential molecular network .
Bub1 and BubR1 are paralogous proteins involved in the spindle assembly checkpoint ( SAC ) , a safety device that monitors the attachment of kinetochores to spindle microtubules and halts mitotic progression until completion of chromosome bi-orientation on the mitotic spindle ( Lara-Gonzalez et al . , 2012; Foley and Kapoor , 2013 ) . Bub1 and BubR1 originated from a gene that was already present in the hypothetical last eukaryotic common ancestor ( LECA ) ( Suijkerbuijk et al . , 2012a ) . After speciation , up to nine distinct duplication events might have occurred , which invariably led to sub-functionalization of the resulting gene products ( Suijkerbuijk et al . , 2012a ) . Human Bub1 and BubR1 are strongly conserved at the sequence and domain level ( Figure 1A ) but play complementary roles in the SAC . Bub1 becomes recruited to kinetochores in prometaphase to provide a platform for additional SAC proteins , including Mad1 , Mad2 , the BubR1/Bub3 complex , and Cdc20 . Bub1 promotes the assembly of a subset of these proteins , Mad2 , BubR1/Bub3 , and Cdc20 , into the SAC effector , the mitotic checkpoint complex ( MCC ) , which targets the anaphase promoting complex/cyclosome ( APC/C ) to inhibit its ability to promote mitotic exit ( Lara-Gonzalez et al . , 2012; Foley and Kapoor , 2013 ) . Bub1 , not in itself a MCC subunit , likely catalyzes MCC assembly by aligning MCC subunits for a profitable interaction ( discussed in Overlack et al . , 2014 ) . Bub1 kinase activity is not required for the SAC ( Sharp-Baker and Chen , 2001; Fernius and Hardwick , 2007; Perera et al . , 2007; Klebig et al . , 2009 ) but contributes to stable kinetochore–microtubule attachments through phosphorylation of histone H2A on Thr120 and subsequent localization of Sgo1 to centromeres and kinetochores ( Kawashima et al . , 2010; Wang et al . , 2011; Caldas et al . , 2013 ) . BubR1 , on the other hand , evolved into an inactive pseudo-kinase ( Suijkerbuijk et al . , 2012a ) and is a crucial subunit of the MCC ( Hardwick et al . , 2000; Fraschini et al . , 2001; Sudakin et al . , 2001 ) . BubR1 contributes to the formation of stable kinetochore–microtubule attachments and checkpoint silencing through kinetochore co-recruitment of protein phosphatase 2A ( PP2A ) ( Suijkerbuijk et al . , 2012b; Kruse et al . , 2013; Xu et al . , 2013; Espert et al . , 2014; Nijenhuis et al . , 2014 ) . 10 . 7554/eLife . 05269 . 003Figure 1 . Mps1 and Bub1 are required for kinetochore localization of BubR1 . ( A ) Similar domain organization of the homologous proteins Bub1 and BubR1 . TPR—tetratrico peptide repeat , B3BD—Bub3 binding domain , B1—Bub1 , BR1—BubR1 . ( B ) Schematic depiction of the outer KT ( KMN network ) . MELT repeats of Knl1 are phosphorylated by the checkpoint kinase Mps1 and recruit Bub1/Bub3 to KTs . It is not clear how BubR1 is recruited to KTs . ( C–D ) Quantitative IP-mass spectrometry analyses showing that the interaction of Bub1 , BubR1 , and Bub3 with KTs is significantly reduced upon inhibition of Mps1 with Reversine . Green- and red-labeled hits indicate respectively proteins whose levels were not strongly affected or were strongly affected in the presence of Reversine . ( E and G ) Representative images of Flp-In T-REx cell lines in BubR1 ( E ) and Bub1 ( G ) RNAi , respectively , after treatment with nocodazole , showing that Bub1 is required for BubR1 KT localization . Scale bar: 10 µm . ( F and H ) Quantification of Bub1 and BubR1 KT levels , respectively , in cells treated as in panel E and G . The graph shows mean intensity , error bars indicate SD . The mean value for non-depleted cells is set to 1 . ( I–J ) Representative images of stable Flp-In T-REx cell lines expressing the indicated GFP-Bub1 constructs ( panel I ) or HeLa cells transfected with the indicated GFP-BubR1 constructs ( panel J ) after treatment with nocodazole . The same images are also shown in Figure 4A and Figure 5C , and quantified in Figures 4B and 5B . Scale bar: 10 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 05269 . 00310 . 7554/eLife . 05269 . 004Figure 1—figure supplement 1 . RNAi quantification and schematic depiction of Bub1 and BubR1 constructs . ( A–B ) Quantification of RNAi-based depletion of Bub1 and BubR1 . Shown are mean and standard deviation from two independent experiments . ( C–D ) Schematic depiction of the main Bub1 and BubR1 deletion constructs used in this study . TPR , tetratrico peptide repeats; B3BD , Bub3 binding domain; BR1 , BubR1; FL , full-length; N , N-terminus . DOI: http://dx . doi . org/10 . 7554/eLife . 05269 . 004 Bub1 and BubR1 have different kinetochore dynamics , likely reflecting their distinct functions in the SAC . Bub1 interacts stably with unattached kinetochores , in agreement with its function as a SAC recruitment platform , while BubR1 turns over rapidly ( t1/2 = 3–20 s ) , likely reflecting its cycle of incorporation into MCC and its release into the cytosol as a soluble APC/C inhibitor ( Howell et al . , 2004; Shah et al . , 2004 ) . Besides different dynamics , another important difference is that kinetochore recruitment of Bub1 is independent of BubR1 , while recruitment of BubR1 is strictly subordinate to Bub1 ( Millband and Hardwick , 2002; Gillett et al . , 2004; Johnson et al . , 2004; Perera et al . , 2007; Logarinho et al . , 2008; Klebig et al . , 2009 ) . The molecular basis for these differences is unclear , because both Bub1 and BubR1 bind a kinetochore-targeting adaptor named Bub3 . Bub3 is a 7-WD40 β-propeller that targets kinetochores by binding to phosphorylated Met-Glu-Leu-ThrP ( MELTP , where TP indicates phosphothreonine ) repeats of the outer kinetochore subunit Knl1 ( a . k . a . Casc5 , Spc105 , Spc7 , AF15q14 , and Blinkin ) ( Kiyomitsu et al . , 2007; Krenn et al . , 2012 , 2014; London et al . , 2012; Shepperd et al . , 2012; Yamagishi et al . , 2012; Primorac et al . , 2013 ) ( Figure 1B ) . Bub3 binds tightly to Bub1 and BubR1 via conserved segments known as Bub3-binding domain ( B3BD ) or GLEBS ( Taylor et al . , 1998; Larsen et al . , 2007 ) . By recognizing MELTP , Bub3 co-recruits Bub1 to kinetochores in Saccharomyces cerevisiae ( Primorac et al . , 2013 ) . In human cells , Bub3 is required for kinetochore recruitment of Bub1 and BubR1 , and consistently the B3BDs of Bub1 and BubR1 are necessary , and in the case of Bub1 also sufficient , for kinetochore targeting of Bub1 and BubR1 ( Taylor et al . , 1998; Logarinho et al . , 2008; Malureanu et al . , 2009; Elowe et al . , 2010; Lara-Gonzalez et al . , 2011; Krenn et al . , 2012 ) . The subordination of BubR1 kinetochore recruitment to the presence of Bub1 suggests that Bub3 may operate differently when bound to Bub1 or BubR1 . In this study , we set out to investigate the molecular basis of this phenomenon and its implications for spindle checkpoint signaling and kinetochore–microtubule attachment .
The SAC kinase Mps1 has been shown to phosphorylate MELT repeats of Knl1 to promote kinetochore recruitment of Bub1 and BubR1 ( Heinrich et al . , 2012; London et al . , 2012; Shepperd et al . , 2012; Yamagishi et al . , 2012; Primorac et al . , 2013; Vleugel et al . , 2013; Krenn et al . , 2014 ) . We precipitated Bub1 or Knl1 ( Vleugel et al . , 2013 ) from mitotic lysates of HeLa cells treated with or without the Mps1 inhibitor Reversine ( Santaguida et al . , 2010 ) . Quantitative mass spectrometry ( see ‘Materials and methods’ ) of proteins associated with Bub1 or Knl1 confirmed the crucial role of Mps1 , as we observed a strong suppression of the interaction of Bub1 , BubR1 , and Bub3 with kinetochores in the presence of Reversine ( Figure 1C–D . Large deviations from a value of 1 for the Reversine/DMSO ratio indicate suppression of binding ) . In HeLa cells treated with nocodazole , which depolymerizes microtubules and activates the SAC , Bub1 decorated kinetochores at essentially normal levels after the depletion of BubR1 ( Figure 1E , quantified in Figure 1F . Quantifications of RNAi-based depletions are shown in Figure 1—figure supplement 1A–B ) . Conversely , BubR1 did not decorate kinetochores after Bub1 depletion ( Figure 1G–H ) . These results confirm that BubR1 requires Bub1 for kinetochore recruitment , in line with previous studies ( Millband and Hardwick , 2002; Gillett et al . , 2004; Johnson et al . , 2004; Perera et al . , 2007; Logarinho et al . , 2008; Klebig et al . , 2009 ) . By monitoring the localization of a GFP-Bub1 reporter construct , we had previously demonstrated that Bub1209-270 , encompassing the B3BD , is the minimal Bub1 localization domain ( Taylor et al . , 1998; Krenn et al . , 2012 ) . Bub1209–270 targeted kinetochores very efficiently even after the depletion of endogenous Bub1 ( Figure 1I ) . We asked if an equivalent GFP reporter construct encompassing the B3BD of BubR1 , BubR1362–431 , was also recruited to kinetochores . BubR1362–431 was not recruited to kinetochores even in the presence of Bub1 ( Figure 1J . Diagrams of Bub1 and BubR1 deletions used in this study are in Figure 1—figure supplement 1C–D ) . Thus , even if Bub1 and BubR1 share a related B3BD to interact with the same kinetochore-targeting subunit ( Bub3 ) and interact in a phosphorylation-dependent manner with Knl1 , the mechanisms of their kinetochore recruitment are different . This raises two crucial questions: ( 1 ) why is the B3BD region of Bub1 sufficient for kinetochore recruitment , while the equivalent region of BubR1 is not ? And ( 2 ) if binding to Bub3 is not sufficient for robust kinetochore recruitment of BubR1 , how is BubR1 recruited to kinetochores ? We will focus sequentially on these questions . To investigate if and how Bub1209–270 and BubR1362–431 modulate the binding affinity of Bub3 for the MELTP repeats of Knl1 , we immobilized on amylose beads a fusion of maltose-binding protein ( MBP ) with residues 138–168 of Knl1 , a region containing a single and functional MELT repeat ( the most N-terminal , and therefore called MELT1; Krenn et al . , 2014 ) . We treated MBP-Knl1MELT1 with or without Mps1 kinase . Next , we incubated MBP-Knl1MELT1 with Bub3 , Bub1209–270/Bub3 , or BubR1362-–431/Bub3 and visualized bound proteins by Western blotting . Bub3 in isolation did not bind MBP-Knl1MELT1 , in agreement with our previous data ( Krenn et al . , 2014 ) . The B3BD of Bub1 strongly enhanced binding of Bub3 to phosphorylated MBP-Knl1MELT1 but not to unphosphorylated MBP-Knl1MELT1 , while the B3BD of BubR1 had a negligible effect ( Figure 2A ) . These results in vitro correlate with the ability of the equivalent B3BD to support ( or not ) kinetochore recruitment in cells ( Figure 1I–J ) . 10 . 7554/eLife . 05269 . 005Figure 2 . The loop regions of Bub1 and BubR1 modulate the interaction of Bub3 with phosphorylated MELT motifs . ( A ) Recombinant Bub3 , Bub1209–270/Bub3 and BubR1362–431/Bub3 were incubated with immobilized MBP-Knl1MELT1 ( residues 138–168 of human Knl1 ) prephosphorylated with Mps1 ( + ) or unphosphorylated ( − ) . Empty lanes ( − ) demonstrate lack of background binding to empty beads . wt , wild type; B3 , Bub3; B1 , Bub1; BR1 , BubR1 . ( B ) Multiple sequence alignments of the Bub3 binding domains ( B3BD ) of human ( Homo sapiens , Hs ) , chicken ( Gallus gallus , gg ) , frog ( Xenopus laevis , Xl ) , and budding yeast ( Saccharomyces cerevisiae , Sc ) Bub1 and BubR1s . Mad3 is the budding yeast BubR1 homolog . ScBub1R314 ( red asterisk ) directly contributes to the interaction with the MELTP peptide . The different Bub1 and BubR1 sequences were aligned manually on the basis of the crystal structures of the B3BDs of Mad3 and Bub1 in complex with Bub3 ( Larsen et al . , 2007; Primorac et al . , 2013 ) . ( C ) Crystal structure of the ScBub1289–359-Bub3-MELTP ternary complex ( Primorac et al . , 2013 ) . N and C indicate the N- and C-terminus , respectively . ( D ) Close-up of the MELTP binding site indicating the role of ScBub1R314 in MELTP binding . ( E ) Schematic depiction of short Bub1 and BubR1 ‘loop swap’ constructs , containing the loop ( different shades of red ) followed by the Bub3-binding domain ( different shades of yellow ) . ( F ) Recombinant BubR1362–431/Bub3 with its own loop ( wt ) or with the Bub1 loop ( B1 ) and recombinant Bub1209–270/Bub3 with its own loop ( wt ) or with the BubR1 loop ( BR1 ) were incubated with immobilized MBP-Knl1MELT1 prephosphorylated with Mps1 ( + ) or unphosphorylated ( − ) as in panel A . Empty lanes ( − ) demonstrate lack of background binding to empty beads . ( G ) Western Blot of immunoprecipitates ( IP ) from mitotic Flp-In T-REx cell lines expressing the indicated GFP-Bub1 and GFP-BubR1 constructs showing the influence of the loop on the ability to pull down the KT-components Knl1 and Mis12 . Tubulin was used as loading control . ( H ) Quantification of the Western blot in ( G ) . In the upper graph , the amounts of co-precipitating BubR1 , Bub3 , Knl1 , and Mis12 were normalized to the amount of GFP-Bub1 bait present in the IP . In the lower graph , the amounts of co-precipitating Bub1 , Bub3 , Knl1 , and Mis12 were normalized to the amount of GFP-BubR1 bait . Values for GFP-Bub1 wt and GFP-BubR1 wt , respectively are set to 1 . The graphs show mean intensity of two independent experiments ( for Mis12 only one ) . Error bars represent SD . DOI: http://dx . doi . org/10 . 7554/eLife . 05269 . 00510 . 7554/eLife . 05269 . 006Figure 2—figure supplement 1 . Validation of recombinant ‘loop swap’ mutants . The indicated constructs were purified and separated by SDS-PAGE after purification . Both wild-type and chimeric constructs interact normally and with similar affinity with Bub3 . DOI: http://dx . doi . org/10 . 7554/eLife . 05269 . 006 Our previous structural and biochemical characterization of the Bub1B3BD /Bub3/MELTP ternary complex of S . cerevisiae demonstrated that while Bub3 carries most of the crucial ( and evolutionarily conserved ) residues involved in high-affinity binding to MELTP , a short segment of Bub1 , the ‘loop’ , contributes to the binding affinity ( Primorac et al . , 2013 ) . The ‘loop’ region of Bub1 or BubR1 is between strands β1 and β2 ( Figure 2B–D ) and precedes the highly conserved core of the B3BD . The loop abuts the binding site for the MELTP peptide and is therefore ideally positioned to modulate the binding affinity of Bub3 for MELTP ( Figure 2C–D ) . Because the loops of Bub1 and BubR1 have quite divergent sequences ( Figure 2B ) , we tested their role in modulating the binding affinity of Bub1209–270/Bub3 or BubR1362–431/Bub3 for immobilized MBP-Knl1MELT1 ( see Figure 2A ) . We swapped the loop regions of Bub1 and BubR1 as schematized in Figure 2E . Recombinant versions of the chimeric mutants Bub1209–270/BR1-loop and BubR1362–431/B1–loop were co-expressed with Bub3 . Both the wild-type and chimeric constructs interacted with apparently similar affinity with Bub3 , excluding gross structural perturbations ( Figure 2—figure supplement 1 ) . We then tested the ability of the recombinant constructs to interact with immobilized MBP-Knl1MELT1 . Bub1209–270 bound tightly and in an Mps1-phosphorylation-dependent manner to MBP-Knl1MELT1 ( Figure 2F ) , while Bub1209–270/BR1–loop bound weakly . Conversely , BubR1362–431/B1–loop bound more strongly to phosphorylated MBP-Knl1MELT1 than did BubR1362–431 ( Figure 2F ) . These results demonstrate a crucial role of the loop region of Bub1 in the recognition of a phosphorylated MELT repeat . IP experiments from stable cell lines expressing GFP fusions of the wild-type or loop-swap mutants in the context of full-length Bub1 or BubR1 recapitulated the results in vitro ( Figure 2G–H ) . Loop swap mutants interacted with Bub3 as efficiently as wild-type counterparts , but the interactions that Bub3 mediates , most notably those with the Knl1 receptor and its associated partner Mis12 , were strongly enhanced when the Bub1 loop region was grafted onto BubR1 . We then tested the localization behavior of these mutants . Kinetochore localization of GFP-Bub1BR1–loop was weaker than that of wild-type Bub1 ( Figure 3A–B ) , whereas kinetochore localization of GFP-BubR1B1–loop was stronger than that of unmodified BubR1 ( Figure 3C–D ) . Indeed , GFP-BubR1B1–loop decorated kinetochores even after the depletion of Bub1 . Thus , when grafted onto BubR1 , the Bub1 loop region is sufficient to confer upon BubR1 the ability to target kinetochores in a Bub1-independent fashion ( Figure 3E–F ) . Although the experiments in vitro were carried out with a single MELT repeat , MELT1 , they are likely to reflect the binding behavior of Bub1 and BubR1 to endogenous Knl1 , which may contain up to 19 MELT repeats . Kinetochore localization of GFP-BubR1B1–loop in cells depleted of Bub1 was inhibited by addition of Reversine , indicating a dependence on Mps1 and MELTP sequences ( Figure 3G–H and Figure 3—figure supplement 1 , panel A ) . Expression of BubR1B1–loop did not overtly affect the kinetochore levels of endogenous Bub1 ( Figure 3—figure supplement 1 , panel B ) , suggesting that the multiple MELTP sequences of Knl1 are not saturated with endogenous Bub1 even in the strong checkpoint-activating conditions of our assay . 10 . 7554/eLife . 05269 . 007Figure 3 . Behavior of the ‘loop swap’ mutants in HeLa cells . ( A ) Representative images of stable Flp-In T-REx cells expressing either GFP-Bub1 wild type ( wt ) or the loop mutant showing that the BR1-loop impairs KT localization . Scale bar: 10 µm . ( B ) Quantification of Bub1 KT levels in cells treated as in panel A . The graph shows mean intensity from three independent experiments . Error bar represents SEM . Values for Bub1 wt are set to 1 . ( C ) Representative images of stable Flp-In T-REx cells expressing either GFP-BubR1 wt or the loop mutant showing that the B1-loop enhances KT localization . Scale bar: 10 µm . ( D ) Quantification of BubR1 KT levels in cells treated as in panel C . The graph shows mean intensity from three independent experiments . Error bar represents SEM . Values for BubR1 wt are set to 1 . ( E ) Representative images of HeLa cells transfected with the indicated GFP-BubR1 constructs , showing that BubR1 B1-loop is independent of Bub1 for its KT localization . In brief , after transfection , cells were depleted of endogenous Bub1 by RNAi , synchronized with a double thymidine block and arrested in mitosis with nocodazole . Scale bar: 10 µm . ( F ) Quantification of BubR1 KT levels in cells treated as in panel E . The graph shows mean intensity from three independent experiments . Error bars represent SEM . Values for BubR1wt in non-depleted cells are set to 1 . ( G ) Representative images of HeLa cells transfected with GFP-BubR1 B1-loop treated as in panel E in the presence ( + ) or absence ( − ) of the Mps1 inhibitor Reversine , showing that BubR1 B1-loop KT localization is dependent on Mps1 . Scale bar: 10 µm . ( H ) Quantification of BubR1 KT levels in cells treated as in panel G . The graph shows mean intensity from two independent experiments . Error bars represent SEM . Values for BubR1wt in non-depleted cells without Reversine ( images are shown in Figure 3—figure supplement 1 , panel A ) are set to 1 . ( I ) Mean duration of mitosis of Flp-In T-REx stable cell lines expressing GFP-BubR1 wt or the loop mutant in the absence of endogenous BubR1 and in the presence of 50 nM nocodazole . Cell morphology was used to measure entry into and exit from mitosis by time-lapse-microscopy ( n > 58 per cell line ) from three independent experiments . Error bars depict SEM . ( J ) Western Blot of immunoprecipitates ( IP ) from mitotic Flp-In T-REx cell lines expressing the indicated GFP-BubR1 constructs showing the influence of the loop on the ability to pull down MCC and APC/C components . Tubulin was used as loading control . ( K ) Western Blot of immunoprecipitates ( IP ) of the APC/C subunit Cdc27 from mitotic Flp-In T-REx cell lines expressing the indicated GFP-BubR1 constructs showing the influence of the loop on the incorporation into APC/C-bound MCC . Tubulin was used as loading control . DOI: http://dx . doi . org/10 . 7554/eLife . 05269 . 00710 . 7554/eLife . 05269 . 008Figure 3—figure supplement 1 . Additional characterization of ‘loop swap’ mutants . ( A ) Representative images of HeLa cells transfected with GFP-BubR1wt treated as in Figure 3E in the presence ( + ) or absence ( − ) of the Mps1 inhibitor Reversine as control for the experiment shown in Figure 3G . The corresponding quantification is shown in Figure 3H . Scale bar: 10 µm . ( B ) Quantification of Bub1 kinetochore levels in cells treated as in Figure 3E . The graph shows mean intensity from three independent experiments . Error bars represent SEM . Values for Bub1 in BubR1wt expressing cells are set to 1 . ( C ) Quantification of the Western Blot in Figure 3J . The amounts of co-precipitating MCC and APC/C components were normalized to the amount of GFP-BubR1 bait present in the IP . Values for GFP-BubR1 wt are set to 1 . The graphs show mean intensity of two independent experiments . Error bars represent SEM . ( D ) Quantification of the Western Blot in Figure 3K . The amounts of co-precipitating proteins were normalized to the amounts present in GFP-BubR1 wt expressing cells . The graphs show mean intensity of two independent experiments . Error bars represent SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 05269 . 008 Next , we asked if GFP-BubR1B1–loop was able to replace the SAC function of BubR1 . In cells depleted of endogenous BubR1 , wild-type BubR1 restored SAC function to a high degree but GFP-BubR1B1–loop failed to do so ( Figure 3I ) . Comparison of immune-precipitates of GFP-BubR1 and GFP-BubR1B1–loop from nocodazole-treated mitotic cells showed much less of the latter associated with MCC and APC/C subunits ( Figure 3J , quantified in Figure 3—figure supplement 1 , panel C ) . Similarly , immune-precipitation ( IP ) of the APC/C demonstrated association of GFP-BubR1 , but not GFP-BubR1B1–loop , with the APC/C ( Figure 3K , quantified in Figure 3—figure supplement 1 , panel D ) . These results demonstrate that sequence divergence in the short loop region of Bub1 and BubR1 has strong functional consequences . In Bub1 , the loop enhances the ability of Bub3 to recognize MELTP sequences of Knl1 . In BubR1 , the precise role of the loop is unknown , but the results in Figure 3 suggest to us a specific role in MCC assembly or in the interaction with the APC/C , as discussed more thoroughly below . The second question , how BubR1 is recruited to kinetochores , has three distinct facets . First , the segment of Bub1 required for kinetochore recruitment of BubR1 should be identified . Second , the segment of BubR1 required for its own kinetochore recruitment should be identified . Third , it should be established if Bub1 and BubR1 interact directly and if kinetochore proteins other than Bub1 play additional roles in BubR1 recruitment . To investigate the role of Bub1 in BubR1 recruitment , we created GFP fusions of several deletion mutants of Bub1 ( Figure 1—figure supplement 1C–D ) and tested concomitantly their kinetochore localization and BubR1 recruitment ( Figure 4A ) . Bub1209–270 , encompassing the B3BD and localizing robustly to kinetochores , did not recruit BubR1 in cells depleted of endogenous Bub1 ( Figure 4A–B ) . A construct containing the tetratrico peptide repeats ( TPRs ) and the B3BD of Bub1 , Bub11–284 , also targeted kinetochores but did not recruit BubR1 ( Figure 4A–B ) . On the contrary , a segment ( Bub1209–409 ) consisting of the B3BD and a ∼140-residue C-terminal extension ( CTE ) , targeted kinetochores and promoted robust kinetochore localization of BubR1 ( Figure 4A–B ) . Thus , Bub1 does not require the CTE ( residues 271–409 ) for its own kinetochore recruitment ( the B3BD , residues 209–270 , is sufficient ) but requires it to recruit BubR1 . A deletion mutant lacking the CTE ( Bub1Δ271–409 ) targeted kinetochores efficiently but failed to recruit BubR1 ( Figure 4C–D ) . 10 . 7554/eLife . 05269 . 009Figure 4 . A minimal BubR1-binding region of Bub1 . ( A and C ) Representative images of stable Flp-In T-REx cell lines expressing the indicated GFP-Bub1 constructs after treatment with nocodazole , showing that Bub1209–409 is sufficient to recruit BubR1 ( panel A ) and that residues 271–409 are essential for this function ( panel C ) . Scale bar: 10 µm . ( B and D ) Quantification of BubR1 KT levels in cells treated as in panels B and D , respectively . The graphs show mean intensity of two independent experiments , the error bars indicate SEM . The mean value for non-depleted cells expressing GFP ( panel B ) or GFP-Bub1209–409 ( panel D ) is set to 1 . ( E ) Western blot of immunoprecipitates ( IP ) from mitotic Flp-In T-REx cell lysates expressing the indicated GFP-Bub1 constructs in the presence or absence of endogenous Bub1 , showing that Bub1209–409 is sufficient to pull down BubR1 . Tubulin was used as loading control . ( F ) Quantification of the Western blot in panel E . The amounts of co-precipitating BubR1 , Bub3 , and Knl1 were normalized to the amount of GFP-Bub1 bait present in the IP . Values for GFP-Bub1 FL in non-depleted cells are set to 1 . The graph shows mean intensity of two independent experiments . Error bars represent SD . ( G ) BubR11–571/Bub3 and Bub11–409/Bub3 interact in size exclusion chromatography , which separates proteins based on size and shape . H6 and TRX are tags used for protein purification and expression . ( H ) BubR11–571 and Bub11–409/Bub3 interact in size exclusion chromatography . ( I ) BubR11–571/Bub3 and Bub1209–409/Bub3 interact in size exclusion chromatography . ( J ) BubR11–571/Bub3 and Bub1271–409 do not interact in size exclusion chromatography . MBP—maltose binding protein , mAu—milliabsorbance unit . DOI: http://dx . doi . org/10 . 7554/eLife . 05269 . 00910 . 7554/eLife . 05269 . 010Figure 4—figure supplement 1 . Additional chromatographic experiments . ( A ) BubR11–571 and Bub1209–409/Bub3 interact in size exclusion chromatography . ( B ) BubR11–571/Bub3 and Bub1209–270/Bub3 do not interact in size exclusion chromatography . The control run of Bub1209–270/Bub3 is missing due to limiting amounts of the protein . ( C ) BubR1222–571/Bub3 and Bub11–280/Bub3 do not interact in size exclusion chromatography . ( D ) BubR1222–571/Bub3 and Bub11–409/Bub3 interact in size exclusion chromatography . mAU—milliabsorbance unit . DOI: http://dx . doi . org/10 . 7554/eLife . 05269 . 01010 . 7554/eLife . 05269 . 011Figure 4—figure supplement 2 . The TPR domain of Bub1 influences KT binding affinity in addition to the loop region . ( A ) Representative images of HeLa cells transfected with the indicated GFP-Bub1FL constructs , showing that the BR1-loop impairs KT localization . However in the absence of endogenous Bub1 considerable binding affinity is regained . Cells were treated as described in Figure 3E . Scale bar: 10 µm . ( B ) Representative images of HeLa cells transfected with the indicated GFP-Bub1209–409 constructs , showing that this construct , which lacks the N-terminal TPR domain , does not regain considerable binding affinity in the absence of Bub1 . Cells were treated as described in Figure 3E . Scale bar: 10 µm . ( C ) Representative images of HeLa cells transfected with the indicated GFP-Bub11–409 constructs , showing that the TPR domain is contributing to regaining KT binding affinity in the absence of endogenous Bub1 . Cells were treated as described in Figure 3E . Scale bar: 10 µm . ( D ) Quantification of Bub1 KT levels in cells treated as in panels A–C . The graph shows mean intensity from two independent experiments . Error bars represent SEM . Values for Bub1FL in non-depleted cells were set to 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 05269 . 01110 . 7554/eLife . 05269 . 012Figure 4—figure supplement 3 . The TPR domain of BubR1 influences kinetochore binding affinity in addition to the loop region . ( A ) Representative images of HeLa cells transfected with the indicated GFP-BubR1362–571 constructs , showing that the B1-loop enhances kinetochore localization . However in the absence of endogenous Bub1 the short loop mutant , which lacks the N-terminal TPR domain , is less efficient than the full-length BubR1 in its localization to kinetochores ( Figure 3E ) . Cells were treated as described in Figure 3E . Scale bar: 10 µm . ( B ) Representative images of HeLa cells transfected with the indicated GFP-BubR11–571 constructs , showing that the TPR domain is contributing together with the loop region to the ability to stay at KTs in the absence of endogenous Bub1 . Cells were treated as described in Figure 3E . Scale bar: 10 µm . ( C ) Quantification of BubR1 KT levels in cells treated as in ( A–B ) . The graph shows mean intensity from at least two independent experiments . Error bars represent SEM . Values for BubR1FL in non-depleted cells are set to 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 05269 . 012 In IP experiments both in the presence and absence of endogenous Bub1 , GFP-Bub1209–409 interacted with BubR1 at levels that were only modestly lower than those of full-length GFP-Bub1 ( Figure 4E–F ) . Conversely , GFP-Bub1209–270 did not interact with BubR1 , in agreement with the inability of this construct to promote BubR1 localization . In this context , it should be noted that the tetratrico peptide repeats ( TPRs ) near the N-terminus of Bub1 and BubR1 ( Figure 1A ) had been initially identified as primary determinants of kinetochore recruitment ( Kiyomitsu et al . , 2007; Bolanos-Garcia et al . , 2011 ) , but later shown to be dispensable ( Krenn et al . , 2012 , 2014 ) , a result confirmed here . In our IP experiments , however , we observe that Bub1 interacts with the outer kinetochore more strongly when the TPRs are present , in agreement with our previous studies ( Krenn et al . , 2012 , 2014 ) . The Bub1 TPR region interacts with a short sequence motif of Knl1 named KI1 motif ( Kiyomitsu et al . , 2007; Krenn et al . , 2012 , 2014 ) . This interaction , whose precise significance is unclear , enhances the SAC response ( Krenn et al . , 2014 ) . Additional evidence of a modest additional role of the TPR in the interaction of Bub1 and BubR1 with kinetochores is presented in Figure 4—figure supplement 2 and Figure 4—figure supplement 3 . Next , we asked if Bub1 binds directly to BubR1 . Bub11–409 and BubR11–571 , both of which target kinetochores ( Taylor et al . , 1998; Vanoosthuyse et al . , 2004; Klebig et al . , 2009; Malureanu et al . , 2009; Elowe et al . , 2010 ) were individually co-expressed with Bub3 in insect cells and purified to homogeneity . In size-exclusion chromatography ( SEC ) experiments , in which the elution volume reflects macromolecular mass and shape , BubR11–571/Bub3 bound stoichiometrically to Bub11–409/Bub3 ( Figure 4G ) , thus demonstrating a direct interaction in the absence of other proteins . In these experiments in vitro , BubR11–571 did not require Bub3 to bind Bub11–409/Bub3 ( Figure 4H ) , and therefore in the following SEC experiments , we used BubR1 and BubR1/Bub3 interchangeably ( whereas Bub1 was poorly expressed and largely insoluble in insect cells without Bub3 ) . Although this result may suggest that Bub3 is not required for the interaction of BubR1 with Bub1/Bub3 , we show and discuss in the context of Figure 6 that a functional B3BD is required for the interaction of BubR1 with Bub1 in living cells . In agreement with the ability of Bub1209–409 to recruit BubR1 to kinetochores , Bub1209–409/Bub3 formed a stoichiometric complex with BubR11–571/Bub3 ( Figure 4I ) or BubR11–571 ( Figure 4—figure supplement 1 , panel A ) , further showing that the Bub1 TPR region is dispensable . Neither the isolated B3BD of Bub1 ( Bub1209–270/Bub3 , Figure 4—figure supplement 1B ) nor the isolated CTE ( Bub1271–409 , Figure 4J ) bound BubR11–571/Bub3 , indicating that both regions contribute to BubR1 binding in vitro . Additional SEC experiments supporting this conclusion are shown in Figure 4—figure supplement 1 , panels C–D . To identify a minimal kinetochore-targeting region of BubR1 , we created GFP fusions of several deletion mutants of BubR1 ( Figure 1—figure supplement 1C–D ) and tested their localization to kinetochores . BubR1362–571 , which contains the B3BD and a ∼140-residue CTE , localized to kinetochores in a Bub1-dependent manner ( Figure 5A–B ) . Shorter fragments of BubR1 , including the isolated B3BD ( BubR1362–431 , see also Figure 1J ) and the isolated CTE ( BubR1432–571 ) , did not localize to kinetochores ( Figure 5B–C ) . In agreement with the kinetochore recruitment assay , SEC experiments showed binding of BubR1362–571 but not of the isolated B3BD ( BubR1362–431/Bub3 ) or the isolated CTE ( BubR1432–571 ) to Bub11–409/Bub3 ( Figure 5D–F ) . 10 . 7554/eLife . 05269 . 013Figure 5 . A minimal Bub1-binding region of BubR1 . ( A and C ) Representative images of HeLa cells transfected with the indicated GFP-BubR1 constructs . Cells were treated as described in Figure 3E . BubR1362–571 is the minimal construct that is able to localize to KTs in the presence of Bub1 . Scale bar: 10 µm . ( B ) Quantification of BubR1 KT levels in cells treated as in panels A and C . The graph shows mean intensity of at least two independent experiments , error bars depict SEM . Values for GFP-BubR1 FL in non-depleted cells are set to 1 . ( D ) BubR1362–571 and Bub11–409/Bub3 interact in size exclusion chromatography . ( E ) BubR1362–431/Bub3 and Bub11–409/Bub3 do not interact in size exclusion chromatography . ( F ) BubR1432–571 and Bub11–409/Bub3 do not interact in size exclusion chromatography . mAU—milliabsorbance unit . DOI: http://dx . doi . org/10 . 7554/eLife . 05269 . 013 Our results predict that BubR1362–571 and Bub1209–409 ought to be sufficient for the Bub1/BubR1 interaction in vitro . Indeed , BubR1362–571 and Bub1209–409/Bub3 interacted stoichiometrically in SEC runs ( Figure 6A ) . A summary of the properties of the crucial Bub1 and BubR1 segments discussed in the last two sections is presented in Figure 6B . An alignment of the interacting domains of Bub1 and BubR1 ( residues 209–409 and 362–571 , respectively; the alignment was obtained with programs Muscle [Edgar , 2004] and JPRED [Cole et al . , 2008] ) shows that their sequences are structurally equivalent ( Figure 6—figure supplement 1A ) . Both start with the B3BD , continue with a segment predicted to adopt a helical conformation and end with a region predicted to lack defined secondary structure . We surmise that the ability of modern-day Bub1 and BubR1 to form heterodimers using these structurally equivalent ( ‘pseudo-symmetric’ ) segments may reflect the ability of their ancestor to form homodimers , similarly to what is observed in cohesins . 10 . 7554/eLife . 05269 . 014Figure 6 . A pseudo-symmetric Bub1–BubR1 interaction . ( A ) The identified minimal constructs BubR1362–571 and Bub1209–409/Bub3 interact in size exclusion chromatography . ( B ) Summary of the behavior of the indicated Bub1 and BubR1 constructs . ( C and E ) Representative images of HeLa cells transfected with the indicated GFP-BubR1 constructs showing that neither BubR1Δ432–484 ( panel C ) , which lacks the predicted helical segment of the C-terminal extension , nor BubR1E409K+E413K ( panel E ) , which is not able to bind Bub3 , are able to localize to KTs . Cells were treated as in Figure 3E . For BubR1Δ432–484 two different expression levels are depicted in the non-depleted condition . Scale bar: 10 µm . ( D and F ) Quantification of BubR1 KT levels in cells treated as in panels C and E , respectively . The graph shows mean intensity from at least two independent experiments . Error bars represent SEM . Values for BubR1FL in non-depleted cells are set to 1 . ( G ) Domain organization of LacI-GFP-Bub1 constructs . ( H ) LacI-Bub1wt recruits BubR1 to the Lac-Operator , whereas Bub1E252K , which cannot bind Bub3 , does not . DOI: http://dx . doi . org/10 . 7554/eLife . 05269 . 01410 . 7554/eLife . 05269 . 015Figure 6—figure supplement 1 . Alignment of the Bub1 and BubR1 interaction domains . ( A ) Multiple sequence alignments of the interacting domains of Bub1 and BubR1 from four different species ( Homo sapiens , Hs ) , chicken ( Gallus gallus , Gg ) , frog ( Xenopus laevis , Xl ) , and budding yeast ( Saccharomyces cerevisiae , Sc ) . Mad3 is the budding yeast BubR1 homolog . ( B ) Representative images of HeLa cells transfected with the indicated GFP-BubR1 constructs , showing that also the short Bub3 binding domain mutant is not able to localize to KTs . Cells were treated as described in Figure 3E . Scale bar: 10 µm . ( C ) Quantification of BubR1 KT levels in cells treated as in panel B . The graph shows mean intensity from two independent experiments . Error bars represent SEM . Values for BubR1362–571 in non-depleted cells are set to 1 . ( D ) Quantification of co-localization . Data represent normalized mean intensities and standard deviations from 12 cells . BubR1 was identified at the ectopic locus on 12/12 cells for the wild type and on 0/12 cells for the E252K mutant . DOI: http://dx . doi . org/10 . 7554/eLife . 05269 . 015 Deletion of residues 432–484 of BubR1 ( GFP-BubR1Δ432–484 ) in the predicted helical region impaired kinetochore localization of BubR1 ( Figure 6C–D ) . Additionally , mutations in the B3BD of BubR1 ( E409K + E413K ) known to impair Bub3 binding ( Taylor et al . , 1998; Larsen et al . , 2007 ) prevented kinetochore localization of full-length BubR1 ( Figure 6E–F ) and of BubR1362–571 ( Figure 6—figure supplement 1 , panels B–C ) . Thus , Bub3 binding is necessary for efficient kinetochore localization of BubR1 even if it may appear to be dispensable for the interaction in vitro ( Figure 4H ) . We note that in those SEC binding experiments in vitro in which we used isolated BubR1 rather than BubR1/Bub3 , Bub3 might have exchanged from the Bub1/Bub3 complex to reconstitute BubR1/Bub3 . Alternatively , the relatively high protein concentrations in the SEC experiments ( 5–15 µM ) may effectively compensate for reduced binding affinity when BubR1 is devoid of Bub3 . Regardless of the precise explanation , BubR1 clearly requires Bub3 for efficient kinetochore localization . This requirement may reflect a direct contribution to the interaction with Bub1 or alternatively a residual ability of BubR1/Bub3 to bind phosphorylated motifs on Knl1 or other kinetochore proteins . To try distinguishing between these possibilities , we targeted a Lac repressor ( LacI ) fusion of Bub1 to an ectopic Lac operator ( LacO ) site on the chromosome distinct from centromeres . LacI-Bub1 recruited BubR1 to the ectopic site ( Figure 6G–H , quantified in Figure 6—figure supplement 1D ) , thus suggesting that kinetochores are not required for the Bub1/BubR1 interaction , in agreement with a previous study in Schizosaccharomyces pombe ( Rischitor et al . , 2006 ) . This result suggests that Bub1/Bub3 is sufficient for recruitment of BubR1/Bub3 , and that the role of Bub3 in recruitment of BubR1 might be direct and not through Knl1 . Mutation of the Bub3-binding site of the LacI-Bub1 fusion ( Bub1E252K ) prevented BubR1 recruitment , suggesting that Bub3 is required for a robust interaction between Bub1 and BubR1 on both sides of the complex ( Figure 6G–H ) . BubR1 is important for the SAC and for establishing bi-orientation . We therefore asked if perturbing its kinetochore localization impaired these processes . HeLa cells depleted of endogenous BubR1 failed to arrest in mitosis in the presence of nocodazole ( Figure 7A ) . SAC proficiency was re-established upon expression of GFP-BubR1 but not GFP-BubR1E409K+E413K . GFP-BubR1Δ432–484 , on the other hand , restored a robust SAC response . IP experiments indicated that both GFP-BubR1E409K+E413K and GFP-BubR1Δ432–484 were unable to interact with Bub1 or Knl1 , as expected based on their localization behavior . However , GFP-BubR1E409K+E413K was also significantly impaired in its ability to interact with two additional MCC subunits ( Mad2 and Cdc20 ) and was even more dramatically impaired in its ability to interact with the APC/C . Conversely , GFP-BubR1Δ432–484 interacted with the MCC subunits ( including Bub3 , as expected ) and with the APC/C at levels that were comparable to those of the wild-type GFP-BubR1 control ( Figure 7B ) . Thus , two distinct mutants of BubR1 both impaired in kinetochore localization have uncorrelated behaviors with regard to SAC proficiency . On the other hand , because GFP-BubR1E409K+E413K does not bind Bub3 , while GFP-BubR1Δ432–484 does ( Figure 7B , quantified in Figure 7C ) , it appears that at least in human cells BubR1 needs to bind Bub3 to become incorporated into the MCC . 10 . 7554/eLife . 05269 . 016Figure 7 . Functional characterization of BubR1 mutants . ( A ) Mean duration of mitosis of Flp-In T-REx stable cell lines expressing GFP-BubR1 wt or the indicated mutants in the absence of endogenous BubR1 and in the presence of 50 nM nocodazole . Cell morphology was used to measure entry into and exit from mitosis by time-lapse-microscopy ( n > 44 per cell line per experiment ) from at least three independent experiments . Error bars depict SEM . ( B ) Western blot of immunoprecipitates ( IP ) from mitotic Flp-In T-REx cell lines expressing the indicated GFP-BubR1 constructs . Tubulin was used as loading control . ( C ) Quantification of the Western Blot in Figure 7B . The amounts of co-precipitating proteins were normalized to the amount of GFP-BubR1 bait present in the IP . Values for GFP-BubR1 wt are set to 1 . The graphs show mean intensity of two independent experiments . Error bars represent SEM . ( D ) Analysis of cold-stable microtubules in cells expressing the indicated GFP-BubR1 constructs . ( E ) Western Blot of immunoprecipitates ( IP ) from mitotic Flp-In T-REx cell lines expressing the indicated GFP-BubR1 constructs . The asterisk represents an unspecific band recognized by the PP2A antibody . DOI: http://dx . doi . org/10 . 7554/eLife . 05269 . 016 In addition to playing a role in the SAC , BubR1 is important for establishing bi-orientation in HeLa cells , and this role requires its kinetochore recruitment ( Johnson et al . , 2004; Lampson and Kapoor , 2005; Meraldi and Sorger , 2005; Suijkerbuijk et al . , 2012b; Kruse et al . , 2013; Xu et al . , 2013 ) . Both GFP-BubR1E409K+E413K and GFP-BubR1Δ432–484 , neither of which localizes to kinetochores , failed to complement the deficits in promoting the formation of stable kinetochore–microtubule attachments observed in cells depleted of endogenous BubR1 ( Figure 7D ) . The role of BubR1 in kinetochore–microtubule attachment has been attributed to its interaction with the B56 regulatory subunit of a complex of protein phosphatase 2A ( PP2AB56 ) ( Suijkerbuijk et al . , 2012b; Kruse et al . , 2013; Xu et al . , 2013; Espert et al . , 2014; Nijenhuis et al . , 2014 ) . Recently , this function of BubR1 has been further implicated in SAC silencing ( Espert et al . , 2014; Nijenhuis , et al . , 2014 ) . We asked if the defect of the BubR1Δ432–484 mutant in supporting kinetochore–microtubule attachment correlated with a defective interaction with PP2AB56 . Indeed , the levels of the B56 regulatory subunit in IPs of the GFP-BubR1Δ432–484 mutant were much lower than those of the wild-type protein ( Figure 7E ) .
New genes are frequently created by duplication ( Conant and Wolfe , 2008 ) . After duplication , the paralogs ( i . e . , the genes generated by the duplication event ) may diverge and sub-functionalize , thus allowing specialization in a subset of the functions originally provided by the singleton ( the single ancestor gene before duplication ) ( Conant and Wolfe , 2008 ) . The resolution of adaptive conflicts between different functions of the singleton has been identified as a beneficial consequence of gene duplication ( Hittinger and Carroll , 2007 ) . Detailed molecular illustrations of this process , however , are rare . Our analysis of the mechanism of recruitment of Bub1 and BubR1 illustrates how the duplication of an ancestor gene and subsequent divergence was exploited during evolution to create two gene products with highly diversified and efficient functions , including monitoring checkpoint status through stable recruitment to unattached kinetochores ( Bub1 ) , effecting APC/C inhibition through incorporation into the MCC ( BubR1 ) , stabilization of microtubule attachment ( Bub1 and BubR1 , possibly through different mechanisms ) , and SAC silencing ( BubR1 ) . The occurrence of nine distinct duplications of the Bub1 and BubR1 ancestor ( Suijkerbuijk et al . , 2012a ) likely reflects an extreme evolutionary pressure to separate functions that were originally condensed in a single gene . Investigating to what extent sub-functionalization after each of the nine distinct duplication events of the Bub1 and BubR1 ancestor followed similar or divergent paths will be an interesting direction for future studies . Previously , the loss of kinase activity specifically in BubR1 but not Bub1 was identified as a manifestation of the divergence of these proteins ( Vleugel et al . , 2012; Suijkerbuijk et al . , 2012b ) . Here , we have considerably extended our understanding of this divergence by showing that the loop motifs in the B3BDs of Bub1 and BubR1 modulate the interaction of Bub3 with MELTP . A suboptimal loop region for MELTP binding in BubR1 makes it depend on an alternative mechanism for kinetochore localization . This alternative is the direct interaction with Bub1 , extensively characterized in Figures 4–6 . By swapping loop motifs , we created a gain-of-function mutant of BubR1 that can bind kinetochores in the absence of Bub1 , and a loss-of-function mutant of Bub1 severely impaired in autonomous kinetochore binding . The failure of the BubR1 gain-of-function mutant to complement the depletion of wild-type BubR1 strongly suggests that the evolutionary divergence of Bub1 and BubR1 , with its specific effects on kinetochore recruitment , is functionally relevant . The mechanism of BubR1 localization depends on a direct , pseudo-symmetric hetero-dimerization interaction with Bub1 ( Figure 8A ) . It involves equivalent segments of Bub1 and BubR1 comprising their B3BD/GLEBS motif and a CTE whose first part is predicted to be helical . The interaction of Bub1 and BubR1 requires that each has a bound Bub3 . Thus , Bub3 has at least three distinct functions: ( 1 ) it recruits Bub1 to MELTP motifs of Knl1; ( 2 ) it contributes to the dimerization of Bub1 with BubR1 required for kinetochore recruitment of the latter; and ( 3 ) in complex with BubR1 , it has an additional hitherto unknown function in the SAC . This function was exposed by the behavior of the BubR1E409K+E413K mutant . The SAC defect observed with this mutant is unlikely to be a consequence of impaired kinetochore localization , because another kinetochore-localization impaired mutant , BubR1Δ432–484 , was checkpoint proficient . We surmise that BubR1-bound Bub3 is involved in an unknown aspect of the SAC mechanism downstream of kinetochores , possibly having to do with MCC formation or APC/C binding , as recently proposed ( Han et al . , 2014 ) . In analogy with the role of the Bub1 loop motif in modulating the function of Bub3 as a MELTP receptor , we speculate that the BubR1 loop motif influences the specificity of BubR1-bound Bub3 for additional SAC-relevant targets . The fact that the ‘loop swap’ BubR1 mutant is unable to sustain the SAC provides evidence for this hypothesis . The properties of the loop region in the recently discovered protein BuGZ , which contains a B3BD/GLEBS motif that interacts with Bub3 , are also of interest ( Jiang et al . , 2014; Toledo et al . , 2014 ) . 10 . 7554/eLife . 05269 . 017Figure 8 . Extension of the template model . ( A ) Model of the Bub1–BubR1 interaction . The upper part shows the described KT recruitment mechanism of BubR1/Bub3 , which in turn recruits the phosphatase PP2A , to a Bub1/Bub3 complex on Knl1 . The lower part depicts a close-up of the identified pseudo-symmetric Bub1–BubR1 interaction , which involves equivalent segments of Bub1 and BubR1 comprising the Bub3 binding domain and a C-terminal extension whose first part is predicted to have a helical fold in both proteins . The presence of Bub3 on both proteins seems to be essential for this interaction , although due to different reasons ( for more explanations see text ) . The TPR regions of human Bub1 and BubR1 bind to non-conserved short motifs of Knl1 named KI1 and KI2 , respectively ( Kiyomitsu et al . , 2007; Krenn et al . , 2012 , 2014 ) . ( B ) Extension of the template model . Mad1/C-Mad2 at KTs is known to act as a template for the establishment of the Cdc20/C–Mad2 interaction . This seems similar to the BubR1 recruitment mechanism , wherein Bub1/Bub3 recruits BubR1/Bub3 through a pseudo-symmetric interaction . Ultimately , the entire MCC ( BubR1/Bub3 and Cdc20/C-Mad2 ) may represent the copy of a KT template consisting of Bub1/Bub3 and Mad1/C-Mad2 . DOI: http://dx . doi . org/10 . 7554/eLife . 05269 . 017 Recruitment of PP2AB56 by BubR1 has been recently implicated in SAC silencing through a mechanism ultimately impinging on dephosphorylation of the MELTP motifs ( Espert et al . , 2014; Nijenhuis et al . , 2014 ) . The apparent checkpoint proficiency of BubR1Δ432–484 may indicate that BubR1 localization to kinetochores is not essential for checkpoint function , as it implies that substantial amounts of MCC can be generated even when BubR1/Bub3 cannot be recruited to kinetochores . However , we note that this mutant interacts only weakly with PP2AB56 and might therefore carry an additional checkpoint-silencing defect obscuring an underlying SAC defect . Furthermore , we cannot exclude that residues in the deleted segments ( 432–484 ) of BubR1 are normally involved in an intra-molecular control mechanism that couples the activation of wild-type BubR1 to kinetochore recruitment . Whether or not BubR1 kinetochore recruitment is important for SAC function , it is clearly essential for kinetochore–microtubule attachment and bi-orientation . Previously , it has been shown that BubR1 promotes bi-orientation through recruitment of PP2AB56 ( Suijkerbuijk et al . , 2012b; Kruse et al . , 2013; Xu et al . , 2013 ) , which counteracts Aurora B activity and thus stabilizes kinetochore–microtubule attachments ( Lampson and Kapoor , 2005; Foley et al . , 2011 ) . Our results demonstrate that BubR1Δ432–484 , which cannot interact with Bub1 and localize to kinetochores , has a strong defect in kinetochore–microtubule attachment that correlates with a defective interaction with PP2AB56 . Dissecting the requirement for kinetochore recruitment of the SAC subunits is instrumental for distinguishing their roles in the SAC from their roles in chromosome bi-orientation ( Brady and Hardwick , 2000; De Antoni et al . , 2005; London et al . , 2012; Shepperd et al . , 2012; Yamagishi et al . , 2012; Nijenhuis et al . , 2013; London and Biggins , 2014; Moyle et al . , 2014 ) . By describing the mechanism of BubR1 recruitment and the role of the loop motif of Bub1 and BubR1 in modulating the affinity for kinetochores , this study fills an important gap . We and others have previously shown that two conformers of Mad2 , O-Mad2 and C-Mad2 , form an asymmetric conformational dimer ( Luo et al . , 2004; De Antoni et al . , 2005 ) . In this reaction , Mad1 acts as a stable placeholder for C-Mad2 ( De Antoni et al . , 2005 ) . Once at kinetochores , the Mad1/C-Mad2 complex recruits a high-turnover cytosolic form of O-Mad2 and converts it into the active C-Mad2 form , which targets Cdc20 , thus overcoming a rate-limiting step towards the formation of MCC ( De Antoni et al . , 2005; Mapelli et al . , 2007; Simonetta et al . , 2009 ) . This scheme , summarized in Figure 8B , identifies Mad1/Mad2 as a ‘template’ for the establishment of the Cdc20/C-Mad2 complex , a structural ‘copy’ of the Mad1/Mad2 complex ( De Antoni et al . , 2005; Musacchio and Salmon , 2007 ) . Even if directed primarily towards a different function ( SAC signaling rather than chromosome bi-orientation ) , we note that this pattern is remarkably similar to that emerging from the mechanism of BubR1 recruitment by Bub1 , in which a stable Bub1/Bub3 complex at kinetochores recruits a rapidly cycling BubR1/Bub3 complex . The MCC , made of Cdc20/C-Mad2 and BubR1/Bub3 , can be interpreted as a ‘copy’ of kinetochore-bound ‘templates’ made of Mad1/C-Mad2 and Bub1/Bub3 complexes . Whether such templates engage in a complex at kinetochores is unclear but plausible . Even if we did not identify Mad1 in our mass spectrometry analysis of Bub1 , we recently identified both Mad1 and Bub1 in precipitates of the N-terminal segment of Knl1 ( Krenn et al . , 2014 ) . Furthermore , Mad1 and Bub1 have been shown to interact directly in S . cerevisiae and Caenorhabditis elegans ( Brady and Hardwick , 2000; London et al . , 2012; Moyle et al . , 2014 ) . Our future studies will aim to investigate the significance of the copy–template molecular relationship for SAC signaling and chromosome bi-orientation .
Plasmids were derived from the pCDNA5/FRT/TO-EGFP-IRES , a previously modified version ( Krenn et al . , 2012 ) of the pCDNA5/FRT/TO vector ( Invitrogen , Carlsbad , CA ) . To create N-terminally-tagged EGFP Bub1 and BubR1 truncation constructs , Bub1 and BubR1 sequences were obtained by PCR amplification from the previously generated pCDNA5/FRT/TO-EGFP-Bub1-IRES and pCDNA5/FRT/TO-EGFP-BubR1-IRES vector , respectively ( Krenn et al . , 2012 ) and subcloned in frame with the GFP-tag . Mutations and deletions within the Bub1 and BubR1 constructs were generated by standard site-directed mutagenesis or by a mutagenesis protocol ( Liu and Naismith , 2008 ) . All Bub1 constructs were RNAi resistant ( Kiyomitsu et al . , 2007 ) . BubR1-expressing constructs were made siRNA-resistant by changing the sequence targeted by the RNAi oligos to ‘AACGTGCCTTCGAGTACGAGA’ . pCDNA5/FRT/TO-based plasmids were used for generation of stable cell lines , as well as for transient transfection . All plasmids were verified by sequencing . HeLa cells were grown in DMEM ( PAN Biotech , Aidenbach , Germany ) supplemented with 10% FBS ( Clontech , part of Takara Bio group , Shiga , Japan ) , penicillin and streptomycin ( GIBCO , Carlsbad , CA ) , and 2 mM L-glutamine ( PAN Biotech ) . For all plasmid transfections of HeLa cells , X-tremeGENE transfection reagent ( Roche , Basel , Switzerland ) was used at a 3:1 ratio with plasmid DNA . Flp-In T-REx HeLa cells used to generate stable doxycycline-inducible cell lines were a gift from SS Taylor ( University of Manchester , Manchester , England , UK ) . Flp-In T-REx host cell lines were maintained in DMEM with 10% tetracycline-free FBS ( Clontech ) supplemented with 50 μg/ml Zeocin ( Invitrogen ) . Flp-In T-REx HeLa expression cell lines were generated as previously described ( Krenn et al . , 2012 ) . Gene expression was induced by addition of 0 . 2–0 . 5 µg/ml doxycycline ( Sigma , St . Louis , MO ) for 24 hr siBUB1 ( Dharmacon , part of GE Healthcare , Piscataway , NJ; 5′-GGUUGCCAACACAAGUUCU-3′ ) or siBUBR1 ( Dharmacon; 5′-CGGGCAUUUGAAUAUGAAA-3′ ) duplexes were transfected with Lipofectamine 2000 ( Invitrogen ) at 50 nM for 24 hr . For experiments in HeLa cells , cells were synchronized with a double thymidine arrest 5 hr after transfection with siRNA duplexes . In brief , after washing the cells with PBS they were treated with thymidine for 16 hr and then released into fresh medium . 3 hr after the release , 50 nM siRNA duplexes were transfected again . 5 hr after transfection , cells were treated with thymidine for 16 hr and afterwards released in fresh medium . Unless differently specified , nocodazole ( Sigma–Aldrich ) was used at 3 . 3 µM . MG132 ( used at 5–10 µM ) was obtained from Calbiochem , thymidine ( 2 mM ) was purchased from Sigma–Aldrich . Reversine ( Calbiochem , part of EMD Biosciences , Darmstadt , Germany ) was used at 0 . 5 µM . To generate mitotic populations for immunoprecipitation experiments , cells were treated with 330 nM nocodazole for 16 hr . Mitotic cells were then harvested by shake off and lysed in lysis buffer ( 150 mM KCl , 75 mM Hepes , pH 7 . 5 , 1 . 5 mM EGTA , 1 . 5 mM MgCl2 , 10% glycerol , and 0 . 075% NP-40 supplemented with protease inhibitor cocktail [Serva , Heidelberg , Germany] and PhosSTOP phosphatase inhibitors [Roche] ) . Extracts were precleared using a mixture of protein A–Sepharose ( CL-4B; GE Healthcare ) and protein G-Sepharose ( rec-Protein G-Sepharose 4B; Invitrogen ) for 1 hr at 4°C . Subsequently , extracts were incubated with GFP-Traps ( ChromoTek , Martinsried , Germany; 3 µl/mg of extract ) for 3 hr at 4°C . Immunoprecipitates were washed with lysis buffer and resuspended in sample buffer , boiled and analyzed by SDS-PAGE and Western blotting using 4–12% gradient gels ( NuPAGE Bis-Tris Gels , Life technologies , Carlsbad , CA ) . For Cdc27 IPs cells were synchronized by addition of the CDK1-inhibitor RO3306 ( Calbiochem ) for 15 hr and subsequently released into 330 nM nocodazole for 2–3 hr before harvesting by shake off . Cells were lysed in lysis buffer ( described above ) , and extracts were precleared with protein G-Sepharose for 1 hr at 4°C . Afterwards , extracts were incubated with 1 . 5 μg/mg of the Cdc27 primary antibody ( mouse monoclonal , BD ) for 2 hr at 4°C . Subsequently , protein G-Sepharose was added for 4 hr at 4°C . Immunoprecipitates were washed and analyzed as described above . The following antibodies were used: anti-GFP ( in house made rabbit polyclonal antibody; 1:1000–3000 ) , anti-Mis12 ( in house made mouse monoclonal antibody; clone QA21-74-4-3; 1:1000 ) , anti-Knl1-N ( in house made rabbit polyclonal SI0787 antibody; 1:1000 ) , anti-Bub1 ( rabbit polyclonal; Abcam , Cambridge , UK; 1:5000 ) , anti-BubR1 ( mouse monoclonal; BD; 1:1000 ) , anti-Bub3 ( mouse monoclonal; BD; 1:1000 ) , anti-Tubulin ( mouse monoclonal; Sigma; 1:8000 ) , anti-Apc7 ( in house made rabbit polyclonal antibody SI0651 , 1:500 ) , anti-Cdc20 ( mouse monoclonal , Santa Cruz , Dallas , TX , 1:500 ) , anti-Mad2 ( in house made mouse monoclonal antibody , clone AS55-A12 , 1:500 ) , anti-Cdc27 ( mouse monoclonal , BD; 1:1000-3000 ) , anti-PP2AB56α ( rabbit polyclonal; Bethyl , Montgomery , TX; 1:1000 ) . Secondary antibodies were anti-mouse ( Amersham , part of GE Healthcare ) and anti-rabbit ( Amersham ) affinity purified with horseradish peroxidase conjugate ( working dilution 1:10000 ) or Protein G with horseradish peroxidase conjugate ( Life technologies ) ( working dilution 1:6000 ) . After incubation with ECL Western blotting system ( GE Healthcare ) , images were acquired with ChemiBIS 3 . 2 ( DNR Bio-Imaging Systems , Jerusalem , Israel ) in 16-bit TIFF format . Levels of images were adjusted using ImageJ software and then cropped and converted to 8-bit . Unmodified 16-bit TIFF images were used for quantification with ImageJ software . Measurements were graphed with Excel ( Microsoft , Seattle , WA ) and GraphPad Prism version 6 . 0 for Mac OS X ( GraphPad Software , San Diego California USA ) . Cells were plated on a 24-well µ-Plate ( Ibidi , Martinsried , Germany ) . Drugs were diluted in CO2 Independent Medium ( Gibco ) and added to the cells 1 hr before filming . Cells were imaged every 20 to 30 min in a heated chamber ( 37°C ) on a 3i Marianas system ( Intelligent Imaging Innovations Inc . , Göttingen , Germany ) equipped with Axio Observer Z1 microscope ( Zeiss ) , Plan-Apochromat 40×/1 . 4NA oil objective , M27 with DIC III Prism ( Zeiss , Oberkochen , Germany ) , Orca Flash 4 . 0 sCMOS Camera ( Hamamatsu , Hamamatsu City , Japan ) and controlled by Slidebook Software 5 . 5 ( Intelligent Imaging Innovations Inc ) . For cells expressing the GFP-BubR1 proteins , only cells in which kinetochores were visible were considered for the analysis . HeLa and Flp-In T-REx HeLa cells were grown on coverslips precoated with poly-D-Lysine ( Millipore , 15 µg/ml ) and poly-L-Lysine ( Sigma ) , respectively . For the experiments with HeLa cells , cells were synchronized with a double thymidine block and after release from that arrested in prometaphase by the addition of 330 nM nocodazole for 3 hr . For all other experiments , asynchronously growing cells were arrested in prometaphase by the addition of nocodazole for 3–4 hr and fixed using 4% paraformaldehyde . Cells were stained for Bub1 ( mouse , ab54893 , 1:400 ) , BubR1 ( rabbit , Bethyl A300-386A , 1:1000 ) , CREST/anti-centromere antibodies ( Antibodies , Inc . , Davis , CA , 1:100 ) , diluted in 2% BSA-PBS for 1 . 5 hr . Goat anti-human and chicken anti-rabbit Alexa Fluor 647 ( Invitrogen ) , goat anti-rabbit and anti-mouse RRX , and donkey anti-human Alexa Fluor 405 ( Jackson ImmunoResearch Laboratories , Inc . , West Grove , PA ) were used as secondary antibodies . DNA was stained with 0 . 5 µg/ml DAPI ( Serva ) , and coverslips were mounted with Mowiol mounting media ( Calbiochem ) . Cells were imaged at room temperature using a spinning disk confocal device on the 3i Marianas system equipped with an Axio Observer Z1 microscope ( Zeiss ) , a CSU-X1 confocal scanner unit ( Yokogawa Electric Corporation , Tokyo , Japan ) , Plan-Apochromat 63× or 100×/1 . 4NA Oil Objectives ( Zeiss ) , and Orca Flash 4 . 0 sCMOS Camera ( Hamamatsu ) . Images were acquired as z-sections at 0 . 27 µm . Images were converted into maximal intensity projections , exported , and converted into 8-bit . Quantification of kinetochore signals was performed on unmodified 16-bit z-series images using Imaris 7 . 3 . 4 32-bit software ( Bitplane , Zurich , Switzerland ) . After background subtraction , all signals were normalized to CREST . At least 138 kinetochores were analyzed per condition . Measurements were exported in Excel ( Microsoft ) and graphed with GraphPad Prism 6 . 0 ( GraphPad Software , San Diego California USA ) . For analysis of cold-stable microtubules , cells that were synchronized with a single Thymidine arrest , released for 6 . 5 hr and kept for 4 hr in 5 µM MG132 , were incubated for 5 min on ice in medium with 10 mM HEPES pH 7 . 5 and then directly fixed in 4% PFA . Cells were stained for Tubulin ( mouse , Sigma T9026 , 1:5000 ) and CREST . DNA was labeled with DAPI . CREST staining was used to identify kinetochores in image z-stacks to count kinetochores attached to cold-stable microtubules . Each kinetochore was classified as ‘attached’ or ‘not attached’ depending on whether a microtubule fiber ended at the kinetochore . An average of 120 kinetochores was counted per cell , and seven cells were analyzed for each condition . MBP-Knl1138–168-H6 and MBP-Knl1138–168-H6 constructs ( MELT1 ) were obtained by sub-cloning into a pGEX vector backbone in which the coding sequence for GST was replaced with that for MBP . Expression was carried out in BL21 RIL strain at 25°C and by using 1 mM IPTG for 2 . 5 hr to induce expression . Cell pellets were re-suspended in three pellet volumes of 50 mM HEPES-NaOH pH 7 . 5 , 250 mM KCl , 2 mM DTE , 10% glycerol , protease inhibitor mix ( Serva ) . Cells were lysed by sonication , and the lysates were centrifuged at 100000×g for 1 hr at 4°C . Recombinant products were isolated from the lysate by using the HisTrap ( GE Healthcare ) column , followed by buffer exchange using a desalting column ( GE Healthcare ) . Purified proteins were concentrated to about 3 mg/ml and frozen in liquid nitrogen . GST-BubR1362–431/Bub3 and GST-Bub1209–270/Bub3 constructs were obtained by sub-cloning the coding sequences for Bub3 and for the indicated segments of Bub1 into pFLMultiBac vector ( Trowitzsch et al . , 2010 ) . Expression was carried out by infection of Tna38 insect cells ( Hashimoto et al . , 2012 ) for 72 hr at 27°C . Viruses of the constructs were generated as described ( Trowitzsch et al . , 2010 ) . Insect cells were harvested by centrifugation at 1500 rpm for 30 min in a Sorvall RC 3BP+ ( Thermo Scientific , Carlsbad , CA ) centrifuge with Rotor H6000A , and the pellets were frozen in liquid nitrogen and stored at −80°C . 1 g of cell pellet was re-suspended in 10 ml lysis buffer ( 50 mM Tris–HCl pH 8 . 0 or 8 . 5 , 150 mM KCl , 2 mM DTE , DNAse , PMSF , protease inhibitors [Serva] ) . Cells were lysed by sonication and the lysate centrifuged at 100000×g for 1 hr at 4°C . The supernatant was filtered through Nalgene bottle-top filter . 1 . 5 ml of GSH bead slurry ( GE Healthcare ) was added to 50 ml of cleared lysate . After 1 hr at 4°C on a rotating wheel , the beads were recovered by centrifugation and washed with lysis buffer . Bead-bound complexes in 50 ml of lysis buffer were retrieved from the GSH beads by addition of GSH-Prescission protease ( produced in house ) for 14 hr at 4°C . Eluates were concentrated using Amicon concentrators ( 3 kDa cutoff ) , diluted with 20 mM Tris–HCl , 2 mM DTE to a final KCl concentration of 50 mM , and further purified using 1 ml HiTrap QFF column ( GE Healthcare ) . Peak fractions were collected and concentrated down to a volume of 2 ml and further purified by size exclusion chromatography using the S75 16/60 column ( GE Healthcare ) . Peak fractions were collected , concentrated to about 3 mg/ml , frozen in small aliquots in liquid nitrogen , and stored at −80°C . Sequences of loop swap constructs were as follows:Bub1209–270RRVITISKSEYSVHSSLASKVDVEQVVMYCKEKLIRGESEFSFEELRAQKYNQRRKHEQWVNBub1209–270-BubR1 loopRRVITTRKPGKEEGDPLSKVDVEQVVMYCKEKLIRGESEFSFEELRAQKYNQRRKHEQWVNBubR1362–431INHILSTRKPGKEEGDPLQRVQSHQQASEEKKEKMMYCKEKIYAGVGEFSFEEIRAEVFRKKLKEQREAEBubR1362–431-Bub1 loopINHILSISKSEYSVHSSLAQRVQSHQQASEEKKEKMMYCKEKIYAGVGEFSFEEIRAEVFRKKLKEQREAE Sequences coding for H6-BubR1 , H6-Bub1 , and untagged Bub3 constructs were sub-cloned into pFLMultiBac vectors and baculoviruses were generated . Baculovirus expressing Bub3-TRX was generated by the Dortmund Protein Facility ( DPF ) using the pOPIN vector system {Berrow:2007cy} . Bub1/Bub3 and BubR1/Bub3 complexes were generated by co-infection and co-expression at 27°C for 72 hr . Insect cells were harvested by centrifugation at 1500 rpm for 30 min in a Sorvall RC 3BP+ ( Thermo Scientific ) centrifuge with Rotor H6000A , the pellets were frozen in liquid nitrogen and stored at −80°C . 1 g of cell pellet was re-suspended in 10 ml Lysis buffer ( 50 mM HEPES-KOH pH 7 . 5 , 150 mM KCl , 15 mM imidazole , 2 mM DTE , 0 . 05% Tween20 , PMSF , protease inhibitors [Serva] ) . Cells were lysed by sonication and centrifuged at 100000×g for 1 hr at 4°C . The supernatant was filtered through Nalgene bottle-top filter . The complexes were isolated from the cleared lysate on a 5-ml HisTrap ( GE Healthcare ) column . Peak fractions were pooled concentrated using Amicon concentrators and further purified in GF buffer ( 50 mM Hepes-KOH pH 7 . 5 , 150 mM KCl , 2 mM DTE , 0 . 05% Tween20 ) by size exclusion chromatography using S200 16/60 column ( GE Healthcare ) . Peak fractions were pooled , concentrated to typically 3 to 5 mg/ml , and frozen in liquid nitrogen . Sequences encoding H6-TRX-BubR1 , H6-Bub1-MBP , and untagged Bub3 constructs were sub-cloned into pFLMultiBac vectors and baculoviruses were generated . All constructs , apart from H6-Bub1209–409-MBP , which was co-expressed with untagged Bub3 , were expressed individually in insect cells at 27°C for 72 hr . Insect cells were harvested by centrifugation at 1500 rpm for 30 min in a Sorvall RC 3BP+ ( Thermo Scientific ) centrifuge with Rotor H6000A , the pellets were frozen in liquid nitrogen and stored at −80°C . 1 g of cell pellets were re-suspended in 10 ml Lysis buffer ( 50 mM Tris–HCl pH 7 . 5 , 150 mM KCl , 0 . 5 mM β-mercaptoethanol , 0 . 05% NP40 , PMSF , protease inhibitors [Serva] ) . Cells were lysed by sonication and centrifuged at 100000×g for 1 hr at 4°C . The supernatant was filtered through Nalgene bottle-top filter . The complexes were isolated from the cleared lysate on a 5-ml TALON column ( Clontech ) . Peak fractions were pooled , concentrated , and further purified in GF buffer ( 50 mM Tris–HCl pH 8 . 0 , 150 mM KCl , 2 mM TCEP ) by size exclusion chromatography on a S75 16/60 or S200 16/60 column . Peak fractions were pooled , concentrated to typically 3 to 5 mg/ml , frozen in small aliquots in liquid nitrogen , and stored at −80°C . Proteins tested for interactions were diluted to 15 µM in 150 µl reactions in GF buffer ( 50 mM Tris–HCl pH 8 , 150 mM KCl , 1 . 5 mM TCEP , 0 . 05% Tween20 ) and incubated at 20–22°C for 1 hr on a rotating wheel . 100 µl of the resulting incubation was analyzed by size exclusion chromatography on a Superose6 10/300 column at a flow rate of 0 . 4 ml/min and collected in 300 µl fractions . Eluates were analyzed by SDS PAGE and Coomassie staining . 5 µg of MBP-Knl1MELT1-H6 protein was phosphorylated with Mps1 kinase ( TTK , Life technologies ) in a 35 µl reaction in 12 . 5 mM Tris–HCl pH 7 . 5 , 35 mM KCl , 10 mM MgCl2 , 0 . 5 mM EGTA , 0 . 005% Triton X-100 , 2 mM TCEP , 0 . 5 µM Okadaic acid , at 30°C for 1 hr , 1200 rpm . BSA was added to a final concentration of 2 mg/ml , and the reaction was incubated with 20 µl Amilose resin ( New England Biolabs , Ipswich , MA ) for 1 hr at room temperature . The beads were washed with 50 mM HEPES-NaOH pH 7 . 5 , 150 mM KCl , 0 . 05% Tween-20 , 2 mM DTE . 20 µl Knl1-bound resin was then incubated in 50 µl of a 400 nM solution of prey proteins ( e . g . , H6-BubR1362–431/Bub3 or H6-Bub1209–270/Bub3 or loop swap constructs ) in 50 mM HEPES-NaOH pH 7 . 5 , 50 mM KCl , 0 . 05% Tween-20 , 2 mM DTE , 10% glycerol , 4 mg/ml BSA , for 1 hr at room temperature . Unbound proteins were removed with 50 mM HEPES-NaOH pH 7 . 5 , 150 mM KCl , 0 . 05% Tween-20 , 2 mM DTE . 20 µl of the resin was boiled in 70 µl of sample buffer , separated on 10% gel , and blotted with anti-Bub3 ( mouse monoclonal , BD , 1:1000 ) or anti-MBP ( mouse monoclonal , New England Biolabs , 1:10000 ) antibodies . Cells were adapted to Lys-0/Arg-0 ( Light ) medium or Lys-8/Arg-10 ( Heavy ) medium for 2 weeks . Cells were synchronized in mitosis by a 24-hr thymidine block , followed by a 14-hr treatment with nocodazole . After harvesting the mitotic population , cells were split in the presence of either 500 nM Reversine for 30 min or with DMSO as a control . LAP-BUB1 or LAP-KNL1 expression was induced for 24 hr using doxycycline and cells were harvested and mixed , followed by immunoprecipitation and mass spectrometry . Cells were lysed at 4°C in hypertonic lysis buffer ( 500 mM NaCl , 50 mM Tris–HCl [pH 7 . 6] , 0 . 1% sodium deoxycholate , 1 mM DTT ) including phosphatase inhibitors ( 1 mM sodium orthovanadate , 5 mM sodium fluoride , 1 mM β-glycerophosphate ) , sonicated , and LAP-tagged proteins were coupled to GFP-trap ( ChromoTek ) for 1 hr at 4°C . Purifications were washed three times with high-salt ( 2 M NaCl , 50 mM Tris–HCl ( pH 7 . 6 ) , 0 . 1% sodium deoxycholate , 1 mM DTT ) and low-salt wash buffers ( 50 mM NaCl , 50 mM Tris–HCl ( pH 7 . 6 ) , 1 mM DTT ) and subsequently eluted in 2 M Urea , 50 mM Tris-HCL ( pH 7 . 6 ) , 5 mM IAA . Samples were loaded on a C18 reverse phase column and ran on a nano-LC system coupled to a mass spectrometer ( LTQ-Orbitrap Velos; Thermo Fisher Scientific ) via a nanoscale LC interface ( Proxeon Biosystems , now Thermo Fisher Scientific ) , as described in Suijkerbuijk et al . ( 2012b ) . U2OS LacO cells ( a gift from S Janicki ) were grown in DMEM supplemented with 8% FBS ( Clontech ) , hygromycin ( 200 µg/ml ) , pen/strep ( 50 µg/ml ) , and L-glutamine ( 2 mM ) . Cells were transfected with the indicated constructs for 48 hr using Fugene HD according to the manufacturer's protocol . Asynchronously growing cells were arrested in prometaphase by the addition of nocodazole ( 830 nM ) for 2–3 hr . Cells plated on 12-mm coverslips were fixed ( with 3 . 7% paraformaldehyde , 0 . 1% Triton X-100 , 100 mM Pipes , pH 6 . 8 , 1 mM MgCl2 , and 5 mM EGTA ) for 5–10 min . Coverslips were washed with PBS and blocked with 3% BSA in PBS for 1 hr , incubated with primary antibodies ( GFP-booster [Chromotek] , rabbit-anti-BUBR1 [Bethyl] and CREST/anti-centromere antibodies [Cortex Biochem , Inc . ] ) for 16 hr at 4°C , washed with PBS containing 0 . 1% Triton X-100 , and incubated with secondary antibodies ( goat-anti-rabbit Alexa Fluor 568 and goat anti-human Alexa Fluor 647 ) for an additional hour at room temperature . Coverslips were then washed , incubated with DAPI for 2 min , and mounted using antifade ( ProLong; Molecular Probes , Eugene , OR ) . All images were acquired on a deconvolution system ( DeltaVision RT; Applied Precision , part of GE Healthcare ) with a 100×/1 . 40 NA U Plan S Apochromat objective ( Olympus , Shinjuku , Tokyo , Japan ) using softWoRx software ( Applied Precision ) . | The genetic material within our cells is arranged in structures called chromosomes . Before a cell divides it makes an accurate copy of all of its DNA . The genetic material then needs to be equally split so that both daughter cells have a complete set of chromosomes . As the cell prepares to divide , each chromosome—consisting of two identical sister chromatids—lines up on a structure known as the spindle , which is made of filaments called microtubules . Cells have a sophisticated safety mechanism known as the spindle assembly checkpoint to ensure that chromosomes have time to correctly line up on the spindle before the cell can divide . Once this checkpoint is satisfied , the microtubules pull the sister chromatids apart so that each daughter cell receives one chromatid from each pair . The microtubules attach to the chromosomes through a large protein complex known as the kinetochore that assembles on each sister chromatid . The spindle assembly checkpoint monitors the attachment of the kinetochores to the microtubules; and two proteins , called Bub1 and BubR1 , play an essential role in this process . These proteins bind to another protein called Bub3 that is also part of the spindle assembly checkpoint . Although Bub1 and BubR1 are very similar , they do not appear to perform the same roles , but the precise molecular details of their differences remain unclear . In this study , Overlack , Primorac et al . studied Bub1 and BubR1 in human cells . The experiments show that Bub1 can be recruited to kinetochores in the absence of BubR1 , but BubR1 will only move to kinetochores when Bub1 is present . Furthermore , BubR1 needs to bind to Bub1 directly to move to the kinetochores . Overlack , Primorac et al . also identified a region in Bub1 that binds to Bub3 , and which is considerably different in BubR1 . When this region of Bub1 was grafted into BubR1 , the resulting protein was able to bind kinetochores even in the absence of Bub1 . The genes that encode the Bub1 and BubR1 proteins originate from a single ancestor gene that was duplicated during evolution . Therefore , the findings of Overlack , Primorac et al . show how the duplication of a gene can be beneficial for cells by creating products that have different roles in cells . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"biochemistry",
"and",
"chemical",
"biology",
"cell",
"biology"
] | 2015 | A molecular basis for the differential roles of Bub1 and BubR1 in the spindle assembly checkpoint |
Despite its being historically conceptualized as a motor expression site , emerging evidence suggests the ventral pallidum ( VP ) plays a more active role in integrating information to generate motivation . Here , we investigated whether rat VP cue responses would encode and contribute similarly to the vigor of reward-seeking behaviors trained under Pavlovian versus instrumental contingencies , when these behavioral responses consist of superficially similar locomotor response patterns but may reflect distinct underlying decision-making processes . We find that cue-elicited activity in many VP neurons predicts the latency of instrumental reward seeking , but not of Pavlovian response latency . Further , disruption of VP signaling increases the latency of instrumental but not Pavlovian reward seeking . This suggests that VP encoding of and contributions to response vigor are specific to the ability of incentive cues to invigorate reward-seeking behaviors upon which reward delivery is contingent .
The tendency to seek rewards , like many adaptive behaviors , is influenced by multiple dissociable decision-making processes . These decision-making processes have different costs and benefits and may be differentially vulnerable to perturbations that contribute to psychopathology . The ventral pallidum ( VP ) is a critical node in the neural circuitry underlying reward-related behaviors ( Creed et al . , 2016; Smith et al . , 2009 ) , including relapse to drug and alcohol seeking ( Farrell et al . , 2018; Kalivas and Volkow , 2005; Saunders et al . , 2015 ) . Neurons in the VP are known to respond to a variety of reward-related stimuli , including primary rewards ( Itoga et al . , 2016; Tindell et al . , 2006 ) , Pavlovian cues predicting reward delivery ( Smith et al . , 2011; Tindell et al . , 2005; 2009 ) , cues predicting reward availability ( Richard et al . , 2016 ) , and cues indicating specific appropriate reward-seeking actions ( Ito and Doya , 2009; Tachibana and Hikosaka , 2012 ) . Further , activity in VP is critical for normal levels of cue-elicited reward seeking across a variety of paradigms ( Leung and Balleine , 2015; Mahler et al . , 2014; Prasad and McNally , 2016 ) . Yet , there is little consensus on the primary role of VP in driving reward-seeking behaviors , and how VP signaling contributes to reward seeking across behavioral domains and rewarding outcomes . While historically the VP has been suggested to act as a motor expression site for a ‘limbic-motor’ interface ( Heimer et al . , 1982; Mogenson et al . , 1980 ) due to its close connections with motor output regions , more recent work has suggested a greater role for VP in reward processing itself . For instance , VP responses to cues have been argued to contribute to reward seeking by representing expected reward value ( Tachibana and Hikosaka , 2012 ) , but have also been suggested to encode state and action values ( Ito and Doya , 2009 ) . Previously , we showed that VP neurons encode the incentive value of cues , as defined by their ability to invigorate instrumental reward-seeking actions ( Richard et al . , 2016 ) . Here , we examined VP encoding of cue-driven reward seeking in two different behavioral models , in which cues generate superficially similar behavioral patterns based on similar levels of and variations in reward expectancy , after training with distinct underlying reward contingencies . The first model consists of Pavlovian conditioning , in which an auditory tone predicts the delivery of sucrose to a reward port . While delivery of the reward is not contingent on the animal’s behavior during the cue , over conditioning the animal learns to approach the reward port during the cue , prior to reward delivery . The second model consists of a discriminative stimulus task , similar to that reported previously ( Richard et al . , 2016 ) , but modified to generate superficially similar behavior to that following Pavlovian conditioning . In this task , the animal is presented with an auditory cue ( the discriminative stimulus , DS ) , which indicates the availability of sucrose reward from the port if the animal enters the reward port during the cue period . We found that although VP neurons respond robustly to reward-related cues in both tasks , these cue responses are only robustly predictive of reward seeking vigor when reward delivery is contingent upon that behavior . This suggests that VP neuron responses to cues contribute to reward seeking , not by driving locomotor vigor more generally , but by encoding an underlying decision process that is reflected in the latency of instrumental but not Pavlovian responses .
To assess the degree to which VP encoding of cue-elicited reward seeking depends on the underlying task structure , we trained rats in either an instrumental task ( the ‘DS task’ ) or in Pavlovian conditioning . In the DS task , entry into the reward port during the DS ( an auditory cue lasting up to 10 s ) resulted in delivery of liquid sucrose ( 10% ) , whereas port entries during an alternative auditory cue ( the non-reward stimulus [NS] ) had no programmed consequences . During Pavlovian conditioning , presentations of one auditory cue ( the CS+ ) predicted delivery of liquid sucrose at the end of the cue , irrespective of the animal’s port entry behavior , whereas presentations of an alternative auditory cue ( the CS– ) did not predict sucrose delivery . In both tasks , rats learned to enter the reward delivery port more quickly ( Figure 1A–B and Figure 1—figure supplement 1A–B ) and frequently ( Figure 1C–D and Figure 1—figure supplement 1C–D ) during the reward-related cue ( DS or CS+ ) than the control cue ( NS or CS– ) . They were trained until they made port entries during at least 70% of reward cue presentations ( DS or CS+ ) and less than 30% of control cue presentations ( NS or CS– ) . Rats trained in the two tasks did not differ in the number of training days required to meet training criteria ( Figure 1E; t ( 10 ) =0 . 1249 , p=0 . 9031 , Bayes factor of 2 . 112 in favor of the null ) . Once the rats met the training criteria , they were implanted with drivable electrode arrays aimed at the VP . Most recording sites were centered in middle to slightly caudal locations in VP ( 0 to 0 . 24 mm behind bregma ) , though one rat trained in the Pavlovian task had electrodes centered more caudally ( ~0 . 5 mm caudal to bregma ) . All comparisons between encoding in the instrumental and Pavlovian tasks were conducted with and without this subject , to ensure that our conclusions were not biased on the basis of recording location . The results reported here include this subject . Overall , many VP neurons were responsive to cues , and at the time of port entry and/or reward delivery in both tasks ( Figure 2C , Figure 2—figure supplement 1A–D , Figure 3A and C ) . Because we were primarily interested in determining the degree to which VP neurons encoded the learned or incentive value of cues in the two tasks , we focused the bulk of our analysis on explicit cue responses . The DS elicited increases in activity in 60% of VP units ( 189/314 ) and decreases in activity in 27% of units ( 85/314 ) . A smaller , but still robust , population of VP neurons responded to the CS+ , including 48 . 7% of VP neurons that were excited by the CS+ ( 191/392 ) and 20 . 9% of VP neurons that were inhibited ( 82/392 ) . While the proportion of cue-responsive neurons differed across the two tasks ( x2 = 30 . 34 , p<0 . 001 ) , we next assessed whether these cue responses encoded learned value or behavioral responses similarly . To determine what information about the cues is encoded in the two tasks , we first assessed whether changes in VP neuron firing distinguished between the reward-related cue ( DS or CS+ ) and the control cue ( NS or CS– ) in each task ( Figure 3A–D ) . Most DS excited neurons were significantly more excited by the DS than the NS ( Figure 2—figure supplement 1A; 132/189 , 69 . 8% ) , whereas only 23 . 5% of DS-inhibited neurons were more inhibited by the DS than the NS ( Figure 2—figure supplement 1B; 20/85 , 6% of the whole population; Figure 3B ) . As a population , DS-excited neurons responded significantly more to the DS and the NS ( Figure 3E; t[197]=15 . 43 , p<0 . 001 ) and DS-inhibited neurons were more inhibited by the DS than by the NS ( Figure 3G; t[84]=–3 . 10 , p=0 . 0027 ) , suggesting that both response types encode cue value . VP neurons also encoded learned cue value in the Pavlovian task: the majority of CS+ excited neurons were more excited by the CS+ than by the CS– ( Figure 2—figure supplement 1C; 64%; 123/191; 31 . 4% of the population ) , though only a small fraction of CS+ inhibited neurons were more inhibited by the CS+ than by the CS– ( 8 . 3%; 7/84;~2% of the whole population; Figure 3D and Figure 2—figure supplement 1D ) . As a population , CS+ excited neurons were significantly more excited by the CS+ than by the CS– ( Figure 3G; t[190]=18 . 6890 , p<0 . 001 ) and CS+ inhibited neurons were also more inhibited by the CS+ than by the CS– ( Figure 2H; t[83]=-−2 . 72 , p=0 . 01 ) , indicating that , like VP responses in the instrumental task , both excitations and inhibitions predict cue value after Pavlovian conditioning . To further assess the ability of post-cue VP neuron firing to predict cue identity , we conducted receiver operating characteristic ( ROC ) analyses to assess the detection of either the DS or the CS+ versus their respective control cues . For each unit , we calculated the area under the ROC curve ( auROC ) ; auROCs greater than five indicate units that had greater firing rates on reward cue than control cue trials . We found no difference in the population auROC distributions for the DS versus the CS+ for predicting cue identity ( Figure 3I; t[704]=1 . 22 , p=0 . 22; Bayes factor of 5 . 706 favoring the null ) , suggesting that VP firing encodes cue value equally in both tasks . Post-cue firing was significantly more predictive of cue identity than pre-cue firing for both the instrumental ( t[313]=11 . 8213 , p<0 . 001 ) and the Pavlovian task ( t[391]=11 . 0872 , p<0 . 001 ) . While cue responses were equally predictive of cue identity for the instrumental and Pavlovian tasks , DS excited neurons were more likely to be inhibited by the control cue ( the NS ) than CS+ excited neurons were to be inhibited by the CS– ( Figure 3J; x2 = 63 . 9923 , p<0 . 001 ) , perhaps suggesting a more active role for response inhibition by the neurons in the instrumental task . Beyond representing the predictive value of the DS , VP neurons have been previously shown to encode both the likelihood and latency of an instrumental reward-seeking response ( Richard et al . , 2016 ) . Here we assessed , whether VP cue responses would do so similarly for superficially similar behavioral responses to the DS and the CS+ . First , we assessed the magnitude of VP cue responses on trials with and without a behavioral response from those sessions with a sufficient number of trials without a response to allow reliable assessment of the difference ( at least five trials ) . Because responding in these two tasks was more likely to approach 100% than in our previous experiment ( Richard et al . , 2016 ) , probably due to the simpler , pre-potent nature of the port entry responses as opposed to a lever press , this trial criterion was lower than previously used , which may have reduced our ability to detect significant differences at an individual neuron level . When assessed individually ( Figure 4A ) , about 20% of DS-excited neurons are significantly more responsive to the DS when it is followed by a response ( 24/130 ) and 16% of DS-inhibited neurons are more responsive on trials with a response ( 14/85 ) . Only ~9% of CS+ excited neurons ( Figure 4B ) were more excited on trials with a response ( 14/160 ) and ~5% of CS+ inhibited neurons are more responsive ( 3/66 ) . As a population , DS-excited neurons ( 130 units ) were significantly more excited on trials when the DS was followed by a response ( Figures 4E and 3I; t[129]=8 . 60 , p<0 . 001 ) . DS-inhibited neurons ( 67 units ) were also more inhibited on trials with a response ( Figure 3F and I; t[66]=–4 . 66 , p<0 . 001 ) . Similarly , CS+ excited neurons ( 174 units ) were more excited on trials with a response ( Figure 4G and J; t[174]=10 . 08 , p<0 . 001 ) , though CS+ inhibited neurons ( 82 units ) did not distinguish between trials with and without a response ( Figure 4H and J; t[81]= . 479 , p=0 . 633; Bayes factor: null is 6 . 46 times more likely than the alternative ) . To assess the predictive ability of VP neuron firing rates in each task , we ran receiver operating characteristic ( ROC ) analysis to assess whether post-cue firing could predict the likelihood of a reward-seeking response , and then compared the distribution of auROCs for each task . Post-DS firing was significantly more predictive of port entry likelihood than post-CS+ firing ( Figure 4C and D; t[704]=3 . 49 , p<0 . 001 ) , though post-cue firing was predictive of port entry likelihood for both the instrumental task ( t[313]=3 . 708 , p<0 . 001 ) , and the Pavlovian task ( t[391]=2 . 697 , p=0 . 0073 ) . Overall , post-cue firing in both tasks is predictive of reward seeking likelihood , but at least to some degree , DS responses are more robustly predictive of port entry likelihood than responses to the CS+ . Given the differences in the VP encoding of response likelihood in the two tasks , we next wanted to examine whether post-cue response differentially predicted response vigor in the two tasks . To do so , we examined whether cue responses on a given trial were predictive of the latency of the animal to respond on that trial , and whether this differed for instrumental versus Pavlovian port entries , by running Spearman rank correlations on individual neurons . The post-DS firing rate of 62 units ( 19 . 75% ) significantly predicted the animal’s latency to make a port entry ( Figure 5A ) . By contrast , only 14 units ( 3 . 57% ) had post-CS+ firing rates that significantly predicted port entry latency ( Figure 5C ) , a much lower proportion of the population ( DS versus CS+ % correlated: x2 = 47 . 477 , p<0 . 001 ) . To determine the degree to which the proportion of latency-predicting neurons in the Pavlovian task was meaningful , we ran the same analysis on 1000 shuffled iterations of the data and determined the number of units that significantly predicted response latency in these artificial datasets . The proportion of units with true post-DS firing that predicted the real latency was well outside the distribution of significantly correlated units from the shuffled datasets ( Figure 5—figure supplement 1E ) , whereas the proportion of units in the Pavlovian task was not ( Figure 5—figure supplement 1F ) . In addition , when we assessed the number of units with post-cue firing rates that were more predictive of the real latency than of the shuffled latencies ( Figure 5—figure supplement 1C and D ) , we found only 9 units ( 2 . 8% ) that predicted latency in the Pavlovian task , versus 61 in the instrumental task ( 19 . 42% ) . Furthermore , while latency-predicting neurons are , on average , significantly excited by presentations of the DS ( Figure 5D and G; t[61]=7 . 08 , p<0 . 001; 48/62 units excited , 77 . 4% ) , latency-predicting neurons in the CS+ task are not ( Figure 5E and H; t[13]=2 . 00 , p=0 . 06650; 7/14 units excited; Bayes factor of 1 . 27 in favor of the alternative ) . Latency-predicting neurons are more likely to be excited by the cue in the instrumental task than in the Pavlovian task ( x2 = 4 . 294 , p=0 . 038246 ) . For the population , the distribution of correlation coefficients was significantly more negative in the instrumental task than in the Pavlovian ( Figure 5A and C; t[703]=2 . 0493 , p=0 . 04 ) , though the distribution is more negative in the 300 msec post-cue than in the 300 msec pre-cue in both the instrumental t[313]=–10 . 81 , p<0 . 001 ) and Pavlovian tasks ( t[390]=–9 . 52 , p<0 . 001 ) . This suggests that while latency encoding in the Pavlovian task is much weaker , VP firing rates do encode some information relating to the subsequent port entry latency . To determine when this information is first encoded by VP neurons , we assessed correlation coefficients ( Figure 5F ) and the proportion of units with significant correlations ( Figure 5I and J ) in 50 msec windows starting from 0 . 5 s before to 1 s after cue onset . Surprisingly , we found that the distribution of correlation coefficients in the Pavlovian task shifted negatively in the 50 msec window prior to cue onset ( Figure 5F; q=3 . 532 , p<0 . 05 ) , suggesting that the weak relationship between post-CS+ firing and latency may be accounted for by trial-by-trial variation in firing rates , rather than in phasic cue responses . By contrast , the distribution of correlation coefficients in the DS task did not significantly differ from 0 until the window 50–100 msec post-DS ( Figure 5F; q=4 . 171 , p<0 . 05 ) , when phasic cue excitations are occurring in the bulk of correlated neurons . Overall , encoding of the latency of cue-elicited reward seeking by individual VP units and in the population is much stronger in the instrumental task than in the Pavlovian task . These differences are not explained by greater variability in post-cue firing or in latency in the DS task . Variability in post-cue firing did not differ when comparing the whole population ( Figure 5—figure supplement 2A; t[704]=–1 . 16 , p=0 . 24; Bayes factor of 6 . 10 favoring the null ) and was greater in the Pavlovian task when only the cue excited neurons were considered ( Figure 5—figure supplement 2B; t[298]=2 . 75 , p=0 . 0063 ) . In addition , post-cue port entry latency is more variable in the Pavlovian task ( Figure 5—figure supplement 2C and D; z ( 95354 ) =−5 . 82 , p<0 . 001 ) , indicating that weak correlations between post-CS+ firing and latency are not due to low variability in post CS+ port entry latency . To further assess whether neural encoding differences could be accounted for by subtle locomotor differences between the two tasks , we conducted frame-by-frame video tracking analysis to assess velocity and distance from the reward port during the pre- and post-cue periods ( Figure 5—figure supplement 3A–H ) , as well as the timing of post-cue movement onsets ( Figure 5—figure supplement 4 ) from a subset of sessions . Overall , although we found that velocity increased and distance from the port decreased shortly after cue onset for both tasks , instrumentally trained rats moved at greater velocities post-cue ( F[1 , 118]=23 . 166 , p<0 . 001 ) and were positioned at further distances from the port at cue onset ( F[1 , 118]=6 . 99 , p=0 . 009 ) . We also assessed movement onsets to determine whether greater encoding of latency in instrumentally conditioned rats might be due to a greater incidence of movement onsets during the immediate post-cue period ( Figure 5—figure supplement 4 ) . Rats trained under Pavlovian versus instrumental contingencies did not differ in the likelihood of movement onset during the first 300 ms post cue in which we assessed neural firing ( 13 . 3% of post-DS movement onsets and 8 . 3% of post-CS+ movement onsets; X2=0 . 345 , p=0 . 56 ) , and we did not observe a significant difference in movement-onset times ( F[1 , 74]=1 . 365 , p=0 . 24 ) . To determine whether subtle differences in movement ( velocity and movement onset ) and/or non-movement ( distance ) variables drive differences in neural encoding between the two tasks , we assessed whether the relationships between post-cue firing and latency in individual neurons were predicted by the relationship between post-cue firing in these same units and trial-by-trial a ) distance from port , b ) post-cue velocity , or c ) movement onset latency . We found that the only significant single predictor of latency encoding was distance encoding , in that neurons that had greater post-cue firing rates on trials with shorter port entry latencies , also had greater post-cue firing on trials where the rat was closer to the port at cue onset ( Figure 5—figure supplement 5A; F[1 , 140]=16 . 295 , p<0 . 001 ) , similar to results previously reported for the nucleus accumbens ( McGinty et al . , 2013 ) . This relationship between encoding of proximity and vigor depended on training history ( F[1 , 140]=4 . 55 , p=0 . 034 ) , as distance encoding in the rats trained under a Pavlovian contingency was not significantly predictive of latency encoding ( F[1 , 44]=2 . 09 , p=0 . 15 ) , whereas distance encoding robustly predicted latency encoding in instrumentally trained rats ( F[1 , 96]=35 . 632 , p<0 . 001 ) . Notably , latency encoding was not significantly predicted by the degree to which neurons encoded post-cue velocity ( Figure 5—figure supplement 5B; F[1 , 140]=–0 . 335 , p=0 . 56 ) or movement-onset latencies ( Figure 5—figure supplement 5C; F[1 , 140]=1 . 83 , p=0 . 17 ) . Proximity encoding has been suggested to reflect pre-cue variables such as attention or ‘task engagement’ , but may also reflect encoding of other information derived from distance , such as expected effort ( Hamid et al . , 2016; McGinty et al . , 2013; Nicola , 2010 ) . That the relationship between distance and latency encoding was modulated by training history suggests a greater coupling between these variables and action in the instrumental task . Given that VP neurons differentially encode the likelihood and vigor of cue-elicited reward-seeking behavior , we next wanted to assess the functional contributions of VP activity to performance in these tasks . As in the electrophysiology experiments , we trained rats in either the DS task or in Pavlovian conditioning . By the end of training , rats made port entries at significantly shorter latencies during the reward cue than during the control cue ( main effect of cue: F[1 , 28]=25 . 314 , p<0 . 001 ) , regardless of training group ( interaction of cue and training: F[1 , 28]=0 . 144 , p=0 . 707; Figure 6—figure supplement 1A , B and F ) . Rats in both training groups increased their probability of port entry preferentially during the reward-related cue versus the control cue ( main effect of cue: F[1 , 28]=951 . 681 , p<0 . 001 ) , with rats trained under the instrumental contingency reaching a slightly higher response probability during the reward cue once they met training criteria ( interaction of cue and training: F[1 , 28]=5 . 962 , p=0 . 021; Figure 6—figure supplement 1C , D and G ) . On average , rats trained under the Pavlovian conditioning required more days to reach training criteria ( 10 . 35 ± 0 . 23 ) than those trained under an instrumental contingency ( 9 . 07 ± 0 . 24; t[29]=3 . 839 , p<0 . 001; Figure 6—figure supplement 1E ) . In the instrumental task , inactivation of VP with low doses of the GABA agonists baclofen and muscimol reduced the probability of port entry during the cue period ( Figure 6B and Figure 6—figure supplement 2B; main effect of treatment: F[4 , 130]=28 . 666 , p<0 . 001; Sidak , p<0 . 001 ) , and did so more strongly during the DS period than during the NS ( interaction between treatment and cue identity: F[4 , 130]=7 . 205 , p<0 . 001; DS Sidak , p<0 . 001; NS Sidak , n . s . ) . VP inactivation also increased port entry latency during the cue ( Figure 6D and Figure 6—figure supplement 2A; main effect of treatment: F[4 , 18 . 254]=4 . 574 , p=0 . 01 ) also selectively during the DS ( interaction between drug and cue: F[4 , 23 . 879]=4 . 354 , p=0 . 009; DS Sidak , p=0 . 022; NS Sidak , n . s . ) . Inactivation of VP during the Pavlovian task similarly reduced the probability of port entry during the cue ( Figure 6C and Figure 6—figure supplement 3B; main effect of treatment: F[4 , 112]=9 . 904 , p<0 . 001; Sidak p<0 . 001 ) , and did so selectively during the CS+ period ( interaction between treatment and cue identity: F[4 , 112]=2 . 622 , p=0 . 039; CS+ Sidak , p<0 . 001; CS- Sidak , n . s . ) . In contrast to the instrumental task , VP inactivation during the Pavlovian task had no significant effect on port entry latency ( Figure 6E and Figure 6—figure supplement 3A; main effect of treatment: F[4 , 21 . 406]=1 . 059 , p=0 . 401 ) , regardless of cue identity ( interaction between treatment and cue identity: F[4 , 18 . 061]=1 . 742 , p=0 . 185 ) . In both tasks , no drug treatment had any significant effect on port entry behavior during the inter-trial interval ( main effect of treatment , instrumental task: F[4 , 65]=2 . 295 , p=0 . 07; Pavlovian task: F[4 , 63]=1 . 526 , p=0 . 205 ) . VP receives a number of inputs that may contribute to reward-cue excitations that have not been previously studied , including glutamatergic inputs from a variety of cortical and cortical-like structures ( Fuller et al . , 1987; Kelley et al . , 1982; Maslowski-Cobuzzi and Napier , 1994; Reep and Winans , 1982; Záborszky et al . , 1984 ) , as well as from the lateral hypothalamus ( Grove , 1988 ) , midline thalamic nuclei , and subthalamic nucleus ( Fuller et al . , 1987 ) , dopaminergic inputs from the midbrain ( Maslowski-Cobuzzi and Napier , 1994; Napier and Potter , 1989 ) and substance P inputs from the nucleus accumbens ( Napier et al . , 1995 ) . Therefore , we next sought to assess the contributions of glutamatergic , dopaminergic and substance P signaling in the VP to the likelihood and vigor of cue-elicited reward seeking after instrumental or Pavlovian training . Infusions of either a mixture of the glutamate receptor antagonists CNQX and MK801 or the relatively non-selective dopamine receptor antagonist flupenthixol had similar effects to VP inactivation with GABA agonists in both tasks . Glutamate or dopamine blockade reduced the probability of port entry during the cue ( Sidak p<0 . 001 for both ) , though only the effect of glutamate blockade was selective to the DS ( Figure 6B and Figure 6—figure supplement 1B; DS Sidak , p<0 . 001; NS Sidak , p=n . s . ) whereas flupenthixol reduced the probability of port entry during both cue types ( DS Sidak , p<0 . 001; NS Sidak , p<0 . 003 ) . Glutamate or dopamine antagonism selectively increased port entry latency during the DS ( CNQX-MK801 Sidak , p=0 . 005; flupenthixol Sidak , p=0 . 022 ) , but not during the NS . In the Pavlovian task , glutamate or dopamine antagonism , like inactivation , significantly reduced port entry probability during the CS+ ( Figure 6C and Figure 6—figure supplement 2B; CNQX-MK801 Sidak , p<0 . 001; flupenthixol Sidak , p=0 . 043 ) , but not during the CS– , and had no effect on port entry latency ( Figure 6E and Figure 6—figure supplement 2A ) . Infusions of the NK-1 antagonist WIN51708 , used to block the effects of substance P , had no effect on port entry probability or latency in either task . Because , for many behaviors , VP functionality appears to be organized topographically ( Root et al . , 2015 ) , we mapped our behavioral effects for each subject at each microinfusion site in the horizontal plane ( Figure 6—figure supplements 2 and 3 ) to determine whether the null effects on latency in the Pavlovian task were due to mixed effects at distinct neuroanatomical locations . This does not appear to be the case because the vast majority of our infusion sites were located in the caudal half of VP , in and around the hedonic ‘hot spot’ ( Ho and Berridge , 2013; Smith and Berridge , 2005; 2007 ) , and we found inconsistent changes in post-CS+ port entry latency even at this particular anatomical location ( Figure 6—figure supplement 3A ) .
Our results demonstrate that VP neurons do not drive movement invigoration generically , but encode a variable that is manifested by port entry latency after instrumental , but not after Pavlovian training . Yet , VP has not been classically construed as a motor output region ( Heimer et al . , 1982; Mogenson et al . , 1980 ) without cause . Manipulations of VP alter a wide range of motor behaviors including general locomotion ( Churchill and Kalivas , 1999; Kitamura et al . , 2001; Napier and Chrobak , 1992; Root et al . , 2015 ) , though reports of locomotor activation linked to VP may be more related to activity in nearby structures such as the rostral preoptic area ( Zahm et al . , 2014 ) . The differences in VP encoding of the latency of instrumental and Pavlovian approach behaviors that we report here are inconsistent with a general role in motor invigoration , but perhaps VP signaling serves to invigorate a specific subset of behavioral responses . For instance , similar cue-related signals in the nucleus accumbens and in midbrain dopamine neurons have been proposed to promote ‘flexible approach’ , in which animals must navigate toward their goal location from a flexible starting location ( McGinty et al . , 2013; Nicola , 2010 ) . It is unlikely , however , that promotion of flexible approach accounts for the differences in encoding that we report here for superficially similar approach behaviors . VP neuron activity may contribute to at least some forms of flexible approach , but only when that approach reflects a specific underlying decision variable . VP activity has been suggested to signal many different decision-making variables including action values or state values ( Ito and Doya , 2009 ) . One critical difference between cue responses in the instrumental versus Pavlovian tasks may be the degree to which state values are linked to action values ( Averbeck and Costa , 2017 ) . While a stronger link between state and action values in the instrumental task may explain a more robust relationship between VP cue responses and reward-seeking behaviors , the existing literature does not make clear predictions about how changes in state value should alter response latency in these two conditioning paradigms , or about how response latency might be altered separately from response probability after Pavlovian conditioning . An alternative variable that predicts a more specific relationship between VP activity and the animals’ reward-seeking vigor or motivation is incentive value , or the degree to which cues have the ability to activate motivational states ( Bindra , 1978; Robinson et al . , 2014 ) , including those that invigorate ongoing reward-seeking actions . We and others have shown that this property of incentive cues requires normal activity in VP neurons ( Leung and Balleine , 2015; Prasad and McNally , 2016; Richard et al . , 2016 ) . Our finding that latency-encoding neurons are more likely to also encode proximity in instrumentally trained rats supports the hypothesis that these firing patterns reflect an underlying variable that is related to both proximity and response vigor . Proximity-encoding signals have been hypothesized to reflect encoding of information derived from distance , such as reward expectancy or expected effort , but may also be influenced by pre-cue factors such as ‘task engagement’ or motivational state at cue onset ( Howe et al . , 2013; McGinty et al . , 2013; Nicola , 2010 ) , consistent with an incentive motivational role for this firing ( Ahrens et al . , 2016; Richard et al . , 2016; Smith et al . , 2011 ) . Within this framework , VP neuron activity may represent the incentive value of both the instrumental DS and the Pavlovian CS+ , but the latency to approach the reward location following Pavlovian cues does not reflect this incentive value ( Ahrens et al . , 2016; Chang et al . , 2015; Robinson and Flagel , 2009 ) . This is supported by our finding that proximity encoding is similar in the two tasks , but is only correlated with latency encoding in the instrumental task . Alternatively , incentive value may be low in the Pavlovian task , and therefore the corresponding encoding less apparent , consistent with weaker VP population-level representation of reward cues in this task in general . We should note that VP activity both encodes and contributes to the likelihood of behavioral responses following the Pavlovian cue , though this encoding is weaker than in the instrumental task . VP activity may contribute to response likelihood in the Pavlovian task via the same underlying circuits and processes as in the instrumental task , or through distinct neural and psychological mechanisms , such as thencoding of incentive value versus expected reward value ( Chan et al . , 2016; Tindell et al . , 2009 ) , as some evidence suggests that reward value and motivation rely on dissociable neural mechanisms even within VP ( Creed et al . , 2016 ) . Because rostral and caudal VP neurons have different morphologic and electrophysiologic properties ( Bengtson and Osborne , 2000; Kupchik and Kalivas , 2013 ) , and differentially modulate reward-related behaviors ( Chang et al . , 2017; Ho and Berridge , 2014; Johnson et al . , 1993; Mahler et al . , 2014; McBride et al . , 1999; Panagis et al . , 1995; Smith and Berridge , 2005 ) , we previously assessed neural responses in the DS task in rostral , middle and caudal subregions of VP ( Richard et al . , 2016 ) . We observed similar population encoding at rostral , middle and caudal recording sites in VP , in that most neurons in each subregion exhibited excitations following the DS , and post-cue firing in 18–25% of neurons predicted response latency . Although neurons throughout VP appear to encode cue value and the subsequent reward seeking similarly across the rostrocaudal axis , discrete populations may be differentially involved in dissociable behavioral responses to these reward cues ( Milton and Everitt , 2010 ) , including their ability to act as conditioned reinforcers ( Mahler et al . , 2014; Torregrossa and Kalivas , 2008 ) or to generate instrumental reward-seeking or other reward-related motivational states ( Leung and Balleine , 2013 ) . Here , we aimed to reduce the potential influence of subregional heterogeneity on differences in encoding or functional effects between the two tasks by focusing our recording and injections in the same middle to slightly caudal area of VP , in and around the VP ‘hot spot’ . Whether the differences reported here hold at more rostral regions of VP , remains an open question . VP has previously been conceptualized primarily as a major output structure of the nucleus accumbens , which has been demonstrated to encode the value of reward-predictive cues ( Ambroggi et al . , 2008 , 2011; Day et al . , 2006 ) as well as both proximity to the response operandum and the vigor of subsequent reward-seeking actions ( McGinty et al . , 2013; Nicola et al . , 2004 ) . We previously showed that the timing of nucleus accumbens and VP cue responses is inconsistent with the idea that VP cue responses are a reflection of upstream activity changes in the accumbens ( Richard et al . , 2016 ) . In addition , chemogenetic ‘disconnection’ of VP and accumbens during Pavlovian conditioning results in elevated sign-tracking behavior , suggesting a competitive interaction between these two sites ( Chang et al . , 2018 ) . Since nucleus accumbens inputs to VP are primarily GABAergic , the likeliest mechanism by which accumbens activity changes could result in VP excitations is via disinhibition: yet we showed previously that post-cue excitations in VP neurons occur well before accumbens inhibitions . Further , we show here that blocking the actions of substance P , the main known source of excitatory drive from the accumbens to the VP ( Napier et al . , 1995 ) , has no effect on the likelihood or latency of cue-elicited reward seeking . By contrast , local disruption of either glutamate or dopamine signaling was effective in reducing the likelihood and speed of cue-elicited reward seeking , indicating a role for both neurotransmitter systems in VP contributions to the invigoration of reward seeking . VP neurons integrate glutamatergic inputs from a variety of brain areas implicated in reward learning and cue-elicited motivation ( Fuller et al . , 1987; Grove , 1988; Kelley et al . , 1982; Maslowski-Cobuzzi and Napier , 1994; Reep and Winans , 1982; Záborszky et al . , 1984 ) , including the medial pre-frontal cortex ( mPFC ) ( Capriles et al . , 2003; Ishikawa et al . , 2008a , 2008b; Moorman and Aston-Jones , 2015; Peters et al . , 2009; Stefanik et al . , 2013 ) and the basolateral amygdala ( BLA ) ( Ambroggi et al . , 2008; Ishikawa et al . , 2008b; Jones et al . , 2010a , 2010b; McDonald , 1991; Perry and McNally , 2013; Záborszky et al . , 1984 ) . Excitatory drive from the BLA onto VP neurons has been shown to be modulated by local dopamine following stimulation of the ventral tegmental area ( VTA ) ( Maslowski-Cobuzzi and Napier , 1994; Napier and Potter , 1989 ) . Importantly , inactivation of the VTA reduces the likelihood and increases the latency of instrumental responding ( Fischbach-Weiss et al . , 2018 ) , including DS-elicited reward seeking ( Yun et al . , 2004 ) . The mechanisms by which dopamine and glutamate inputs might interact in VP to modulate response probability or vigor are an important area of future study . Here we demonstrate that VP neurons selectively encode and contribute to the vigor of cue-elicited reward-seeking actions when those actions are trained via an instrumental contingency , rather than via Pavlovian conditioning . These results indicate that VP encoding of vigor is not related to motor invigoration or reward expectancy per se , but to the ability of reward-related cues to invigorate actions upon which reward is contingent . Whether VP neurons signal the value of work or the incentive properties of cues more generally requires further investigation . Together , our results show that VP cue responses do not merely reflect motor invigoration , but encode a motivation signal that may be integrated in the VP .
Male and female Long Evans rats ( n = 54; Envigo ) , weighing 250–275 grams at arrival , were individually housed in a temperature- and humidity-controlled colony room on a 12 hr light/dark cycle . Starting one day prior to the initiation of training in either task and until they met criteria for responding to the reward-related cue , rats were food restricted to ~18–20 g/rat/day , and the amount of food was adjusted daily to maintain rats at ~90% of their free-feeding body weight . All experimental procedures were approved by the Institutional Animal Care and Use Committee at Johns Hopkins University and were carried out in accordance with the guidelines on animal care and use of the National Institutes of Health of the United States . Rats were randomly assigned one of the following two auditory cues as their conditioned stimulus ( CS+ ) for training and testing: ( 1 ) white noise or ( 2 ) 2900 Hz tone . Rats received the alternate auditory cue as their CS– . During conditioning sessions , the CS+ and CS– , each lasting 10 s , were presented on a pseudorandom variable interval schedule with a mean inter-trial interval ( ITI ) of 50 s . At 8 s after the CS+ onset , 13 mL of 10% sucrose was delivered into the sucrose delivery port over a period of 2 s . Rats underwent daily conditioning until they met final response criteria ( port entries on at least 70% of CS+ presentations and less than 30% of CS– presentations ) prior to being implanted with electrode arrays or undergoing pharmacological manipulations of VP . Rats were trained to perform a modified DS task , similar to those described previously ( Ghazizadeh et al . , 2012; Richard et al . , 2016 ) . Rats in the modified DS task group were randomly assigned one of the following two auditory cues as their DS for training and testing: ( 1 ) white noise or ( 2 ) a 2900 Hz tone . Rats received the alternate auditory cue as their NS ( neutral stimulus ) . Entries into the sucrose delivery port during the DS presentation resulted in liquid sucrose ( 0 . 13 ml , 10% ) delivery and termination of the DS cue . Port entries during the NS presentation or during the inter-trial interval ( ITI ) had no programmed consequences . Rats underwent the following sequential training stages ( 1 ) DS only , up to 60 s , ( 2 ) DS only , up to 30 s , ( 3 ) DS only , up to 20 s , ( 4 ) DS only up to 10 s , and ( 5 ) final stage , DS and NS ( Figure 1—figure supplement 1 ) . Once rats met preliminary DS response criteria at each stage ( port entries on at least 60% of DS presentations ) , they were advanced to the next stage on the following day . During the final stage of the DS task , the DS and NS were presented on a pseudorandom variable interval schedule with a mean ITI of 50 s , and each trial lasted up to 10 s . Rats were trained to final criteria ( port entries on at least 70% of DS presentations and less than 30% of NS presentations ) prior to being implanted with electrode arrays or undergoing pharmacological manipulations of VP . During surgery , rats were anesthetized with isoflurane ( 5% ) and placed in a stereotaxic apparatus , after which surgical anesthesia was maintained with isoflurane ( 0 . 5–2 . 0% ) . Rats received pre-operative injections of carprofen ( 5 mg/kg ) , topical lidocaine for analgesia and cefazolin ( 75 mg/kg ) to prevent infection . Guide cannulae , electrodes and microdrives were secured to the skull with bone screws and dental acrylic . All rats were given at least 7 days to recover prior to any microinjections or tethering . Electrophysiological recording was conducted as described previously ( Ambroggi et al . , 2008; Ghazizadeh et al . , 2010; 2012; Nicola et al . , 2004; Richard et al . , 2016 ) . Rats were connected to the recording apparatus ( Plexon Inc , TX ) , consisting of a head stage with operational amplifiers , cable and a commutator to allow free movement during recording . Rats were run for 1 . 5 hr daily sessions , and recording was started after rats regained performance criteria ( port entries during at least 70% of CS+ or DS presentations and less than 30% of control cue presentations ) following surgery . The microdrive carrying the electrode arrays was lowered by 80–160 µm at the end of each session with satisfactory behavior ( >60% of DS or CS+ presentations with a response ) , in order to obtain a new set of neurons for each session that was included in the analysis . For a subset of recording sessions , overhead video was captured at 30 frames per s ( Figure 5—figure supplement 1A and B ) . For those subjects for which video was available ( n = 4 , 2 instrumental and 2 Pavlovian ) , Noldus Ethovision software was used to track the position of a red LED mounted on the recording headstage located on top of the rats’ heads during the session , with the maximum number of recorded units located in VP for each subject . The position of the LEDs was transformed into a position coordinate , with missing data points filled in using linear interpolation . An experimenter who was blind to conditioning group viewed all pre- and post-cue periods to adjust for errors in tracking or interpolation manually . These position coordinates were used to determine frame-by-frame velocity ( cm/s ) and distance from the reward port ( cm ) during the pre- and post-cue periods ( Figure 51—figure supplement 1C–H ) . Velocity data were used to determine the latency of movement onsets following each cue presentation , by finding the first of at least five consecutive frames with velocity values greater than the 95% confidence interval of baseline velocity ( 5 s pre-cue; Figure 5—figure supplement 2 ) . To assess differences in these variables across the two tasks , we fit linear mixed-effects models for each movement variable with a fixed effect for training history and a random effect for subject . To determine the degree to which neural encoding of port entry latency was related to encoding of these response characteristics , we first computed Spearman’s rank correlation coefficients for each unit and trial-by-trial velocity ( 0 to 0 . 5 s and 0 . 5 to 1 s post-cue , cm/s ) , distance from the port ( cm ) and movement onset latency ( s ) . Then , we fit a linear mixed-effects model of the correlation coefficients relating neural firing to port entry latency , with fixed effects for training history and the correlation coefficients for velocity , distance and movement onset latencies , as well as a random effect for subject . Drug microinjections were administered bilaterally in a 0 . 3 µl volume on test days spaced at least 48 hr apart , and counterbalanced for drug order across rats . On test days , solutions were brought to room temperature , and were infused at a rate of 0 . 3 µl per minute using a syringe pump attached via PE-20 tubing to stainless steel injectors ( Plastics One , 29-gauge ) that extended 2 mm beyond the end of the guide cannulae into VP . Injectors were left in place for 1 min to allow diffusion of the test solution , after which the experimenter replaced the obturators and immediately placed the rat in the testing chamber to begin the session . After reaching criteria in either task and prior to test sessions , rats received microinjections of vehicle immediately prior to a final training session , to habituate them to the testing procedure . Following each test day , rats received at least one drug-free retraining session to ensure that initial performance criteria were met . In order to test the importance of VP activity and the role of specific neurotransmitter systems in cue-elicited behavior , rats ( n = 43; Pavlovian conditioning , n = 23; DS task , n = 20 ) received bilateral VP infusions of the following drugs ( amounts given are per hemisphere ) : a mixture of the GABAA agonist muscimol and the GABAB agonist baclofen ( 10 ng each in 0 . 3 µl 0 . 15 M saline ) to inactivate VP; a mixture of the AMPA/kainate receptor antagonist CNQX ( 450 ng ) and the NMDA receptor antagonist MK-801 ( 2 µg in 0 . 3 µl 90% saline/10% DMSO vehicle ) ; the relatively non-selective dopamine receptor antagonist , flupenthixol ( 15 µg in 0 . 3 µl saline ) ; and the NK-1 receptor antagonist WIN57108 ( 10 ng in 0 . 3 µl 90% saline/10% DMSO vehicle ) to block the actions of substance P . Because WIN51708 and CNQX were initially dissolved in a small amount of DMSO , we also compared the effects of saline vehicle with the 90% saline/10% DMSO mixture to ensure that there were no differing effects of the vehicle solutions themselves , and found no significant effects ( all ps > 0 . 05 ) . On the basis of histological analysis , a total of 31 rats were included in the analysis of the microinjection effects , 17 from the Pavlovian task and 14 from the instrumental task . Effect sizes were not determined a priori , but we aimed to include at least 12 subjects with accurate bilateral cannulae placements in each behavioral group , based on established standards . To compare the effects of VP inactivation or receptor blockade on responses to Pavlovian versus instrumental cues , we primarily focused our analysis on the likelihood and latency of port entry during each cue type . To account for some missing data ( e . g . when port entry probability went to zero , no latency data were available ) , we analyzed the data using linear mixed models ( drug X cue type [CS+ versus CS- or DS versus NS] ) , followed by pairwise comparisons with Sidak corrections . To achieve this , we used Akaike’s information criteria to determine the best-fitting covariance model . Depending on the best-fitting covariance model ( Verbeke and Molenberghs , 2009 ) , the degrees of freedom may be a non-integer value . In both tasks , the best fitting model for the assessment of latency was a first-order antedependence model and the best fitting model for the assessment of probability was an identity model . We also analyzed the number of port entries during the inter-trial interval using linear mixed models with an identity covariance structure , followed by pairwise comparisons with Sidak corrections . Animals were deeply anesthetized with pentobarbital and electrode sites were labeled by passing a DC current through each electrode . All rats were perfused intracardially with 0 . 9% saline following by 4% paraformaldehyde . Brains were removed , post-fixed in 4% paraformaldehyde for 4–24 hr , cryoprotected in 25% sucrose for >48 hr , and sectioned at 50 um on a microtome . We verified the location of injection or recording sites using two methods . One set of tissue was stained with cresyl violet and analyzed using light microscopy . On the alternating set of sections , we performed immunohistochemistry for substance P ( SP ) to better demarcate the boundaries of VP . Sections were washed in PBS with bovine serum albumin and triton ( PBST ) for 20 min , and incubated in 10% normal donkey serum in PBST for 30 min , before incubating in primary antibody ( rabbit anti-SP 1:6500 Immunostar #20064 , RRID: AB_572266 ) in PBST overnight at room temperature . Sections were then washed with PBST three times , incubated in 2% normal donkey serum in PBS for 10 min , and incubated for 2 hr in secondary antibody in PBS ( Alexa Fluor 594 donkey anti-rabbit 1:500 Thermo Fisher #A21207 , RRID:AB_141637 ) . Sections were then washed with PBS three times , mounted on coated glass slides in PBS , air-dried , coverslipped with Vectashield mounting medium with DAPI , and imaged on a fluorescent microscope . The dorsoventral location of recording sites was determined by subtracting the distance driven between recording sessions from the final recording location . Units that were recorded during sessions when the recording sites were determined to be localized outside of the VP were excluded from analysis . | Sounds or other cues associated with receiving a reward can have a powerful effect on an individual’s behavior or emotions . For example , the sound of an ice cream truck might cause salivation and motivate an individual to stand in a long line . Cues may prompt specific actions necessary to receive a reward , for example , approaching the ice cream truck and paying to get an ice cream . This is called instrumental conditioning . Some cues predict reward delivery , without requiring a specific action . This is called Pavlovian conditioning . Pavlovian cues can still prompt actions , such as approaching the truck , even though the action is not required . But exactly what happens in the brain to generate these actions during the two types of learning , is unclear . Learning more about these reward-driven brain mechanisms might help scientists to develop better treatments for people with addiction or other conditions that involve compulsive reward-seeking behavior . Currently , scientists do not know enough about how the brain triggers this kind of behavior or how these processes lead to relapse in individuals who have been abstinent . Basic studies on the brain mechanisms that trigger reward-seeking behavior are needed . Now , Richard et al . show that a greater activity in neurons , or brain cells , in a part of the brain called the ventral pallidum predicts a faster response to a reward cue . In the experiments , some rats were trained to approach a certain location when they heard a particular sound in order to receive sugar water , a form of instrumental conditioning . Another group of rats underwent Pavlovian training and learned to expect sugar water every time they heard sound even if they did nothing . Both groups learned to approach the sugar water location when they heard the cue , despite the different training requirements . Richard et al . measured the activity of neurons in the ventral pallidum when the rats in the two groups heard the reward-associated sound . The experiments showed that the amount of activity in the brain cells in this area predicted whether a rat would approach the sugar-water delivery area and how quickly they would approach the reward after hearing the cue . The predictions were most reliable for rats that had to do something to get the sugar water . When Richard et al . reduced the activity in these cells they found the rats took longer to approach the reward source , but only when this action was required to receive sugar water . The experiments show that the ventral pallidum may provide the motivation to undertake reward-seeking behavior . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"neuroscience"
] | 2018 | Ventral pallidal encoding of reward-seeking behavior depends on the underlying associative structure |
Premature infants are highly vulnerable to aberrant gastrointestinal tract colonization , a process that may lead to diseases like necrotizing enterocolitis . Thus , spread of potential pathogens among hospitalized infants is of great concern . Here , we reconstructed hundreds of high-quality genomes of microorganisms that colonized co-hospitalized premature infants , assessed their metabolic potential , and tracked them over time to evaluate bacterial strain dispersal among infants . We compared microbial communities in infants who did and did not develop necrotizing enterocolitis . Surprisingly , while potentially pathogenic bacteria of the same species colonized many infants , our genome-resolved analysis revealed that strains colonizing each baby were typically distinct . In particular , no strain was common to all infants who developed necrotizing enterocolitis . The paucity of shared gut colonizers suggests the existence of significant barriers to the spread of bacteria among infants . Importantly , we demonstrate that strain-resolved comprehensive community analysis can be accomplished on potentially medically relevant time scales .
Infection by potentially pathogenic and antibiotic-resistant bacterial strains is a major source of disease in hospitalized patients . However , the spread of bacteria among patients is hard to track because most methods cannot distinguish between closely related strains . Strain transmission is especially important during colonization of newborns , a process that is critical for proper development ( Arrieta et al . , 2014 ) . Premature infants , in particular , are highly susceptible to aberrant colonization , as their microbiome is often disrupted by antibiotic treatments ( Greenwood et al . , 2014 ) and since the source of colonists likely includes the hospital environment ( Brooks et al . , 2014; Taft et al . , 2014 ) . Necrotizing enterocolitis ( NEC ) is a common and life-threatening gastrointestinal disease that primarily affects hospitalized premature infants . Recent data indicate that ∼7% of infants born weighing <1 . 5 kg develop NEC ( Neu and Walker , 2011 ) . Various observations support a microbial role in NEC , including the high incidence of pneumatosis intestinalis ( gas in the bowel wall ) in affected infants and resolution of symptoms in a majority of patients after antibiotic therapy and bowel rest ( Grave et al . , 2007; Morowitz et al . , 2010; Carlisle and Morowitz , 2013 ) . NEC is characterized by intestinal inflammation and commonly progresses to necrosis , sepsis , and death . Risk factors may include feeding with artificial infant formula , blood transfusion , infant genetics , and overall health status ( Mally et al . , 2006; Schnabl et al . , 2008; Neu and Walker , 2011; Wan-Huen et al . , 2013 ) . Such factors might be expected to give rise to a fairly constant disease incidence rate . However , NEC is commonly reported to occur in outbreaks ( Boccia et al . , 2001; Meinzen-Derr et al . , 2009 ) , suggesting involvement of a contagious microorganism . A review of 17 published outbreaks of NEC did not identify a reproducible pattern of bacterial infection ( Boccia et al . , 2001 ) . Cultivation-based approaches to identify and track medically relevant organisms can be labor intensive , biased , and inefficient . Yet , sequencing of the genomes of these cultured organisms can distinguish between strains with divergent phenotypes such as antibiotic susceptibility and virulence ( Didelot et al . , 2012 ) , and has enabled analysis of pathogen dispersal ( Chin et al . , 2011; Köser et al . , 2012; Snitkin et al . , 2012; He et al . , 2013 ) . An alternative approach uses 16S rRNA gene sequencing to identify organisms without cultivation ( Brooks et al . , 2014; Taft et al . , 2014 ) , but the taxonomic resolution is limited , and distinct strains cannot be differentiated or tracked . Nonetheless , the method has been used to compare gut bacterial populations in fecal samples from infants with and without NEC . The results have been inconclusive . Some studies have identified no differences between cases and controls ( Normann et al . , 2013 ) , while others have reported positive , but divergent findings ( Wang et al . , 2009; Mshvildadze et al . , 2010; Mai et al . , 2011; Claud et al . , 2013; Morrow et al . , 2013 ) . In contrast , whole community DNA sequencing methods ( metagenomics ) can profile microbial communities with strain resolution and probe the metabolic potential of community members ( Tyson et al . , 2004; Gill et al . , 2006; Kuczynski et al . , 2012 ) . Applied to series of samples , the approach can document shifts in community structure and identify responses to medical treatments , increasing age , altered diet , and changing health status . Unlike 16S rRNA gene surveys , metagenomics does not rely on previously established information ( e . g . conserved sequences that guide PCR-based rRNA-based detection ) and is less likely to miss community members ( e . g . organisms with unusual rRNA sequences , phage , and plasmids ) . Compared to cultivation-based methods , the metagenomic approach provides a relatively unbiased view of community composition and thus may be particularly helpful when an unknown microorganism is the cause of a disease ( Relman , 2011 ) . A year ago , the power of such an approach was demonstrated in a retrospective analysis of banked samples from patients affected during a 2011 outbreak of a severe diarrheal illness caused by Shiga-toxigenic Escherichia coli ( Loman et al . , 2013 ) . More recently , shotgun sequencing of bacterial DNA present within cerebrospinal fluid enabled the diagnosis and treatment of leptospirosis in a critically ill child with meningitis ( Wilson et al . , 2014 ) . Among gastrointestinal diseases with a possible microbial origin , NEC is somewhat unique as it is relatively common and because samples that provide information about gut consortia can be collected prior to the development of symptoms . This is because infants at risk for NEC are typically hospitalized for weeks to months in the neonatal intensive care unit ( NICU ) and onset of the disease occurs over a defined time period . Only one small study that we are aware of has analyzed metagenomic sequence data from infants with and without NEC ( Claud et al . , 2013 ) , but assembly of the sequences was not attempted . Recently , a group of infants developed NEC over a short time period in the NICU of Magee-Womens Hospital of the University of Pittsburgh Medical Center . Here , we investigated the degree to which specific microbial strains were shared among co-hospitalized infants and whether the disease could be attributed to a single infectious agent . Because the analysis required confirmation that the same strain was present in multiple infants , we deployed a genome-resolved sequencing-based approach . Our analyses included consideration of the fastest evolving features of genomes ( e . g . prophage and the CRISPR/Cas loci ) to maximize strain resolution . We also investigated strain-level metabolic potential and evaluated population heterogeneity for one abundant and widespread species . Genome-resolved approaches are typically slow and bioinformatics intensive because the data sizes are massive , simultaneous reconstruction of genomes for multiple community members is complex , and comparative and metabolic analyses for hundreds of genomes are challenging . We applied a new analysis system to resolve data into genomes and analyze the metabolic potential . To the best of our knowledge , this study is the first to provide comprehensive , genome-resolved analysis of gut bacterial communities in co-hospitalized patients . The core methods are fast enough to make them useful in some clinical settings , and we anticipate that analysis time can be substantially decreased with future developments .
When it became apparent that the incidence of NEC was increasing , we selected five infants who had developed NEC and five controls for comprehensive microbial community analysis . Ultimately , during the summer of 2014 , nine infants were diagnosed with NEC ( Bell's stage II or III ) . The total number of NEC cases was 10 , as one of the affected infants developed recurrent NEC . This incidence rate was 2 . 5 times higher than average for this NICU . For affected infants #2 ( who developed NEC twice ) , #3 , and #8 , multiple fecal samples collected prior to the onset of symptoms were available . Two additional infants , #9 ( not premature ) and #10 , were enrolled after diagnosis and treatment . The other infants who developed NEC were not enrolled in our study . Four infants ( #1 , #4 , #6 , and #7 ) did not develop NEC . Infant #5 was not diagnosed with NEC but had a single bloody stool on day of life ( DOL ) 20 and was treated with antibiotics for a suspected urinary tract infection . All infants were hospitalized concurrently within the same NICU ( i . e . synchronous controls ) , and several were matched also according to gestational age . The selection of samples for sequencing was aimed to provide dense sampling around diagnosed NEC cases , from both the diagnosed infants as well as co-hospitalized infants who did not develop NEC ( see sampling schedule in Figure 1 and additional medical details in Supplementary file 1 ) . Bacterial load in each sample was quantified by ddPCR ( see ‘Materials and methods’ section ) . The estimated load was in general agreement with previous measurements in full-term infants of similar postnatal ages ( De Leoz et al . , 2014 ) . Notably , the variation in the number of microbes per gram feces did not exceed a 100-fold across all samples . Infants who developed NEC did not show a consistent trend of change in bacterial load prior to or following diagnosis ( Figure 1 ) . 10 . 7554/eLife . 05477 . 003Figure 1 . Overview of the sampling of infants affected by necrotizing enterocolitis ( red ) and controls ( blue ) and microbial cell loads based on droplet digital PCR ( ddPCR ) quantification of fecal samples . For ddPCR , standard deviations for triplicates are plotted within each data point . Also shown are necrotizing enterocolitis ( NEC ) diagnosis times ( vertical red lines ) and periods of antibiotic administration: green: ampicillin + cefotaxime , orange: vancomycin + cefotaxime , and blue: ampicillin + gentamycin ( see Supplementary file 1 ) . Black boxes indicate metagenomic samples for which insufficient sample remained for ddPCR . EGA: estimated gestational age . DOI: http://dx . doi . org/10 . 7554/eLife . 05477 . 003 DNA was extracted from up to nine samples per infant and sequenced using an Illumina HiSeq2500 at the University of Illinois . Overall , we analyzed 55 samples from the 10 infants ( Figure 1 and Supplementary file 2 ) . Between 2 . 22 and 7 . 35 Gbp of trimmed data from each sample was assembled independently . This enabled at least 4× coverage for genomes of organisms that comprised more than ∼0 . 2% to ∼0 . 6% of each community . For the 10 datasets , we assembled 181 . 2 Gbp of read sequence information . In total , 1 . 35 Gbp of genome sequence was generated on scaffolds >1000 bp ( see ‘Materials and methods’ section and Supplementary files 2 and 3 ) . The genome reconstruction strategy involved a user assigning scaffolds to organisms using online binning tools ( see ‘Materials and methods’ section ) . Genome bins were defined based on a combination of a phylogenetic profile , GC content , and coverage . These bins were then verified independently using emergent self organizing maps ( ESOMs ) that clustered either tetranucleotide composition or time series abundance pattern information . Genome completeness and purity were evaluated based on the inventory of ribosomal proteins and 51 genes expected to be in single copy in any genome ( see ‘Materials and methods’ section for details ) . A total of 509 bacterial genomes ( including multiple genomes for the same organism in different samples ) were recovered , with average read coverage of between 2 and 1148 . Between 1 and 23 bacterial genome bins were detected per sample , and overall 260 near-complete genomes were reconstructed ( see ‘Materials and methods’ section ) . Scaffolds identified as putative phage or plasmids based on their encoded genes were assigned to 328 bins ( Supplementary file 3 ) . Overall , between 86% and 98% of reads generated for each sample was assigned to a genome bin ( Supplementary file 2 ) . In order to assess the extent of strain dispersal among the hospitalized infants , genome bins with >0 . 5 Mbp of sequence were compared by aligning the single copy genes sequences . When these were too fragmented for conclusive results , entire genome bins were aligned . Genome bins that were >98% identical across >90% of bin length were considered indistinguishable . Manual curation of assemblies was performed in some cases to eliminate disagreements due to scaffolding errors that are introduced occasionally during assembly ( see ‘Materials and methods’ section ) . Remarkably , very few bacterial strains occurred in more than one infant and no strain was shared by all infants who developed NEC ( Figure 2 ) . In contrast , and as could be expected , identical genotypes were almost always detected in samples from the same infant , providing reassurance regarding the validity of our methods ( Figure 2 ) . 10 . 7554/eLife . 05477 . 004Figure 2 . An overview of the distribution of 144 of the 149 tracked strains in the 55 samples from 10 infants ( five rare organisms were not included for space reasons ) . White boxes indicate that the strain was absent; shading intensity increases with increased organism abundance . Note the persistence of specific genotypes within infants and the almost complete lack of overlap in strains between infants . The few strains shared between infants are highlighted in red . Colors associated with organism names indicate the broader organism classification: green are Firmicutes , orange are Gammaproteobacteria , red are Epsilonproteobacteria , pink are Betaproteobacteria , and blue are Actinobacteria . Red lines indicate antibiotic administration associated with necrotizing enterocolitis diagnoses , blue lines indicate antibiotic administration for other reasons . DOI: http://dx . doi . org/10 . 7554/eLife . 05477 . 004 Specifically , of the 149 strains compared , only four were shared by two or more infants , and only three of these were identified in infants who developed NEC . A Klebsiella oxytoca strain was present in infants #1 and #6 , neither of whom developed NEC . Clostridium sporogenes was present in infants #3 and #5 , but occurred at very low abundance in infant #3 . Two strains were more widely distributed: a Clostridium butyricum strain was detected in infants who did ( infants #3 , #8 ) and did not ( infants #1 , #5 , #6 ) develop NEC but was missing from infant #2 , who developed NEC . C . butyricum has no predicted type III or type VI secretion system genes and no identified toxin-producing genes ( Supplementary file 4 ) . Thus , this strain seems unlikely to be a pathogen or the cause of NEC . Lastly , a Clostridium paraputrificum strain with a moderate predicted pathogenicity potential ( Supplementary file 4 ) was shared by infants #2 , #5 , and #8 , and also occurred in one sample from infant #6 , although the predominant strain in this infant ( who did not develop NEC ) was different ( Figure 2 ) . C . paraputrificum was not detected in infant #3 ( NEC case ) . Both C . paraputrificum and C . butyricum have been previously suggested as potential causative agents in NEC ( Waligora-Dupriet et al . , 2005 ) . Interestingly , although the colonizing strains were almost always distinct , infants often shared bacteria of the same genus or species . At high abundance in multiple infants , including two who developed NEC , were members of the genus Veillonella . However , multiple distinct strains and species were present across infants ( Figure 3 ) . A Veillonella strain was very abundant in infant #2 prior to development of NEC ( Supplementary file 3 ) but disappeared after the first antibiotic treatment , to be replaced by different Veillonella species ( Figure 2 ) . In the other infants who developed NEC , Veillonella was either absent ( infant #8 ) or present as a different strain altogether ( infant #3 ) . The results likely rule out a Veillonella strain as a single , shared agent of NEC . 10 . 7554/eLife . 05477 . 005Figure 3 . A phylogenetic tree ( RAXML; black dots indicate bootstrap values of ≥80% ) for predicted RuBisCO Form IV ( RuBisCO-like ) proteins involved in methionine salvage . This protein was chosen for analysis because it is well studied and is not one of the 51 single copy ( and generally highly conserved ) genes used in other analyses . Colored dots identify the infant , while the number indicates the sample of origin . Red boxes highlight infants who developed necrotizing enterocolitis ( NEC ) . Although Veillonella were prominent in many samples , sequence analysis revealed many distinct strains/species over the study cohort . Strain shifts occurred following antibiotic administration ( e . g . in infant #2 ) , but identical sequences were often detected in series of samples from the same infant . Note infants affected by NEC do not share the same strains/species . DOI: http://dx . doi . org/10 . 7554/eLife . 05477 . 005 Another organism found in multiple infants , often at high abundance ( Figure 2 and Supplementary file 3 ) , was Enterococcus faecalis . This organism is common in fecal samples from both premature and term infants ( Chang et al . , 2011; Costello et al . , 2013; Vallès et al . , 2014 ) . Interestingly , it appears that the host–E . faecalis relationship , as it pertains to the infant gut , is nuanced . Although E . faecalis has repeatedly been identified as a source of neonatal infection ( Stoll et al . , 2002; Härtel et al . , 2012 ) , it also has been studied as a potential probiotic ( Nueno-Palop and Narbad , 2011 ) , with beneficial properties related to modulation of innate immunity ( Wang et al . , 2014 ) . Furthermore , links between mobile genomic elements and enterococcal virulence are well described ( Gilmore et al . , 2013 ) . These considerations suggest that strain-level variation in E . faecalis is significant and potentially clinically relevant . For E . faecalis , we reconstructed 30 near-complete genomes ( Supplementary file 3 ) for multiple strains ( Figure 2 ) . Alignment of the longest of the single copy genes tracked , the ∼2500 bp DNA gyrase subunit A ( gyrA ) gene , illustrates five distinct sequence types for this gene alone ( Figure 4A ) . Strains recovered from infants #2 and #7 and also strains recovered from infants #3 and #5 ( early samples ) could not be distinguished by this locus . Notably , reconstructed 16S rRNA gene sequences were identical in these strains , illustrating that the limited taxonomic resolution of this locus prevents its use in studies of strain dispersal . 10 . 7554/eLife . 05477 . 006Figure 4 . Strain differences in recovered Enterococcus faecalis genomes . ( A ) Alignment of the ∼2500 Enterococcus faecalis gyrA nucleotide sequences from all infants to that from infant #3 , sample 1 revealing five distinct types ( gray bars are scaffolds; SNPs are vertical black lines ) . Shown below are a tiny subset of reads from infant #3 , sample 4 with SNPs that match nucleotides in the gyrA sequences from E . faecalis in another infant; all SNPs are consistent with a strain very similar to that in infants #2 and #7 ( although derivation of some reads from other strains cannot be ruled out ) . ( B ) Phylogenetic representation illustrating two distinct Cas1 sequence types . ( C ) Inventory of 51 single copy genes showing that the 30 E . faecalis genomes are near-complete and providing information about encoded CRISPR and Cas . DOI: http://dx . doi . org/10 . 7554/eLife . 05477 . 00610 . 7554/eLife . 05477 . 007Figure 4—figure supplement 1 . Alignments showing single nucleotide polymorphisms ( vertical colored lines on gray bars that represent the sequences ) in the Histidyl-tRNA synthetase genes that distinguish from Enterobacter cloacae strains across samples and infants . Small numbers to the left of each gray bar indicate the samples of origin . Dashed black lines separate samples from before and after antibiotic administration in infants #2 , #5 , and #7 . Note the presence of different ( although often closely related ) strains in different infants and the presence of two distinct Enterobacter cloacae genotypes in most infants . Also note the persistence of strains in infants #2 and #5 , through antibiotic administration . DOI: http://dx . doi . org/10 . 7554/eLife . 05477 . 00710 . 7554/eLife . 05477 . 008Figure 4—figure supplement 2 . Aspartyl-tRNA synthetase from Klebsiella pneumoniae strains in samples from infants #4 , #5 , #6 , #7 , and #8 . Note the strain switch in Klebsiella pneumoniae following treatment of infant #5 . DOI: http://dx . doi . org/10 . 7554/eLife . 05477 . 008 A genomic region of interest for strain-level studies is the CRISPR/Cas locus . This locus can be one of the fastest evolving regions of bacterial genomes and thus can potentially provide high-resolution insight into strain distinction , as well as shared ancestry ( Tyson and Banfield , 2008 ) . All 30 well-sampled E . faecalis genomes encode a CRISPR spacer-repeat array that lacks proximal Cas proteins and some genomes ( in infants #3 , #5 , #8 ) encode an additional locus with proximal Cas genes ( Figure 4B , C ) . Given that Cas proteins are required for CRISPR-Cas function , strains in infants #2 , #6 , #7 , and #9 that lack Cas proteins altogether , have lost CRISPR-Cas-based phage immunity . This pattern of loci with and without Cas proteins has been reported previously in E . faecalis ( Palmer and Gilmore , 2010 ) . Different Cas1 sequences ( types a and b ) and a different repeat sequence were identified in E . faecalis from infants #3 , #5 before antibiotic treatment , and #8 , compared to the strain in infant #5 after antibiotic treatment ( Figure 4B ) . The repeat-spacer arrays in the loci with Cas1 type a are identical in the genotypes of E . faecalis in infant #3 and in early samples from infant #5 ( Figure 5A ) , reinforcing the very high similarity of these populations deduced from single copy gene sequence comparisons ( Figure 2 ) . As often happens in CRISPR loci ( Tyson and Banfield , 2008 ) , a block comprising six spacers and flanking repeats has been excised in the strain from infant #8 and three novel spacers have been added at the growing tip , versus two in infants #3 and #5 ( Figure 5A ) . Shared spacers at the older end ( distant from the Cas ) imply that the strains in infants #3 , #5 , and #8 had a recent common ancestor . 10 . 7554/eLife . 05477 . 009Figure 5 . Comparison of CRISPR loci in Enterococcus faecalis genomes . ( A ) The CRISPR-Cas loci in infants #3 , #5 ( early strain ) , and #8 and ( B ) the CRISPR locus lacking adjacent Cas proteins . The first defective repeats are shown in blue , other repeats are in green . The CRISPR loci are expanded below . In A , two versus three spacers have been added to the young end of the loci ( left side , adjacent to Cas ) in infants #3 , #5 versus #8 , respectively . In B , scaffolds encoding the loci are shown as horizontal gray bars ( polymorphisms in the multi-sequence alignment are small vertical tic marks ) . The same color indicates shared sequences . Blue boxes to the left indicate that the genome encodes Cas proteins . Both loci ( A and B ) are identical in infants #3 and #5 . DOI: http://dx . doi . org/10 . 7554/eLife . 05477 . 009 The Cas-less CRISPR array is flanked by DNA-related and antibiotic resistance-related genes , and differs in length considerably among strains . The repeat for the locus without Cas proteins is identical to that of the loci with type a Cas1 . All first repeats are defective , but the polymorphisms are only shared by strains in infants #3 , #5 , and #8 . However , the repeat-spacer array distinguishes the genotype in infants #3 and #5 from that in infant #8 ( Figure 5B ) . The loci in the strains in infants #2 and #7 are probably the same ( the sequence from infant #7 is not shown due to very partial recovery ) . Both the single copy gene and CRISPR-Cas analysis suggested that E . faecalis in infants #2 and #7 are very closely related . Similarly , the strains in infant #3 and in early samples from infant #5 are almost identical ( a single SNP in the surveyed gene set distinguished the sequences ) . To gain better understanding of the type and extent of genomic differences between the recovered E . faecalis genomes , and specifically of these closely related genome pairs , we mapped reads from eight samples , representative of the eight different genotypes reported in Figure 2 , to a 1 Mbp E . faecalis scaffold recovered from infant #9 , sample 1 ( one third of the recovered genome ) . Multiple alignment of the consensus sequences from mapping of each sample provided a view of sequence variability across strains ( Figure 6A; a similar alignment for C . paraputrificum strains is shown in Figure 6—figure supplement 1 ) . The analysis revealed many SNP locations and small indels that were spread across the entire length of the sequence , as well as a small number of longer ( 20–30 Kbp ) indel regions . These regions included among other things a sucrose metabolism operon , mobile elements , and genes related to Fe-S protein biogenesis . 10 . 7554/eLife . 05477 . 010Figure 6 . Alignment view of genome-wide differences in Enterococcus faecalis strains . Consensus sequence for the alignments ( shown at the top of each alignment ) represents the calculated order of the most frequent nucleotide residues . Alignments were done in Geneious v7 . 1 . 7 ( Kearse et al . , 2012 ) , using MAFFT v7 . 017 ( Katoh et al . , 2002 ) with default parameters . Samples are ordered by similarity . For each sample , SNPs and indel locations relative to the multiple alignment are marked by black lines or boxes . ( A ) Reads from eight samples , from which different Enterococcus faecalis strains were recovered , were mapped to a 1 Mbp E . faecalis scaffold ( scaffold 0 ) recovered from infant #9 , sample 1 . Shown is a multiple alignment of the consensus sequences derived for each sample from these mappings . Multiple SNPs and short indels are detected throughout the sequence . Several larger indels are also detected . ( B ) Enlarged view of a region in A showing a large indel locus . This view distinguishes sets of extremely closely related strains ( i . e . strains in infants #7 and #2; strains in infants #3 and #5 [early samples] ) from more distant strains . ( C ) Pairwise alignment of consensus sequences derived from read mapping to an E . faecalis scaffold ( scaffold 2962 ) recovered from infant #5 , sample 2 distinguishes closely related strains in infants #3 and #5 ( early samples ) . ( D ) Pairwise alignment of consensus sequences derived from read mapping to an E . faecalis scaffold ( scaffold 17 ) recovered from infant #7 , sample 3 distinguishes closely related strains in infants #2 and #7 . The region missing in the assembly from the other infants corresponds to a mobile element . DOI: http://dx . doi . org/10 . 7554/eLife . 05477 . 01010 . 7554/eLife . 05477 . 011Figure 6—figure supplement 1 . Alignment view of genome-wide differences in Clostridium paraputrificum strains . This organism is one of the very few for which a single strain ( found for example in infant #5 , sample 1 shown here ) was detected in multiple infants . Consensus sequence for the alignments ( shown at the top of each alignment ) represents the calculated order of the most frequent nucleotide residues . Alignments were done in Geneious v7 . 1 . 7 ( Kearse et al . , 2012 ) , using MAFFT v7 . 017 ( Katoh et al . , 2002 ) with default parameters . Samples are ordered by similarity . For each sample , SNPs and indel locations relative to the multiple alignment are marked by black lines or boxes . Reads from three samples , from which different Clostridium paraputrificum strains were recovered ( infant #5 samples 1 and 7 , infant #6 sample 3 ) were mapped to three 100–200 Kbp scaffolds ( infant #5 sample 1 , scaffolds 3 and 8 , and infant #5 sample 7 , scaffold 41 ) . Shown is a multiple alignment of the consensus sequences derived for each sample from these mappings . The different strains are very closely related , yet multiple SNPs and short indels are detected throughout the sequence . Large indels shown in the bottom panels are both associated with mobile elements . DOI: http://dx . doi . org/10 . 7554/eLife . 05477 . 011 The high degree of similarity between the closely related genotypes ( infants #2 and #7 , infants #3 , and early samples of infant #5 ) is evident from inspection of the multiple alignment ( Figure 6B ) . While , on average , sequence pairs were ∼90% identical across the scaffold length , those of infants #2 and #7 and those of infants #3 and #5 ( early samples ) showed identity of 99 . 3% and 98 . 7% , respectively . However , a more comprehensive comparison of these genome pairs revealed differences in the genotypes in both cases . Interestingly , a ∼1 Kbp region distinguishes strains in infants #3 and #5 ( Figure 6C ) , whereas a prophage insertion separates those in infants #2 and #7 ( Figure 6D ) . Both of these regions were missing from the genome recovered from infant #9 , sample 1 and thus could not be discovered by read mapping to that genome . These differences are very subtle , and especially in the case of the prophage , might have arisen after colonization . While the above analysis characterized differences in the major strains detected in the hospitalized infants , we further investigated whether E . faecalis strains from one infant occur at low or even trace levels in other infants . This was done by analysis of sequence polymorphisms in reads that map to the gyrA gene in each assembly . While in all infants colonized by E . faecalis , some polymorphic locations could be identified , in most infants , polymorphisms matching strains of other infants were undetectable ( infants #2 , #6 , #8 , #9 ) . For the abundant population in sample 2 from infant #9 , analysis of >20 , 000 reads indicated that the maximum abundance level of a strain detected in another infant was <0 . 01% ( Supplementary file 5 ) . However , in infant #3 , ∼0 . 12% of reads have sequences consistent with derivation from the genotype in infants #2 and #7 , which are indistinguishable at this locus ( Figure 4A ) . The gyrA analysis indicated that this genome is even more prominent in early-collected samples from infant #5 ( 3–9% of reads; Supplementary file 5 ) . Thus , compared to other E . faecalis strains , the population in infants #2 and #7 is relatively widely distributed . The evidence from all infants is that multiple strains and species are present in the NICU . The intriguing pattern of mostly infant-specific E . faecalis could have arisen due to a very small number of colonizing E . faecalis cells . However , investigation of population-level sequence variation revealed some reads with shared polymorphisms that are not characteristic of the strains in the other infants . This , and the detection of the dominant strain from infants #2 and #7 in other infants indicate that multiple E . faecalis inoculation events occur during colonization . The lack of overlap in strain genotype could indicate the existence of infant-specific strain sources ( e . g . mothers ) , and barriers that prevent spread of those populations to other infants in the NICU . Even if dispersal occurs , the founding population may preclude establishment of later introduced populations . Alternatively , strains could be dispersing freely , and strain dominance could reflect strong selection in the gastrointestinal tract ( possibly imposed by microbial community context and/or human genetics ) leading to the establishment of a single , most adapted strain . Another model worth considering would involve stochastic acquisition of a strain from a set of populations that is so large that it is improbable that any two infants would initially acquire the same strain . As there is ongoing input of strains over the colonization period , the observation of one ( usually ) highly dominant population still suggests some barrier to establishment of other populations . Previous studies that were done at lower resolution than achieved in the current study have pointed to a high abundance of Proteobacteria as a factor in NEC development ( Wang et al . , 2009; Mai et al . , 2011; Torrazza et al . , 2013 ) . To see if that pattern applied in the current study , we collapsed our organism identifications to the phylum level ( Figure 7 ) . Proteobacteria are abundant in most infants , but Proteobacteria representation in the communities did not distinguish those infants who developed NEC from those that did not . In fact , the relative abundance of Proteobacteria declines in infant #2 prior to both NEC diagnoses . Abundances are generally low in infant #3 , and consistently high in infant #8 . 10 . 7554/eLife . 05477 . 012Figure 7 . Stacked bar plot of community composition across samples and infants after organism identifications were collapsed to the phylum level to allow comparison to prior studies . Red lines indicate necrotizing enterocolitis diagnoses . DOI: http://dx . doi . org/10 . 7554/eLife . 05477 . 012 Given that no single organism could be associated with all NEC cases , we considered the possibility that overall metabolic imbalance was a contributing factor . Owing to the high quality and completeness of many of the recovered genomes , we are able with this type of data , to go beyond strain identification and search for unique characteristics in the predicted gene content of colonizers of NEC-affected infants . However , clustering analyses of the detailed profiles of genomically encoded metabolic capacities ( Supplementary file 4 ) and inspection of individual patterns failed to distinguish the capacities of microbial communities in infants who did and did not develop NEC . Of course , other considerations , such as differences in gene expression levels , may play an important role in NEC , but cannot be studied with DNA sequence information . While no metabolic imbalance was found , many pathogens were detected in fecal samples of infants who developed NEC . For example , Enterobacter cloacae is abundant in infant #2 and is predicted to have many toxin and type VI secretion system genes and an extensive antibiotic resistance repertoire ( Supplementary file 4 ) . The resistance genes may explain why E . cloacae remained abundant after antibiotic treatment ( sequence analyses indicate that the same genotype persisted through the treatments; Figure 2 and Figure 4—figure supplement 1 ) . ESOMs used to validate the binning also provide an overview of the community composition , and in the case of infant #2 , also highlight the almost complete dominance of E . cloacae in response to antibiotics ( Figure 8 ) . Note that these maps do not reflect organism abundances , although very small areas can indicate genomes that were partially sampled due to low sequence coverage ( for abundance information see Supplementary file 3 and Figure 8—figure supplement 1 ) . 10 . 7554/eLife . 05477 . 013Figure 8 . Microbial community composition , community complexity , and an overview of binning for samples from infant #2 . The diagrams are unit repeats of a tetranucleotide emergent self organizing map; points coded to reflect the bin assignment of the scaffold verify the binning ( see ‘Materials and methods’ section ) . Vertical red lines separate samples before and after antibiotic administration to treat necrotizing enterocolitis ( NEC ) ( two instances ) . Organisms are listed primarily in order of abundance in the first sample . Note that , with the exception of the dominant member , Enterobacter cloacae , species representation changed dramatically following antibiotic administration . The Veillonella strain varied ( numbers differentiate areas that represent different populations ) . DOI: http://dx . doi . org/10 . 7554/eLife . 05477 . 01310 . 7554/eLife . 05477 . 014Figure 8—figure supplement 1 . Rank abundance curves describing the microbial community ( exclusive of phage and plasmids ) in infant #2 . Colors correspond with those used in emergent self organizing maps ( see Figure 8 and Supplementary file 3 ) . Details are available in Supplementary file 3 . NEC: necrotizing enterocolitis . DOI: http://dx . doi . org/10 . 7554/eLife . 05477 . 01410 . 7554/eLife . 05477 . 015Figure 8—figure supplement 2 . An overview of the microbial communities from infant #3 . The red line separates samples collected before and after antibiotic treatment for necrotizing enterocolitis . Shown are rank abundance curves for all samples and time series emergent self organizing maps for two samples , which were used to refine the binning ( see ‘Materials and methods’ section ) . Note the prominence of Veillonella parvula , Enterococcus faecalis , and Citrobacter KTE32 in samples prior to diagnosis , and the loss of Veillonella and other less abundant species following antibiotic administration . E . faecalis and Citrobacter KTE32 strains persist through treatment , but the Staphylococcus epidermidis-related strains before ( 7 ) and after treatment ( 7′ ) are distinct . DOL: day of life . DOI: http://dx . doi . org/10 . 7554/eLife . 05477 . 01510 . 7554/eLife . 05477 . 016Figure 8—figure supplement 3 . Overview community composition for infant #8 , who developed necrotizing enterocolitis 1 day after collection of the last sample . ( A ) Time series + GC content emergent self organizing maps ( ESOMs ) were used to fine-tune binning and provide an overview of community composition . Points in the ESOM are color coded to indicate genome bin , the name for which is given to the right . ( B ) Time series abundance patterns for the relatively well-sampled bacteria; brown shading over numbers indicates sample pairs collected on the same day . The communities were dominated by bacteria closely related to Enterobacter cloacae ( yellow ) and Klebsiella pneumoniae ( brown ) . ( C ) Expanded view of the low abundance part of B . Several organisms were present at low abundance; some appeared a few days prior to the necrotizing enterocolitis diagnosis . Clostridium was detected but the genome sampling was so low that it was not included in the figure ( see Supplementary file 3 ) . DOL: day of life . DOI: http://dx . doi . org/10 . 7554/eLife . 05477 . 016 Many other potential pathogens are present prior to the NEC diagnoses in infant #2 ( see Figure 8 and Supplementary file 4 ) . Of interest due to their predicted gene complement are an Actinomyces strain that dominated the community prior to the first event and Clostridium difficile , which occurs in both samples ( at low abundance ) collected immediately prior to onset of NEC . C . difficile has been implicated as a cause of NEC ( Han et al . , 1983; Pérez-González et al . , 1996 ) , although its role is controversial ( Boccia et al . , 2001; Mshvildadze et al . , 2010 ) . Notably , the genome of the organism in infant #2 encodes tcdABCDE genes characteristic of toxigenic strains ( Dingle et al . , 2014 ) and Clostridial binary toxin B/anthrax toxin PA family proteins are affiliated with this organism . Also of potential significance in infant #2 , from the perspective of its genetic repertoire , were Clostridium perfringens , Streptococcus salivarius , Enterococcus faecium , and E . faecalis ( Supplementary file 4 ) . Prominent in the communities of infant #3 prior to NEC diagnosis were Veillonella parvula ( a strain predicted to have minimal pathogenicity , see Supplementary file 4 , and unique to this infant; Figure 3 ) , E . faecalis , K . oxytoca , and a Citrobacter related to strain KTE32 ( Figure 8—figure supplement 2 ) . The same strains of E . faecalis and Citrobacter persist through treatment , likely reflecting their large repertoire of antibiotic resistance genes ( Supplementary file 4 ) . Citrobacter and Klebsiella have many toxin and type VI secretion system genes , and thus may have contributed to disease in this infant . Pseudomonas aeruginosa was only detected after antibiotic treatment , and also has many predicted type III and toxin genes , as well as type VI secretion system genes ( Supplementary file 4 ) . Other potentially significant bacteria were strains of C . sporogenes and Paenibacillus . Interestingly , the communities in infant #3 included Bifidobacterium ( MSTE12-related ) , an organism that is often considered to be a beneficial commensal and not frequently observed in premature infants ( Butel et al . , 2007 ) . Infant #8 developed NEC 1 day after collection of the last sample . The communities in the two pairs of samples from different times on the same day ( Figure 8—figure supplement 3 ) contain generally similar organisms , but rapid abundance shifts occur , consistent with general observations over whole day periods . Especially prominent in samples from infant #8 were a Klebsiella pneumoniae-related strain ( Figure 4—figure supplement 2 ) and an E . cloacae ( Figure 4—figure supplement 1 ) strain . C . perfringens has a notable inventory of predicted pathogenicity-related genes . E . coli , present in the three samples collected prior to diagnosis , may have contributed to intestinal inflammation , given that it has a large inventory of type III and type VI secretion system genes and many toxin-encoding genes ( Supplementary file 4 ) . Samples from infants #9 and #10 ( both of whom developed NEC ) were collected only after diagnosis . Notable in the post-treatment communities from infant #9 were E . faecalis , Candida parapsilosis ( see below ) , and some Staphylococcus and Streptococcus . A variety of Lactobacilli and E . coli were prominent in infant #10 . In infants who developed NEC , the prominence of many potentially pathogenic organisms is striking . Although our results do not suggest that a single organism ( abundant or not ) caused NEC in the studied infants , bacteria that may have contributed to NEC were present . Infants who were not diagnosed with NEC were likewise colonized by a wide variety of potentially pathogenic bacteria and some strains were even shared by NEC cases and controls . If gut colonization by pathogenic bacteria is a significant factor in the development of NEC , other health and/or environmental attributes may ultimately determine which infants become sick . Due to the small number of cases studied to date and the large number of potentially important variables , a reliable model that predicts NEC development without over-fitting cannot be constructed at this point . However , accumulation of additional data may enable the construction of such a model in the future . The gastrointestinal tract can host a complex mixture of mobile elements , including phage , viruses , plasmids , and conjugative transposons , that can transfer virulence and antibiotic resistance factors ( Salyers et al . , 2004; Schjørring et al . , 2008; Minot et al . , 2011 ) . To consider the possibility that a mobile element , moving around the NICU ( potentially independently of the host bacterium ) , was the common factor leading to NEC , we compared all sequences from all samples that were binned as plasmid-like , phage or phage-like , or of unknown origin . We commonly found essentially identical sequences in different samples from the same infant and a few identical plasmids and phage were detected in different infants ( e . g . one complete , circular plasmid from infant #6 that is affiliated with a Clostridium species , based on sequence similarity ( Supplementary file 3 ) , also occurs in infants #5 and #8 ) . However , no mobile elements were shared by all sick infants , or by all infants diagnosed with NEC ( see ‘Materials and methods’ section ) . We leveraged the fact that some bacteria have CRISPR loci to determine whether bacteria colonizing the gastrointestinal tract of newborns have CRISPR-Cas-conferred immunity to co-occurring phage . This is important because phage sensitivity could explain rapid shifts in organism abundance . Our analyses focused on E . faecalis because it was abundant and widely distributed over the infant cohort . In no case did we identify an E . faecalis CRISPR spacer with a perfect match to any phage that coexisted in the same community ( Supplementary file 6 ) . However , we detected imperfect matches between E . faecalis CRISPR spacers and phage in the same sample , and some spacers matched perfectly to phage in other samples from the same infant and to phage in another infant . One spacer in the CRISPR locus of E . faecalis from infant #3 targets a prophage integrated into the genome of E . faecalis from infant #5 ( Supplementary file 6 ) . These observations indicate recent exposure of E . faecalis to phage populations related to those that coexisted in the NICU during the study period and suggest phage sensitivity of bacterial populations in this early gut colonization period . Notable in samples 2 and 3 from infant #1 were Enterobacteriales phage , the genomes of which were 95× and 30× more abundant than the genome of the probable Klebsiella host ( Supplementary file 3 ) . Interestingly , ddPCR shows that the period of phage proliferation corresponded to an increase in overall cell numbers by a factor of 10 over ∼5 days ( Figure 1 ) . K . oxytoca must account for this increase in bacterial cell numbers because it is essentially the only species in early-sampled communities . We infer that phage predation moderated the Klebsiella bloom and probably facilitated the subsequent establishment of the more complex community in later-collected samples . Our data provided a unique opportunity to study at high resolution possible factors that may have been responsible for the cluster of diagnosed NEC cases in the summer of 2014 . We were able to eliminate the possibility that a single bacterial strain was the causative agent in all cases , and also did not find any support for a causative role of specific mobile elements , or particular metabolic functions . In light of these findings , we turned to statistical characterization of the disease cluster . Surprisingly , despite the frequent reference to disease outbreaks in the literature , the statistical significance of disease clusters is rarely studied ( Turcios-Ruiz et al . , 2008; Meinzen-Derr et al . , 2009 ) . Here , we analyzed 67 months of monthly counts of NEC diagnoses in the NICU of Magee-Womens Hospital to determine whether the apparent outbreak in the summer of 2014 was a statistically significant anomaly ( Figure 9A ) . Monthly statistics are collected for other purposes and are based on different criteria for NEC , as outlined by the Vermont Oxford Network ( VON ) . Infant #3 , and another infant not enrolled in our study , were excluded due to lack of pneumatosis or pneumoperitoneum on X-rays . 10 . 7554/eLife . 05477 . 017Figure 9 . Statistical evaluation of the clustering of necrotizing enterocolitis cases during 2009–2014 . ( A ) The number of diagnosed necrotizing enterocolitis ( NEC ) cases meeting the stringent Vermont Oxford Network ( VON ) criteria over 67 months . Gray shading highlights the studied period . ( B ) Observed frequency of each value of monthly NEC cases in collected data ( blue ) ; expected frequency from a Poisson ( red ) and negative binomial ( NB; green ) distributions that were fit to the observed data using maximum likelihood parameter estimation ( Poisson: λ = 1 . 90 , NB: r = 5 . 81 , p = 0 . 75 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 05477 . 017 No seasonal or otherwise periodic patterns in NEC diagnoses were observed , and no correlation between the number of NEC cases and daily average of hospitalized infants in the NICU was detected . Data were modeled using Poisson and negative binomial ( NB ) distributions and maximum likelihood estimates of the corresponding parameters were extracted ( Poisson: λ = 1 . 90 , NB: r = 5 . 81 , p = 0 . 75 ) . Data were somewhat over-dispersed relative to the Poisson distribution ( Figure 9B ) and fit the negative binomial distribution modestly better ( R2 = 0 . 85 for Poisson model , R2 = 0 . 95 for NB model ) , in line with a potential dependency between diagnosed cases . While the eight cases from summer 2014 that met VON criteria are undoubtedly at the high end of the scale , they could not be established as statistically significant , assuming these underlying Poisson or NB distributions . Inclusion of additional sick infants who do not meet the VON criteria could change the conclusion , but unfortunately monthly statistics for all diagnosed cases were unavailable . Results were essentially unchanged when data were normalized to the average daily NICU occupancy . Future study of clusters of NEC events should be tested for their statistical significance to evaluate whether consideration of a single infective agent is appropriate . The approach used here can , in a cost- and time-effective manner , generate very good draft genomes . For example , two near-complete Veillonella genomes ( >4 Mbp ) were assembled into 12 and 15 pieces , four E . faecalis genomes were reconstructed into 27–43 pieces , two Actinomyces into 21 and 24 pieces , one Negativicoccus succinicivorans genome into 12 pieces , and one Citrobacter strain genome into 23 pieces . Notably , some genomes reconstructed in this study represent organisms with no closely related sequenced relatives . For example , we achieved many near-complete genomes for bacteria related to Tissierella sp . LBN 295 ( for which only four partial gene sequences are available in NCBI ) . This bacterium is more distantly related to , but currently profiled as related to , Clostridium ultunense . We also reconstructed a draft genome for an organism that we infer is related to Peptococcus niger . Both the Tissierella and P . niger genomes have been further curated ( see ‘Materials and methods’ section ) . Also , we reconstructed hundred of genomes that , although similar to previously sequenced genomes , are different in potentially important ways ( e . g . in antibiotic resistance potential and pathogenicity factors ) . Interestingly , we reconstructed an ∼12 . 7 Mbp draft genome of C . parapsilosis , a microbial eukaryote ( fungus ) that was highly abundant in the gut of infant #9 after treatment for NEC ( Figure 10A ) . The genome shares >99% identity with the genome of the CDC317 isolate . Alignment of our genome with the CDC317 genome verified the overall accuracy of our assembly , a notable finding given that very few microbial eukaryote genomes have been recovered previously from metagenomic data ( Figure 10—figure supplement 1 ) . 10 . 7554/eLife . 05477 . 018Figure 10 . Medically important organisms were revealed by genome-resolved analyses . The emergent self organizing maps illustrate bin accuracy ( dashed boxes show the periodicity of the maps ) and rank abundance curves ( lower right ) indicate community structure . ( A ) Candida parapsilosis was present in infant #9 after treatment for necrotizing enterocolitis . Due to the large genome size , Candida parapsilosis accounts for the majority of DNA in this sample . ( B ) Streptococcus agalactiae ( also known as group B streptococcus , GBS ) was detected , albeit at low abundance , in infant #1 . It is likely that the GBS caused the septic episode . ( C ) Overview of the metabolic potential for two organisms showing very different inventories of type III , VI secretion system , toxin , and antibiotic resistance genes . DOI: http://dx . doi . org/10 . 7554/eLife . 05477 . 01810 . 7554/eLife . 05477 . 019Figure 10—figure supplement 1 . Mauve genome alignment ( Darling et al . , 2010 ) of the CDC317 Candida parapsilosis genome and the genome reconstructed in the current study from infant #9 , sample 1 showing overall synteny and high sequence identity . DOI: http://dx . doi . org/10 . 7554/eLife . 05477 . 019 The ability of our approach to uncover organisms of clear medical relevance ( and not detected otherwise ) is also illustrated for infant #1 , who developed early onset sepsis . Although present at very low abundance ( ∼0 . 2% ) , we identified an organism whose 16S rRNA gene sequence shares 99% identity with Streptococcus agalactiae ( group B streptococcus , GBS ) , in samples 4 and 5 ( Figure 10B and Supplementary file 3 ) . The infant's mother tested positive on her GBS surveillance swab and the placenta was found by pathologists to contain trace amounts of GBS . However , the newborn blood culture was negative for GBS . Likely , this organism caused the episode of early onset sepsis in this infant . The ability to profile the genome for clinically relevant traits such as pathogenicity and antibiotic resistance is illustrated for K . oxytoca from infant #1 ( Figure 10C ) . The low predicted pathogenicity potential of C . butyricum is shown for contrast ( Figure 10C and Supplementary file 4 ) . The time between collection of the last fecal sample and receipt of DNA sequence information from all 55 samples was 10 days . For a group of three samples , as might be collected in a clinical setting , the time required for DNA extraction is a couple of hours and for sequencing , about 2 days ( the sequencing speed is method dependent ) . Sequence assembly , annotation , and data import into our metagenomics analysis system required 6–8 hr/sample and the time required for a first round binning that largely defined community composition and metabolic potential ( generated automatically ) was ∼10 to ∼60 min/sample . ESOM-based bin confirmation ( which may not be needed for some applications ) requires 1–2 hr/sample . Thus , new bioinformatics approaches tested here enable comprehensive genome-based analyses on a timescale that approaches that required for some clinical applications , for example involving patients with long-term health issues . The analysis time will decrease as sequencing technologies continue to improve and with automation of binning steps . This will make strain-resolved analysis clinically relevant for a wider range of applications , potentially including acute illnesses like NEC . We applied newly developed methods to rapidly and extensively resolve into genomes , sequence data from gastrointestinal tract-associated microbial communities from premature infants . All microbial communities in all infants sampled prior to onset of NEC harbored organisms with significant pathogenicity potential . However , we found no evidence for one common , abundant ( or even minor ) genomically distinct infective agent . If bacteria contribute to NEC , the effect more likely is due to exposure to a variety of potentially dangerous hospital-associated bacteria . A major finding is that the dominant population of each bacterium acquired by each infant was generally genotypically distinct . Extremely closely related organisms were only identified in a handful of cases , and occurred in both sick and healthy infants . Yet , the fact that they were identified at all , and the detection of shared minor strains of E . faecalis in a few cases , confirms that dispersal can occur among infants in the NICU . Overall , we suspect the existence of significant barriers that limit establishment of strains during the early stages of colonization of the premature infant gastrointestinal tract .
Fecal samples for enrolled infants were collected as available . From the pool of all available samples , we selected for sequencing 55 samples from which sufficient amounts of DNA were extracted . Our selection of samples from infants who did and did not develop NEC was aimed to provide dense sampling around dates leading to and following cases of NEC diagnosis in the NICU . The sampling schedule is shown in Figure 1 , and additional medical details for all infants are provided in Supplementary file 1 . The study was performed with approval from the University of Pittsburgh Institutional Review Board under protocol number PRO10090089 , and written parental consent was obtained on behalf of the neonates . Sequencing reads of 160 bp in length were processed with Sickle ( Joshi and Fass , 2011 ) ( v1 . 33; available at https://github . com/najoshi/sickle ) to trim both ends to remove low quality base calls . After trimming , reads were assembled with idba_ud ( Peng et al . , 2012 ) ( v1 . 1 . 1; available at http://i . cs . hku . hk/∼alse/hkubrg/projects/idba_ud/ ) using default settings . Resulting scaffolds >1000 bp were annotated . We used prodigal ( Hyatt et al . , 2010 ) ( v2 . 60; available at https://github . com/hyattpd/Prodigal/releases/tag/v2 . 60 ) to predict genes using default settings for metagenomics gene prediction . Protein sequences were searched against KEGG ( Kanehisa et al . , 2014 ) ( KEGG FTP Release 2014-07-07; available at http://www . kegg . jp/kegg/download/ ) , UniRef100 ( release 2014_07; available at ftp://ftp . uniprot . org/pub/databases/uniprot/previous_releases/release-2014_07/ ) , and UniProt ( Leinonen et al . , 2004 ) ( same as UniRef ) using USEARCH ( Edgar , 2010 ) ( v7 . 0 . 1001; available at http://www . drive5 . com/ ) . Additionally , reciprocal best-blast hits were determined . All matches with bit scores greater than 60 were saved , and reciprocal best hits with a bit score greater than 300 were also cataloged . We identified 16S rRNA sequences using Infernal ( v1 . 1; available at http://infernal . janelia . org/ ) using default settings . The rRNA genes were predicted using Infernal ( Nawrocki and Eddy , 2013 ) and tRNAs using tRNAscan_SE ( Lowe and Eddy , 1997 ) ( v1 . 23-r2; available at http://lowelab . ucsc . edu/tRNAscan-SE/ ) . Scaffolds , gene predictions , and all associated annotations were uploaded to ggKbase . berkeley . edu for binning and analysis ( http://ggkbase . berkeley . edu/project_groups/necevent_samples ) . We estimated detection sensitivity for bacterial populations using the data size per sample and assuming a genome size of ∼3 Mbp . For example , the sample with the least amount of data was from infant #4 , sample 4 ( 2 Gbp ) . This amount of data would allow detection ( 4× coverage ) of an organism with a 3 Mbp genome that comprised 0 . 6% of the sample . Evaluation of genome completeness relied in part on the number of expected single copy genes that were identified per bin . A bin was classified as very good if the genome size was not vastly different from that of genomes of closely related organisms and most single copy genes were present in one , and only one copy . The accuracy of our genome completeness statistics was somewhat affected by genes that were split by scaffold ends . Partial genes ( <50% of the gene ) were not counted . As the accuracy of binning depended in part on the quality of the phylogenetic profile , we tested two approaches . First , we inventoried the best matches of proteins on each scaffold by comparison to the UniRef100 database . As this did not provide sufficient taxonomic resolution , we adopted a second approach in which the profiles were established by comparison to the much larger UniProt database . In both cases , the phylogenic classification required that ≥50% of predicted proteins on a fragment had shared affiliation at some taxonomic level . If ≥50% of predicted proteins had best matches to the same species in the database , that scaffold was profiled as that species . If ≥50% had best matches to the same genus ( but not the same species ) , the profile assigned was of that genus . This process continued , until each scaffold had been assigned a profile at some taxonomic level . Some scaffolds were assigned the profile ‘unknown’ because ≤50% of predicted proteins had hits to the same domain ( these scaffolds were often from viruses and plasmids ) . Binning was carried out via an online interface within ggKbase ( http://ggkbase . berkeley . edu/ ) . When using this interface , the user selects a group of genome fragments ( scaffolds ) based on a specific phylogenetic profile , and/or scaffold coverage and/or GC information . The amount of sequence information , the number of expected single copy genes , the number of ribosomal proteins , and bin coverage statistics are displayed for the selection . For human microbiome research , usually the choice of scaffolds is first based on phylogenetic profile and is then fine-tuned by selection of a specific subset of scaffolds based on their coverage and/or GC content . If the bin size and single copy gene inventory are appropriate , the group of scaffolds is then binned . Following one round of binning ( 10–60 min/sample ) , typically ∼1 Mb of sequence information per sample was left unassigned to any organism or phage/plasmid . Typically , the identity of genomically sampled organisms was determined based on overall sequence similarity to previously known genomes . In many cases we reconstructed partial or complete 16S rRNA genes and used this sequence information to inform organism classifications , although the presence of these genes in multiple copies often resulted in misbinning of small scaffolds encoding this gene ( see notes in Supplementary file 3 ) . An advantage of the presence of multiple copies of the 16S rRNA gene per genome is that it can allow us to detect populations that are otherwise at such low abundance that they would be invisible based on their overall genome coverage . The 16S rRNA scaffolds were the only parts of some very low abundance genomes detected for this reason . The correctness of the assignment of scaffolds to genomes was verified with emergent self organizing maps ( ESOMs ) , a clustering tool ( Ultsch and Moerchen , 2005 ) that was applied to scaffold tetranucleotide frequency information ( Dick et al . , 2009 ) . In most cases , data points assigned to the same bin clustered into clearly defined and generally strongly bounded regions of ESOMs , supporting the accuracy of the binning method . Some bin adjustments were made based on the ESOM analyses . When the approach described above was insufficient to resolve bins for closely related species/strains ( e . g . Enterobacteriales in infant #8 ) , we constructed ESOMs using patterns of abundance of the organisms over the time series of samples from an infant ( Sharon et al . , 2013 ) , in combination with GC content . This led from minor to substantial improvements in bin purity and completeness . Up to eight near-complete genomes ( ≥94% of expected single copy genes identified ) were reconstructed per sample , and 221 near-complete genomes were reconstructed over the study . This accounting under-represents the completeness of the analysis because the presence of multiple highly related Enterobacteriales genotypes in many samples resulted in partial and fragmented recovery of specific conserved ribosomal proteins . When including Enterobacteriales genomes of the expected size but lacking these specific ribosomal proteins , 260 near-complete genomes were reconstructed . Rank abundance curves were constructed based on coverage . For this analysis , coverage values were normalized to account for differences in data size per sample . Strain comparison was done for genomes with >0 . 5 Mbp of recovered sequences , and was mostly based on alignment of sequences for 51 predicted single copy genes , many of which were ribosomal proteins . For cases with inconclusive results , mostly due to highly fragmented or very partial genomes in which many of these genes were missing , entire genome bins were aligned . In a few cases , mostly when verification of very small differences was required , manual curation of results based on inspection of read mapping to the regions in question was performed to detect local mis-assemblies . Geneious software v7 . 1 . 7 ( Kearse et al . , 2012 ) was used to align individual single copy gene sequences and for manual curation of the CRISPR loci . We used the online CRISPR spacer and repeat finder tools to recover spacer and repeat sequences ( http://crispr . u-psud . fr ) . Scaffolds longer than 5000 bp that were unbinned or were binned as plasmid , phage , or mobile elements , were extracted and aligned to each other ( using nucmer [Delcher et al . , 2002] ) . Scaffolds that were 99% identical across 90% of their length were considered closely related . An overview of the metabolic potential associated with genomes reconstructed in this study was established by searching the functional predictions for specific annotation terms . The number of genes that have the selected annotation terms is displayed in a table format in which rows are genomes and columns list the number of genes in each category ( see Supplementary file 4 ) . The search and exclusion terms for each functional category can be found via the ggKbase list function . Genomes of C . parapsilosis ( infant #9 ) , a species related to C . ultunense ( infant #3 ) , a Clostridiales from infant #6 , a V . parvula-related strain ( infant #3 ) , an Actinomyces species ( infant #4 ) , and a N . succinicivorans strain ( infant #5 ) were chosen for curation because they were significant and/or of very good draft quality . The curation used programs for correcting mis-assemblies and improving assemblies ( Sharon et al . , in preparation ) , which were identified through read mapping as follows . First , all reads were mapped to the genomes using bowtie2 ( http://www . nature . com/nmeth/journal/v9/n4/full/nmeth . 1923 . html ) with the --sensitive option . Next , short deletions ( which we found to be common in idba-ud assemblies ) in the assembled sequences were identified based on the read mappings . Last , all regions on the genomes with exceptionally low coverage were checked by collecting reads that map to those regions and their mate pairs and re-assembling them . Improvement of assemblies was achieved through read-mapping based identification of scaffolds that could be elongated or connected to other scaffolds . Both elongations and connections were achieved through local assembly of reads that were mapped to the analyzed regions and their mate pairs . For the Candida genome , our pipeline corrected 106 mis-assemblies ( about one mis-assembly for every 120 Kbp ) and reduced the number of scaffolds from 401 to 348 . To quantify bacterial load in infant fecal samples , ddPCR was performed on the Bio-Rad QX200 platform using EvaGreen-based chemistry ( Bio-Rad , Hercules , CA ) . A conserved , approximately 150 bp region flanking the V7 region of the 16S rRNA gene was targeted , as it has been successfully used in other probe-based qPCR assays in the past ( 1048f: GTGSTGCAYGGYYGTCGTCA , 1194r: ACGTCRTCCMCNCCTTCCTC [Ramirez-Farias et al . , 2009; Kennedy et al . , 2014] ) . Sample gDNA was diluted to 1:1000 and used as template in a PCR reaction consisting of 0 . 25 µl of 10 µM forward and reverse primer , 12 . 5 µl of 2× ddPCR EvaGreen Supermix ( Bio-Rad ) , and 12 µl of template , totaling 25 µl . This PCR mix was used to create droplets following the manufacture's instructions . Thermocycling parameters were: ( 1 ) 95°C for 10 min , ( 2 ) 95°C for 30 s , ( 3 ) 61°C for 30 s , ( 4 ) 72°C for 30 s , ( 5 ) 40 cycles ( go to steps 2–4 ×39 ) , ( 6 ) 98°C for 10 min , and ( 7 ) 12°C forever . All ramp rates were at 2 . 5°C/s . Each reaction was done in triplicate . Analysis of the ddPCR data was conducted with the QuantaSoft software package ( Bio-Rad ) and negative/positive thresholds set manually ( just above the negative population ) . To calculate cell density , the copies/µl output from QuantaSoft was normalized by grams of fecal mass used for each gDNA extraction reaction . To broadly correct for copy number , the assumption of four copies per bacteria was used ( Hospodsky et al . , 2012 ) . The sequence information can be accessed via NCBI , accession # SRP052967 . All metagenomic data associated with this study can be accessed via the ggKbase NECEvent project: http://ggkbase . berkeley . edu/project_groups/necevent_samples . Note that this is a ‘live data’ repository , so that errors found after publication will be corrected and more highly curated assemblies may be available . A snapshot of the published dataset is also available for download . | The spread of potentially harmful bacteria is a major source of disease in patients staying in hospitals . Premature babies—born before 37 weeks of pregnancy—can be particularly vulnerable to these infections because their organs may not yet be fully developed . Also , young babies do not have the fully established populations of beneficial microbes that help to protect us from dangerous bacteria . Necrotizing enterocolitis—a life-threatening disease that can cause portions of the bowel to die—is mostly seen in extremely premature babies . Although it is not known what causes this serious condition , research has suggested that a contagious microbe may be responsible . The development of methods that can sequence DNA from whole communities of microbes , known as metagenomics , allows researchers to identify the presence of individual strains of bacteria within these communities . This makes it possible to compare and contrast the strains of bacteria present in both diseased and healthy individuals , to help identify the bacteria responsible for a disease . Here , Raveh-Sadka et al . used a metagenomics approach to study the communities of microbes present in premature babies in a hospital unit during an outbreak of necrotizing enterocolitis . The study found that very few bacterial strains were present in more than one baby , suggesting that bacterial strains are not readily transferred between the babies while they are in the hospital . Furthermore , Raveh-Sadka et al . reveal that no single bacterial strain was shared among all the babies who developed necrotizing enterocolitis . These findings indicate that necrotizing enterocolitis is not caused by a single strain of bacterium . Instead , if bacteria do contribute to the disease , it maybe that it is caused by a variety of potentially harmful bacteria colonizing the gut at the cost of beneficial bacteria . In future , better understanding of the barriers that limit the transfer of bacteria between premature babies could help inform efforts to reduce the spread of infections between patients in hospitals . | [
"Abstract",
"Introduction",
"Results",
"and",
"discussion",
"Materials",
"and",
"methods"
] | [
"microbiology",
"and",
"infectious",
"disease"
] | 2015 | Gut bacteria are rarely shared by co-hospitalized premature infants, regardless of necrotizing enterocolitis development |
Although most Drosophila species lay eggs in overripe fruit , the agricultural pest Drosophila suzukii lays eggs in ripe fruit . We found that changes in bitter taste perception have accompanied this adaptation . We show that bitter-sensing mutants of Drosophila melanogaster undergo a shift in egg laying preference toward ripe fruit . D . suzukii has lost 20% of the bitter-sensing sensilla from the labellum , the major taste organ of the head . Physiological responses to various bitter compounds are lost . Responses to strawberry purées are lost from two classes of taste sensilla . Egg laying is not deterred by bitter compounds that deter other species . Profiling of labellar transcriptomes reveals reduced expression of several bitter Gr genes ( gustatory receptors ) . These findings support a model in which bitter compounds in early ripening stages deter egg laying in most Drosophila species , but a loss of bitter response contributes to the adaptation of D . suzukii to ripe fruit .
A major agricultural pest has recently emerged in dramatic fashion . Drosophila suzukii , endemic to Southeast Asia , invaded California in 2008 ( Hauser , 2011 ) . It moved rapidly across the United States and has now emerged in Northern Europe as well ( Asplen et al . , 2015; Cini et al . , 2012; Deprá et al . , 2014; Dos Santos et al . , 2017; Walsh et al . , 2011 ) . D . suzukii is a threat to a wide variety of fruit crops , including strawberries , blueberries , peaches , cherries , and grapes ( Burrack et al . , 2013; Lee et al . , 2011; Mazzi et al . , 2017 ) . Whereas related species such as Drosophila melanogaster lay eggs in fermenting fruit that is of no commercial value , D . suzukii lays eggs in ripe fruit , leading to larval infestations and microbial infections that destroy crops ( Karageorgi et al . , 2017; Lee et al . , 2011; Walsh et al . , 2011 ) . The adaptation of D . suzukii to ripe fruits has been accompanied by the evolution of a large serrated ovipositor , which penetrates the surface of ripe fruit and deposits an egg ( Atallah et al . , 2014; Green et al . , 2019 ) . A recent study showed that changes in the olfactory and mechanosensory systems contribute to the adaptation of D . suzukii to its new niche ( Karageorgi et al . , 2017 ) . The study also suggested the possibility that taste may play a role , a suggestion that we explore in the present study . Plants produce a vast diversity of toxins to defend themselves against insect attack ( Biere et al . , 2004; Frost et al . , 2008; Fürstenberg-Hägg et al . , 2013; Ibanez et al . , 2012; War et al . , 2012 ) . Many of these compounds are secondary metabolites that taste bitter to humans ( Dagan-Wiener et al . , 2017; Drewnowski and Gomez-Carneros , 2000; Keast et al . , 2003; Wiener et al . , 2012 ) . Insects in turn have evolved mechanisms for detecting these bitter compounds and avoiding them; these compounds thus deter feeding and oviposition ( Briscoe et al . , 2013; Chen et al . , 2019; Pontes et al . , 2014; Salloum et al . , 2011; Sellier et al . , 2011; Wada-Katsumata et al . , 2013 ) . Since levels of bitter compounds differ among various stages of fruit ripening , it seems plausible that the sensitivity of an insect to different bitter compounds may influence its choice of a ripening stage on which to lay eggs ( Batista-Silva et al . , 2018; Cheng and Breen , 1991; Taghadomi-Saberi et al . , 2018 ) . As a corollary , it seems conceivable that changes in bitter perception might contribute to the shift of oviposition preference in D . suzukii . Although there has been little , if any , previous analysis of the bitter taste system in D . suzukii , bitter taste in D . melanogaster has been studied in detail ( Delventhal and Carlson , 2016; Dweck and Carlson , 2020; Liman et al . , 2014; Ling et al . , 2014; Scott , 2018; Weiss et al . , 2011 ) . Bitter-sensing neurons are housed in taste sensilla in the labellum ( one of the mouthparts ) , the tarsal segments of the legs , and the pharynx ( Chen and Dahanukar , 2017; Delventhal and Carlson , 2016; Dweck and Carlson , 2020; Lee et al . , 2010; Lee et al . , 2015; Ling et al . , 2014; Marella et al . , 2006; Meunier et al . , 2003; Moon et al . , 2009; Poudel and Lee , 2016; Rimal et al . , 2020; Sang et al . , 2019; Weiss et al . , 2011 ) . Although there are several kinds of taste receptors , bitter responses depend largely on the Gr ( gustatory receptor ) family ( Clyne et al . , 2000; Joseph and Carlson , 2015; Liman et al . , 2014; Scott , 2018 ) . Many Gr genes have been found to be required for response to individual bitter compounds ( Dweck and Carlson , 2020; Lee et al . , 2010; Lee et al . , 2015; Moon et al . , 2009; Poudel and Lee , 2016; Rimal et al . , 2020; Sang et al . , 2019; Weiss et al . , 2011 ) . Moreover , expression of certain Gr genes in sugar-sensing neurons confers response to bitter compounds ( Dweck and Carlson , 2020; Shim et al . , 2015; Sung et al . , 2017 ) . Here , we analyze bitter taste and its role in the evolution of oviposition behavior in D . suzukii . First we measure the preferences of D . suzukii and related species ( Figure 1A ) for strawberries at a variety of ripening stages . We then show that a mutant of D . melanogaster with reduced bitter response has a shift in oviposition preference like that of D . suzukii . Anatomical analysis of D . suzukii shows that it has lost 20% of its bitter-sensing sensilla from the labellum . Physiological analysis of D . suzukii and its close relative Drosophila biarmipes reveals that the shift to ripe fruits has been accompanied by a loss of many bitter responses , including responses to individual bitter compounds and to strawberry purées . Likewise , D . suzukii lays eggs on substrates with bitter compounds that deter oviposition in D . melanogaster and D . biarmipes . Finally we characterize the labellar transcriptomes of all three species and find that D . suzukii has reduced expression of a number of bitter taste receptor genes . Taken together , these results provide an unprecedented view of how the bitter taste system of an invasive crop pest evolved in its shift to a new ecological niche .
In a natural environment , female fruit flies seeking an oviposition site often have a wide range of choices . A given plant may simultaneously bear fruit at stages ranging from green to ripe to overripe , with fermenting fruit on the ground underneath . To determine which stages are most and least preferred by D . suzukii we used a multiple-choice oviposition paradigm . We collected strawberries from a field in Connecticut , USA , and separated them into seven stages: white-green , mature first blush , light red , dark red , ripe , early fermented , and fermented . From fruit at each stage we generated a purée , from which we prepared an agar plate . We then tested a stock of D . suzukii that also originated from a field in Connecticut . Flies were allowed to choose oviposition sites in the dark ( Figure 1B ) . Whereas D . melanogaster laid the most eggs on the purée of the fermented stage of strawberry ( Stage 7 , Figure 1C ) , D . suzukii females laid the fewest eggs on this fermented stage ( Figure 1D ) . Rather , D . suzukii laid more eggs on the white-green and ripe stages ( Stages 1 and 5 ) . We also tested a third species , D . biarmipes , which is phylogenetically much closer to D . suzukii than to D . melanogaster ( Figure 1A ) , and did not find strong preferences ( Figure 1E ) . We note that this species laid a smaller number of eggs than the other two species in this experiment . These results from our multiple-choice paradigm confirm and extend previous studies ( Bernardi et al . , 2017; Karageorgi et al . , 2017; Lee et al . , 2011; Olazcuaga et al . , 2019; Shrader et al . , 2019 ) showing that D . suzukii has an oviposition preference for early maturation stages , including both ripe fruit and earlier ripening stages , unlike D . melanogaster and many other drosophilids . We asked whether taste plays a role in the oviposition differences between D . melanogaster and D . suzukii . For this purpose we tested the oviposition preference of D . melanogaster and D . suzukii for ripe and overripe strawberry in a two-choice assay ( Figure 2A ) . D . melanogaster preferred the overripe fruit , whereas D . suzukii preferred the ripe fruit , as expected ( Figure 2B ) . We then tested D . melanogaster mutant for Gr33a ( gustatory receptor ) , a receptor that is expressed in many taste neurons and is required for behavioral and physiological responses to many bitter tastants ( Dweck and Carlson , 2020; Moon et al . , 2009 ) . Surprisingly , Gr33a2 showed a shift in preference similar to that of D . suzukii ( Figure 2C; the genetic background control is w1118 Canton-S , p<0 . 0001 , n = 18 , Wilcoxon signed-rank test ) . We confirmed this shift with another allele , Gr33a3 , and a different source of strawberries ( Figure 2D , p<0 . 01 , n = 19–20 ) . These results support a role for bitter taste in the oviposition preference between early and late ripening stages . One hypothesis suggested by these results is that the adaptation of D . suzukii to ripe fruit has been accompanied by a loss of bitter responses . We next investigated the anatomical basis of taste in D . suzukii . We examined three organs that make direct contact with potential oviposition sites: the labellum , the legs , and the ovipositor . These organs all harbor sensilla that could differ in number , structure , or position from those in closely related Drosophila species with different oviposition preferences . We first examined the labellum , the main taste organ of the fly head , via scanning electron microscopy ( SEM ) . Three types of taste sensilla were identified: short ( S ) , intermediate ( I ) , and long ( L ) ( Figure 3A–D ) . S sensilla are present on the most medial region ( Figure 3A , white dots ) ; I sensilla are found more laterally ( Figure 3A , arrowheads ) ; L sensilla ( Figure 3A , arrows ) are located between S and I sensilla . Corresponding classes with similar distributions are found in D . melanogaster ( Shanbhag et al . , 2001; Stocker , 1994; Weiss et al . , 2011 ) . Taste sensilla in both species fall into two classes distinguishable by the morphology of their tips: straight ( Figure 3B ) and forked ( Figure 3C ) . In D . melanogaster , the straight tip and each prong of the forked tip have been shown to contain a terminal pore ( Nayak and Singh , 1983 ) . Two other sensilla lie near the periphery ( Figure 3A , asterisks ) in both species . They are ~17 μm long and taper to a fine tip with no pore , arguing against a function in taste . A striking difference in sensillum morphology was found between D . suzukii and D . melanogaster: sensilla in D . suzukii are much longer . S sensilla of D . suzukii are ~43–53 μm long compared to ~20–30 μm in D . melanogaster; I sensilla are ~57–63 μm vs . 30–40 μm; L sensilla are ~73–100 μm vs . ~ 40–50 μm . D . suzukii has fewer labellar sensilla . On each half-labellum of D . suzukii and D . biarmipes there are 27 , rather than 31 , sensilla as in D . melanogaster . The numbers of S sensilla and I sensilla are each reduced by two ( Figure 3D ) . Unlike D . melanogaster , the region between I0 and L7 sensilla lacks sensilla in both D . suzukii and D . biarmipes . The positions of the remaining S and I sensilla do not correspond precisely to those of D . melanogaster sensilla , but the overall spatial patterns are similar , providing an opportunity for a comparative analysis of their functions . Next we examined the 4th and 5th segments of the female foreleg in D . suzukii by light microscopy . We identified three putative taste sensilla on the 4th segment and four on the 5th segment ( Figure 3E ) . All of these sensilla , except f4c , are arranged in pairs , such that lateral sensilla have a symmetric counterpart on the medial surface of the leg . These taste sensilla are similar in morphology and position to those in D . melanogaster and D . biarmipes . We adopt the nomenclature used for D . melanogaster , for example , ‘f’ indicates ‘female , ’ and ‘4’ indicates the fourth tarsal segment ( Ling et al . , 2014; Meunier et al . , 2003; Zhang et al . , 2011; Zhang et al . , 2010 ) . Ovipositors have taste function in larger flies ( Merritt and Rice , 1984 ) . Although the ovipositor is often referred to as a taste organ in D . melanogaster ( Stocker , 1994 ) , there is little , if any , evidence to support a taste function in this species . The saw-like ovipositor of D . suzukii is larger and facilitates egg laying in ripening fruit that other drosophilid species cannot use ( Atallah et al . , 2014; Harris et al . , 2014; Lee et al . , 2011 ) . We hypothesized that it might have evolved a taste function lacking in D . melanogaster . We examined the D . suzukii ovipositor by SEM and identified four types of structures on each vaginal plate ( VP ) : trichoid sensilla ( TS ) , long bristles ( LB ) , thorn bristles type I ( TB1 ) , and thorn bristles type 2 ( TB2 ) ( Figure 3—figure supplement 1; structures described in legend; terminology from Hodgkin and Bryant , 1978; Lauge , 1982 ) . We did not observe a pore at the tip of any of these structures , suggesting that they do not function in taste . Thus , of the three D . suzukii organs that make contact with potential oviposition sites , the labellum and legs but not the ovipositor have a repertoire of sensilla whose morphology is characteristic of taste sensilla . We focused on them for a functional analysis . Since we had found that bitter taste contributes to the difference in oviposition preference between D . suzukii and D . melanogaster ( Figure 2 ) , we analyzed the coding of bitter taste in D . suzukii . Bitter taste is the interface between drosophilids and many plant secondary metabolites that are toxic to insects ( Briscoe et al . , 2013; Dweck and Carlson , 2020; Pentzold et al . , 2017; Weiss et al . , 2011 ) . A wide variety of insect species have undergone evolutionary shifts that allow them to specialize on particular plant hosts that are toxic to other species , thereby reducing competition ( Whiteman and Pierce , 2008 ) . To examine bitter taste coding in female D . suzukii , we systematically measured electrophysiological responses of all 27 labellar sensilla to a panel of 16 bitter compounds , that is , 432 sensillum-tastant combinations , in an analysis comprising >3100 recordings . The compounds are structurally diverse and include naturally occurring alkaloids , terpenoids , and phenolic compounds . They also include DEET ( N , N-Diethyl-meta-toluamide ) , the most widely used insect repellent worldwide ( Diaz , 2016 ) . We found that L sensilla of D . suzukii showed little or no response to any tested bitter compound ( Figure 4 ) . Two S sensilla , S3 and S7 , also showed little response to bitter compounds ( n < 10 spikes/s to all tastants ) . I sensilla responded to a subset of bitter compounds , and most S sensilla responded to different subsets . The strongest responses were from several S sensilla to escin ( ESC ) and aristolochic acid ( ARI ) , ~60 spikes/s in each case ( Figures 4 and 5A ) . Some bitter compounds , such as DEET and saponin ( SAP ) , elicited little or no response from any sensillum . How does bitter coding of D . suzukii compare to that in other species ? We carried out a comparable analysis in D . biarmipes , examining the same 432 sensillum-tastant combinations ( >2700 total recordings ) . We also took advantage of a dataset that was generated previously in our laboratory for D . melanogaster and that is comparable to those obtained with our current methods ( one-way ANOSIM test of distinguishability , R = 0 . 58 , p=0 . 19; Weiss et al . , 2011 , Dweck and Carlson , 2020 ) . We found that some basic organizational principles are conserved among all three species . All three show a paucity of bitter responses among L sensilla , and in all species there are two S sensilla that show little , if any , response to the bitter compounds ( Figure 4 and Figure 4—source data 1 ) . A number of S sensilla appeared more broadly tuned than I sensilla in each species . Different compounds elicited the strongest responses from different species: ESC and ARI in D . suzukii , ESC in D . biarmipes , and caffeine ( CAF ) , umbelliferone ( UMB ) , theophylline ( TPH ) , and SAP in D . melanogaster . Interestingly , the strongest responses to ESC in D . suzukii are from S1 , S4 , and S9; corresponding sensilla show similar responses in D . biarmipes ( S1 , S4 , and S9 ) , but in D . melanogaster none of the S sensilla show such strong responses to ESC ( Figures 4 and 5A ) . D . melanogaster differs markedly from the other two species in its strong responses of I sensilla , that is , the responses of I8 , I9 , and I10 to CAF , UMB , and TPH . These responses are virtually absent in D . suzukii and D . biarmipes , even at higher concentrations ( Figures 4 and 5A , Figure 4—figure supplement 1 , and Figure 4—source data 1 ) . D . suzukii differs from both D . melanogaster and D . biarmipes in having little or no response to DEET or SAP ( Figures 4 and 5C , D ) . By contrast , D . suzukii has evolved stronger responses to ARI than are observed in either of the other species ( Figure 4 ) . To determine the number of functional classes of sensilla on the labellum of D . suzukii , we performed a hierarchical cluster analysis . Sensilla fell into four functional classes ( Figure 4—figure supplement 2A ) . All L sensilla clustered together with two S sensilla ( ‘S-c’ sensilla ) to form a class that showed little or no response to any of the tested bitter compounds . The other three classes consisted uniformly of either S or I sensilla . We carried out a similar cluster analysis of D . biarmipes ( Figure 4—figure supplement 2B ) and then compared the results from both species to an earlier analysis of D . melanogaster ( Figure 4—figure supplement 2C ) . All three species have a cluster consisting of all L sensilla and two S sensilla . In each species the remaining S sensilla divide into two classes , which we will refer to as S-a and S-b , but the functional characteristics of these S classes vary across species . In D . suzukii , the S-a class contains four members and was broadly tuned , responding to 13 of the 16 tested bitter compounds with a mean spike frequency of >10 spikes/s . S-b contains three members and responded to only four compounds at >10 spikes/s . In D . biarmipes , S-a also contains four members and is broadly tuned . S-b contains three members and responded to only two bitter compounds with a response greater than 10 spikes/s . In D . melanogaster , S-a contains six S sensilla and S-b contains three . S-a and S-b are both broadly tuned , responding to 9 and 15 of the 16 bitter compounds , respectively , with a spike frequency >10 spikes/s . I sensilla all fall into a single class , I-a , in both D . suzukii and D . biarmipes . In D . melanogaster , the I sensilla fall into two classes , I-a and I-b , which respond to non-overlapping subsets of tastants . These results , taken together , reveal that functional classes of taste neurons and their tuning breadths expanded or contracted during the evolution of the three species . Having characterized labellar sensilla of the three species , we next asked whether there were functional differences among species that could contribute to their oviposition preferences . We measured electrophysiological responses of the entire ensemble of labellar sensilla of all three species to extracts of ripe and overripe strawberry . The response of S sensilla to ripe strawberry was low in all three species ( Figure 6A , left ) . However , total spike input was lower in D . suzukii and D . biarmipes than in D . melanogaster in both S-a and S-b ( Figure 6B , left; p<0 . 05 , one-way ANOVA followed by Tukey's multiple comparison test , n = 5 ) . The response to overripe strawberry also differed among species ( Figure 6A , right column ) . Whereas all S sensilla of D . melanogaster responded , there was little or no response of any S-a or S-b sensilla of D . suzukii . Specifically , the responses of D . melanogaster to overripe strawberry in S-a and S-b sensilla were 13 ± 0 . 6 spikes/s and 14 ± 0 . 5 spikes/s , respectively ( Figure 6—figure supplement 1A ) . In D . suzukii , the corresponding responses were 1 . 0 ± 0 . 2 spikes/s and 0 . 0 ± 0 spikes/s . Moreover , since D . suzukii has two fewer S-a sensilla than melanogaster , the difference in total spike input is even greater: 78 spikes/s compared to 4 spikes/s ( Figure 6B; note the scale in the left and right panels of Figure 6B are different; see also Figure 6—figure supplement 1C , D ) . Interestingly , the response of D . biarmipes to overripe strawberry is intermediate . S-b sensilla do not respond in D . biarmipes ( Figure 6A and Figure 6—figure supplement 1A ) ; S-a show a response , but lower than that of D . melanogaster . The total spike input is 40 spikes/s ( Figure 6B , right ) . Based on the spike amplitudes , the responses of S sensilla to overripe strawberry appeared to represent the activity of the bitter-sensitive neuron in these sensilla . As a test of this notion , we measured the response of S sensilla to overripe strawberry in D . melanogaster mutant for Gr33a . We found that the response was eliminated or severely reduced , in each of three S sensilla tested: S5 , which is an S-b sensillum , and S6 and S7 , which are of the S-a class ( Figure 6C , D ) . Response was reduced in each of two independently generated Gr33a alleles . The L class of sensilla do not contain bitter-sensing neurons , and the responses we have measured represent response to sugars , salts , and other compounds . L sensilla gave a greater response to ripe strawberry in D . suzukii and D . biarmipes than in D . melanogaster ( Figure 6—figure supplement 1A , B , p<0 . 05 , one-way ANOVA followed by Tukey's multiple comparison test , n = 5 , for both A and B ) . The I class of sensilla contain bitter-sensing neurons but we are unable to resolve their spikes from the spikes of neurons that respond to other compounds . I sensilla gave greater total input to ripe strawberry in D . suzukii than D . melanogaster as well ( Figure 6—figure supplement 1B , p<0 . 05 , one-way ANOVA followed by Tukey's multiple comparison test , n = 5 , for both A and B ) . Principal component analysis ( PCA ) showed that all three species are distinguishable based on their responses to extracts of ripe or overripe strawberry ( Figure 6—figure supplement 2A , B , ANOSIM based on Bray-Curtis similarity; R = 0 . 88 , p<0 . 0001 for ripe strawberry; R = 0 . 99 , p<0 . 0001 for overripe strawberry ) . Taken together , these results indicate that labellar taste response to extracts of ripe and overripe strawberry have changed in D . suzukii compared to the other species . We analyzed coding of bitter tastants in the tarsal segments of the female forelegs , focusing on the same panel of 16 bitter tastants and 7 sensilla of the two most distal segments of all three species , that is , 336 tastant-sensillum combinations in all . As in the labellum , different sensilla responded to different subsets of tastants , and different tastants elicited responses from different subsets of sensilla ( Figure 7A and Figure 7—source data 1 ) . Some sensilla such as f5s responded to a number of tastants in all species , whereas others such as f5a and f4b responded to none in any species . f5v responded to several bitter compounds in D . biarmipes , but not in D . suzukii or D . melanogaster ( Figure 7A , B ) . D . suzukii shows a striking loss of response to certain bitter compounds . Coumarin ( COU ) and DEET both elicit excitatory responses from f5b , f4s , and f4c in both D . melanogaster and D . biarmipes , but few , if any , excitatory responses ( n > 0 spikes/s ) from any sensilla in D . suzukii; interestingly , both tastants appear to inhibit f5s ( Figure 7A , C , D ) . A hierarchical cluster analysis based on the bitter responses elicited from these sensilla identified classes of bitter-sensing neurons and shows that they have been dynamic over evolutionary time . Specifically , the analysis identified three classes in D . suzukii and D . melanogaster; each class contains the same sensilla in these two species ( Figure 7—figure supplement 1 ) . The tarsal sensilla in D . biarmipes fall into five classes . In all three species , one class consists of sensilla that responded to none of the tested tastants . In D . melanogaster and D . suzukii , this class contains three sensilla , f5a , f5v , and f4b; in D . biarmipes , the class contains only two , f5a and f4b , as f5v has evolved a different profile and falls into a separate class . In all three species f5s is the sole member of a class . All three species have another class that includes f5b and f4s; in D . melanogaster and D . suzukii , this class includes f4c , but in D . biarmipes f4c has evolved a different response profile and is the unique member of another class . The oviposition preference shift observed in Gr33a mutants ( Figure 2 ) , the loss of bitter-sensing sensilla in D . suzukii ( Figure 3 ) , and the loss of response to certain bitter compounds in the D . suzukii labellum ( Figures 4 and 5B–D ) and tarsi ( Figure 7 ) together suggested the hypothesis that bitter compounds could play a role in the oviposition differences between species . We wondered if there were any bitter compounds in ripe fruit that deterred oviposition in D . melanogaster but not in D . suzukii . We assessed the egg laying behavior of D . suzukii , D . melanogaster , and D . biarmipes to the 16 bitter taste compounds using a two-choice oviposition assay ( Figure 8A ) , initially at 0 . 5 mM concentrations . D . melanogaster avoided laying eggs on COU , lobeline hydrochloride ( LOB ) , DEET , and denatonium benzoate ( DEN ) , and D . biarmipes avoided COU , LOB , DEN , and sparteine sulfate salt ( SPS ) . Remarkably , D . suzukii oviposition was not deterred by any of these bitter compounds ( Figure 8B ) . To confirm and extend our finding that D . suzukii lacks oviposition avoidance of the five bitter compounds that elicited responses from either of the other two species , we tested higher concentrations of all five compounds . D . suzukii again showed no deterrence at either of the higher concentrations of any tested compound ( Figure 8C; one-way ANOVA followed by Dunnett's multiple comparison test , p>0 . 05 ) . Together these results demonstrate that D . suzukii has lost oviposition deterrence to at least some bitter compounds that deter its close relatives . This behavioral difference may represent an adaptation that facilitates the ability of D . suzukii to lay eggs on earlier ripening stages . We wondered if there were differences in gene expression between the taste systems of D . suzukii and its relatives , perhaps even differences in the expression of bitter receptors . Since the most striking anatomical and physiological differences we had found were in the labellum , we profiled the labellar transcriptomes of the three species . We carried out high-throughput sequencing of polyadenylated labellar RNA samples and obtained a total of 100–130 million paired-end reads from each species , deriving from a total of three biological replicates in each case . As a test of the purity of our labellar RNA samples , we asked whether they contained transcripts from pharyngeal taste neurons , which are anatomically close to the labellar neurons ( Figure 9—figure supplement 1A ) . Ionotropic receptor ( IR ) gene expression in the labellum and pharynx has been characterized in D . melanogaster previously ( Chen and Dahanukar , 2017; Koh et al . , 2014; Sánchez-Alcañiz et al . , 2018 ) . In the D . melanogaster samples , we detected the expression of most labellar IRs ( Figure 9—figure supplement 1B , C , blue ) but none of the pharyngeal-specific IRs ( Figure 9—figure supplement 1B , C , red , Supplementary file 1 ) . The same pharyngeal-specific IRs were also absent from the D . suzukii and D . biarmipes samples , suggesting that our labellar RNA samples contain little , if any , pharyngeal RNA ( Supplementary file 2 ) . Similarly , nearly all Grs and odorant binding proteins ( Obps ) previously detected in the labellum via GAL4 drivers or microarrays ( Galindo and Smith , 2001; Jeong et al . , 2013; Koganezawa and Shimada , 2002; Sánchez-Gracia et al . , 2009; Weiss et al . , 2011; Yasukawa et al . , 2010 ) were also detected in our D . melanogaster transcriptome ( Figure 9—figure supplement 2A , B , blue ) . These included 24 Grs found previously ( Weiss et al . , 2011 ) to be expressed in bitter-sensing neurons ( Supplementary file 1 ) . Grs whose expression was clearly detected in the labellum by RNAseq also included eight sugar-sensitive Grs – Gr5a , Gr61a , Gr64a , Gr64b , Gr64c , Gr64d , Gr64e , and Gr64f – consistent with several earlier studies of their expression ( Dahanukar et al . , 2001; Dahanukar et al . , 2007; Jiao et al . , 2007 ) . We also identified labellar IRs , Grs , and Obps that had not previously been found to be expressed in the labellum ( Figure 9—figure supplement 1C , gray and Figure 9—figure supplement 2A , B gray; Supplementary file 1 ) . To compare the transcriptomes , we considered those genes for which an ortholog was annotated in all three species . Among such genes , more than 4500 showed a discrepancy in the coding sequence length across the three species orthologs . We inspected the read coverage of nearly a quarter of these genes; most appeared to be misannotated or truncated in the D . suzukii genome ( version 1 . 0 ) . We manually fixed the annotation of the genes inspected ( n ~ 1000 ) and excluded the other genes from the analysis ( ~3500 ) . Additionally , we expanded the set of D . suzukii genes by annotating 86 chemosensory-related genes that had been missing or misannotated ( <10% of all reannotated genes ) . Altogether we analyzed the labellar expression levels of more than 6000 genes . We detected transcripts from 4200 to 4500 genes in each species ( ≥10 Transcripts Per Million ( TPM ) ; Supplementary file 2 ) . The labellar transcription profile of D . suzukii is more closely related to that of D . biarmipes than that of D . melanogaster , as determined by a hierarchical cluster analysis ( Figure 9A ) . This finding is consistent with the phylogenetic relationship among these species ( Figure 1A ) . We analyzed the relationship among the transcriptomes by PCA , which confirmed that each species has a distinct transcriptome ( Figure 9B ) . The first component separates all three species ( Figure 9B ) . Intriguingly , the second component clearly separated D . suzukii from its relatives but showed unexpected similarity between D . melanogaster and D . biarmipes . Such separation is reminiscent of the difference between the ecological niche occupied by D . suzukii and those of other Drosophila species . We next performed a pairwise comparison between D . suzukii and D . melanogaster and between D . suzukii and D . biarmipes . We found 162 genes differentially expressed between D . suzukii and both of the other two species , as determined by the following conservative statistical criteria: |log2 Fold Change| > 2 , and adjusted p-value<0 . 01 across all of four different differential expression ( DE ) analysis pipelines ( Supplementary file 3 , Supplementary file 4; see Materials and methods ) . Of these 162 genes , 13% were associated with the GO term ‘sensory perception of chemical stimulus , ’ a fivefold enrichment compared to the set of all genes expressed in the labellum of any species ( adjusted p-value=2 . 99E-5 ) . Altogether , the results suggest a molecular basis for the evolutionary shift between D . suzukii and its relatives . Gr gene expression in the D . suzukii labellum showed a reduction compared to its D . melanogaster and D . biarmipes counterparts . Of 38 Grs whose expression was detected in this study , seven are expressed at levels fourfold lower or less in D . suzukii than in D . melanogaster ( Figure 9C , log2FC<-2 , adjusted p-value<0 . 01 , Supplementary files 1–4 ) . Interestingly , five of these , Gr8a , Gr22e , Gr22f , Gr32a , and Gr98d , have been found previously to be expressed in bitter-sensing neurons ( Weiss et al . , 2011 ) . There were 10 Grs expressed at levels fourfold lower or less in D . suzukii than in D . biarmipes , including four Grs expressed in bitter-sensing neurons , Gr22f , Gr39b , Gr47a , and Gr59b ( Figure 9D and Supplementary file 4 ) . By contrast , no Grs were expressed at levels fourfold higher in D . suzukii than in either of the other species . Gr22f is a particularly striking case . Its expression was detected in both D . melanogaster and D . biarmipes , but was undetectable in D . suzukii by RNAseq even with 50 million paired-reads for a sample . To confirm that Gr22f is virtually absent from the D . suzukii labellar transcriptome , we carried out RT-PCR experiments . Consistent with the RNAseq results , a Gr22f product was amplified by RT-PCR from D . melanogaster and D . biarmipes labellar RNA , but little , if any , product was observed from a D . suzukii preparation ( Figure 9—figure supplement 3A ) . We confirmed the severely reduced levels of Gr22f expression in D . suzukii by performing RT-PCR with three additional Gr22f primer sets ( Figure 9—figure supplement 3B ) . Interestingly , mutation of Gr22f in D . melanogaster reduces the response to DEN in the S-b sensilla ( Sung et al . , 2017 ) . This phenotype is reminiscent of the reduced response to DEN in the S-b sensilla of D . suzukii , relative to D . melanogaster . Perhaps an evolutionary loss of Gr22f receptor expression accounts for this loss of DEN responses in D . suzukii . A detailed genetic analysis of Gr22f in taste and oviposition behaviors of D . melanogaster could be highly informative . The IR co-receptor genes IR76b and IR25a were expressed at similar levels across the three species ( Supplementary file 3 , Supplementary file 4 , that is , they did not meet the statistical criteria ) . We note that the comparable expression of these genes , which are broadly expressed in taste neurons ( Sánchez-Alcañiz et al . , 2018 ) , as well as the comparable expression of the pan-neuronal genes elav and nsyb , argues against the possibility that the reduced expression of certain Grs in D . suzukii is a simple consequence of fewer neurons or more non-neuronal cells in the D . suzukii labellum . By contrast , four IR genes , IR11a , IR40a , IR60a , and IR76a fell below the detection level in D . suzukii and D . biarmipes but were readily detected in the labellum of D . melanogaster . Curiously , in all three replicates of the D . suzukii labellar transcriptome , IR21a was expressed more abundantly than any other IR , including the co-receptor genes . IR21a was expressed ~85 times more abundantly in D . suzukii than in D . melanogaster . In D . biarmipes , IR21a was the second most abundant IR , after the co-receptor IR76b . IR21a has been implicated in cool sensing , raising interesting questions about the regulation and function of this receptor ( Ni et al . , 2016 ) . Members of other chemosensory-related gene families are also differentially expressed ( Figure 9C , D ) . Unlike bitter receptor genes , however , the number of these other genes that are expressed at higher levels in D . suzukii is nearly identical to the number expressed at lower levels , when compared to either D . melanogaster or D . biarmipes . The differentially expressed genes include 73 members ( out of 136 detected ) of the Obp , chemosensory protein ( Che ) , pickpocket ( Ppk ) , cytochrome P450 enzyme ( Cyp ) , and glutathione S transferase ( Gst ) families ( Supplementary file 3 , Supplementary file 4 ) . Of these , 26 are differential expressed in D . suzukii compared to both its relatives . We speculate that some Cyps may contribute to the adaptation of D . suzukii by metabolizing toxic compounds in early ripening stages . We note finally an observation that may have significance for pest control: Cyp6g1 and Cyp12d1-p are more abundant in D . suzukii than in D . melanogaster ( 100- and 25-fold , respectively ) and D . biarmipes ( 5- and 20-fold , respectively ) . Overexpression of either gene in D . melanogaster increases resistance to insecticides , including Dichlorodiphenyltrichloroethane ( DDT ) ( Daborn et al . , 2007; Festucci-Buselli et al . , 2005 ) .
Early ripening stages of fruits differ in their physicochemical parameters from those of overripe stages ( Ménager et al . , 2004 ) . We have focused on plant secondary metabolites that taste bitter to humans and that are aversive and toxic to many insects ( Biere et al . , 2004; Dagan-Wiener et al . , 2017; Dweck and Carlson , 2020; Ibanez et al . , 2012; Lee et al . , 2010; Moon et al . , 2009; Pontes et al . , 2014; Poudel and Lee , 2016; Weiss et al . , 2011; Whiteman and Pierce , 2008; Wiener et al . , 2012 ) . The profiles of these metabolites are dynamic , changing as the fruit develops ( Cheng and Breen , 1991; Oikawa et al . , 2015 ) . For example , levels of flavonoids , many of which taste bitter to humans , decline as a function of developmental stage in strawberries ( Cheng and Breen , 1991 ) . We have found six lines of evidence supporting a model in which a loss of bitter responses in D . suzukii has contributed to its novel oviposition preference: Taken together these six lines of evidence support the notion that loss of bitter taste from D . suzukii contributes to its evolutionary shift in oviposition preference . We do not claim that the loss of bitter responses is the only gustatory change that facilitated the evolutionary transition of D . suzukii to oviposition on ripe fruit . Sugar responses , for example , may also have changed and may contribute to the transition , a possibility that deserves investigation . Nor is the gustatory system the only sensory system that has adapted in D . suzukii: the olfactory and mechanosensory systems have also adapted ( Karageorgi et al . , 2017 ) . However , our results suggest a major role for bitter sensation in the shift of D . suzukii to a new niche . It is striking that so much of the evolutionary plasticity we have found – anatomical , physiological , and molecular – is in the peripheral taste system , that is , in taste organs . A priori one might have imagined that taste organs could have retained their underlying molecular and cellular underpinnings , with the plasticity occurring exclusively in the central processing of taste input . In fact , a recent study found that evolution of Drosophila mating preferences emerged from evolution of a central circuit , with the peripheral detection mechanisms remaining conserved ( Seeholzer et al . , 2018 ) . Although we have found extensive changes in the peripheral taste system , we suspect there may also be changes in central circuit mechanisms . For example , we note that D . suzukii has lost oviposition response to LOB , to which it has retained much of its physiological response , suggesting the possibility of changes in central circuitry . It seems likely that evolution has operated at a variety of levels in the shift of D . suzukii to its new niche . D . biarmipes is much closer to D . suzukii phylogenetically than to D . melanogaster ( Figure 1A ) . However , D . biarmipes did not show the oviposition preference for early ripening stages observed for D . suzukii ( Figure 1C ) . Rather , D . biarmipes showed an intermediate phenotype , as it did in another study using different oviposition assays ( Karageorgi et al . , 2017 ) . The D . biarmipes taste system also appears intermediate , in the sense that some phenotypes resemble those of D . suzukii and some those of D . melanogaster . D . biarmipes is like D . suzukii in that both have four fewer taste sensilla than D . melanogaster . D . biarmipes is like D . melanogaster in that both show oviposition avoidance to several bitter compounds that did not affect D . suzukii ( Figure 8 ) . The S-a sensilla of D . biarmipes are like those of D . melanogaster in that they retain a substantial response to overripe strawberry , but the S-b sensilla are like those of D . suzukii in that they have lost this response ( Figure 6 ) . One interpretation of all these results , taken together , is that evolutionary changes in a common ancestor of D . suzukii and D . biarmipes provided a foundation for further adaptations that allowed D . suzukii to occupy its current niche . D . suzukii exemplifies a broad theme in drosophilid evolution: the successful adaptation of a variety of species to a variety of niches . While D . suzukii has adapted to occupy ripening stages not exploited by other drosophilids , other Drosophila species have adapted to particular host species . For example , Drosophila sechellia has specialized on the noni fruit ( Morinda citrifolia ) , which is toxic to other species , and Drosophila erecta has specialized on screw pine fruit ( Pandanus spp . ) ( Jones , 1998; Linz et al . , 2013; Whiteman and Pierce , 2008 ) . Are the features of adaptation that we have observed in D . suzukii also found in these other species ? D . suzukii differs from D . sechellia and D . erecta in that it has lost taste sensilla from the labellum; D . sechellia and D . erecta have retained the canonical numbers and map positions of taste sensilla defined originally in D . melanogaster ( Dweck and Carlson , 2020 ) . However , commonalities in adaptation mechanisms are also suggested by our results . First , analysis of the D . sechellia genome suggested that the rapid loss of 12 bitter Grs contributed to a loss of taste responses to bitter secondary metabolites of noni fruit ( McBride , 2007; McBride et al . , 2007 ) . Our results in D . suzukii using RNAseq support this notion that loss of bitter Gr expression may contribute to a shift in evolutionary niche . Second , oviposition preference in D . sechellia was found to depend on two genes expressed in the legs , Obp57d ( odorant binding protein ) and Obp57e , leading to the suggestion that an evolutionary change in tarsal taste response contributed to its shift in oviposition preference ( Matsuo et al . , 2007 ) . Our results now establish precedent via direct electrophysiological recording for such a change in tarsal response profiles . In fact , our results indicate how functional classes of taste neurons and their tuning breadths expanded or contracted during the evolution of the three species we examined . Plasticity was not restricted to a particular type of taste sensillum or taste organ . We have found a shift from overripe toward ripe preference in D . melanogaster mutants whose bitter taste responses are reduced compared to wild type . We also found that D . suzukii has bitter taste responses that are reduced in many ways relative to D . melanogaster . What is the link between bitter taste and oviposition preference ? One simple model to explain our results is that bitter compounds in early ripening stages deter oviposition in D . melanogaster . Loss of bitter response in Gr33a mutants of D . melanogaster or in D . suzukii would reduce detection of these deterrent compounds , and may thereby contribute to a shift toward oviposition on ripe fruits . Thus the loss of bitter-sensing sensilla , bitter Grs , and physiological responses of the remaining sensilla would all represent adaptations that allow D . suzukii to occupy a niche whose bitter compounds deter competition from other Drosophila species . Direct evidence to support this model comes from electrophysiological recordings of S sensilla , the only class of sensilla in which activity can be confidently attributed to bitter-sensing neurons . In the case of both S-a and S-b sensilla , responses to ripe purée of strawberry are severely reduced in D . suzukii compared to D . melanogaster ( Figure 6B ) . However , although the loss of response in D . suzukii to bitter compounds in early ripening stages seems likely to contribute to the oviposition shift , further investigation will be required to fully understand the role of bitter taste in the shift . One might have expected an increased response of D . suzukii to overripe fruit . However , the response of S-a and S-b sensilla to overripe purées is also reduced in D . suzukii . This reduced response to overripe strawberry might by itself , according to the simplest model , be expected to favor a countervailing preference for overripe fruit . This finding illustrates that a full appreciation of the role of bitter taste in the evolutionary shift will require a better understanding of the role of bitter neurons in driving oviposition behaviors , in two respects . First , previous work has shown that the influence of tastants on oviposition decisions is complex ( Joseph et al . , 2009; Joseph and Heberlein , 2012; Schwartz et al . , 2012; Yang et al . , 2008 ) . Bitter-sensing neurons are diverse in their specificities ( Delventhal and Carlson , 2016; Dweck and Carlson , 2020; Weiss et al . , 2011 ) , and the activation of different bitter neurons may have distinct effects , or even opposing effects , on behavioral circuits at certain concentrations or in certain contexts ( Joseph et al . , 2009; Joseph and Heberlein , 2012; Schwartz et al . , 2012; Yang et al . , 2008 ) . Bitter neurons of D . melanogaster and D . suzukii are tuned differently and may be sensitive to different natural cues; D . suzukii could conceivably have acquired a new response to a bitter compound in ripe strawberry , perhaps in an I sensillum , that favors a shift toward ripe fruits . Clearly , further work will be required to understand which of the evolutionary changes in bitter coding we have observed affect oviposition choices and the mechanisms by which they affect them . Second , we emphasize that bitter neurons operate in a larger context; their activities contribute to , but do not alone dictate , oviposition responses . As an illustration , S sensilla in D . melanogaster gave a greater response to overripe than ripe purée . If bitter-sensing neurons of D . melanogaster detect deterrent cues in an overripe fruit , why do these flies lay eggs on it ? Oviposition decisions are likely made based on an evaluation of many cues , both negative and positive , and it seems likely that positive cues detected by other neurons of D . melanogaster – for example , by sugar neurons of the taste system or by neurons of other sensory modalities – predominate in the overripe fruit we have tested . By contrast , in a natural environment in which overripe fruits become increasingly covered with diverse populations of microbes , bitter neurons may provide a warning system that detects toxins , responds strongly , and inhibits oviposition . Our results lay a foundation for a wide variety of avenues for future investigation . What specific bitter compounds in ripe or overripe strawberries influence oviposition decisions of each species in a natural context ? Are the most influential compounds present in other fruits ? We have tested individual compounds and purees , but we do not know the identities or quantities of the compounds that a fly encounters while exploring a fruit in nature . Which bitter receptors respond to these compounds , and is their expression reduced in D . suzukii ? Might the receptors that respond to these compounds have undergone evolutionary changes in their functional characteristics ? Finally , how is information about bitter compounds integrated with information about sugars , other tastants , and other cues to guide oviposition , and have there been evolutionary adaptations in the taste circuitry of D . suzukii ? In summary , we have identified gustatory innovations – anatomical , physiological , behavioral , and molecular – in D . suzukii . Our results support a major role for gustation in the altered oviposition preferences of D . suzukii . Taken together our study provides , for the first time to our knowledge , new understanding of how the gustatory system of an invasive pest species has adapted in its evolutionary adaptation to a new niche .
D . melanogaster Canton-S , D . suzukii , and D . biarmipes were reared on corn syrup and soy flour culture medium ( Archon Scientific ) at 25 °C and 60% relative humidity in a 12:12 hr light-dark cycle . D . suzukii stock was collected in Connecticut . D . biarmipes stock ( 14023–0361 . 04 ) was obtained from the Drosophila Species Stock Center . Gr33a2 is described in Dweck and Carlson , 2020; Gr33a3 is an independent allele generated by the same means in the same study . Ripening stages of strawberries were collected from Lockwood Farm , Connecticut Agricultural Experiment Station , Hamden , CT . Strawberries used in the single experiment shown in Figure 2D were from Elm City Market , New Haven , CT; in this case overripe strawberries were obtained by leaving ripe strawberries at room temperature for 3 days . Bitter 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 . All tastants were prepared fresh and used for no more than 1 day . For electrophysiological recordings , tastants were tested at the following concentrations unless otherwise indicated: ARI , 1 mM; azadirachtin ( AZA ) , 1 mM; berberine chloride ( BER ) , 1 mM; CAF , 10 mM; coumarin ( COU ) , 10 mM; DEET , 10 mM; DEN , 10 mM; ESC , 10 mM; gossypol from cotton seeds ( GOS ) , 1 mM; ( - ) -LOB , 1 mM; saponin from quillaja bark ( SAP ) , 1%; D- ( + ) -sucrose octaacetate ( SOA ) , 1 mM; SPS , 10 mM; strychnine nitrate salt ( STR ) , 10 mM; TPH , 10 mM; UMB , 10 mM . All compounds were stirred for 24 hr . THE and UMB were additionally heated to increase their solubility , then cooled and tested while in solution . These experiments were carried out in a cage ( 24 . 5 cm x 24 . 5 cm x 24 . 5 cm ) that was equipped with seven Petri dishes ( 60 mm × 15 mm , Falcon ) . Each Petri dish was filled with 1% agar containing 10% w/v purée of one of the ripening stages . One hundred 5- to 7-day-old flies ( 80 females and 20 males ) were placed in each cage . Experiments were carried out in a climate chamber ( 22°C , 60% humidity , in the dark ) . The number of eggs was counted after 24 hr . The positioning of the oviposition plates was randomized in each replicate . The two-choice oviposition assay was modified from Joseph et al . , 2009 , except that corn meal food was replaced with 1% agar containing 100 mM sucrose . Oviposition plates consisted of plastic Petri dishes ( 60 mm × 15 mm , Falcon ) , which were divided into two halves; each half contained either sucrose or sucrose mixed with a bitter compound . Fifty flies ( 40 females and 10 males ) , when 5- to 7-day-old , were placed into an oviposition cage ( Genesee Scientific ) without anesthesia through a small funnel that fits in the lid of the cage , and left for 24 hr in the dark . Experiments were carried out in a climate chamber ( 22°C , 60% humidity ) . Eggs on each substrate were counted . An oviposition preference index was calculated as follows: ( number of eggs on sucrose substrate – number of eggs on sucrose+bitter substrate ) / ( total number of eggs on both substrates ) . Flies were fixed in a solution of 0 . 1 M sodium cacodylate , 2% paraformaldehyde , and 2 . 5% glutaraldehyde for 2 hr in microporous specimen capsules ( Electron Microscopy Sciences ) . Flies were then dehydrated in a graded series of ethanol washes until they were incubated overnight in 100% ethanol . Ethanol-dehydrated flies were dried in a Leica CPD300 critical point dryer . Flies were then glued to metallic pegs with graphite conductive adhesive ( Electron Microscopy Sciences ) . Samples were then coated in 2 nm of iridium with a Cressington Sputter Coater and imaged in a Hitachi SU-70 SEM . Electrophysiological recordings were performed with the tip-recording method ( Hodgson et al . , 1955 ) , with some modifications; 5- to 7-day-old mated female flies were used . Flies were immobilized in pipette tips , and the labellum or the female foreleg was placed in a stable position on a glass coverslip . A reference tungsten electrode was inserted into the eye of the fly . The recording electrode consisted of a fine glass pipette ( 10–15 µm tip diameter ) and connected to an amplifier with a silver wire . This pipette performed the dual function of recording electrode and container for the stimulus . Recording started the moment the glass capillary electrode was brought into contact with the tip of the sensillum . Signals were amplified ( 10x; Syntech Universal AC/DC Probe; http://www . syntech . nl ) , sampled ( 10 , 667 samples/s ) , and filtered ( 100–3000 Hz with 50/60 Hz suppression ) via a USB-IDAC connection to a computer ( Syntech ) . Action potentials were extracted using Syntech Auto Spike 32 software . Responses were quantified by counting the number of spikes generated over a 500 ms period after contact . Different spike amplitudes were sorted; we did not convolve all neurons into a single value . However , in nearly all recordings in this study the great majority of the spikes were of uniform amplitude ( e . g . , Figures 5 and 7B–D ) , and those were the spikes whose frequencies we report . Responses to the TCC diluent alone were subtracted . Labella were meticulously hand-dissected from approximately one-hundred 5-day-old D . melanogaster , D . suzukii , and D . biarmipes females . The tissues were collected and mechanically disrupted in lysis buffer ( ‘RTL lysis buffer’ from Qiagen ) . Labellar RNA was extracted using the hot acid phenol procedure . Three biological replicates were produced for each species . Libraries were prepared using KAPA mRNA HyperPrep Kit ( Kapa Biosystems ) and sequenced on an Illumina HiSeq 2500 sequencer by the Yale Center for Genome Analysis . Thirty to fifty million 75 bp paired-end reads were obtains per sample . Raw reads are accessible at the Genbank SRA database ( BioProject accession number PRJNA670502 ) . Reads were aligned to the D . melanogaster genome ( BDGP6 ) , D . suzukii genome ( version 1 . 0 ) , or the D . biarmipes genome ( version 2 . 0 ) using TopHat ( version 2 . 1 . 1 ) . Cufflinks ( version 2 . 2 . 1 ) was used to generate de novo GTF files for each species and quantify D . melanogaster labellar transcripts ( Ensemble annotation version 100 ) ( Figure 9—figure supplement 1 and 2 ) . IGV , Integrative genomics viewer ( version 2 . 5 . 3 ) , was used to inspect the read coverage of genes of interest . For quantification , only the coding sequence ( CDS ) of genes was considered and CDS with length differences across species larger than the read length were discarded . Reads were remapped to the curated CDS transcriptomes and counted using HTseq ( version 0 . 6 . 1 ) . Read 1 and read 2 were analyzed separately . Differential expression ( DE ) analysis was carried between D . suzukii and D . melanogaster and between D . suzukii and D . biarmipes using four different pipelines: ( i ) DESeq2 ( version 1 . 26 . 0 ) using ashr for Log Fold Change ( LFC ) shrinkage ( Stephens , 2017 ) ; ( ii ) edgeR ( version 3 . 28 . 1 ) ; ( iii ) NOIseq ( version 2 . 31 . 0 ) with counts normalized by length and read depth ( TPM ) ; ( iv ) NOIseq with counts normalized with SCBN ( scale-based normalization , version 1 . 4 . 0 ) , a recent method optimized for cross species DE analysis ( Zhou et al . , 2019 ) . The two latter approaches were used to estimate the number of false positive candidates related to minor differences in transcript length . In the case of duplicated genes , the closest ortholog was kept . If this could not be determined , the most abundant was used . Only significant hits ( |Log2FC| ≥ 2 , adjusted p-value≤0 . 01 ) common to all DE analysis methods were considered . The hierarchical clustering of DESeq2 and edgeR result matrices was performed using default settings of the pvclust package in R with default settings . By default , reliability of the branching was assessed by generating 1000 bootstrap samples by random sampling . PCA plot was generated using the prcomp and ggbiplot packages in R with DESeq2 and edgeR results and default settings . The gene ontology ( GO ) analysis was performed using GOrilla . RT-qPCR cDNA was made from 300 ng of labellar RNA as template from using EpiScript ( Lucigen ) . Two biological replicates were prepared per species . PCR was carried out with Apex master mix ( Genesee Science ) using 15 ng of cDNA . Primers used in Figure 9—figure supplement 3A were the following: Primers used in Figure 9—figure supplement 3B: Hierarchical cluster analyses were performed using Ward’s method with PAST ( Paleontological Statistics Software Package for Education and Data Analysis; Hammer et al . , 2001 ) . This technique organizes the data into clusters based on the response profiles of each sensillum to the panel of tastants . Euclidean distances were calculated according to Ward’s classification method for the hierarchical cluster analysis . Other statistical tests were performed in GraphPad Prism ( version 6 . 01 ) . All error bars are SEM . *p<0 . 05 , **p<0 . 01 , ***p<0 . 001 , ****p<0 . 0001 . | A new agricultural pest has recently emerged in the United States and Northern Europe . The invasive species is a type of fruit fly that normally lives in Southeast Asia called Drosophila suzukii ( also known as the spotted wing Drosophila ) . This fly poses a threat to fruit crops – including strawberries , blueberries , cherries , peaches and grapes – because , while other fruit flies lay eggs in overripe fruit , D . suzukii lays eggs in ripe fruit , leading to agricultural losses . This shift in where fruit flies prefer to lay their eggs is related to changes in the senses of smell and touch , and taste could also play a role . Insects have evolved mechanisms that dissuade them from eating or laying eggs in plants with high levels of toxins , which taste bitter . If D . suzukii is less sensitive to bitter tastes than other flies , this could help explain why it lays eggs in just-ripe fruit , since the levels of certain bitter compounds are higher in the early stages of ripening than later on . To figure out if this is the case , Dweck et al . studied different species of fruit fly . Compared to Drosophila melanogaster ( a fruit fly common in America and Europe that is regularly used in scientific studies ) , D . suzukii had fewer bitter taste receptor neurons on the major taste organ of the fly head . These receptor neurons were also less responsive to a variety of bitter compounds . Next , Dweck et al . tested whether D . melanogaster and D . suzukii showed different preferences for where to lay their eggs by offering them strawberry purées made from fruit at different ripening stages . In this experiment , D . suzukii preferred to lay its eggs on purées made from unripe or just-ripe strawberries , while D . melanogaster showed a preference for fermented ( overripe ) purée . Furthermore , when D . melanogaster flies were genetically modified so that they became less sensitive to bitter taste , they preferred to lay their eggs in ripe ( rather than overripe ) fruit , similar to D . suzukii . These results suggest that taste has a major role in the egg laying preferences of D . suzukii . Further research is needed to determine which bitter compounds influence egg-laying decisions in each species of fruit fly , and what receptors respond to these compounds . However , Dweck et al . ’s results lay the groundwork for new approaches to reducing D . suzukii’s impact on agriculture . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"neuroscience"
] | 2021 | Evolutionary shifts in taste coding in the fruit pest Drosophila suzukii |
Electrophysiological data disclose rich dynamics in patterns of neural activity evoked by sensory objects . Retrieving objects from memory reinstates components of this activity . In humans , the temporal structure of this retrieved activity remains largely unexplored , and here we address this gap using the spatiotemporal precision of magnetoencephalography ( MEG ) . In a sensory preconditioning paradigm , 'indirect' objects were paired with 'direct' objects to form associative links , and the latter were then paired with rewards . Using multivariate analysis methods we examined the short-time evolution of neural representations of indirect objects retrieved during reward-learning about direct objects . We found two components of the evoked representation of the indirect stimulus , 200 ms apart . The strength of retrieval of one , but not the other , representational component correlated with generalization of reward learning from direct to indirect stimuli . We suggest the temporal structure within retrieved neural representations may be key to their function .
Associative memory in animals and humans provides a model of the environment . Retrieval of such memories , driven by cues or occurring autonomously , is suggested as central to a wide variety of processes and functions , including online and offline planning and model-learning ( Sutton , 1991; Moore and Atkeson , 1993; Foster and Wilson , 2006; Johnson and Redish , 2007; Hasselmo , 2008; Lisman and Redish , 2009; Gupta et al . , 2010; van der Meer et al . , 2010; Jadhav et al . , 2012; Wimmer and Shohamy , 2012; Pfeiffer and Foster , 2013; Singer et al . , 2013 ) , cognitive search ( Kurth-Nelson et al . , 2012; Todd et al . , 2012; Morton et al . , 2013 ) , mental time travel ( Hopfield , 2010; Schacter et al . , 2012 ) , memory maintenance and consolidation ( Marr , 1971; Nádasdy et al . , 1999; Káli and Dayan , 2004; Kuhl et al . , 2012; Deuker et al . , 2013 ) as well as temporal expectation ( Sakai and Miyashita , 1991; Rainer et al . , 1999 ) . Retrieval is classically linked to reinstantiation of a particular distributed spatial pattern of neural activity mirroring that evoked by the original experience of the object or context being retrieved ( Tulving and Thomson , 1973; Nyberg et al . , 2000; Hoffman and McNaughton , 2002; Polyn et al . , 2005; Johnson and Rugg , 2007; Gelbard-Sagiv et al . , 2008; Danker and Anderson , 2010; Rissman and Wagner , 2012; Miller et al . , 2013; Kuhl and Chun , 2014 ) . However , electrophysiology experiments robustly demonstrate that when an object is directly experienced , the evoked pattern of neural activity evolves rapidly over tens to hundreds of milliseconds ( Makeig et al . , 1997; Schmolesky et al . , 1998 , 1998; Näätänen and Winkler , 1999; VanRullen and Thorpe , 2001; Rossion and Jacques , 2008; Schneider et al . , 2008; Cichy et al . , 2014 ) . This implies that direct experience of an object evokes multiple distinct spatial patterns of neural activity in sequence . However , these distinct spatial patterns have never been identified independently at retrieval . At retrieval , recent studies have explored the fast evolution of neural representation . EEG studies provide evidence that some information is retrieved as early as 300 ms following a cue ( e . g . , Johnson et al . , 2008; Yick and Wilding , 2008; Wimber et al . , 2012 ) . Manning et al . ( 2011 ) , ( 2012 ) , using electrocorticography , and Jafarpour et al . ( 2014 ) , using MEG , showed that oscillatory patterns are also reinstated during retrieval; in two of these studies the predominance of low oscillatory frequencies in reinstatement suggests a potential spectral signature . However , the dynamics of representation during direct experience of an object have never been tied to the dynamics of retrieval . It is not known which of the patterns evoked in sequence by direct experience are reinstantiated during retrieval , what the temporal relationship is in their retrieval , or what functional significance this has . Recent advances in multivariate methods for MEG have greatly improved our ability to discern fast-changing distributed representations in humans ( Carlson et al . , 2013; Cichy et al . , 2014; Jafarpour et al . , 2013 , 2014; van de Nieuwenhuijzen et al . , 2013; Sandberg et al . , 2013 ) . Here , we apply these methods to a simple sensory preconditioning task adapted from Wimmer and Shohamy ( 2012 ) . Sensory preconditioning is a well-established paradigm in which subjects first form an association between two stimuli ( ‘direct’ or Sd and ‘indirect’ or Si ) and then form an association between the direct stimulus and a reward ( Brogden , 1939 ) . Generalization of value to the indirect stimulus is evidence of retrieving the learned association ( Gewirtz and Davis , 2000 ) . Using fMRI , Wimmer and Shohamy ( 2012 ) showed that neural representations of the associated indirect stimulus are reinstated when direct stimuli are presented during the Reward-learning phase , and this retrieval is linked to the generalization of value from direct to indirect stimuli . This suggests that reinstatement through the learned associative link may be part of the mechanism for value updating . Our aim here is to explore the temporal structure of this reinstatement , which may help to shed light on the mechanisms of value updating as well as providing general insight into the dynamics of representations during retrieval . We therefore examined retrieval in the same paradigm , using MEG to gain temporal precision . We show that the neural representation of the indirect stimulus can be decomposed into at least two temporal components with distinct properties , and these are retrieved at different times during the Reward-learning phase . The retrieval of only one of these components is correlated with a behavioral measure of the generalization of value across the learned associations .
We used a slightly modified version of the behavioral task employed by Wimmer and Shohamy ( 2012 ) . This involved three phases ( Figure 1A ) . In the Association phase , subjects watched visual stimuli appearing sequentially at the center of the screen . The stimuli alternated between photographs ( ‘Si’ ) and circular fractals ( ‘Sd’ ) , with a short blank fixation interval between each stimulus . Each Si came from one of three categories ( face/body/scene ) , and each unique Si was deterministically followed by a unique Sd , thus establishing a pairing between Si and Sd images . As in Wimmer and Shohamy ( 2012 ) , debriefing revealed that subjects were not aware of the Si–Sd pairings . There were two unique Si in each category; making for a total of six unique Si and six unique Sd stimuli used for later phases ( along with six additional unique Si and six additional unique Sd that functioned as dummies for the Association phase cover task and were included in the imaging analysis ) . 10 . 7554/eLife . 04919 . 003Figure 1 . Task design and behavior . Subjects participated in a sensory preconditioning task comprising three phases: Association , Reward and Decision . ( A ) In the Association phase , subjects were exposed to pairs of stimuli ( presented sequentially ) . One member ( called Si ) of each pair was taken from one of three classes ( faces , bodies , and scenes ) ; the other member ( Sd ) was a fractal . In the Reward phase , some of the fractals ( labelled Sd+ ) were paired with reward; the others ( labelled Sd− ) were not . Through the pairing , this implicitly established a separation between Si+ and Si− . In the Decision phase , subjects chose between Si+ and Si− within the same category , or between Sd+ and Sd− . All photos shown are from pixabay . com and are in the public domain . ( B ) In the Decision phase , subjects displayed a strong preference for Sd+ over Sd− ( p = 6 . 9 × 10−4 , one-sample t-test ) . There was no preference at the group level for Si+ over Si− , but we exploited the variability between subjects for value-related analyses . The change in relative liking from before to after the experiment was more positive for Sd+ than Sd− ( p = 0 . 04 , one-sample t-test ) ; but there was no significant difference between the changes for Si+ and Si− . Bar heights show group means and dots show individual subjects . Error bars show standard error of the mean . DOI: http://dx . doi . org/10 . 7554/eLife . 04919 . 003 In the Reward phase , of the two Sd images associated with a category of Si images , one , which we therefore call Sd+ was followed by a reward on 14 out of 18 presentations ( and otherwise by a neutral outcome , a blue square ) ; the other , which we call Sd- was always followed by a neutral outcome . By virtue of the prior pairing , this established an Si+ and Si− for each category . In the Decision phase , subjects were faced with pairwise choices between an Si+ and an Si− , or an Sd+ and an Sd− . The two items always had the same category ( face/body/scene ) for Si , or associated category for Sd . Subjects exhibited a strong preference for Sd+ over Sd− ( p = 6 . 9 × 10−4 ) , but as a group showed no evidence of preferring Si+ over Si− ( p = 0 . 9 ) ( Figure 1B ) . Neural activity was recorded by magnetoencephalography ( MEG ) during all three phases . We first explored where in space and time the MEG signal carried information about the Si stimuli being presented in the Association phase . Using one-way ANOVA , we found that the raw amplitude , in single time bins , of the event-related field ( ERF ) at many individual sensors was significantly related to the Si category ( Figure 2 ) . ( The significance threshold was set to 95% of peak-level over space and time from 100 random category label shuffles , to correct conservatively for multiple comparisons . ) 10 . 7554/eLife . 04919 . 004Figure 2 . Event-related field ( ERF ) discriminates between categories ( face/body/scene ) at time of Si presentation . Sensors became category-discriminative in two waves . ( A ) The first time , relative to stimulus onset , when the relationship between ERF amplitude and category membership became significant by ANOVA ( significance threshold set at 95% of peak-level ( across all sensors and all time ) log10 ( p ) of 100 shuffles ) at each of 275 sensors . Many occipital and temporal sensors first became predictive of Si category between 90 and 230 ms post stimulus onset , followed by some parietal and frontal sensors ranging from 330–550 ms post stimulus onset . Open circles indicate the sensors that never reached 95% peak-level . ( B ) Histogram of how many sensors first became significantly discriminative at each time following stimulus presentation . DOI: http://dx . doi . org/10 . 7554/eLife . 04919 . 004 Next , we built a multivariate linear SVM classifier , which combined the reports of multiple sensors ( Figure 3A ) . As in many previous studies ( cf . Norman et al . , 2006; Cichy et al . , 2014 ) , the extra sensitivity achieved by combining multiple features supported the use of multivariate analysis to track neural representations ( Figure 3—figure supplement 1 ) . We constructed null distributions at each time bin by repeating this procedure 100 times with randomly shuffled category labels . At 200 ms post-stimulus , the 95th percentile of the null distribution was 35 . 0% accuracy , and the median was 33 . 7% ( deviating from 1/3rd only due to the finite number of shuffles ) . 10 . 7554/eLife . 04919 . 005Figure 3 . Multivariate analysis reveals two temporal components of evoked response to visual stimuli . ( A ) Multivariate decoding performed well to predict the category of photograph ( Si ) in the Association phase . Cross-validated linear SVM prediction accuracy using all 275 sensors at each time bin is shown . A pattern of two distinct peaks in classifier accuracy around 200 ms and 400 ms after Si onset is evident . ( B ) At 200 ms after Si onset , there was no difference in representational similarity between same-category and different-category Si objects ( left panel , p = 0 . 2 by t-test between subjects ) . At 400 ms , representational similarity was higher for same-category than different-category objects ( right panel , p = 5 × 10−7 ) . F1–F4 , B1–B4 and S1–S4 refer to the unique faces , bodies and scenes presented during the Association phase . ( C ) When discriminating fractal identity ( i . e . , a 6-way classification problem of stimuli with no natural categories ) , performance was sharply peaked before 200 ms after fractal onset . Shaded area shows standard error of the mean . DOI: http://dx . doi . org/10 . 7554/eLife . 04919 . 00510 . 7554/eLife . 04919 . 006Figure 3—figure supplement 1 . Univariate classification using best sensor . We tested the capacity of the most discriminative single sensor ( selected separately for each subject ) to predict the Si category , using linear support vector machines ( SVM ) with a single feature . The accuracy of this univariate classifier peaked at 47 . 4 ± 1 . 3% in cross-validation ( red trace ) . ( When using a nearest-mean univariate classifier rather than a univariate SVM , accuracy peaked at 45 . 6 ± 1 . 9% . ) We constructed independent null distributions at each time bin by repeating this procedure 100 times with randomly shuffled category labels . At 200 ms post-stimulus , the median of the null distribution was 37 . 0% accuracy ( greater than 1/3rd due to allowing the best sensor for each subject ) , while the 95th percentile of the null distribution was 38 . 6% . Blue line shown is multivariate SVM performance , from Figure 3A , for comparison . DOI: http://dx . doi . org/10 . 7554/eLife . 04919 . 00610 . 7554/eLife . 04919 . 007Figure 3—figure supplement 2 . Multivariate classification of Si for individual subjects . ( A ) Classification accuracy in predicting Si category in Association phase for individual subjects . ( B ) We fit regression models to each subject's accuracy curve ( between 200 ms and 400 ms ) , with constant , linear , and quadratic terms . This histogram shows the estimated betas on the quadratic term . Positive beta indicates positive curvature of the accuracy curve between 200 ms and 400 ms . No individual subject reached Bonferroni-corrected significant betas on the quadratic term of the regression . DOI: http://dx . doi . org/10 . 7554/eLife . 04919 . 00710 . 7554/eLife . 04919 . 008Figure 3—figure supplement 3 . Nearest-mean multivariate classifiers , under a variety of distance metrics , underperform SVM but extract a similar pattern of multiple peaks in classification performance . Compare to SVM applied to the same classification problem in Figure 3B , blue trace . DOI: http://dx . doi . org/10 . 7554/eLife . 04919 . 00810 . 7554/eLife . 04919 . 009Figure 3—figure supplement 4 . Decoding outcome identity . At the time of outcome , there was a strong neural representation of the identity of the outcome itself ( the coin or blue square ) . Together with Figure 3D , this suggests that the neural signal at time of Sd and outcome strongly encoded a representation of the on-screen stimulus . DOI: http://dx . doi . org/10 . 7554/eLife . 04919 . 00910 . 7554/eLife . 04919 . 010Figure 3—figure supplement 5 . Generalization of instantaneous representational patterns over time , with finer temporal binning . Here we trained classifiers on every time bin relative to the onset of Si in the Association phase , and tested at every time bin relative to the same onsets . For this figure we binned the data into 8 ms bins rather than the 20 ms bins used in the rest of the paper . Each cell of this grid shows cross-validated prediction accuracy , so the diagonal is equivalent to Figure 3B , blue trace ( except that this figure has finer temporal binning ) . Later classifiers generalized better over time than earlier classifiers . We note the possibility that the 200 ms peak of classification might be decomposed into further sub-peaks ( white and black arrows ) ; however , we were unable to statistically separate these sub-peaks , due to variability between subjects . The peak at 400 ms is evident ( blue arrow ) . Absolute classification accuracy is lower than with more coarsely binned data , likely due to a poorer signal to noise ratio . DOI: http://dx . doi . org/10 . 7554/eLife . 04919 . 01010 . 7554/eLife . 04919 . 011Figure 3—figure supplement 6 . Image statistics . Image types varied in low-level visual properties as well as shape . The methods we used are agnostic as to the kinds of features that drove the neural representation of category . DOI: http://dx . doi . org/10 . 7554/eLife . 04919 . 011 We observed two distinct peaks in multivariate classification performance , one centered approximately around 200 ms and the other around 400 ms post-stimulus onset . Although these peaks had measurable width , for simplicity , we will henceforth refer to them as ‘200 ms’ and ‘400 ms’ . To test more formally for two distinct peaks in classification , we asked whether there was significant concavity in the evolving classification accuracy in the interval from 200 to 400 ms , by regressing the classification accuracy against linear and quadratic functions of time . At the group level , the quadratic term was significantly different from zero ( p = 0 . 02 ) . We also performed this regression on the accuracy curves from individual subjects; many subjects trended toward a positive quadratic term , but none reached significance at a Bonferroni-corrected threshold ( Figure 3—figure supplement 2 ) . Finally , to rule out any peculiarities in the SVM algorithm being responsible for two distinct peaks in classification accuracy , we also repeated the same analysis at the group level with a variety of nearest-mean classifiers and found the same pattern ( Figure 3—figure supplement 3 ) . Given past observations and ideas about separate post-stimulus phases encoding qualitatively different kinds of stimulus information ( Schmolesky et al . , 1998; Lamme and Roelfsema , 2000; Riesenhuber and Poggio , 2000; Engel et al . , 2001; Bar , 2003; Cichy et al . , 2014 ) , we asked if these two peaks had different representational similarity structure . We calculated representation similarity matrices ( Kriegeskorte et al . , 2008 ) , which reflect the similarity in activation patterns between each pair of unique stimuli . We found that at 200 ms , the activity patterns evoked by stimuli within a category were no more similar than those evoked by stimuli in different categories ( Figure 3B , left panel; p = 0 . 2 , paired t-test between subjects ) ; whereas at 400 ms , patterns within a category were substantially more similar than between categories ( Figure 3B , right panel; p = 5 × 10−7 ) . This is consistent with the idea that the dominant coding of stimulus information changes between 200 and 400 ms . Further supporting the idea that the later component of the ERF had a relatively more dominant coding of categorical information , we found that the cross-validated performance of a linear SVM in a 6-way discrimination of fractal identity was sharply peaked at 160 ms post-stimulus onset , and lacked a substantial second peak ( Figure 3C ) . We note that a shift in the timing of the early peak from ∼200 ms to ∼160 ms could be consistent with previous observations ( Bobak et al . , 1987; Cichy et al . , 2014 ) that the precise timing of each wave of representation is sensitive to the particular stimuli concerned . During the Reward phase , Sd ( fractals ) and outcomes ( coin/blue square ) were presented . We confirmed it was possible to predict the identity of fractals ( cf . Figure 3C ) and outcomes ( Figure 3—figure supplement 4 ) reliably based on the MEG signal . However , the main intention of our study was to examine whether the activity evoked by these stimuli contained information about the Si stimulus with which the Sd had been associated . To this end , we trained classifiers on neural responses to Si in the Association phase ( exactly as above , but using all trials because cross-validation was not necessary ) , and tested these classifiers on neural responses elicited in the Reward phase when Sd was presented . The classifier was considered to be correct if it reported the category label of the Si that had previously been paired with this Sd . We performed this train-on-Si , test-on-Sd procedure for every pair of times relative to the onsets of Si ( in the Association phase ) and Sd ( in the Reward phase ) , leading to a 2-D grid of classification accuracies ( Figure 4A ) . These 2-D grids were then smoothed with a 2-D Gaussian kernel ( σ = 30 ms ) . 10 . 7554/eLife . 04919 . 012Figure 4 . Early and late components of associated object representation retrieved at time of cue and outcome , respectively . During the Reward phase , the 200 ms component of the Si representation was retrieved for an extended period from shortly after Sd was presented , while the 400 ms component of Si representation was retrieved around the time the outcome was presented . ( A ) Classifiers trained around 200 ms after Si presentation in Association phase and tested around 400 ms after Sd presentation in Reward phase decode the object category previously associated with the Sd . Photo is from pixabay . com and is in the public domain . ( B ) Classifiers trained around 400 ms after Si presentation and tested 70 ms after outcome presentation decode the object category previously associated with the Sd . In A and B , black outlines show p = 0 . 05 peak-level significance thresholds ( empirical null distribution generated by 1000 random permutations of training category labels , see Methods for more details ) . ( C ) Peak classification accuracy in the 200 ms and 400 ms rows of A and B . By 2-way ANOVA , there was no main effect of 200 ms vs 400 ms or of Sd vs outcome , but there was a significant interaction ( p = 0 . 04 ) . Error bars show standard error of the mean . DOI: http://dx . doi . org/10 . 7554/eLife . 04919 . 012 We observed that the classifiers trained around 200 ms post-Si presentation achieved above-chance accuracy in predicting which Si category had previously been associated with the presented Sd ( the 95th percentile of the peak-level achieved in 200 random shuffle tests is shown as a solid black line in each panel ) . This effect was above chance from 270–530 ms following presentation of Sd . In other words , the spatial pattern of brain activity present 200 ms after presentation of Si in the Association phase was partially reinstantiated 270–530 ms after presentation of Sd in the Reward phase . Note that the randomization of Si–Sd pairings across subjects makes exceedingly unlikely the possibility that some visual features of Si happen to be shared with the associated Sd and might therefore carry a shared neural signature . We also applied the same set of classifiers to the activity evoked by presentation of outcome ( coin or neutral blue square ) that followed each Sd in the Reward phase . The classifiers trained around 400 ms after Si achieved above-chance accuracy in predicting the Si category previously associated with the Sd presented on this trial ( Figure 4B ) . This effect was strongest at 70 ms following presentation of the outcome , meaning that the spatial pattern of activity present 400 ms after presentation of Si in the Association phase was at least partially reinstantiated 70 ms after presentation of the outcome in the Reward phase . Since the outcome always appeared 3500 ms after Sd in each trial , 70 ms after presentation of outcome was equivalently 3570 ms after presentation of Sd . Since all the information necessary to retrieve Si was carried by Sd , some of the retrieval process might occur before onset of the outcome . Two-way ANOVA revealed no significant main effects of 200 ms vs 400 ms or Sd vs outcome but a significant interaction ( p = 0 . 04; Figure 4C ) . That is , the peak accuracy following Sd was higher for the 200 ms than the 400 ms classifier , while the peak accuracy following outcome was higher for the 400 ms than the 200 ms classifier , implying a double dissociation in the component that was more strongly retrieved at Sd vs outcome . Both forms of cross-classification were very much less accurate than ( linear ) classification of the identity of the Sd ( fractals ) or outcome ( coin/blue square ) from the activity directly evoked by these stimuli ( cf . Figure 3C and Figure 3—figure supplement 4 ) . To investigate which MEG sensors carried retrieved information , we again trained classifiers on Si-evoked data and tested on Sd– or outcome-evoked data ( i . e . , cross-classification ) . However , rather than using all 275 sensors , we repeated the procedure for 2000 iterations using a different random subset of 50 sensors each time . To investigate the retrieval identified in Figure 4A , B , we restricted analysis to 60 × 60 ms temporal ROIs centered on the peaks of cross-classification in Figure 4A , B , and averaged over these temporal ROIs . For each sensor , each iteration of this procedure thus yielded a single classification accuracy . We could then calculate how accurate the cross-classification was on average when a given sensor participated in classification . The average of these data across subjects are shown in Figure 5 , separately for Sd- and outcome-evoked data . To test whether these spatial patterns were significantly different , we again used a linear SVM with cross-validation to predict whether each pattern originated from Sd–or outcome-evoked data . Each pattern was mean-subtracted to avoid any trivial classification based on overall higher cross-classification performance for Sd- than outcome-evoked data . Prediction accuracy reached 71 . 2% , which was greater than chance by one-tailed binomial test ( p = 0 . 002 ) . 10 . 7554/eLife . 04919 . 013Figure 5 . Contributions of sensors to retrieval . To explore which brain areas carried the information about Si that was retrieved at the time of Sd and outcome , we copied the procedure of training linear category classifiers on presentation of Si , and predicting the category at the time of Sd or outcome—but instead of using all 275 sensors , we repeated the analysis 2000 times using subsets of 50 sensors randomly selected on each iteration . The contribution of sensor s was taken to be the mean of all prediction accuracies ( within 60 × 60 ms temporal ROIs containing the peak time bins ) achieved using an ensemble of 50 sensors that included s . Intriguingly , the information about the category of Si retrieved at the time Sd was presented emerged primarily from occipital sensors ( A ) , while the information about the category of Si retrieved at the time the outcome was shown appeared more strongly in parietal and temporal sensors ( B ) . In the difference between the two conditions , no individual sensor survived correction for multiple comparisons . However , a linear SVM was reliably able to classify whether a spatial pattern belonged to Sd or outcome ( 71 . 2% accuracy , p = 0 . 002 by one-sided binomial test against chance classification ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04919 . 013 Finally , we were intrigued by the apparent retrieval of only the late ( 400 ms ) and not the early ( 200 ms ) component of the Si representation during outcome presentation . The representational similarity analysis in Figure 3B suggested that this 400 ms component might preferentially encode stimulus category . Thus , we speculated the value of the associated Si category , rather than the value of a particular Si stimulus , might be updated when the outcome appears . This could provide a potential explanation for the lack of group-level behavioral preference for Si+ over Si− during the subsequent Decision phase , since each Si category contained both an Si+ and an Si− , with equal presentations . This hypothesis predicts that , although at the group level there might be no significant retrieval of the 200 ms component of Si representation during outcome presentation , the subjects who did retrieve the 200 ms component of Si might have a positive preference for Si+ over Si− . ( Meanwhile , a preference for Si− over Si+ should be unrelated to retrieval . ) We therefore plotted the correlation between behavioral preference and accuracy of Si-trained classifier in predicting the associated category of the Sd stimulus presented on this trial . This analysis was split according to whether subjects preferred Si− over Si+ ( Figure 6A ) or Si+ over Si− ( Figure 6B ) . Remarkably , in subjects preferring Si+ over Si− , reinstatement of the 200 ms component of Si was strongly correlated with behavioral preference . Shuffling subject identities yielded a null distribution of peak log10 p-values for the correlation of classifier accuracy with behavioral preference . The 400 ms classifier showed no substantial positive correlation with behavioral preference ( Figure 6C ) , while the 200 ms classifier showed a corrected-significant peak in correlation strength ∼400 ms after the onset of the outcome ( Figure 6D ) . The raw data driving these correlations are also shown in Figure 6E , F . 10 . 7554/eLife . 04919 . 014Figure 6 . Retrieval of early component of Si representation predicts value updating across subjects . At the group level , only the 400 ms component was significantly retrieved at the time of outcome ( cf . Figure 4B ) . However , at the single-subject level , the degree of retrieval of the 200 ms component correlated with value updating . As in Figure 4B , the accuracy of classifiers trained at each time bin around Si ( in the Association phase ) was tested at each time bin around the time of outcome ( in the Reward phase ) to predict the category of the Si associated with the Sd preceding the outcome . In each time*time bin , this accuracy was regressed , across subjects , against the behavioral preference for Si+ over Si− from the Decision phase ( i . e . , P ( Si+ ) ) . As we only explored positive correlations , one-tailed log10 p-values of the regression are reported . ( A ) In subjects who preferred Si− over Si+ , there were no correlations between the degree of preference and the degree of reinstatement of Si at outcome . ( B ) In subjects who preferred Si+ over Si− , there was a strong correlation between the degree of preference and the degree of reinstatement . This correlation peaked at around 400 ms after outcome onset . ( C , D ) Red and blue traces show single rows of panels A and B at 200 and 400 ms . Significance was tested by randomly shuffling subject identities to obtain a null distribution of peak-level log10 p-values . Thresholds are shown at 95% of the null distribution of the peak-level of 200 and 400 ms rows , and at 95% of the null distribution of peak-level of all rows . ( E , F ) Raw classification accuracies underlying the correlations in A–D , when training at 200 ms after Si onset and testing at 400 ms after outcome onset . Each point is a subject . DOI: http://dx . doi . org/10 . 7554/eLife . 04919 . 014
We used a sensory preconditioning paradigm to explore the temporal structure of the retrieval of representations through associative links . We found that presenting photographs ( Si , in three categories ) elicited an evolving representation with two temporally distinct components: one around 200 ms and the other around 400 ms after stimulus onset . The earlier component was reinstated when a fractal ( Sd ) previously paired with the Si was presented . The later component was reinstated when a rewarding or neutral outcome was presented following Sd . Although at the group level there was no significant reinstatement of the earlier component at the time of outcome , between subjects the degree of reinstatement of this earlier component correlated with the degree of subsequent value generalization . Our results fit comfortably with the large body of literature showing that retrieval ( which is notably unconscious here and in Wimmer and Shohamy , 2012 , as contrasted with conscious retrieval that is more commonly studied ) induces reinstantiation of at least some aspects of the pattern of neural activity evoked by the original presentation . For instance , in the fMRI study whose design we copied ( Wimmer and Shohamy , 2012 ) , univariate methods were used to show the equivalent of Si category retrieval during the Reward phase . Equally , ERP studies have found neural signals as early as 300 ms following a retrieval cue that are different depending on which information is retrieved or whether the information is retrieved ( Johnson et al . , 2008; Yick and Wilding , 2008 ) . Further , using MEG , Jafarpour et al . ( 2014 ) identified reinstatement of a pattern of oscillatory activity appearing approximately 180 ms following presentation of the retrieved item . This pattern was reinstated approximately 500 ms following the retrieval cue , slightly later than the 400 ms we observed . Multivariate pattern analysis provides a much more powerful microscope than traditional univariate analysis for detecting distributed patterns encoding neural representations ( Norman et al . , 2006 ) . Combining MVPA with MEG enables tracking the fast time-evolution of these representations ( Schmolesky et al . , 1998; Jafarpour et al . , 2013; Cichy et al . , 2014 ) . Using these methods we have extended previous findings on retrieval to now establish a mapping between the dynamics of object representation and the dynamics of retrieval in this behavioral paradigm . We identified two temporal components of object representation that were retrieved at different times . The earlier component of Si representation , which appeared roughly 200 ms following Si presentation , was first detectable 270 ms following presentation of Sd . This is consistent with past ERP studies showing similar timing , which have been taken as suggesting that reactivation is mediated by hippocampus ( Bosch et al . , 2014 ) . The prolongation of this representation from 270–530 ms may represent averaging ( over trials or subjects ) of temporally abrupt retrievals , or a sustained information retrieval . By contrast , the late component of Si representation re-appeared 70 ms following outcome presentation . The outcome did not provide any additional information about Si category , so the representation of Si must have been sustained in some form through the ( fixed ) delay between Sd and outcome . This raises questions such as where the information about Si was held during the delay , and what are the implications of this timing . For the former , we were only able to detect a representation of Si when it took the form of a spatial pattern of activity mirroring the pattern at presentation of Si . Thus information might have been online in the activity of , for instance , prefrontal neurons ( Fuster , 2001; Wang et al . , 2006 ) , but in a different form from that inspired by Si itself ( Sakai and Miyashita , 1991; Rainer et al . , 1999 ) . Alternatively , it might have been stored in short-term synaptic weight changes ( Hempel et al . , 2000; Seung , 2003; Florian , 2007; Mongillo et al . , 2008 ) . Supporting the idea of these ∼200 ms and ∼400 ms components as distinct representational periods , we note the following . First , there was a decrease in classification accuracy between these periods . Second , classifiers trained on one epoch had low accuracy in the other epoch ( Figure 3—figure supplement 5 ) , suggesting information about the stimuli was coded differently between epochs . Third , the epochs had different similarity structure with respect to the stimulus categories ( Figure 3B ) . Fourth , the patterns from the two epochs were doubly dissociated in terms of their retrieval at Sd vs outcome ( Figure 4 ) , while the time period between the two peaks ( i . e . , around 300 ms post-stimulus ) was not strongly retrieved either at Sd or outcome ( Figure 4 ) . In terms of timing , the relatively precise epoch of retrieval of Si following the presentation of the outcome may reflect the point of strongest overlap between a variety of timings in individual subjects . Alternatively , it may be that a representation that is latent became detectable as soon as more power arose in the visual-evoked ERF due to onset of the outcome . Yet another possibility is expectations of the next stimulus partly drive representations in the first 10 s of milliseconds after a visual onset , before the present stimulus is processed . The low accuracy in classifying retrieved representations ( ∼35% ) compared to evoked responses ( ∼60% ) might imply that retrieved representations ( perhaps especially those that subjects are not consciously aware of ) were weak compared to evoked representations . It is also possible that Si representations were only retrieved on a subset of trials , weakening the average signal . Finally , it is possible that retrieved representations had a distributed spatial pattern that was only partly overlapping with the evoked representation , making it more difficult to detect with pattern classifiers trained on evoked activity . We exploited the distinct temporal components of retrieval to help elucidate the neural underpinnings of value generalization through associations . In both our study and in the similar design of Wimmer and Shohamy ( 2012 ) , behavioral evidence of sensory preconditioning rests wholly on stimulus-specific retrieval ( since the rewards associated with each category are balanced ) . If the 400 ms component of Si representation preferentially encodes information about category rather than specific stimuli , as suggested by our representational similarity analysis , retrieval of solely this component at outcome time might cause value learning to be assigned to categories rather than individual stimuli . This hypothesis would explain our finding that the subjects who retrieve the 200 ms component at outcome show behavioral evidence of sensory preconditioning . Under this interpretation , the correlation that Wimmer and Shohamy found in BOLD between retrieved stimulus representations and behavior between subjects may also have been driven by the 200 ms component of the stimulus representation; these temporally precise signals could not be distinguished using fMRI . Although the particular representations online at the time of reward were probably driven by quirks of this task design ( since other sensory preconditioning experiments have found robust group-level preference for Si paired with rewarded Sd ( e . g . , Seidel , 1959 ) ) , the finding is of general importance because it suggests that the exact timing of reward relative to fast-evolving neural representational structures is crucial to value updating and credit assignment . Like Wimmer and Shohamy ( 2012 ) , we have compared a behavioral value generalization measure against the output of a neural classifier trained on the category of Si , rather than the identity of an individual Si . The latter would give a more direct test of the idea that subjects who retrieve a representation of the specific Si paired with the particular Sd viewed on this trial drive larger value updates . Although it is in principle possible to train a classifier to distinguish between individual exemplars of an Si category , this did not reach a sufficiently high level of performance in our hands , perhaps limited by the relatively small number of training samples per unique stimulus . Future experiments could also employ Si+ and Si− stimuli that are more neurally distinguishable . We noted in the ‘Introduction’ a large number of proposals for the use of associative information both at the time of decision ( online ) or when a decision is not imminent ( offline ) . Offline and online processes may share similar mechanisms ( Doll et al . , 2014 ) , and in some cases the division between offline and online mechanisms is blurred . For example , retrieving elements of past experiences may serve as part of the process of planning in advance for the next time related situations are encountered ( Dragoi and Tonegawa , 2011 , 2013 ) , similar to the psychological notion of implementation intentions ( Gollwitzer , 1999 ) . Some theoretical methods ( e . g . , the successor representation ( Dayan , 1993 ) and beta-models [Sutton , 1995] ) shift a portion of the burden of online calculations using offline updates to carefully structured representations . In sensory preconditioning , it is an open question whether generalized values are updated offline ( either during the Reward phase or in between the Reward and Decision phases ) , retrieved through associative links at the time of decision , or a mix of both . In animals the vulnerability of sensory preconditioning to extinction ( Gewirtz and Davis , 2000 ) hints at an online mechanism , but it is equally possible that extinction drives offline value updates through the same generalization mechanism as acquisition . Indeed , although our description of the reinstatement of Si suggests that it arises through a distinct process of retrieval , we cannot distinguish this from the subtly different possibility hinted by these ideas that the representation of Sd changed through the associative learning so that it more closely resembles that of Si . In animals , the temporal structure of retrieval appears to subserve complex memory ( Sirota et al . , 2003; Schwindel and McNaughton , 2011 ) , learning and decision-making processes , especially in hippocampus and hippocampal–cortical interactions . Rodents retrieve representations of past and future locations , actions , and rewards ( Johnson and Redish , 2007; van der Meer et al . , 2010; Steiner and Redish , 2014 ) ; the timing of this retrieval is tightly structured and likely encodes critical information in the decision-making computation . In humans , frontal theta power ( Hsieh et al . , 2011 ) and patterns of activity in hippocampus ( Ezzyat and Davachi , 2014; Hsieh et al . , 2014 ) are implicated in coding temporal order within sequences of stimuli . Applying methods from the present work could be useful to establish a finer grained map of the representations used in complex memory and decision processes . Important to understanding the retrieval dynamics in this behavioral paradigm is the shift we observed in the dominant coding of information in evoked responses from 200 ms to 400 ms post-stimulus . Information in the visual system up to 200 ms post-stimulus may hew closely to the form of the stimulus that was presented ( Tanaka and Curran , 2001; VanRullen and Thorpe , 2001; Liu et al . , 2002; Schiff et al . , 2006; Rossion and Jacques , 2008 ) . This is consistent with our finding that spatial patterns of activity evoked by different exemplars within a category were relatively distinct and that individual fractals were better classified at this time bin . Conversely , brain activity later than 200 ms post-stimulus is often found to include contextual and other sources of information ( Kok , 2001; Tsivilis et al . , 2001; Schiff et al . , 2006; Garrido et al . , 2007; Sanguinetti et al . , 2014 ) . In particular , the N400 component of the event-related potential ( ERP ) in EEG extends from roughly 250–500 ms post-stimulus and appears to be driven at least partly by the medial temporal lobe , which is functionally coupled to sensory cortices ( Bar , 2004 ) . The N2pc component of the ERP , which occurs earlier from roughly 200–300 ms post-stimulus , has also been tied to contextually-sensitive processing ( Conci et al . , 2006; Schiff et al . , 2006 ) , and originates from lateral temporal and parietal sources ( Hopf et al . , 2000; Oostenveld et al . , 2001 ) . Although information about the category of our stimuli is directly available in their visual form , one interpretation of our observation of more consistent category information at 400 ms is that this reflects such contextually-sensitive processing happening based on lateral and top-down functional connections ( MacKay and Bowman , 1969; Rao and Ballard , 1999; Ullman , 2000; Engel et al . , 2001; Bledowski et al . , 2006; Garrido et al . , 2007; Friston and Kiebel , 2009; Kourtzi and Connor , 2011 ) . Finally , we note that timing of event-related signals depends strongly on stimulus properties ( e . g . , Bobak et al . , 1987 ) . Multivariate classification also yields different timings in the peaks of classification depending on the specific kinds of categories evaluated ( Cichy et al . , 2014 ) . Thus the particular temporal structure of evoked responses is most likely specific to the stimuli used . Mapping this structure for a given task and stimuli can be leveraged to probe the dynamics of retrieval . In summary , neural retrieval of representations through associative links is central for memory and decision-making . Here we provide evidence that the dynamical structure within retrieval is functionally relevant for value-guided decision making . Analyzing the fine temporal structure of representations also increases the potential for studying temporally rich retrieval processes such as replay and planning in humans , which were previously confined to animal recordings .
Twenty-nine adults participated in the experiment , recruited from the UCL Institute of Cognitive Neuroscience subject pool . Three were excluded before the start of analysis for large movement or myographic artifacts . Of the 26 remaining , age quartiles were 18 . 7 , 19 . 5 , 21 . 3 , 26 . 7 , 41 . 4 years; 14 were female , and 1 was left-handed . All participants had normal or corrected-to-normal vision and had no history of psychiatric or neurological disorders . All participants provided written informed consent and consent to publish prior to start of the experiment , which was approved by the Research Ethics Committee at University College London ( UK ) , under ethics number 1825/005 . Participants performed three phases of a simple behavioral task ( copied almost exactly from Wimmer and Shohamy , 2012; but with timings set to be faster for MEG ) designed to induce and measure sensory preconditioning . The task was coded in Cogent ( Wellcome Trust Centre for Neuroimaging , United Kingdom ) , running in MATLAB 7 . 14 ( Mathworks , Natick , Massachusetts ) . Before the experiment , participants rated 78 images , one at a time , using a visual analog scale to indicate how much they subjectively liked each image , ranging from ‘Strongly Dislike’ to ‘Strongly Like’ . These images consisted of 60 photos ( 20 faces , 20 body parts , 20 scenes ) , and 18 fractals . Luminance and contrast varied between images ( Figure 3—figure supplement 6 ) . Four of each photo category and 12 fractals were then selected to be used in the experiment . For each subject we chose the stimuli whose liking ratings were closest to neutral; different subjects therefore saw different images in the experiment . In the first ( ‘Association’ ) phase of the experiment , each of the 12 selected photos ( ‘Si’ , indirect stimuli ) were deterministically paired with a different fractal pattern ( ‘Sd’ , direct stimuli ) . Two of each Si category were ‘dummies’ for the cover task , and two were ‘real’ stimuli . Subjects viewed Si and Sd images sequentially while performing a cover task of pressing one button in response to rightside-up images and a different button for upside-down images , with the button response mapping randomized across subjects . Dummies had a 50% chance of being upside-down , and real stimuli were never upside-down . Dummies were not presented in subsequent phases . In each trial , subjects saw an Si for 1750 ms , followed by an interstimulus-interval ( ISI ) of 1000 ms , followed by the paired Sd for 1750 ms , followed by an intertrial-interval ( ITI ) of 2500 ms . Every nine trials , each of the six real Si stimuli was presented once , and one of each of the dummy Si stimuli in each category was presented once ( both reals and dummies were always followed by the paired Sd ) . The order was randomly permuted over every 9 trials , and this was repeated 12 times , for a total of 108 trials . In debriefing at the end of the experiment , no subject reported being aware of any pairing between Si and Sd indicating the effectiveness of the cover task; the Si–Sd association was implicit . No subject reported being aware that the dummies did not appear in later phases . In the second ( ‘Reward’ ) phase , subjects were taught that some of the fractals ( Sd+ ) were worth money , while others ( Sd− ) were not . In each conditioning trial , subjects saw an Sd for 2000 ms , followed by an ISI of 1500 ms , and then either a reward ( image of a one pound sterling coin ) or no-reward ( blue square ) for 2000 ms , followed by an ITI of 3000 ms . Each Sd appeared 18 times , for a total of 108 trials . Sd− were never rewarded , while Sd+ were rewarded 14 out of 18 times that they appeared . The cover task was to press one button for any Sd or for no-reward , and a different button for reward ( meaning that in an unrewarded trial , the same button was to be pressed twice; while in a rewarded trial two different buttons should be pressed ) . Pressing the correct button to ‘pick up’ the coin led to actually receiving this money at the end of the experiment ( divided by a constant factor of ten ) ; subjects were informed of this . Through the unique pairing between Si and Sd , conditioning implicitly established Si+ ( previously paired with Sd+ ) and Si− ( previously paired with Sd− ) . The pairing was such that each Si category contained one Si+ and one Si− . In the third ( ‘Decision’ ) phase , in each trial subjects made a pairwise choice between either two Sd images or two Si images . The two Si images were always of the same category ( face/body/scene ) : one Si+ and one Si−; likewise , the two Sd images , an Sd+ and an Sd− , had always been previously paired with the same Si category . Subjects were instructed that they would receive monetary reward for choosing the correct stimulus , but , as in Wimmer and Shohamy ( 2012 ) , were given no instructions about how to identify the correct stimulus ( except to choose the one they thought was more lucky ) . They actually received these rewards at the end of the experiment , again divided by ten . In addition to the money earned within the task , subjects received a flat compensation of £10 . Each pairwise choice was repeated 4 times for a total of 24 trials . Any preference for Si+ over Si− would provide evidence of sensory preconditioning . After the experiment , subjects again provided subjective liking ratings on a visual analog scale , this time for each Si and Sd actually used in the experiment ( excluding dummies ) . Decision-phase preferences for Sd+ , Sd− , Si+ , and Si− were measured by averaging the four binary responses for each pair , and performing a one-sample t-test between subjects on the mean response against 50% . Similar results could be obtained by treating the first choice of each subject for each pair as an independent draw from a Bernoulli distribution and comparing the results to p = 0 . 5 . Changes in subjective liking ratings from Pre-Liking to Post-Liking phases were differences on an arbitrary scale ( pixels in the visual analog scale ) and were linearly de-trended as subjects showed a robust tendency to increase all ratings at the end of the experiment compared to the beginning ( many subjects reported in debriefing that they liked most of the stimuli more because they were more familiar at the end of the experiment ) . MEG was recorded continuously at 600 samples/second using a whole-head 275-channel axial gradiometer system ( CTF Omega , VSM MedTech , Canada ) , while participants sat upright inside the scanner . Continuous head localization was recorded with three fiducial coils at the nasion , left pre-auricular , and right pre-auricular points . The task script sent synchronizing triggers ( outportb in Cogent ) which were written to the MEG data file . A projector displayed the task on a screen ∼80 cm in front of the participant . Participants made responses on a button box using either thumbs or index fingers as they found most comfortable . All analysis was performed in MATLAB . Some analyses used SPM12b ( Wellcome Trust Centre for Neuroimaging , United Kingdom ) . Data were first converted to SPM12 format using spm_eeg_convert . Each event was then epoched , using spm_eeg_epochs , to 1000 ms segments from −400 ms to +600 ms relative to the event , based on the triggers recorded from the task script . All timings were corrected for one frame ( 1/60 s ) of lag between triggers and refreshing of the projected image , measured using a photodiode outside the task . The 600 samples in each epoch were then reduced to 50 time bins by averaging together each consecutive 12 samples . Thus , the time bins were spaced every 20 ms and represented the average raw signal of the 12 samples within that 20 ms . Pre-stimulus bins were treated as baseline . We built three-way classifiers for the category of the Si stimuli . Classifiers were trained based on the activity evoked by the presentation of the Si stimuli in the Association phase , and used to classify the activity associated with the presentation of the Sd and outcome stimuli in the Reward phase . Classifiers were built for each time bin following Si presentation , and tested on each time bin following Sd and outcome presentation during the Reward phase , giving rise to ( Association ) time* ( Reward ) time maps of classification performance . Support vector machine ( SVM ) classification analyses were performed with the svmtrain/svmpredict routines from libsvm ( National Taiwan University , Taiwan; http://www . csie . ntu . edu . tw/∼cjlin/libsvm ) . Each feature used for classification ( i . e . , a sensor at a time bin ) was independently z-transformed before classification . Results are reported with linear kernels . The regularization parameter C was tuned to optimize cross-validation performance in cross-validation of Association-phase data ( C = 105 ) but was then fixed for all further analyses . Cross-validation was tested using leave-one-out , k-fold ( 5 , 10 , or 20 ) , or repeated random subsampling ( 50 or 100 independent subsamples with 10% of samples left-out ) , without any difference in results between methods . In Figure 4 , we show 2-dimensional maps where the dimensions are times relative to two different events . To generate statistical significance thresholds for these maps , we recalculated these maps many times with independently shuffled category labels for the stimuli . Each shuffle yielded a map that contained no true information about the stimuli , but preserved overall smoothness and other statistical properties . The peak levels of each of these maps were extracted , and the distribution of these peak levels formed a nonparametric empirical null distribution . The 95th percentile of this distribution is reported as the significance threshold . Representational similarity between two different trials was measured by correlation between the patterns of activation over sensors , at the same time bin relative to stimulus onset . Classifiers trained on Association-phase data were used directly to predict Reward-phase data without any tuning to optimize cross-classification performance . All ( Association ) time* ( Reward ) time maps of classification performance were smoothed by a 2-D Gaussian kernel ( σ = 30 ms ) for display and for calculating peak-level shuffling statistics . In the interest of reporting our work as completely as possible , we discuss a set of analyses that were based on relevant hypotheses , but did not lead to significant results . 1 ) An important issue in the analysis of retrieved representations is to make sure that what are apparently retrieved representations are not in fact coincidences in the representation of the retrieved object and the retrieval cue . In the analyses in the main paper , this is controlled by randomizing Si–Sd pairings between subjects . We attempted another way of controlling for this , by training a classifier on all subjects' ( except one ) Sd-evoked data ( using the category labels of the associated Sis ) , and testing on the left-out subject . If this procedure , repeated across left-out subjects , would produce an above-chance prediction of the Si category associated with the displayed Sd , this would imply that the Sd-evoked data contain a real representation of Si . Unfortunately when we attempted this , the group-level prediction of Si category did not reach significance . We speculate this is because the category representation differs substantially between subjects ( supported by Sandberg et al . ( 2013 ) ) ; an issue that the analysis in the main paper is immune to because classifiers are trained separately for each subject . 2 ) Wimmer and Shohamy ( 2012 ) regressed their neural signal against within-category differences in behavioral preference . For example , if one subject in the Decision phase preferred the face paired with the rewarded fractal , but did not prefer the scene paired with the rewarded fractal , then he or she was more likely to have a large fusiform face area activation during presentation of the face-paired fractal in the Reward phase than to have a large parahippocampal place area activation during presentation of the scene-paired fractal . We attempted the same analysis but no correlation with neural decoding reached significance . In our hands collapsing within categories to look at between-subject variance in total value updating appeared more statistically powerful . Along similar lines , we also trained classifiers to distinguish individual stimuli in the Association phase ( e . g . , a particular face , rather than the category of faces—so the classifier learned about 12 distinct categories ) , and applied these classifiers to activity at the time of outcome in the Reward phase . The classifier was treated as ‘correct’ if it predicted the identity of the photograph that had been previously associated with the fractal presented on this trial of the Reward phase . We then correlated the resulting correctness ratings against the behavioral preference for Si+ over Si− in the Decision phase ( just as in Figure 6 of the main paper , but classifying individual stimuli rather than categories ) . However , these correlations did not reach shuffle-corrected significance . This may be a result of the difficulty of classifying many individual stimuli with relatively few trials . 3 ) We wondered if , when photos ( Si ) were presented during the Decision phase , it would be possible to identify neural signals containing information about the paired fractal ( Sd ) . It is possible that this could represent an online retrieval of value information about Sd to guide the choice about Si . However , we could not detect above-chance classification of either associated Sd when pairs of Si were presented during the Decision phase . We suspect the patterns of representation may be more difficult to disentangle when two stimuli are shown on-screen at the same time . | Seeing an object triggers a complex and carefully orchestrated dance of brain activity . The spatial pattern of the brain activity encoding the object can change multiple times even within the first second of seeing the object . These rapid changes appear to be a core feature of how the brain understands and processes objects . Yet little is known about how these patterns unfold through time when we remember an object . Remembering , or retrieving information about objects , is how we use our knowledge of the world to make good decisions . It is not clear whether , during remembering , there are rapid changes in the patterns similar to those that happen when directly seeing an object . Mapping brain activity during remembering could help us understand how stored information can guide decisions . Using recently developed methods in brain imaging and statistics , Kurth-Nelson et al . found that two distinct patterns of brain activity appeared when viewing particular objects . One occurred around 200 milliseconds after viewing an object , and the other appeared a bit later , by about 400 milliseconds . Later , when remembering the object , these patterns reappeared in the brain , but at different points in time . Furthermore , these two patterns had distinct roles in learning associated with the objects to guide later decisions . This work shows that rapid changes in the pattern of neuronal activity are central to how stored information is retrieved and used to make decisions . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"neuroscience"
] | 2015 | Temporal structure in associative retrieval |
Female Aedes aegypti mosquitoes are deadly vectors of arboviral pathogens and breed in containers of freshwater associated with human habitation . Because high salinity is lethal to offspring , correctly evaluating water purity is a crucial parenting decision . We found that the DEG/ENaC channel ppk301 and sensory neurons expressing ppk301 control egg-laying initiation and choice in Ae . aegypti . Using calcium imaging , we found that ppk301-expressing cells show ppk301-dependent responses to water but , unexpectedly , also respond to salt in a ppk301-independent fashion . This suggests that ppk301 is instructive for egg-laying at low-salt concentrations , but that a ppk301-independent pathway is responsible for inhibiting egg-laying at high-salt concentrations . Water is a key resource for insect survival and understanding how mosquitoes interact with water to control different behaviors is an opportunity to study the evolution of chemosensory systems .
A female Ae . aegypti mosquito must take a blood-meal to develop a batch of eggs . Many strains of Ae . aegypti are anthropophilic , meaning that they target humans as their preferred blood source ( Ponlawat and Harrington , 2005; McBride et al . , 2014 ) . Once proteins and other nutrients in the blood have been converted into eggs , a female mosquito must find a suitable body of freshwater to lay these eggs ( Day , 2016 ) . Larval and pupal stages of Ae . aegypti are aquatic and thus the choice of egg-laying site is a primary determinant of offspring survival ( Christophers , 1960; Powell and Tabachnick , 2013 ) . In the field , Ae . aegypti often lay eggs in small containers of freshwater , such as drinking water storage containers , discarded tires , clogged gutters , and other by-products of human settlement , likely contributing to their pernicious ability to adapt and colonize human settlements across the globe ( Kraemer et al . , 2015 ) . The act of choosing an egg-laying site in many insects , including Ae . aegypti , is a deliberate and evaluative behavior . A female Ae . aegypti mosquito with a fully developed batch of eggs will use elevated humidity and bacterial volatiles to locate water at a distance ( Bentley and Day , 1989; Barbosa et al . , 2010 ) . Once in close-range , a mosquito contacts water to evaluate its suitability for egg-laying by sensing a wide variety of cues that include salinity , food , bitter toxins , and animal-derived chemical signals indicating the density of conspecific larvae and pupae or the presence of predators ( Hudson , 1956; Benzon and Apperson , 1988; Bentley and Day , 1989; Zahiri and Rau , 1998; Pamplona et al . , 2009; Afify and Galizia , 2015; Day , 2016; Zuharah et al . , 2016 ) . The physiological mechanisms by which a mosquito senses the presence of water and evaluates its composition to safely nurture her offspring remain unknown . Here , we show that the DEG/ENaC channel ppk301 is required for mosquitoes to exploit freshwater egg-laying substrates . When ppk301 mutant females contact water , they do not lay eggs as readily as wild-type animals and are more likely to make aberrant decisions between freshwater and saltwater at concentrations that impair offspring survival . We developed a CRISPR-Cas9-based genetic knock-in strategy to build genetic tools for labeling and imaging molecularly defined populations of neurons and generated a reporter line expressing the QF2 transcriptional activator from the endogenous ppk301 locus . We found that ppk301 is expressed in sensory neurons in legs and proboscis , appendages that directly contact water , and that sensory afferents from these ppk301-expressing neurons project to central taste centers . Using in vivo calcium imaging with the genetically encoded calcium sensor GCaMP6s at the axonal terminals of ppk301-expressing neurons in the ventral nerve cord , we found that ppk301 neurons respond to freshwater and that this response was almost entirely abolished in ppk301 mutant animals . Surprisingly , these projections were also activated by salt in both wild-type and ppk301 animals . We propose that ppk301 drives egg-laying at low-salt concentrations and that egg-laying is inhibited by a ppk301-independent pathway at high-salt concentrations .
We observed Ae . aegypti egg-laying behavior in the laboratory to understand the range of sensory information that is integrated to guide the final decision to lay an egg . A female Ae . aegypti mosquito with a fully developed batch of eggs will first evaluate a potential egg-laying site by physical contact with some combination of sensory tissues that include all three pairs of legs and her proboscis . If a site is deemed suitable , she will lay eggs singly on the moist substrate above the waterline ( Hudson , 1956 ) ( Figure 1A , Video 1 ) . When we used a mesh barrier to block access to liquid water , females did not lay eggs , even though they had free access to the moist filter paper substrate ( Figure 1B , C ) . Together , this suggests that direct contact with liquid water is necessary to stimulate egg-laying in Ae . aegypti and we set out to probe the molecular and cellular basis of this water-sensation to better understand this critical mosquito behavior . As a consequence of their global spread , Ae . aegypti are faced with diverse habitats with a wide variety of potential egg-laying sites . For example , they can be found in abundance in a number of coastal regions rich with standing saltwater ( Ramasamy et al . , 2014; de Brito Arduino et al . , 2015 ) . To mimic the choice between freshwater and seawater in the lab , we developed a two-choice assay in which individual blood-fed females were placed in a container with a divided Petri dish filled with deionized water on one side and varying concentrations of a chemically defined artificial seawater solution ( Kester et al . , 1967 ) on the other ( Figure 1D , Supplementary file 1 ) . Mosquitoes showed no significant preference between deionized water and dilute seawater up to 10% , with individual mosquitoes either picking a solution at random or distributing their eggs between both solutions ( Figure 1D ) . However , they showed a strong dose-dependent aversion to higher concentrations of seawater , with an IC50 of 12 . 25% seawater ( Figure 1D , E ) . Females showed near-complete aversion to 25–100% seawater ( Figure 1D , E ) . These choices have consequences for the offspring . When we measured survival of offspring reared in varying concentrations of seawater , we found dose-dependent lethality ( LD50 = 25 . 23% ) ( Figure 1E ) . To simplify the stimulus , we used sodium chloride ( NaCl ) , the predominant salt in artificial seawater , in all subsequent experiments . Females showed dose-dependent inhibition of egg-laying with increasing concentrations of NaCl ( Figure 1F ) when they were only given access to a single concentration , suggesting that the preference for freshwater may be driven in part by an aversion to laying eggs in saltwater . Similar to artificial seawater , NaCl produced a dose-dependent decrease in offspring survival ( Figure 1G ) . This demonstrates that the female mosquito’s choice of freshwater or saltwater correlates with the survival of her offspring , making this an essential decision for the propagation and fitness of the species . In a search for genes that control Ae . aegypti freshwater egg-laying , we reasoned that animals carrying a mutation in an egg-laying preference gene would fail to lay eggs in freshwater , inappropriately lay in saltwater , or both . pickpocket ( ppk ) genes are members of the Degenerin/ENaC channel superfamily and encode cation channel subunits that function as putative mechanoreceptors and chemoreceptors for a wide array of stimuli including pheromones , liquid osmolality , and salt ( Adams et al . , 1998; Chalfie , 2009; Cameron et al . , 2010; Chen et al . , 2010; Zelle et al . , 2013 ) . Through manual curation of geneset annotations , we predict that the Ae . aegypti genome ( Matthews et al . , 2018 ) contains 31 ppk genes . These are unevenly distributed across its three chromosomes . Whereas chromosome 1 has only three ppk genes , chromosome 3 has 22 , including three clusters of 3–7 ppk genes ( Figure 2A ) . The predicted protein products of many ppk genes have clear 1-to-1 orthologues in Drosophila melanogaster and the malaria mosquito Anopheles gambiae , while others are in species-specific expansions ( Figure 2B ) . Examination of tissue-specific transcript abundance using previously published RNA-seq data ( Matthews et al . , 2016; Matthews et al . , 2018 ) reveals a broad range of expression patterns in adult tissues , including in legs , proboscis , and other sensory tissues ( Figure 2C ) . In a previous study ( Kistler et al . , 2015 ) , we generated CRISPR-Cas9 mutations in four ppk genes ( marked with arrowheads in Figure 2A , B ) that we considered as candidates for controlling egg-laying . We first measured the ability of these mutant strains to blood-feed and found that all four were attracted to and engorged fully on the blood of a live human host ( Figure 3A ) . To test egg-laying behavior , we introduced single blood-fed females of each strain into egg-laying vials containing a small amount of water and a filter paper as an egg-laying substrate . One mutant , ppk301 , an orthologue of the Drosophila melanogaster ppk28 low-osmolality sensor ( Cameron et al . , 2010; Chen et al . , 2010 ) , showed a defect in egg-laying . Fewer than 40% of ppk301 mutants laid more than 10 eggs ( Figure 3B ) . To exclude the possibility that this defect was due to an inability to convert blood into developed embryos , we counted the number of mature eggs in ovaries and confirmed that there was no difference between wild-type and ppk301 mutants ( Figure 3C ) . To investigate freshwater egg-laying preference in the ppk mutants , we introduced single blood-fed female mosquitoes into individual chambers containing two Petri dishes , one filled with freshwater ( 0 mOsm/kg NaCl ) and the other with 200 mOsm/kg NaCl . Even when ppk301 mutant animals laid eggs , they laid fewer eggs than wild-type on water and more eggs than wild-type on 200 mOsm/kg NaCl ( Figure 3D ) . We assayed these four ppk mutant strains across a range of NaCl concentrations and found a significant reduction in aversion to salt solution only in the ppk301 mutant , as measured by the proportion of animals laying eggs primarily on salt solution ( Figure 3D–F ) . Together , these data support the conclusion that mutations in ppk301 disrupt freshwater egg-laying in two distinct ways: by dramatically reducing the drive to lay eggs in suitable , low-salt , substrates and reducing the aversion to concentrations of NaCl that are lethal to their offspring . If ppk301 mutant animals fail to sense water , which normally triggers mosquitoes to lay an entire clutch of eggs in a short timespan , we hypothesized that these mutants would show a delay in onset and a reduced egg-laying rate when housed in close proximity to water for many days . To ask if ppk301 mutant animals will lay eggs on water given sufficient time , we introduced individual blood-fed females into egg-laying vials containing water and monitored the number of eggs laid per female per day over 7 days . While the vast majority of wild-type and ppk301 heterozygous animals laid all their eggs within the first 2 days of being introduced into egg-laying vials , ppk301 mutant animals did not . Instead , the mutants showed increased variability in the time of egg-laying initiation and a tendency to spread egg-laying out over many days ( Figure 3G–J ) . This slow , sustained , and variable egg-laying behavior on freshwater is consistent with a defect in sensing the water that triggers egg-laying . If ppk301 directly senses the osmolality or salinity of liquid , we would expect it to be expressed in the sensory appendages that contact water . At the inception of this project , genetic tools for labeling , monitoring , and manipulating neurons in Ae . aegypti did not exist . To address this gap , we developed new genetic tools in the mosquito to label all ppk301-expressing neurons and to image neuronal activity at sensory neuron terminals using the genetically encoded calcium sensor GCaMP6s ( Chen et al . , 2013 ) . We adapted an approach in which a T2A ‘ribosomal skipping’ peptide is used to express multiple independent protein products from a single RNA transcript ( Diao and White , 2012; Daniels et al . , 2014 ) . We first tested the efficiency of T2A in Ae . aegypti by generating a transgene containing a membrane-targeted mCD8:GFP fusion protein and a nuclear-targeted dsRed:NLS fusion protein separated by T2A and driven from the Ae . aegypti polyubiquitin promoter ( Anderson et al . , 2010 ) ( Figure 4A ) . Confocal microscopy revealed complete subcellular separation of the two fluorophores in individual larval body-wall cells ( Figure 4A ) . This demonstrates that T2A functions efficiently to prevent peptide bond formation in Ae . aegypti and can be used as a tool to independently express multiple gene products from a single locus . To build a flexible system for expressing a diverse array of effector transgenes , we employed the Q-binary expression system for transgene amplification ( Potter et al . , 2010 ) , which has been successfully implemented in An . gambiae malaria mosquitoes ( Riabinina et al . , 2016; Afify et al . , 2019 ) . We used CRISPR-Cas9 with the same guide RNA used to generate the ppk301 mutant to introduce an in-frame T2A sequence into the ppk301 locus followed by the QF2 transcriptional activator ( Figure 4B ) . This is predicted to cause a loss-of-function mutation in ppk301 and also express QF2 in all ppk301-expressing cells . We also generated two transgenic QUAS effector strains . The first is a QUAS response element driving the expression of both cytosolic dTomato and GCaMP6s ( 15x-QUAS-dTomato-T2A-GCaMP6s ) , which allows us to simultaneously label neurons and image their activity with the genetically encoded calcium sensor GCaMP6s ( Chen et al . , 2013 ) ( Figure 4C ) . The second drives expression of membrane-bound GFP ( 15x-QUAS-mCD8:GFP ) , allowing us to reveal the complete morphology of the neurons in which it is expressed ( Figure 4I ) . We looked for expression of the ppk301>dTomato-T2A-GCaMP6s reporter in the appendages that contact water during egg-laying ( Figure 4D ) and found that sensory neurons innervating trichoid sensilla in the labellar lobes of the proboscis ( Figure 4E ) were labeled . We also found labeling in the legs , primarily in the distal tarsal segments ( Figure 4F ) . The mosquito central nervous system consists of a brain and a ventral nerve cord ( Ito et al . , 2014; Smarandache-Wellmann , 2016; Court et al . , 2017 ) ( Figure 4G ) . To facilitate our understanding of the neuroanatomy of Ae . aegypti in general , and the projection pattern of the ppk301 driver line in particular , we built a three-dimensional female mosquito brain atlas , in which we identified and named the major neuropils in accordance with nomenclature established by the Insect Brain Name Working Group ( Ito et al . , 2014 ) ( Figure 4H and http://mosquitobrains . org ) . Projections of ppk301-expressing sensory neurons in the head extend processes specifically to the subesophageal zone ( Figure 4H , J ) . Two nerves enter each hemisphere of the subesophageal zone , arising from the proboscis and the pharynx ( Figure 4J ) . Additionally , each leg sends projections into the ventral nerve cord , with nerves running into each neuromere ( Figure 4K ) . Both brain and ventral nerve cord innervation patterns are consistent with these neurons mediating taste sensation ( Scott , 2018 ) . We noted that projections of ppk301-expressing neurons are also present in the male brain and ventral nerve cord ( Figure 4—figure supplement 1A–C ) , consistent with a role for ppk301 in behaviors other than egg-laying . Labeling in both males and females was absent in genetic controls expressing either QF2 driver or QUAS effector alone ( Figure 4—figure supplement 1D–K ) . We hypothesized that if ppk301-expressing neurons were promoting egg-laying , they should be maximally activated by water , and we set out to test this by functional calcium imaging in response to freshwater and behaviorally relevant concentrations of NaCl . We developed an in vivo calcium imaging preparation with GCaMP6s ( Chen et al . , 2013 ) in ventral nerve cord sensory afferents of a mosquito presented with water or NaCl solutions on a single foreleg ( Figure 5A , B ) . This appendage was chosen because it most frequently contacts water during egg-laying ( Figure 1A , Video 1 ) . We imaged the prothoracic segment of the ventral nerve cord with two-photon microscopy , using a custom fluidics device to deliver and retract liquids to the foreleg and compared responses to different stimuli within individual animals . ppk301-expressing neuronal projections in the ventral nerve cord were identified by dTomato expression ( Figure 5C ) , which was also used to determine the region of interest for calcium imaging analysis . We observed low GCaMP6s fluorescence at baseline and an increase in fluorescence in every trial where water was presented ( Figure 5C–E ) , with no apparent desensitization across trials ( data not shown ) . Only the ipsilateral side of the ventral nerve cord innervated by the stimulated leg showed activation ( data not shown ) . In D . melanogaster , cells that express the ppk301 orthologue ppk28 respond to water but are inhibited by high osmolality solutions including NaCl ( Cameron et al . , 2010; Chen et al . , 2010; Jaeger et al . , 2018 ) . The population response in ppk301-expressing afferents in the mosquito is functionally different , showing robust responses to water and strong activation by salt solutions ( Figure 5D–F ) . To determine whether ppk301 is required for the neuronal responses to water or salt , we performed imaging experiments in a ppk301 mutant background and found that the response to water is almost entirely abolished in the ppk301 mutant ( Figure 5D–F ) . In contrast , 200 and 300 mOsm/kg NaCl solutions still elicited a strong response in the ppk301 mutant , albeit with a small reduction in amplitude ( Figure 5D–F ) . These data show that ppk301 is required for the activation of these neurons by freshwater and that a ppk301-independent pathway activates these neurons in response to salt . This suggests that mosquito ppk301-expressing afferents are either multi-modal ( Zocchi et al . , 2017 ) and each neuron responds to both water and salt , or that there is a functionally heterogeneous population of ppk301-expressing afferents with distinct neurons responding to water and salt . To test if ppk301-expressing neurons are activated by any solution with increased osmolality , we performed imaging experiments with two additional solutes that are not ionic salts , L-serine and D- ( + ) -cellobiose . We delivered all stimuli at 200 mOsm/kg , the osmolality of NaCl that elicited the peak response . We found that ppk301-expressing neurons did not respond more strongly to L-serine or D- ( + ) -cellobiose than they did to water alone ( Figure 6A–B ) . Thus , activity in these neurons does not only track with osmotic pressure . Intrigued by these findings , we performed egg-laying preference assays where mosquitoes were given a choice to lay eggs on freshwater and either L-serine or D- ( + ) -cellobiose . Although female mosquitoes found 200 mOsm/kg NaCl highly aversive , they did not discriminate between freshwater and L-serine or D- ( + ) -cellobiose at 200 mOsm/kg ( Figure 6C–D ) . Thus , the behavioral aversion to salt is unrelated to osmotic pressure . The observation that ppk301-expressing cells are activated by both water and salt is intriguing because wild-type females avoid laying eggs specifically in high-salt solutions . To understand how female mosquitoes interact with these different stimuli , we monitored real-time behavior of individual females offered either water or 300 mOsm/kg NaCl over 40 min by scoring their contact with liquid and individual egg-laying events . Both wild-type and ppk301 mutant mosquitoes contacted water , but only wild-type females consistently transformed these touches into egg-laying events ( Figure 7A ) . When offered 300 mOsm/kg salt solution , both wild-type and ppk301 mutant mosquitoes touched liquid , but neither genotype reliably laid eggs ( Figure 7A ) . These results are consistent with a model ( Figure 7B ) in which activation of ppk301 cells by water is a permissive signal for rapid and reliable egg-laying . Animals lacking ppk301 fail to detect the water activation signal and show delayed and intermittent egg-laying . Activation of ppk301-expressing neurons by high salt is not sufficient to drive egg-laying , suggesting that ppk301-expressing cells gate egg-laying at low NaCl concentrations , but as NaCl concentrations increase , an independent noxious salt-sensing pathway is recruited that overrides the activation of the tarsal ppk301-expressing cells imaged in this study to inhibit egg-laying ( Figure 7B ) . We predict mutations that disrupt this noxious salt sensor would yield mosquitoes that show indiscriminate egg-laying on a high-salt substrate .
Ae . aegypti mosquitoes preferentially lay eggs in freshwater and avoid saltwater . We annotated the ppk ion channel gene family in Ae . aegypti and made targeted mutations in several candidate genes , identifying a mutation in ppk301 that disrupts freshwater egg-laying preference and behavior . ppk301 is expressed in sensory neurons of the tarsi and proboscis , tissues that contact water during egg-laying . ppk301-expressing tarsal sensory neurons are activated by both water and high salt . This response is not driven solely by osmotic pressure , because two other solutes L-serine and D- ( + ) -cellobiose presented at the same osmolality were indiscriminable from water . In a ppk301 mutant background , responses to water are almost entirely abolished but responses to salt remain . Together , these data suggest a model in which activation of ppk301-expressing neurons by water is required for egg-laying while a ppk301-independent pathway encodes aversion to high concentrations of salt ( Figure 7B ) . This study identifies an important component of the circuitry regulating egg-laying behavior and provides an entry point into understanding the most important parenting decision a female mosquito makes . What sensory cues and modalities activate ppk301-expressing neurons ? Sensory neurons in the D . melanogaster proboscis that express the orthologous ppk28 gene are maximally activated by water but inhibited by dissolved solutes ( Cameron et al . , 2010; Chen et al . , 2010; Jaeger et al . , 2018 ) . In contrast , the afferents of the ppk301-expressing cells in the mosquito tarsi are not tuned specifically to osmolality but rather encode multimodal responses to liquid ( ppk301-dependent ) and to high concentrations of salt ( ppk301-independent ) . Whether or not individual neurons within this population exhibit distinct tuning properties remains to be determined . In addition to the tarsi , ppk301 is expressed in the proboscis and pharynx . It will be important to evaluate whether these populations of ppk301-expressing cells are strictly tuned to water , as are D . melanogaster ppk28-expressing labellar sensory neurons , and lack the multimodal responses seen in the tarsal ppk301-expressing neurons . How does ppk301 regulate egg-laying ? We favor a model in which ppk301-dependent activation of sensory neurons by liquid provides a permissive signal required for egg-laying , while high concentrations of salt activate noxious salt pathway ( s ) that can override this signal and prevent egg-laying . Together , these opposing signals are integrated to sculpt egg-laying preference and allow a mosquito to exploit freshwater substrates while preventing eggs from being laid on high-salt substrates that are lethal to their offspring . It will be interesting to discover the biophysical mechanisms by which ppk301 neurons respond to water . ppk301 encodes a single subunit of a trimeric DEG/ENaC channel ( Jasti et al . , 2007 ) . One possibility is that ppk301 ion channels are homomers that respond to osmotic pressure , similar to D . melanogaster ppk28 ( Cameron et al . , 2010; Chen et al . , 2010 ) . Alternatively , ppk301 could encode a promiscuous subunit that could be part of multiple heteromeric PPK channels with different ligands that account for some of the multimodal tuning properties seen in ppk301-expressing tarsal sensory neurons . In this scenario , genetic ablation of ppk301 could result in a novel set of PPK channels that are composed of the remaining subunits , thereby shifting the composition of channels on the membrane and influencing overall neuronal tuning properties . Water- and salt-sensing are critically important because all animals must regulate fluid and ion homeostasis for the proper functioning of their physiological systems . Terrestrial insects such as mosquitoes are small and have a large surface-area-to-volume ratio , meaning that they are susceptible to desiccation . Therefore , both male and female adult mosquitoes must seek out liquid for survival . It is not known if sensory systems used for determining the physical properties of liquids during fluid ingestion in male and female mosquitoes are the same as those used by the female for evaluating water quality during egg-laying . Our data show that ppk301 is expressed in both male and female appendages , and that neurons expressing ppk301 project to similar areas of the brain and ventral nerve cord . Studying the effects of ppk301 mutations on liquid ingestion in adult Ae . aegypti will provide insight into whether the same molecules and sensory neurons are being used in distinct behavioral contexts . It will also be interesting to investigate if the activity of ppk301 neurons is modulated by physiological state , including the presence of developed eggs or elevated thirst due to dehydration or desiccation stress . Previous studies have shown the existence of a canonical ‘water’ cell in the apical labral chemoreceptor sensilla of Ae . aegypti ( Werner-Reiss et al . , 1999 ) , but we do not see ppk301 expression in this tissue , indicating that there may be additional osmosensitive receptors encoded in the mosquito genome . A complete cellular and molecular understanding of salt sensation in mosquitoes remains to be elucidated and the ‘noxious salt sensor’ proposed in our model remains to be discovered . We note that a distributed coding of salt taste in which distinct populations of sensory neurons are modulated by salt is not unprecedented . A recent study in Drosophila reveals that in the proboscis , salt can activate or inhibit every class of gustatory receptor neuron ( Jaeger et al . , 2018 ) . Salt is also a key component of blood and understanding the mechanisms of salt sensation during blood-feeding , egg-laying , and other critical mosquito behaviors is an important future goal ( Werner-Reiss et al . , 1999; Sanford et al . , 2013 ) . The evolution of egg-laying preference both within Ae . aegypti and in other mosquito species is a critical determinant of the types of environments and ecological niches that they are able to exploit . Indeed , some strains of Ae . aegypti have begun to exploit brackish water for egg-laying ( Ramasamy et al . , 2014; de Brito Arduino et al . , 2015 ) . Such an adaptation requires co-evolution of the shifting preference of females for salty water and larval tolerance of high salinity environments into which they are deposited . Investigating the underlying genetic changes that support the coordination of these shifting phenotypes in both larvae and adults will be important for understanding how Ae . aegypti adapts to novel breeding sites across the globe , which has greatly increased their capacity as a disease vector . The advances described here depended critically on new genetic tools we developed that did not previously exist in this non-model insect . These techniques now make it possible to visualize neuronal anatomy and activity within molecularly identified cell types and to identify the genetic and neural circuit substrates of many mosquito behaviors . These behaviors contribute to the spread of deadly pathogens and understanding their underlying biology could contribute to mosquito control efforts . The development of similar tools and reagents in other insect species of public health , agricultural , or ethological interest will broaden our view of insect biology and facilitate comparative studies of the genes and circuits underlying evolutionary adaptations in insects .
Aedes aegypti wild-type laboratory strains ( Liverpool ) were maintained and reared at 25–28°C , 70–80% relative humidity with a photoperiod of 14 hr light: 10 hr dark as previously described ( DeGennaro et al . , 2013 ) . Adult mosquitoes were provided constant access to 10% sucrose . For routine strain maintenance , animals were blood-fed on live mice or human subjects . Adult females of all genotypes were blood-fed on a single human subject during mutant generation , for behavioral assays , and for calcium imaging . Blood-feeding procedures with live hosts were approved and monitored by The Rockefeller University Institutional Animal Care and Use Committee ( IACUC protocol 15772 ) and Institutional Review Board , ( IRB protocol LV-0652 ) . Human subjects gave their written informed consent to participate . All Ae . aegypti mosquitoes used in egg-laying assays were housed in mixed-sex cages and were between 7 and 14 days old . Female mosquitoes were blood-fed by giving them direct access to a human limb introduced in or on the mesh wall of a cage . Fully engorged female mosquitoes were selected within 24 hr of blood-feeding and housed in insectary conditions with ad libitum access to 10% sucrose before being introduced into behavior assays . Unless otherwise indicated , assays were performed beginning 96 hr post-blood-meal and continuing for ~18 hr . 10 blood-fed female mosquitoes were introduced by mouth aspiration into a standard BugDorm rearing cage ( 30 cm3 ) containing a 10% sucrose wick . Two egg-laying cups were present in each cage , each containing deionized water and a 55 mm-diameter partially submerged Whatman filter paper ( Grade 1; 1001–055 ) . In one of the cups , a metal mesh barrier placed ~1 cm above the water line prevented direct access of the mosquitoes to liquid , although the filter paper remained partially submerged and moist . Mosquitoes were allowed to lay eggs overnight ( 18 hr ) , after which the filter papers were dried and eggs counted . Artificial seawater was mixed according to an established recipe ( Kester et al . , 1967 ) with lab-grade chemicals from Sigma-Aldrich . Specific dilutions were made , by volume , with deionized water . Single blood-fed females were introduced into a chamber comprised of a length of transparent acrylic tubing ( inner diameter 9 . 525 cm ) cut to 12 cm in height with a wire mesh grid glued to one end as a ceiling containing a two-sector 90 mm divided Petri dish with 10 mL of deionized water on one side and 10 mL of a specific dilution of artificial seawater on the other . As an egg-laying substrate , a 1 cm tall strip of seed germination paper ( Anchor Paper; SD7615L ) was wrapped around the outer diameter of each half of the Petri dish , partially submerged in the liquid . Animals were introduced by a mouth aspirator through a hole in the top of the chamber , which was then plugged with a cotton ball . Containers were stored under insectary conditions and mosquitoes were allowed to lay eggs overnight ( ~18 hr ) , after which the paper strips were dried and eggs counted . Wild-type mosquitoes were hatched in ‘hatch broth’ consisting of deoxygenated water containing ground Tetramin fish food . Approximately 1 day after hatching , 20 larvae were transferred into a small plastic cup containing 25 mL of a specific dilution of artificial seawater . Cups were examined each day , dead larvae removed , and ground Tetramin pellets added for food as needed . Animals successfully completing the transition to pupal stage by 8 days post-hatch were scored as surviving offspring . A multi-animal egg-laying assay was created out of sheet acrylic , comprising modular trays each with 14 single-animal chambers . Each chamber comprised two angled 50 mm Petri dishes each containing 2 mL of liquid and a 47 mm diameter filter paper ( Whatman , Grade 1 Qualitative Filter Paper ) . We next developed a standardized imaging setup to automatically count eggs from each Petri dish . Each dish was placed onto a LED light panel ( SuperbrightLEDs . com item #2020 ) and images were captured with a Raspberry Pi and associated camera . Images were thresholded and pixels counted for each dish of each chamber . The number of eggs corresponding to each image was determined by dividing the average pixel value of eggs determined from a manually-counted set of 20 test images . The concordance between manual counting and automatic pixel-based counting was r2 = 0 . 96 . A parts list and schematics of the egg-laying chambers and the image acquisition and thresholding code can be found on http://github . com/VosshallLab/MatthewsYoungerVosshall2018/ ( Vosshall Lab , 2019; copy archived at https://github . com/elifesciences-publications/MatthewsYoungerVosshall2018 ) . For studies of the effect of NaCl on egg production ( Figure 1F ) , each dish was filled with the same concentration of NaCl . The osmolality of each solution was measured using a Wescor model 5520 vapor pressure osmometer . Blood-fed female mosquitoes were cold anesthetized and single animals introduced into each chamber by mouth aspiration . Animals were allowed to lay for 18 hr and dishes were imaged to calculate egg numbers . For two-choice assays ( Figure 3D–F ) , assays were performed identically , except that the two dishes contained deionized water and a NaCl solution of a specific osmolality . In Figure 6C–D , the two dishes contained deionized water and either 200mOsm/kg L-serine ( Sigma S4500 ) or D- ( + ) -cellobiose ( Sigma 22150 ) . The position of each solution was varied for each chamber . Experiments were performed anonymized to genotype and data included only for those animals who laid more than 10 eggs . Wild-type mosquitoes were hatched in hatch broth . Approximately 1 day after hatching , 20 larvae were transferred into a small plastic cup ( VWR HDPE Multipurpose Containers; H9009-662 ) containing 25 mL of a specific concentration of NaCl , prepared as above . Cups were examined each day , dead larvae removed , and ground Tetramin pellets added for food as needed . The number of pupae and larvae remaining alive were counted each day , and cumulative survival was calculated for these offspring . To measure the timing of egg-laying across days , individual blood-fed mosquitoes were introduced into egg-laying vials 48 hr after a blood-meal . Mosquitoes were transferred to a fresh vial every 24 hr and eggs from the previous day were counted and recorded . Experiments were performed anonymized to genotype . To measure the timing of egg-laying and liquid touching , single animals were introduced into a 50 mL cell culture flask containing 10 mL of either deionized water or 300 mOsm/kg NaCl , and a 1’ x 2’ strip of seed germination paper , partially submerged . Animals were video recorded for 40 min using a Nikon D7000 SLR with a macro lens . Four flasks were recorded simultaneously , with each set containing 1 replicate of the following conditions: wild-type , 0 mOsm/kg NaCl; wild-type , 300 mOsm/kg; ppk301 , 0 mOsm/kg; ppk301 , 300 mOsm/kg . Videos were manually scored for physical contact with liquid and the appearance of freshly laid eggs . Data on physical touches were recorded in 5 s intervals , with each interval scored as ‘touch’ if a single frame revealed physical contact between any appendage of the mosquito and the liquid . Videos were scored anonymized to genotype and condition . To identify members of the ppk gene family in Ae . aegypti and An . gambiae , we performed two complementary analyses: 1 ) using D . melanogaster ppk sequences as queries , we performed BLASTp against all translated protein-coding genes identified in AaegL5 ( GCF_002204515 . 2_AaegL5 . 0_protein . faa ) and 2 ) ran interproscan v5 . 29 . 68 . 0 ( Finn et al . , 2017 ) against the same database of translated protein-coding genes . We took all genes that were reciprocal best hits with the D . melanogaster ppk family via blastp and were annotated by interproscan as ‘Amiloride-sensitive sodium channel . ’ We repeated this analysis for the An . gambiae , PEST strain geneset downloaded from Vectorbase ( Anopheles-gambiae-PEST_PEPTIDES_AgamP4 . 9 . fa ) . We re-named Ae . aegypti ppk genes by giving them a three-digit identifier corresponding to their chromosomal position . The first digit represents the chromosome ( 1 , 2 , or 3 ) , while the next two digits represent its relative position on that chromosome from the left ( p ) arm to the right ( q ) arm according to coordinates found on NCBI . We next built a phylogenetic tree to visualize the relationship between these genes across these three species . To do this , we selected the longest single isoform for genes predicted to encode multiple protein-coding isoforms and performed multiple sequence alignment using clustal-omega v1 . 2 . 3 ( Sievers et al . , 2011 ) , including a human ASIC channel and Caenorhabditis elegans MEC4 sequence for comparison . A maximum-likelihood-estimate phylogenetic tree was constructed using PhyML v3 . 0 ( Guindon et al . , 2010 ) with default parameters and 100 bootstrap iterations , and manually re-rooted on MEC4 for presentation . A table of all genes and sequences incorporated in this analysis , with previous accessions , is presented in the Supplementary file 1 . To visualize transcript abundance of each predicted ppk gene across tissues , we utilized published data ( Matthews et al . , 2016; Matthews et al . , 2018 ) . Heatmaps were generated as described ( Matthews et al . , 2018 ) and presented as log10 ( TPM +1 ) of the mean expression value for all replicates of the indicated tissue using the heatmap . 2 function of the gplots v3 . 0 . 1 ( Warnes et al . , 2016 ) package in R v3 . 5 . 0 ( 2017 ) . ppk loss-of-function alleles were generated and described previously ( Kistler et al . , 2015 ) . All new strains generated in this paper were injected into wild-type Liverpool embryos at the Insect Transformation Facility at the University of Maryland . PUb-mCD8:GFP-T2A-dsRed:NLS-SV40 was generated by Gibson assembly , using the following fragments: Plasmid backbone and MOS arms from a standard transformation vector ( pMos-3xP3-dsRed ) ; dsRed open reading frame amplified from the same vector by polymerase chain reaction ( PCR ) ; mCD8-GFP open reading frame PCR-amplified from a synthesized vector ( Genscript ) ; Ae . aegypti PUb promoter PCR-amplified from PSL1180-HR-PUb-ECFP ( Addgene #47917 ) . T2A and NLS sequences were added through PCR . 1000 embryos were injected with the plasmids and a Mos helper plasmid . Two stable lines were recovered with qualitatively similar expression patterns . ppk301-T2A-QF2 was generated through CRISPR-mediated homologous recombination into the endogenous locus . We initially attempted to use constructs with gene-specific promoter fragments to drive transgenic constructs , including the broadly expressed polyubitiquin promoter ( Anderson et al . , 2010 ) , but these either did not express at all or were expressed sporadically and mostly in non-neuronal cells ( Bui et al . , 2018 ) . We also had no success with the Gal4/UAS ( Brand and Perrimon , 1993 ) system in Ae . aegypti . We therefore developed techniques to use homologous recombination to knock QF2 into the ppk301 locus . ppk301-T2A-QF2 was generated with a sgRNA targeting exon 2 of the ppk301 locus ( ppk301-sgRNA-1 , target sequence with PAM underlined: GGTTGGCAGTTGAGTCCCGG ) . sgRNA DNA template was prepared by annealing oligonucleotides as described ( Kistler et al . , 2015 ) . In vitro transcription was performed using HiScribe Quick T7 kit ( NEB , E2050S ) following the manufacturer’s directions and incubating for 2 hr at 37°C . Following transcription and DNAse treatment for 15 min at 37°C , sgRNA was purified using RNAse-free SPRI beads ( Ampure RNAclean , Beckman-Coulter A63987 ) , and eluted in Ultrapure water ( Invitrogen , 10977–015 ) . The donor plasmid was constructed by Gibson assembly using the following fragments: homology arms of ~1 kb on either side of the Cas9 cut site ( two base pairs were deleted from the left arm immediately preceding the T2A to maintain the open reading frame ) ; a fragment containing T2A-QF2-SV40 and 3xP3-dsRed , PCR-amplified from a vector derived from pBac-DsRed-ORCO_9kbProm-QF2 ( gift of Chris Potter , Addgene #104877 ) ; a pUC57 vector backbone digested with EcoRI and HindIII . Clones were sequenced verified and midiprepped using an endotoxin free midiprep kit ( Machery-Nagel ) and eluted in Ultrapure nuclease-free water ( Invitrogen ) . 2000 embryos were injected with a mixture of 300 ng/µL Cas9 protein ( PNA Bio ) , 650 ng/µL dsDNA plasmid donor , and 40 ng/µL sgRNA . The progeny of 96 surviving G0 females were screened . Six potential founders were isolated , and one was verified to have a complete and in-frame insertion by PCR with the following primers ( Forward 5’ GTGAGGGTGGTGTCGAATTAACTCTT3’ , Reverse 5’GTTAGGTCAGAGGTATCCCTGAACAT3’ ) . 15x-QUAS-mCD8-GFP was generated from an existing plasmid ( a kind gift from Chris Potter , Addgene #104878 ) . Embryos were injected with the plasmid and a PBac helper plasmid . Two independent lines were recovered . 15x-QUAS-dTomato-T2A-GCaMP6s was generated by Gibson assembly of the following PCR-amplified fragments: Plasmid backbone and Mos arms from PUb-mCD8:GFP-T2A-dsRed:NLS-SV40 ( described above ) ; 15x-QUAS from pBAC-ECFP-15xQUAS_TATA-SV40 ( Addgene #104875 , gift of Chris Potter ) ; dTomato-T2A-GCaMP6s PCR-amplified from a vector synthesized ( Genscript ) . Embryos were injected with the construct and a PBac helper plasmid . Two independent lines were recovered . Dissection of adult brains and immunostaining was modified from previously used protocols ( Siju et al . , 2008; Riabinina et al . , 2016 ) . 6–14 day-old mosquitoes were anesthetized on ice . Heads were carefully removed from the body by pinching at the neck with sharp forceps . Heads were placed in a 1 . 5 mL tube for fixation with 4% paraformaldehyde , 0 . 1 M Millonig's Phosphate Buffer ( pH 7 . 4 ) , 0 . 25% Triton X-100 , and nutated for 3 hr . Brains were then dissected out of the head capsule in ice cold Ca+2- , Mg+2-free phosphate buffered saline ( PBS , Lonza , 17-517Q ) and transferred to a 24-well plate . All subsequent steps were done on a low-speed orbital shaker . Brains were washed in PBS containing 0 . 25% Triton X-100 ( PBT ) at room temperature six times for 15 min . Brains were permeabilized with PBS , 4% Triton X-100 , 2% Normal Goat Serum for ~48 hr ( two nights ) at 4°C . Brains were rinsed once then washed with PBT at room temperature six times for 15 min . Primary antibodies were diluted in PBS , 0 . 25% Triton X-100 , 2% Normal Goat Serum for ~48 hr ( two nights ) at 4°C . The primary antibodies used in this experiment were anti-dmBrp ( mouse; 1:50: NC82 , DSHB ) to label the synaptic neuropil and anti-GFP ( Rabbit: 1:10 , 000; A11122 , Life Technologies ) . Brains were rinsed once then washed in PBT at room temperature six times for 15 min . Secondary antibodies were diluted in PBS , 0 . 25% Triton X-100 , 2% Normal Goat Serum for ~48 hr ( two nights ) at 4°C . The primary antibodies used in this experiment were anti-mouse-Cy5 ( 1:250; Life Technologies A-10524 ) and anti-Rabbit-488 ( 1:500; Life Technologies A-11034 ) . Brains were rinsed once then washed in PBT at room temperature six times for 15 min . Brains equilibrated overnight in Vectashield ( Vector Laboratories H-1000 ) and were mounted in Vectashield . Six- to 14-day-old mosquitoes were anesthetized on ice and the bodies were carefully removed from the heads by pinching at the neck with sharp forceps . The bodies were placed in a 1 . 5 mL tube for fixation with 4% paraformaldehyde , 0 . 1M Millonig's Phosphate Buffer ( pH 7 . 4 ) , 0 . 25% Triton X-100 , and nutated for 3 hr . Ventral nerve cords were dissected out of the body in ice cold PBS and transferred to a 24-well plate . All subsequent steps were the identical to the brain immunostaining protocol described above . ppk301>mCD8:GFP expression was visualized in brains and ventral nerve cords using a Zeiss Inverted LSM 880 laser scanning confocal with a 25x/0 . 8 NA immersion-corrected objective . Glycerol was used as the immersion medium to most closely match the refractive index of the mounting medium Vectashield . Brains were imaged at 2048 × 2048 pixel resolution in X and Y with 0 . 5 µm z-steps for a final voxel size of 0 . 1661 × 0 . 1661×0 . 5 µm . Ventral nerve cords were imaged at 1024 × 1946 pixel resolution in X and Y with 0 . 5 µm z-steps for a final voxel size of 0 . 3321 × 0 . 3321×0 . 5 µm ventral nerve cord images were tiled and stitched with 10% overlap . Confocal images were processed in ImageJ ( NIH ) . 3xP3 was used as a promoter to express fluorescent markers for transgene insertion , and care was taken to distinguish 3xP3 expression from the expression of the ppk301-QF2 driver line . 3xP3 labels the optic lobes , as well as some cells in the dorsal brain . Figure 4J , K and Figure 4—figure supplement 1B , D , F , H , J were cropped to remove 3xP3 expression . In some cases , we saw projections from the optic lobes that traversed the brain . These were sometimes seen in the driver alone and effector alone controls , both of which contain a 3xP3 marker transgene . We speculate that these projections derive from cells marked by 3XP3 expression and are unrelated to ppk301 expression . These patterns of 3xP3 are worth noting , but due to the distance from the subesophageal zone and ventral nerve cord , they do not have an impact on the results of this study . In some brains , we observed faint unilateral projections from the subesophageal zone to the antennal lobe . We examined eight female brains and eight male brains for these projections and saw these projections in 5/8 brains . In males we did not see this projection in any of the eight brains examined . A reference Ae . aegypti brain from a 7 day old wild-type female mosquito was fixed and immunostained for Brp as described above . It was imaged as described above except with 1 μm voxels and as a tiled scan with 10% overlap . Thirty-nine brains were analyzed and the most complete and symmetric brain was chosen to serve as the template . Blind deconvolution was performed with AutoquantX3 software . The female brain was manually annotated using the segmentation and 3D reconstruction software ITK-SNAP . Regions were identified by the anatomical boundaries defined by Brp staining , and by homology to other insects , and named in accordance with revised insect brain nomenclature standards ( Ito et al . , 2014 ) . Structures without clear boundaries were excluded . A surface mesh of each region was exported into the data analysis and visualization software ParaView , which was used to generate the 3D reconstruction shown in Figure 4H . The reference brain raw data and reconstruction are available on the website http://mosquitobrains . org . We have displayed this data as a brain atlas by creating an online ‘Brain Browser’ tool where users can scroll through the brain and highlight different regions . Ae . aegypti brains that are stained with the synaptic protein Brp can be warped onto this standard reference Aedes aegypti brain using the python code ClearMap ( Renier et al . , 2016 ) . To use ClearMap , all images must be acquired with square voxels . The standard brain was imaged with 1 µm voxel size . To use it as a template , all images must either be taken at this resolution , or down sampled to a final voxel size of 1 µm . Additional channels can be warped and registered onto the reference brain , provided that one channel is imaged as described above . The previously characterized Ae . aegypti polyubiquitin promoter ( Pub ) ( Anderson et al . , 2010 ) was used to drive expression of both a membrane-bound variant of GFP ( mCD8:GFP ) and a nuclear localization sequence fused to dsRED by separating these genes with the T2A ribosomal skipping element ( Diao and White , 2012; Daniels et al . , 2014 ) . This construct was expressed in a few cells in the adult brain , as well as in larval body-wall cells . We focused our analysis on the body-wall cells because their flat and compact shape made them amenable to examining the expression of our membrane and nuclear proteins . Larvae were dissected by pinning the head and the tail to Sylgard plates ( DowDupont ) with insect pins , and a long longitudinal cut was made along the dorsal surface of the body wall . The body wall was filleted open with four additional insect pins , and the organs were removed . The animals were fixed for 25 min in 4% paraformaldehyde , 0 . 1 M Millonig's Phosphate Buffer ( pH 7 . 4 ) , 0 . 25% Triton X-100 at room temperature , and then rinsed 3 times in PBS . Larvae were transferred to a 1 . 5 mL tube and washed with PBT 6 times for 15 min while nutating at room temperature . Larvae were then incubated in Vectashield with DAPI overnight and mounted in Vectashield for imaging . Larvae were imaged on a Zeiss Inverted LSM 880 laser scanning confocal with a 40X/1 . 2 NA oil immersion objective . The cells were imaged at a resolution of 2048 × 2048 pixels in a single confocal slice for a pixel size of 0 . 0692 × 0 . 0692 µm . Images were processed in ImageJ ( NIH ) . To visualize sensory neuron cell bodies in ppk301-T2A-QF2 , 15xQUAS-dTomato-T2A-GCaMP6 animals we dissected live sensory tissues using fine forceps , dipped in cold methanol for ~5 s , and mounted on a slide in glycerol . Appendages were viewed on a Zeiss Inverted LSM 880 laser scanning confocal with a 25x/0 . 8 NA immersion corrected objective at a resolution of 1024 × 1024 pixels and a voxel size of 0 . 2076 µm x 0 . 2076 µm x 1 µm . Calcium imaging was performed on an Ultima IV two-photon laser-scanning microscope ( Bruker Nanosystems ) equipped with galvanometers and illuminated by a Chameleon Ultra II Ti:Sapphire laser ( Coherent ) . GaAsP photomultiplier tubes ( Hamamatsu ) were used to collect emitted fluorescence . Images were acquired with a 60X/1 . 0N . A . Long Working Distance Water-Immersion Objective ( Olympus ) at a resolution of 256 × 256 pixels . Calcium imaging experiments were performed on female mosquitoes that were 7–11 days post-eclosion . Mosquitoes were fed a human-blood-meal 96–108 hr prior to imaging and were not giving access to an egg laying substrate so that they were gravid at the time of imaging . Gravid females were anesthetized at 4°C for dissection . The wings were removed and the mosquito was fixed to a custom Delrin plastic holder with UV-curable glue ( Bondic ) . The mosquito was inserted into a hole in the holder , such that the ventral thorax , including all coxae , were exposed above the surface of the holder , with the rest of the mosquito below . The mosquito was secured with a few points of glue ( Bondic ) on the abdomen , thorax and head . The leg that was presented with water remained free of glue to prevent damage to the tissue . Once the mosquito was secured to the plate , one of the forelegs was inserted into a small diameter tube that was secured to the bottom of the plate and could later be attached to the fluidics apparatus used for stimulus delivery ( described below ) . The top of the dish was then filled with external saline . The recipe we used is based D . melanogaster imaging saline:103 mM NaCl , 3 mM KCl , 5 mM 2-[Tris ( hydroxymethyl ) methyl]−2-aminoethanesulfonic acid ( TES ) , 1 . 5 mM CaCl2 , 4 mM MgCl2 , 26 mM NaHCO3 , 1 mM NaH2PO4 , 10 mM trehalose , 10 mM glucose , pH 7 . 3 , osmolality adjusted to 275 mOsm/kg ) . The coxae were gently spread from the midline and secured in dental wax . The cuticle was removed above the prothoracic ganglia using very sharp forceps . Opaque non-neural tissue , primarily fat cells and muscle , was removed if they obstructed the ventral nerve cord . Great care was taken not to damage the ppk301-expressing nerves running from the legs into the ventral nerve cord . These run up the posterior-ventral region of the leg and are extremely superficial . dTomato fluorescence was examined before imaging to verify that the nerves were intact . If an animal did not respond to water , 200 mOsm/kg , or 300 mOsm/kg NaCl it was discarded . The preparation was secured to the stage using a custom laser cut acrylic holder . A single plane through the center of the prothoracic neuropil was scanned at 4 . 22 fps with a 920 nm excitation wavelength imaged through a 680 nm shortpass infrared ( IR ) blocking filter , a 565 nm longpass dichroic and 595/50 nm or 525/70 nm bandpass filters . GCaMP6s and dTomato emission was collected simultaneously for 70 frames per trial . Each concentration was delivered at least three times per animal , and each animal was exposed to a series of either 0 , 200 , and 300 mOsm/kg NaCl , or 200 mOsm/kg NaCl , 200 mOsm/kg L-serine , 200 mOsm/kg D- ( + ) -cellobiose and freshwater . All wild-type and PPK301 mutant animals were reared and imaged in parallel . Imaging remained stable during the duration of the imaging session in all animals that were included in this study . We did not notice a decrease in the response to stimuli over time . Before beginning the experiments using multiple concentrations of salt , we imaged animals with repeated water delivery , and saw no desensitization to the response to water over 10 presentations ( data not shown ) . Liquids were delivered to a single foreleg of the mosquito using a custom built low-volume fluidics device . Piezoelectric diaphragm micropumps and their controller ( Servoflo ) were run by an Arduino using a code written for this purpose . A custom manifold was milled for small volume liquid delivery . The mosquito was illuminated with an IR light and liquid delivery was monitored using an IR camera . The liquid coated the tarsal and tibial segments and was retracted immediately after imaging each sweep . We waited at least 1 min between each trial . We only observed responses in the prothoracic region ipsilateral to the stimulus delivery . All image processing was done using FIJI/ImageJ ( NIH ) . Further processing was done using Excel and Prism ( GraphPad ) . Regions of interest were selected based on the dTomato fluorescence intensity and used for analysis of GCaMP6s signal . All traces with motion , as determined by dTomato fluorescence instability , were discarded . A Gaussian blur with a sigma value of 1 was performed on the GCaMP6s signal . In the calculation of ΔF/F , six frames were averaged before stimulus presentation to determine the baseline fluorescence . To determine Fmax , the average of 3 frames at the peak after stimulus delivery was determined for each sweep . All statistical analyses were performed using Prism ( GraphPad ) or R version 3 . 5 . 0 ( R Development CoreTeam , 2017 ) . Data collected as percentage of total are shown as median with interquartile range and data collected as raw value are shown as mean ± SEM or mean ± SD . Details of statistical methods are reported in the figure legends . All plotted data ( with the exception of raw video files ) are available in Supplementary file 1 , and behavior assay schematics and egg-laying counting image analysis scripts can be found at https://github . com/VosshallLab/MatthewsYoungerVosshall2018 . ( Vosshall Lab , 2019; copy archived at https://github . com/elifesciences-publications/MatthewsYoungerVosshall2018 ) . Plasmids will be made available from Addgene . | When they bite humans , mosquitoes can transmit dangerous diseases . For example , the Aedes aegypti mosquito spreads the viruses that cause yellow fever , Zika and dengue fever . Only the female mosquitoes feed on blood so they can obtain the nutrients they need to develop their eggs . Once they are ready , the insects lay their eggs in carefully selected sites where fresh water collects: if instead they choose places where the water is too salty , their offspring will die . To find a suitable site , a mosquito ‘tastes’ the water by dipping in its legs and mouthparts , which activates the insect’s sensory neurons and sends signals to its brain . However , it remains unclear exactly how the mosquitoes can distinguish between fresh and salty water . To address this question , Matthews , Younger and Vosshall used a combination of genetic and imaging approaches to study female Ae . aegypti mosquitoes . These experiments identified a gene known as ppk301 that is necessary for the mosquitoes to successfully lay their eggs in the right type of water . Mutant Ae . aegypti mosquitoes lacking the ppk301 gene did not properly lay eggs in fresh water even when given the opportunity . Further experiments found that ppk301 was present in specific neurons in the legs and mouthparts of the mosquitoes . In leg neurons , ppk301 played a crucial role in sensing the presence of water and in stimulating the mosquitoes to lay eggs in water containing low levels of salt . However , these cells still responded to salt , even when lacking ppk301: other unidentified genes must therefore also be involved in preventing the mosquitoes from breeding in water that is too salty . Every year , Aedes aegypti and other mosquitoes infect hundreds of millions of people and cause 500 , 000 deaths . Knowing exactly how mosquitoes breed could help to develop traps , repellents and other strategies to stop the insects from multiplying . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"neuroscience",
"genetics",
"and",
"genomics"
] | 2019 | The ion channel ppk301 controls freshwater egg-laying in the mosquito Aedes aegypti |
Respiration , an essential process for most organisms , has to optimally respond to changes in the metabolic demand or the environmental conditions . The branched character of their respiratory chains allows bacteria to do so by providing a great metabolic and regulatory flexibility . Here , we show that the native localization of the nitrate reductase , a major respiratory complex under anaerobiosis in Escherichia coli , is submitted to tight spatiotemporal regulation in response to metabolic conditions via a mechanism using the transmembrane proton gradient as a cue for polar localization . These dynamics are critical for controlling the activity of nitrate reductase , as the formation of polar assemblies potentiates the electron flux through the complex . Thus , dynamic subcellular localization emerges as a critical factor in the control of respiration in bacteria .
Respiration is an essential process for most living organisms . The free energy derived from oxidation of reducing equivalents generated by cell metabolism or taken up from the environment is converted into a protonmotive force ( pmf ) across a membrane that is used , among many processes , to drive adenosine triphosphate synthesis . Although oxygen is the most common terminal electron acceptor of respiratory chains , prokaryotes can make use of alternative acceptors in anoxic environments , such as nitrogen oxides , elemental sulfur and sulfur oxyanions , organic N-oxides , etc . In eukaryotes , respiration is confined to a specific organelle , the mitochondrion and more specifically to its inner membrane which displays remarkably intricate ultra-structural features that impact the respiratory output ( Cogliati et al . , 2013 ) . Respiration involves multimeric membrane-embedded oxidative phosphorylation ( OXPHOS ) complexes which bear several metal cofactors through which electrons are transported . The organization of OXPHOS complexes in the inner mitochondrial membrane has been the subject of intense debate during the last decade after the discovery of supramolecular assemblies by the landmark study of Schägger in 2000 ( Schägger and Pfeiffer , 2000 ) , later substantiated by electron microscopy studies ( for a review , see Vonck and Schafer , 2009 ) . Plasticity of the almost linear mitochondrial OXPHOS chain is ensured by a dynamic equilibrium between isolated complexes and supramolecular assemblies which is , in turn , influenced by both specific lipids and stabilizing factors ( Zhang et al . , 2002; Ikeda et al . , 2013 ) . Such an equilibrium is likely beneficial for the function of the mitochondrion allowing the cell to respond to cellular and environmental cues . Moreover , the extraordinarily low degradation rate of OXPHOS complexes in mitochondria ( Price et al . , 2010; Kim et al . , 2012; Nelson et al . , 2013 ) reinforces the idea that the dynamic association into supercomplexes adds an effective switch to control electron flows in response to environmental changes ( Lapuente-Brun et al . , 2013 ) . The flexibility of energy metabolism is also critical for the adaptation of a number of prokaryotes to varying environments ( Unden et al . , 2014 ) . The utilization of modular OXPHOS chains confers flexibility depending on the available energy source or terminal electron acceptor but also allows modulation of the resulting pmf . So far , the adjustment of respiration to varying environments in prokaryotes is considered to be the result of an intricate transcriptional regulation network that controls the expression of OXPHOS complexes with varying electrogenic capacities . At the same time , there is cumulative evidence for an elaborate spatial organization of macromolecules in bacterial cells . We thus hypothesized that the OXPHOS process could be regulated through the dynamic subcellular localization of complexes in response to the metabolic demand . Here , we investigate the native localization of a major OXPHOS complex under anoxic conditions , the quinol oxidizing nitrate reductase complex from the gut bacterium Escherichia coli , and its dynamics upon changes in the environment by real-time fluorescence imaging in live cells .
The E . coli nitrate reductase complex is composed of three subunits ( NarGHI ) and likely organized in a dimer under physiological conditions ( Bertero et al . , 2003 ) . A di-heme b-type cytochrome subunit , NarI , ensures quinol oxidation and membrane anchoring of the cytoplasmically oriented catalytic dimer , NarGH , where nitrate reduction takes place . The C-terminus of the catalytic subunit NarG protruding from the quaternary structure of the complex was labeled with a green fluorescent protein ( GFP ) . The hybrid gene was expressed under the control of its native promoter activated by FNR and NarL transcriptional factors in response to anaerobiosis and nitrate , respectively and ectopically integrated into the chromosome of the nitrate reductase-deficient strain JCB4023 ( Potter et al . , 1999 ) . The corresponding fusion was functional as verified by cell growth in nitrate-respiring conditions where a functional nitrate reductase complex is mandatory ( Figure 1A ) . Fractionation studies confirmed the membrane localization of the complex ( Figure 1B ) . The in vitro assay of the functional integrity of the NarG-gfpH catalytic module showed that the activity was only slightly affected by the fusion ( Figure 1C ) . Thus , the GFP-labeled nitrate reductase complex is active in the cytoplasmic membrane . 10 . 7554/eLife . 05357 . 003Figure 1 . The GFP-labeled nitrate reductase complex is active and fully assembled . ( A ) Growth curves of E . coli strains expressing untagged or tagged-NarGHI under nitrate-respiring conditions . Cells were grown anaerobically in a minimal medium using glycerol as sole carbon source and nitrate as terminal electron acceptor . JCB4011 strain expresses the untagged NarGHI complex ( ▲ ) , whereas the LCB3635 strain expresses the GFP-tagged NarGHI complex ( ● ) . As a negative control , the nitrate reductase-deficient strain JCB4023 shows no growth under these conditions ( ■ ) . The estimated generation time is about 80 ± 10 min for JCB4011 and 110 ± 10 min for LCB3635 . ( B ) The GFP-tagged NarGHI complex is correctly localized to the membrane . Western blots were performed with antibodies raised against eGFP or IscS as a marker of soluble proteins on soluble ( S ) or membrane ( M ) fractions prepared from nitrate-respiring cells . ( C ) The activity of the NarGH catalytic module is unaffected by the eGFP fusion . Benzyl viologen:nitrate oxidoreductase activity assays were performed on membranes prepared from JCB4011 ( untagged version ) or LCB3635 ( GFP-tagged version ) cells grown under nitrate-respiring conditions and expressed in µmoles of nitrite produced min−1 mg−1 of nitrate reductase . DOI: http://dx . doi . org/10 . 7554/eLife . 05357 . 003 The subcellular localization of the anaerobic GFP-labeled OXPHOS complex was characterized by fluorescence microscopy imaging on exponentially growing E . coli cells under nitrate-respiring conditions ( Figure 2A ) . The fluorescence signal was only detected in the cytoplasmic membrane in agreement with fractionation studies . Interestingly , the fluorescence signal appeared as clusters that predominantly concentrated at the cell poles in nearly 80% of the cells ( Figure 2B–D ) . Furthermore , fluorescence clusters concentrate at the cell poles in nitrate-respiring cells independently of the cell length and not at the division septum ( Figure 2—figure supplement 1 ) . This suggests that there is no relationship between the position of the clusters and the cell cycle . 10 . 7554/eLife . 05357 . 004Figure 2 . The GFP-labeled nitrate reductase complex concentrates at the cell poles under nitrate-respiring conditions . ( A ) Fluorescence images ( top ) and overlays of fluorescence and phase contrast images ( bottom ) are shown for nitrate-respiring and oxygen-respiring cells . The deconvolved image of the fluorescence signal is shown in green , and the cell outline is shown by phase contrast . ( B ) Mean frequency of cells displaying clusters at the cell poles for nitrate-respiring or oxygen-respiring cells . ( C ) A density map was built from a two-dimensional histogram ( relative transversal cluster position vs relative longitudinal cluster position ) of the fluorescence signal observed in nitrate-respiring or oxygen-respiring LCB3635 cells . The color map shows the interpolated density of clusters positions . The dots represent the individual clusters ( a little jittering was added to avoid overlapping of the dots ) . ( D ) The histogram of the fluorescence signal clusters across the transversal axis of nitrate-respiring or oxygen-respiring LCB3635 cells is shown . In both conditions , more than 500 cells were analyzed . Under nitrate-respiring conditions , a strong enrichment of the clusters is observed at the cell poles . Under oxygen-respiring conditions , only few cells exhibit clusters which have an even distribution along the cell axis . DOI: http://dx . doi . org/10 . 7554/eLife . 05357 . 00410 . 7554/eLife . 05357 . 005Figure 2—figure supplement 1 . Spatial distribution of clusters . ( A ) A density map was built from a two-dimensional histogram ( cell length vs relative cluster position ) of the fluorescence signal observed in nitrate-respiring or oxygen-respiring LCB3635 cells . The color map shows the interpolated density of clusters positions . The dots represent the individual clusters ( a little jittering was added to avoid overlapping of the dots ) . ( B ) The ‘poles’ were defined according to the distribution of clusters along the cell axis in Figure 2D . The limit was set to ± 0 . 6 and used subsequently throughout the article . DOI: http://dx . doi . org/10 . 7554/eLife . 05357 . 005 We next evaluated the impact of varying electron routes on the cellular localization of the GFP-labeled nitrate reductase complex . First , fluorescence imaging was performed on exponentially growing cells under oxygen-respiring conditions . Surprisingly , the fluorescence signal was evenly distributed along the cytoplasmic membrane under those conditions ( Figure 2A ) . As compared to nitrate-respiring cells , fewer clusters are present and evenly distributed in oxygen-respiring cells ( Figure 2B–D ) . Second , anaerobic respiration on alternative substrates such as fumarate or trimethylamine N-oxide ( TMAO ) was tested . It induces a significant polar localization of the fluorescence signal albeit lower than under nitrate-respiring conditions ( Figure 3 ) . Fluorescence imaging was eventually analyzed on anoxic fermenting growing cells using glucose as sole carbon source . Strikingly , the fluorescence was evenly distributed as observed under oxygen-respiring conditions ( Figure 3 ) . Detailed analysis of the fluorescence signal distribution in all these metabolic conditions revealed that the formation of fluorescent clusters is systematically associated with polar localization ( Figure 3—figure supplement 1 ) . Thus , the nitrate reductase complex displays a dynamic subcellular localization in response to the metabolic demand . 10 . 7554/eLife . 05357 . 006Figure 3 . Metabolism-dependent localization of the nitrate reductase OXPHOS complex . Mean frequency of LCB3635 cells displaying clusters at the cell poles for various metabolic conditions ( anaerobic respiration with either nitrate , fumarate or TMAO as terminal electron acceptor , aerobic respiration and anoxic fermenting conditions ) . Non-fermentable glycerol was used as sole carbon and electron source in all cases with the exception of fermenting conditions where it was replaced for glucose . Polar localization is only observed under anaerobic respiration whatever the terminal electron acceptor used . DOI: http://dx . doi . org/10 . 7554/eLife . 05357 . 00610 . 7554/eLife . 05357 . 007Figure 3—figure supplement 1 . Comparative analysis of cells displaying clusters at the cell poles vs cells displaying clusters along the cell body upon varying metabolic conditions . ( Gray bars ) Mean frequency of LCB3635 cells displaying clusters of the fluorescence signal at the cell poles . ( Black bars ) Mean frequency of LCB3635 cells displaying clusters of the fluorescence signal along the entire cell body for various metabolic conditions . The conditions tested were: anaerobic respiration with either nitrate , fumarate or TMAO as terminal electron acceptor , aerobic respiration , and anoxic fermenting conditions . DOI: http://dx . doi . org/10 . 7554/eLife . 05357 . 007 A major difference between the metabolic conditions studied above is the resulting pmf , based on the distinct proton transfer capabilities of the OXPHOS complexes involved . The present finding imparts particular significance to the previous reports that pmf can play a role in protein localization in bacteria ( Alcock et al . , 2013; Rose et al . , 2013 ) . Saliently , the dissipation of the electric component of the pmf , the transmembrane potential ( ΔΨ ) , hampers proper localization of proteins involved in cell division , chromosome segregation or cell shape regulation ( Strahl and Hamoen , 2010 ) . However there is , to our knowledge , no report of a protein localization that would be specifically controlled by the other component of the pmf , the proton concentration gradient ( ΔpH ) . To evaluate the participation of the pmf in the subcellular localization of the GFP-labeled nitrate reductase and analyze the contribution of its two components , we first treated nitrate-respiring cells with distinct ionophores . As shown in Figure 4A , the addition of carbonyl cyanide m-chlorophenyl hydrazone ( CCCP ) , a specific proton-ionophore rapidly dissipating the pmf , resulted in a significant reduction of cells displaying polar clusters . Unexpectedly , depolarization of the membrane potential through the addition of the potassium-ionophore valinomycin had no effect , whereas dissipation of ΔpH by the electroneutral anion/OH−-exchanger trichlorocarbanilide ( TCC ) ( Ahmed and Booth , 1983 ) gave rise to significant delocalization of the fluorescent signal . To substantiate the critical role of ΔpH for polar localization , we reasoned that artificial establishment of a proton concentration gradient by the light-driven proton translocation activity of the proteorhodopsin ( PR ) ( Walter et al . , 2007 ) should be sufficient to promote the polar recruitment of the nitrate reductase complex under anoxic fermentative conditions . Indeed , under fermenting conditions , it is considered that the ΔpH is very low as compared to respiring conditions . As shown in Figure 4B , heterologous expression of PR resulted in polar localization of the fluorescence signal in nearly 40% of cells . Thus , we conclude that the proton concentration gradient is a critical cue for polar localization of this anaerobic OXPHOS complex . We next questioned the importance of nitrate reductase activity on its subcellular localization . To this end , a catalytically inactive but stable variant was used . In particular , the H50S substitution in NarG precludes insertion of two metal centers at the active site while x-ray structural analysis of the NarGH50SHI complex revealed the absence of structural changes at the protein surface ( Magalon et al . , 1998; Rothery et al . , 2010 ) . Under nitrate-respiring conditions , the fluorescence signal associated with the H50S variant is uniformly distributed likely due to its inability to generate a ΔpH ( Figure 5A , B and Figure 5—figure supplement 1 ) . In contrast , adding fumarate gives rise to a significant polar localization of the inactive GFP-labeled nitrate reductase complex reinforcing the conclusion that activity of the complex is essential for polar positioning unless a ΔpH is established ( Figure 5A , B and Figure 5—figure supplement 1 ) . Interestingly , polar localization requires both the absence of oxygen and the establishment of a proton concentration gradient to which the nitrate reductase complex may contribute fuelling the localization mechanism . 10 . 7554/eLife . 05357 . 008Figure 4 . Proton gradient is a cue for polar localization of the anaerobic OXPHOS complex . ( A ) The pmf is important for polar localization of the nitrate reductase complex . Shown is the mean frequency of nitrate-respiring LCB3635 cells displaying clusters at the cell poles upon addition of the indicated ionophores . Each value was normalized with respect to that of the untreated cells ( labeled as control ) and expressed in % of the control . Images were taken 15 min after the treatment with the ionophore . ( B ) Establishment of an artificial proton concentration gradient in anoxic fermenting growing cells is sufficient to restore the polar localization . Shown is the mean frequency of fermenting growing LCB3635 cells displaying clusters at the cell poles in the absence ( -PR ) or presence of the proteorhodopsin ( +PR ) . DOI: http://dx . doi . org/10 . 7554/eLife . 05357 . 00810 . 7554/eLife . 05357 . 009Figure 5 . Activity of the nitrate reductase OXPHOS complex is not essential for polar positioning unless a ΔpH is established . ( A ) Fluorescence images of JCB4023/pVA70GFP cells expressing active ( top ) or inactive NarG-H50S variant of the GFP-labeled nitrate reductase ( bottom ) are shown . Cells were grown anaerobically either in glycerol-nitrate ( nitrate ) or glycerol-nitrate-fumarate ( nitrate + fumarate ) medium . ( B ) Mean frequency of cells displaying clusters at the cell poles under nitrate or nitrate–fumarate conditions . ( Black bars ) JCB4023/pVA70GFP cells expressing the active complex . ( Gray bars ) JCB4023/pVA70GFPH50S cells expressing the inactive variant . DOI: http://dx . doi . org/10 . 7554/eLife . 05357 . 00910 . 7554/eLife . 05357 . 010Figure 5—figure supplement 1 . Activity of the nitrate reductase OXPHOS complex is not essential for polar positioning unless a ΔpH is established . Fluorescence images of JCB4023/pVA70GFP cells expressing inactive NarG-H50S variant of the GFP-labeled nitrate reductase and grown anaerobically either in glycerol-nitrate ( nitrate ) or glycerol-nitrate-fumarate ( nitrate + fumarate ) medium . DOI: http://dx . doi . org/10 . 7554/eLife . 05357 . 01010 . 7554/eLife . 05357 . 011Figure 5—figure supplement 2 . Cells expressing the GFP-labeled nitrate reductase complex from the pVA70GFP plasmid show identical metabolism-dependent localization of the complex . ( A ) Overlays of fluorescence and phase contrast images are shown for nitrate-respiring and oxygen-respiring JCB4023 cells expressing the GFP-tagged NarGHI complex from the pVA70GFP plasmid . ( B ) Mean frequency of cells displaying clusters at the cell poles for nitrate-respiring or oxygen-respiring cells . DOI: http://dx . doi . org/10 . 7554/eLife . 05357 . 011 Bacteria exploit the branched character of their OXPHOS chains to respond to varying environments with a great metabolic and regulatory flexibility ( Unden et al 2014 ) . The above-described modification of the localization pattern of the nitrate reductase complex in response to metabolic changes begs the question of its timeframe . First , nitrate-respiring cells were submitted to strong aeration and the fluorescent pattern was followed during the course of the anoxic–oxic transition ( Figure 6A ) . While 80% of the bacterial cells initially displayed a pronounced polar localization of the GFP-labeled nitrate reductase complex , the signal started to change noticeably after 15 min , reaching an almost uniform distribution after one hour of aerobic transition . A similar temporal behavior has been observed under the reverse conditions using aerobically growing cells transferred to anoxic conditions ( Figure 6A ) . In this case , more than 70% of the bacterial cells displayed a polar localization of the fluorescent signal after the transition . Thus , redistribution of the OXPHOS complex triggered by metabolic changes operates on timescales of several tens of minutes . While such timing may reflect the slow diffusion of such large membrane-embedded complexes in the crowded cytoplasmic membrane , we next questioned whether this switch is faster than the changes in nitrate reductase protein content as the result of transcriptional regulation . At first , western-blotting analyses showed nearly unchanged level of GFP-labeled nitrate reductase complex upon anoxic–oxic transition despite the rapid redistribution of the complex in the cytoplasmic membrane ( Figure 6—figure supplement 1A ) . In comparison , an increased level of complex is observed during the oxic–anoxic transition as the result of FNR-mediated regulation . To substantiate these observations , localization upshift experiments were reproduced in cells pre-treated with chloramphenicol ( Cm ) , a protein synthesis inhibitor ( Figure 6B and Figure 6—figure supplement 1B ) . During the anoxic–oxic transition , Cm had nearly no impact on the delocalization kinetics with only slight diminution of the nitrate reductase content . Interestingly , during the oxic–anoxic transition , the fluorescence signal remains uniformly distributed with a concomitant decrease of nitrate reductase content . To rule out any impact on the yielded pmf value under those conditions , the oxic–anoxic upshift experiment was reproduced in presence of PR . As shown in Figure 6B , establishment of a ΔpH was not sufficient to promote polar localization . Altogether , these results support the idea that induction of the expression of at least one gene during the oxic–anoxic transition promotes polar localization of the nitrate reductase complex . 10 . 7554/eLife . 05357 . 012Figure 6 . Metabolism-dependent localization changes occur in a timeframe of several tens of minutes . ( A ) ( Black bars ) Nitrate-respiring JCB4023/pVA70GFP cells were submitted to strong aeration and the fluorescence pattern was evaluated at 15-min intervals . ( Gray bars ) Oxygen-respiring JCB4023/pVA70GFP cells were shifted to anoxic conditions in presence of nitrate . In both cases , shown is the mean frequency of cells displaying clusters at the cell poles . ( B ) ( Black bars ) Upon chloramphenicol ( Cm ) addition , nitrate-respiring 3635/pBAD24 cells were submitted to strong aeration and the fluorescence pattern was evaluated at 30-min intervals . ( Gray bars ) Upon Cm addition , oxygen-respiring 3635/pBAD24 cells were shifted to anoxic conditions in presence of nitrate . ( Shaded bars ) Upon Cm addition , oxygen-respiring 3635/pPR cells were shifted to anoxic conditions in presence of nitrate . In all cases , shown is the mean frequency of cells displaying clusters at the cell poles . DOI: http://dx . doi . org/10 . 7554/eLife . 05357 . 01210 . 7554/eLife . 05357 . 013Figure 6—figure supplement 1 . Metabolism-dependent localization changes occur in a timeframe of several tens of minutes . ( A ) Changes in GFP-labeled nitrate reductase content as estimated by immunoblotting . ( Black bars ) anoxic–oxic transition . ( Gray bars ) oxic–anoxic transition . ( B ) Changes in GFP-labeled nitrate reductase content as estimated by immunoblotting ( Black bars ) anoxic–oxic transition of 3635/pBAD24 cells . ( Gray bars ) oxic–anoxic transition 3635/pBAD24 cells . ( Shaded bars ) oxic–anoxic transition 3635/pPR cells . The mean GFP-labeled nitrate reductase content was normalized using IscS as internal control and expressed as signal ratio intensity . DOI: http://dx . doi . org/10 . 7554/eLife . 05357 . 013 The GFP-labeled nitrate reductase complex displays a dynamic localization pattern leading to the formation of discrete domains in the cytoplasmic membrane . At first , we reasoned that distinct subcellular localization may influence the intrinsic activity of the OXPHOS complex . The activity of the GFP-labeled complex was therefore assessed in membrane vesicles issued from cells that displayed either an even distribution ( oxygen-respiring condition ) or a strong polar enrichment ( nitrate-respiring condition ) of the complex . As shown in Figure 7A , we found no significant variation in the specific activities indicating that subcellular localization has no influence on the intrinsic activity of the OXPHOS complex . 10 . 7554/eLife . 05357 . 014Figure 7 . Polar localization determines integration of the nitrate reductase complex in anaerobic respiratory chains . ( A ) Constant activity of the GFP-tagged complex upon distinct subcellular localization . Quinol:nitrate oxidoreductase activity was measured on membranes prepared from oxygen-respiring or nitrate-respiring JCB4023 cells expressing the GFP-tagged NarGHI complex from the pVA70GFP plasmid . Activities are expressed in µmoles of nitrite produced min−1 mg−1 of nitrate reductase . ( B ) Enhanced nitrite production in cells with enforced polar localization of the GFP-NarGHI complex . At time 0 , nitrate ( 100 mM final concentration ) was added to anoxic fermenting growing LCB3635 cells expressing ( +PR , light grey ) or not expressing ( −PR , black ) the proteorhodopsin . Subsequently , nitrite production was detected in the cell culture over time using the Griess reaction . The indicated values were derived from raw data using a suitable standard curve and represented the means of three distinct experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 05357 . 014 We next hypothesized that the formation of discrete domains through the polar recruitment of the nitrate reductase may influence the overall yield of the electron transport chain . To evaluate the impact of polar localization on the electron flux from primary dehydrogenases to GFP-labeled nitrate reductase , it was essential to ensure an identical composition of the cytoplasmic membrane in terms of OXPHOS complexes . To account for this issue , we used fermenting-growing cells expressing or not PR which have an identical OXPHOS proteome but display distinct localization patterns of the GFP-labeled complex ( Figure 4B ) . Upon addition of nitrate in the growth medium , electron flux through the nitrate reductase complex could be kinetically resolved by quantifying nitrite in the cell culture . As seen in Figure 7B , within the first 10 min following nitrate addition , the rate of nitrite production is significantly higher in PR-expressing cells than in control cells , indicating a direct correlation between subcellular localization and electron flux through the nitrate reductase complex . After one hour , the level of nitrite produced is nearly two orders of magnitude higher in PR-expressing cells . We conclude that environmental conditions ( anaerobiosis and a ΔpH ) promoting polar clustering of the nitrate reductase complex result in a higher efficiency of the associated respiratory chains .
In the past decade , the emerging field of bacterial cell biology has underscored the fact that dynamic subcellular localization is intimately linked to the biological function allowing control of fundamental processes , such as cell division , virulence , motility , or signal transduction ( for reviews , see Kiekebusch and Thanbichler , 2014; Laloux and Jacobs-Wagner , 2014; Nevo-Dinur et al . , 2012; Shapiro et al . , 2009 ) . Similarly , we have shown that the nitrate reductase respiratory complex is submitted to a spatiotemporal regulation in response to environmental conditions which , in turns , potentiates the electron flux through the associated respiratory chain . Nitrate respiration in E . coli is thus controlled by specific subcellular localization of its terminal reductase . Segregation of respiratory complexes within the cytoplasmic membrane has also been reported for components of bacterial aerobic chains ( Johnson et al . , 2004; Lenn et al . , 2008; Werner et al . , 2009; Rexroth et al . , 2011; Llorente-Garcia et al . , 2014 ) . The actual view is that functional consequences are associated with a high level of structural organization of respiratory complexes within bioenergetic membranes ( for a review , see Genova and Lenaz , 2014 ) . Interestingly enough , recent studies have evidenced the existence of functional microdomains in bacterial membranes organizing a specific subset of proteins in space and time surmising a gain-of-function ( Lopez and Kolter , 2010 ) . In the same line of thought , it has recently been shown that clustering of multiple enzymes belonging to the same metabolic pathway into agglomerates accelerates the processing of intermediates ( Castellana et al . , 2014 ) . Clearly , several alternatives can be found for the reported enhanced electron flux in the nitrate reductase-associated respiratory chains such as physical association into supercomplexes or segregating respiratory complexes into microdomains . In both cases , one would hypothesize that the probability of the quinone molecules to encounter the nitrate reductase is higher . Evidences have been provided for such quinone channeling within supercomplexes while the exact mechanism of channeling is not yet understood ( for a review , see Genova and Lenaz , 2014 ) . By providing the first evidence for a spatiotemporal regulation of a respiratory complex in response to environmental conditions , this work establishes a basis for a deeper analysis on how environmental signals are translated into subcellular localization of respiratory complexes to adjust the respiration output .
The E . coli strains and plasmids are described in Table1 , Supplementary file 1 . E . coli strains were grown aerobically at 37°C in defined minimal medium supplemented with 140 mM of glycerol used as sole carbon source and 100 mM nitrate . Anaerobic growth of bacteria is performed in gas tight hungate tubes under Ar atmosphere . For anaerobic growth under respiring conditions , nitrate , fumarate or TMAO were added at 100 mM final concentration and used as terminal electron acceptors . For anoxic fermentative growth , glycerol was replaced by glucose at 40 mM final concentration . The minimal medium is composed of potassium phosphate buffer ( 100 mM ) adjusted to pH 7 . 4 , ammonium sulfate ( 15 mM ) , NaCl ( 9 mM ) , magnesium sulfate ( 2 mM ) , sodium molybdate ( 5 µM ) , Mohr's salt ( 10 µM ) , and calcium chloride ( 100 µM ) . After filtration , casaminoacids ( 0 . 5% ) and thiamine ( 0 . 01% ) were added just before use together with antibiotics , if necessary . The nitrate reductase-deficient JCB4023 strain was used as recipient for integration of the translational narG-egfp fusion at the chromosomal attΦ80 site using the procedure described in ( Haldimann and Wanner , 2001 ) . Introduction of the eGFP at the C-terminus of NarG was performed as described below . The oligonucleotides used in this study are described in Table 2 , Supplementary file 1 . To generate pVA70XN , the 3′ region of narG was PCR-amplified with primers 584 and 585 , whereas the 5′ region of narH was amplified with primers 586 and 447 . Both PCR products were then used as templates for another PCR reaction with primers 584 and 447 yielding a 1200 bp product including the 3′ end of narG followed by XmaI and NotI restriction sites and the native ribosome binding site of narH . The reaction product was then restricted with MunI and AatII , and the resulting fragment was ligated into the pVA70 vector which had been cut with the same enzymes . The entire cloned region was verified by sequencing . To create pVA70GFP , which allows fusion of egfp to the 3′ region of the narG gene , the gfp XmaI-NotI fragment was excised from pEGFP-N1 ( Clontech ) and subsequently ligated into pVA70XN cut with the same enzymes . This plasmid allows the production of the GFP-tagged NarGHI complex . To obtain pFA , which allows integration of a synthetic narG-egfp narHJI operon at the chromosomal attΦ80 site , the entire operon including the pnar promoter was excised from pVA70GFP by restriction with SacI and SalI . The fragment was then ligated into the pAH162 CRIM plasmid which had been cut with the same enzymes . The pFA plasmid was then integrated at the attΦ80 site of the nitrate reductase-deficient strain JCB4023 ( Potter et al . , 1999 ) according to the procedure described in Haldimann and Wanner ( 2001 ) . The resulting LCB3635 strain was maintained in the presence of 6 µg/ml of tetracycline . Cells were harvested in late-exponential phase , washed and resuspended in 40 mM Tris–HCl ( pH 7 . 4 ) , 1 mM MgCl2 . Bacterial cells were broken by passage through a French press . After an initial centrifugation at 20 , 000×g , differential ultracentrifugation at 250 , 000×g allowed the separation between soluble and membrane fractions which were frozen in liquid nitrogen and stored at −80°C until use . Nitrate reductase activity was measured with standard assays using reduced benzyl viologen or menadiol as electron donors ( Jones and Garland , 1977; Giordani et al . , 1997 ) . As the result of chromosomal expression , a low level of the NarGHI complex is observed in membrane vesicles of the LCB3635 strain grown under oxygen-respiring condition . This situation precludes accurate assessment of the quinol:nitrate oxidoreductase activity measured at 260 nm . Therefore , the JCB4023 strain was transformed with pVA70GFP , resulting in higher production of GFP-tagged NarGHI ( eightfold as estimated by immunoblotting , data not shown ) and allowing quinol activity measurements . The metabolism-dependent localization of the nitrate reductase was unaffected by the increased level of GFP-tagged NarGHI complex in the cells ( Figure 5—figure supplement 2 ) . The NarGHI protein concentration was estimated using rocket immunoelectrophoresis as described in Lanciano et al . ( 2007 ) . Western-blots were performed using antibodies raised against eGFP and IscS on samples run on 10% SDS-polyacrylamide gels . Quantitative analysis of the fold change in GFP-labeled nitrate reductase protein levels was achieved by integration of the eGFP chemiluminescence signal normalized with the IscS one using samples containing an equal cell mass . Nitrite levels were determined spectrophotometrically by the Griess reaction . Briefly , 1 ml of the anaerobic cell culture was centrifuged . 500 µl of the cell-free supernatant was mixed with 100 µl of Griess reagent ( 0 . 5% of sulfanilic acid ) and incubated for 5 min at room temperature . After addition of an equal volume of 0 . 6% of N-1-napthylethylenediamine dihydrochloride , the solution was incubated in the dark for 30 min before measuring the absorbance at 525 nm . A standard curve was made with a nitrite solution allowing quantitative measurement . For fluorescence microscopy , cells were grown aerobically or anaerobically to midexponential phase at 37°C , and 2 µl was mounted on microscope slides covered by a thick fresh minimal medium agar pad . In case of anaerobically growing cells , images were taken after a 5–10 min delay shown to be optimal for activation of the GFP moiety during the mounting process . Furthermore using this procedure , we have noticed that the localization pattern was unchanged after 1 hr under the agar pad . The slide was analyzed by microscopy using a Nikon Eclipse TiE PFS inverted epifluorescence microscope ( 100 × oil objective NA 1 . 3 Phase Contrast ) and a Hamamatsu OrcaR2 CCD camera . Images were collected with NIS elements software . Observation of the fluorescence signal of GFP-labeled nitrate reductase under aerobic conditions was performed on cells grown in minimal medium with glycerol as sole carbon source and with nitrate for NarL-mediated nar operon expression . Under this condition , oxygen is preferred to nitrate as terminal electron acceptor ( Unden et al , 2014 ) . For evaluation of the participation of the pmf in the polar localization of the GFP-tagged NarGHI , ionophores were added to nitrate-respiring cells at midexponential phase and images were taken after 15 min of incubation . CCCP , valinomycin , and TCC were used at 100 , 30 , and 200 µM final concentrations , respectively according to ( Ahmed and Booth , 1983 ) . To build an artificial proton gradient across the cytoplasmic membrane , PR was expressed in LCB3635 cells using the pPR plasmid ( Tipping et al . , 2013 ) . Cells transformed with the pBAD24 plasmid were used as control . Transformants were grown under aerobic conditions in a minimal medium supplemented with glucose and ampicillin . At early log phase , PR expression was induced with 0 . 02% arabinose together with all-trans-retinal ( 10 µM ) . At an OD600 of 1 , the culture was shifted to anaerobic conditions and exposed to light for one hour , allowing the establishment of a proton gradient under these anoxic fermenting conditions . Fluorescence images were then taken as described above . Thanks to the red fluorescence emission of PR ( Beja et al . , 2000 ) , uniform distribution of the fluorescence signal is observed at the cell membrane using appropriate filter thus confirming its membrane localization ( data not shown ) . Upon establishment of a polar localization of the GFP-tagged NarGHI complex in PR-expressing cells , nitrate was added at a final concentration of 100 mM and the rate of nitrite production was then determined spectrophotometrically by the Griess reaction as described above . All the objects detection and quantification in images were performed with a new Fiji/ImageJ plugin developed specifically for the treatment of microscopic images of bacterial cells . This plugin is called MicrobeJ and was created by A Ducret in Y Brun laboratory ( http://www . indiana . edu/∼microbej/index . html ) with Fiji software . Schematically , the clusters contrast was enhanced by a FFT band pass filter or by a morphological Top Hat filtering . The procedure includes automatic detection of cell shapes , medial axis determination , measurement of the cell size , and determination of the local fluorescence maxima ( cluster detection ) and of their absolute and relative position . The results of the image analysis were treated with R software ( R Core Team [2014] . R: A language and environment for statistical computing . R Foundation for Statistical Computing , Vienna , Austria . URL http://www . R-project . org ) . An R script yields the number of cells containing at least one cluster with a relative position < −0 . 6 or > +0 . 6 . These regions were defined as the cell poles ( Figure 2—figure supplement 1B ) . The density map was calculated with the R ggplot2 package ( H . Wickham . ggplot2: elegant graphics for data analysis . Springer New York , 2009 ) . The ratio of cells with a least one polar cluster over the total number of cells shown in all the figures was calculated as the mean of at least three independent experiments . In each case , more than 500 cells were analyzed from three independent experiments . In drug treatment experiments , the effect was compared against paired control experiments . All statistical tests were performed by the no parametric Wilcoxon/Mann–Whitney method when appropriate . Tests yielding a p value >0 . 05 were assumed as non-significant difference . One star stands for a p value ≤0 . 05 and two stars stand for a p value ≤0 . 01 . | Respiration occurs at different levels: the body , the organ , and the cells . At the cellular level , it is a molecular process that produces a high-energy molecule called adenosine triphosphate ( ATP ) using the biochemical energy stored in sugars , fatty acids , and other nutrients . Along with the ATP , this process also provides another source of energy to the cell: an electrochemical gradient across the membrane used for a range of processes ranging from the transport of molecules and ions to cell motility . In order to thrive , cells need to quickly respond to cues from the environment or elsewhere in the cell . A cell must therefore have the ability to increase or decrease cellular respiration and the production of ATP to ensure it has an appropriate supply of energy . In bacteria , the protein complexes responsible for cellular respiration are embedded in the cell membrane . In the past decade , research has suggested that large molecules are arranged in a specific way throughout the bacterial cell , which directly influences how they work . Alberge et al . tested this idea by studying the localization of a respiratory complex called nitrate reductase—which is important for generating energy in the absence of oxygen—through the introduction of a fluorescent marker tagged to the complex in the cell membrane of a rod-shaped bacterium called Escherichia coli . This allowed the complex to be tracked when the cells were viewed using a microscope . The experiments revealed that the location of the complex varies depending on how much energy the cell requires . For example , when the cells are in an oxygen-poor environment , the nitrate reductase complex moves towards the poles at each end of the bacterial cells . This allows the cells to produce ATP more efficiently through respiration of nitrate . Alberge et al . show that a ‘proton gradient’ , caused by positively charged hydrogen ions moving through the cell membrane as the result of respiration , controls where the complexes are located in the membrane . Alberge et al . 's findings provide experimental support that dynamic localization of respiratory complexes plays an important role in controlling respiration in bacteria . The next challenge will be to identify the genes that influence the distribution of respiratory complexes throughout the cell , which may help to explain how bacterial cells have adapted to specific environments . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"cell",
"biology",
"microbiology",
"and",
"infectious",
"disease"
] | 2015 | Dynamic subcellular localization of a respiratory complex controls bacterial respiration |
Down syndrome ( DS ) , caused by trisomy of human chromosome 21 ( Hsa21 ) , is the most common cause of congenital heart defects ( CHD ) , yet the genetic and mechanistic causes of these defects remain unknown . To identify dosage-sensitive genes that cause DS phenotypes , including CHD , we used chromosome engineering to generate a mapping panel of 7 mouse strains with partial trisomies of regions of mouse chromosome 16 orthologous to Hsa21 . Using high-resolution episcopic microscopy and three-dimensional modeling we show that these strains accurately model DS CHD . Systematic analysis of the 7 strains identified a minimal critical region sufficient to cause CHD when present in 3 copies , and showed that it contained at least two dosage-sensitive loci . Furthermore , two of these new strains model a specific subtype of atrio-ventricular septal defects with exclusive ventricular shunting and demonstrate that , contrary to current hypotheses , these CHD are not due to failure in formation of the dorsal mesenchymal protrusion .
The formation of a functional four-chambered heart is a complex process and perturbations in its development can lead to congenital heart defects ( CHD ) . These affect almost 1% of the population and are a major cause of morbidity and infant mortality ( Fahed et al . , 2013 ) . The chance of being born with CHD is drastically increased ( to ~50% ) in DS ( Vis et al . , 2009 ) in which a range of heart defects is seen , including ventricular septal defects ( VSD ) and outflow tract abnormalities such as overriding aorta ( OA ) and double outlet right ventricle ( DORV ) . Notably , defects that affect the atrioventricular ( AV ) junction , especially atrio-ventricular septal defects ( AVSD ) ( Freeman et al . , 1998 ) are often seen in DS . AVSD comprise a spectrum of cardiac malformations characterized by a common AV junction , guarded by an essentially common valve , as opposed to separate AV junctions guarded by mitral and tricuspid valves . People diagnosed with AVSD can present with communication between left and right heart chambers ( shunting ) at the atrial level ( ostium primum defect ) or at the ventricular level or with shunting at both atrial and ventricular levels ( Mahle et al . , 2006 ) . The process of forming the AV junction that divides the embryonic heart into a four-chambered structure involves the growth and fusion of a number of tissues from different precursor populations ( Webb et al . , 1998 ) . These include the atrial and ventricular septa ( mainly of myocardial origin ) , the mesenchymal cap of the atrial septum , and the superior and inferior endocardial cushions ( endocardial origin ) . In addition , the dorsal mesenchymal protrusion ( DMP ) first described as the ‘spina vestibuli’ by Wilhelm His the elder in 1880 ( His , 1880; Mommersteeg et al . , 2006 ) makes an important contribution to septation of the AV junction ( Anderson et al . , 2015; Snarr et al . , 2007a; Snarr et al . , 2007b; Webb et al . , 1998 ) . The DMP is a mesenchymal structure at the venous pole of the developing heart derived from the second heart field ( SHF ) and perturbations of its development result in AVSD ( Briggs et al . , 2013; Goddeeris et al . , 2008; Rana et al . , 2014; Tian et al . , 2010; Webb et al . , 1999; Xie et al . , 2012 ) . Abnormalities in the DMP have also been detected in human fetal hearts with trisomy 21 , leading to the hypothesis that malformation of this tissue causes the AVSD in DS ( Blom et al . , 2003 ) . Hsa21 carries 233 protein-coding genes ( genome assembly GRCh38 . p5 ) and it is thought that DS phenotypes result from an increased dosage of one or more of the genes on Hsa21 . The search for dosage-sensitive genes that when present in 3 copies cause DS phenotypes has been approached using both human and mouse genetics . In humans , rare cases of partial trisomy 21 have been used to identify critical regions that contain dosage-sensitive genes that when present in 3 copies cause DS phenotypes ( Delabar et al . , 1993; Korbel et al . , 2009; Korenberg et al . , 1994; Lyle et al . , 2009 ) . Alternatively , mouse strains have been generated to study the pathology and genetics of DS . Hsa21 shares synteny with a large region on mouse chromosome 16 ( Mmu16 ) and with shorter regions on Mmu10 and Mmu17 . The first two DS models generated , Ts65Dn ( Davisson et al . , 1990 ) and Ts1Cje ( Sago et al . , 1998 ) have duplications of regions of mouse chromosome 16 ( Mmu16 ) that are orthologous to Hsa21 , and have been used to identify dosage-sensitive genes contributing to some DS phenotypes ( Baek et al . , 2009; Chakrabarti et al . , 2010; Lana-Elola et al . , 2011; Salehi et al . , 2006; Sussan et al . , 2008 ) . However , both models also have additional aneuploidy ( trisomy of 60 genes on Mmu17 in Ts65Dn and monosomy of 7 genes on Mmu12 in Ts1Cje mice ) , making interpretation of their phenotypes difficult ( Duchon et al . , 2011 ) . Another DS model , the Tc1 mouse , carries a freely-segregating copy of Hsa21 and shows many DS phenotypes , including CHD ( Dunlevy et al . , 2010; O'Doherty et al . , 2005 ) , although the Hsa21 in this strain is not intact and the mice are mosaic for the human chromosome , again making interpretation difficult ( Gribble et al . , 2013 ) . Thanks to recent advances in chromosome engineering , a number of mouse strains with duplications of regions of mouse chromosomes orthologous to Hsa21 have been generated , resulting in partial trisomies , including the most complete mouse model for DS to date ( Dp ( 16 ) 1Yey/+;Dp ( 17 ) 1Yey/+;Dp ( 10 ) 1Yey/+ ) , which carries a duplication of all Hsa21-orthologous regions on Mmu16 , Mmu17 and Mmu10 ( Brault et al . , 2015; Li et al . , 2007; Liu et al . , 2013; Liu et al . , 2011; Olson et al . , 2004; Pereira et al . , 2009; Yu et al . , 2010 ) . However , a comprehensive mapping panel to finely map dosage-sensitive genes in DS has not been available until now . To identify dosage-sensitive critical regions and candidate genes , we now report the generation of a fine mapping panel of 7 partial trisomies of Mmu16 that can be used to identify the genetic basis of DS phenotypes where such genes reside in the region of Mmu16 orthologous to Hsa21 . We use this panel to investigate the cardiac defects in DS , employing high-resolution episcopic microscopy ( HREM ) to analyze the precise three-dimensional ( 3D ) morphology of the developing hearts . This combined approach enabled us to narrow down the critical region for CHD in DS to a minimal region containing 39 protein-coding genes and to show that this region contains at least two dosage-sensitive loci required in three copies to cause CHD . Furthermore using dual-wavelength HREM we show that the DMP develops normally in these DS models , and conclude that CHD in these DS models are not caused by failure in formation or growth of the DMP .
To expedite the identification of dosage-sensitive genes required in 3 copies to cause DS phenotypes , we generated a novel high-resolution ‘mapping panel’ of 7 strains with duplications in Mmu16 ( Figure 1a ) . Of the protein-coding genes on Hsa21 that have orthologues in the mouse , the largest fraction ( ~58% ) is located in the telomeric region of Mmu16 . Thus we used long-range Cre/loxP mediated recombination to engineer the Dp1Tyb mouse strain carrying a duplication from Lipi to Zbtb21 on Mmu16 spanning 23 Mb and 148 coding genes . We then generated 3 further strains with contiguous segmental duplications completely covering the region duplicated in Dp1Tyb: Dp9Tyb ( from Lipi to Hunk ) , Dp2Tyb ( from Mis18a to Runx1 ) and Dp3Tyb ( from Mir802 to Zbtb21 ) . To increase the resolution of the mapping panel further , we generated another 3 strains with duplications breaking up the telomeric region of Mmu16 into three contiguous fragments completely covering the region duplicated in the Dp3Tyb strain: Dp4Tyb ( from Mir802 to Dscr3 ) , Dp5Tyb ( from Dyrk1a to B3galt5 ) and Dp6Tyb ( from Igsf5 to Zbtb21 ) . We recovered live mutant mice from all 7 strains but noted that the yield of Dp1Tyb and Dp3Tyb mice was significantly reduced by 50% and 25% respectively ( Table 1 ) . We also observed hydrocephalus in Dp1Tyb mice around the time of weaning , but not in any other strain ( not shown ) . Comparative genome hybridization ( CGH ) confirmed the expected copy number increase ( from 2 to 3 ) across the duplicated regions of Mmu16 in all 7 strains , with no other copy number changes seen in the genome ( Figure 1b and data not shown ) . These new strains provide a unique resource to study DS-associated phenotypes and to map dosage-sensitive genes causing these phenotypes . To investigate whether duplication of the region of Mmu16 from Lipi to Zbtb21 in the Dp1Tyb strain was sufficient to cause CHD , we used HREM and 3D modeling ( Weninger et al . , 2006 ) , an approach we had previously used successfully to identify CHD in the Tc1 strain ( Dunlevy et al . , 2010 ) . These methods are particularly suited to examination of complex 3D structures , such as the developing heart , overcoming limitations of conventional 2D histological methods . We observed a significant increase of CHD in E14 . 5 Dp1Tyb embryos compared to their wild-type ( Wt ) littermates ( Figure 2a ) . Detailed examination of heart morphology in the mutant embryos revealed a range of defects ( Table 2 ) . About 18% of Dp1Tyb hearts show abnormal arterial trunk arrangements such as OA or DORV with subaortic communication and 62% had a VSD either alone or in combination with other defects ( Figure 2b , c and Videos 1 and 2 ) . Two subtypes of VSD were observed: perimembranous VSD ( pVSD ) , located in the membranous portion of the ventricular septum and muscular or trabecular VSD ( mVSD ) , which opens to the inlet of the right ventricle ( Figure 2c and Videos 3 and 4 ) . Around 25% of Dp1Tyb embryos displayed AVSD presenting two bridging leaflets across the single AV junction and an ‘unwedged’ morphology of the left outflow tract ( Figure 2c and Video 5 ) . Notably , the AVSD in Dp1Tyb mice were associated exclusively with a ventricular shunt and never with an atrial shunt ( Video 6 ) . Thus Dp1Tyb model a subtype of AVSD with a ventricular component , in which the cushions are attached to the leading edge of the atrial septum . Overall , these data show that the Dp1Tyb mouse models the main types of CHD seen in DS . To identify the critical region ( s ) sufficient to cause CHD in DS , we examined heart development in Dp9Tyb , Dp2Tyb and Dp3Tyb mice that between them have duplications covering the entire duplicated region in Dp1Tyb mice ( Figure 1a ) . HREM analysis of embryonic hearts at E14 . 5 revealed that neither Dp9Tyb nor Dp2Tyb showed a significant increase of CHD compared to Wt littermates ( Figure 3a ) . In contrast , the telomeric duplication of 39 coding genes in the Dp3Tyb strain was sufficient to cause a significant increase in embryos with CHD . We observed VSD such as pVSD , mVSD in the inlet portion of the ventricular septum , as well as AVSD ( Figure 3b , Table 2 and Videos 7 and 8 ) . We noted a reduced frequency of outflow tract defects in Dp3Tyb embryos compared to Dp1Tyb , however this difference was not statistically significant . Overall the types of defects observed in Dp3Tyb were very similar in both type and severity to the ones in Dp1Tyb mice , suggesting that all the dosage-sensitive gene ( s ) required to cause CHD in Dp1Tyb mice were located in this shorter region . To further narrow down the critical region for CHD , we analyzed embryonic hearts from the Ts1Rhr strain , which has a duplication that is shorter than that in the Dp3Tyb strain by just 8 genes ( 7 coding genes and 1 micro-RNA gene ) ( Olson et al . , 2004 ) . We and others have previously reported that this strain does not show CHD ( Dunlevy et al . , 2010; Liu et al . , 2011 ) . However , in our earlier study we had examined Ts1Rhr on the mixed 129S8;C57BL/6JNimr genetic background , whilst Dp1Tyb and Dp3Tyb are maintained on the C57BL/6JNimr background . Thus in order to eliminate possible confounding effects of background differences , we backcrossed Ts1Rhr to the C57BL/6JNimr background and used HREM to look for CHD . Once again we found no increased frequency of CHD in the Ts1Rhr strain ( Figure 3c ) . Thus an extra copy of one or more of the 8 genes duplicated in Dp3Tyb but not in Ts1Rhr is required to cause CHD . Next we analyzed Dp4Tyb , Dp5Tyb and Dp6Tyb mice each of which carry a duplication that between them cover the whole of the duplicated region in Dp3Tyb mice . Remarkably , none of these 3 strains showed significant increased rates of CHDs indicating that there are at least two dosage-sensitive loci in the Dp3Tyb mouse that contribute to CHD ( Figure 3c ) . Taken together these data show that a 4 . 9 Mb region of Mmu16 from Mir802 to Zbtb21 is sufficient when in 3 copies to generate cardiac defects similar to those seen in DS . Furthermore , the mapping analysis shows that there are two or more loci within this region that are required in 3 copies to cause CHD , and that at least one of these resides within the 8 genes duplicated in Dp1Tyb but not Ts1Rhr mice . The DMP plays a crucial role in the formation of the AV junction , and defects in its development have been proposed to underlie the AVSD in DS ( Blom et al . , 2003; Briggs et al . , 2012 ) . In order to assess if DMP development was perturbed in the Dp1Tyb mouse model of DS , we first established a method to follow its development during formation of the AV junction . The Isl1 gene is expressed in the SHF and also in the DMP which is derived from it and hence can be used as a marker for this tissue . We visualized expression of Isl1 using 2 different mouse strains: the Isl1Cre ( Cai et al . , 2008 ) strain crossed to Rosa26RLacZ reporter mice ( Soriano , 1999 ) to identify cells that are expressing or had expressed Isl1 , and thus are derived from the SHF; and the Isl1nLacZ strain ( Sun et al . , 2007 ) to visualize ongoing expression of Isl1 . Gene expression studies have traditionally relied on staining individual sections and rendering a 3D expression pattern by compiling histological sections , but this approach results in loss of resolution and is constrained by the chosen sectioning plane . To get around this , we utilized dual-wavelength HREM ( Mohun and Weninger , 2011 ) to visualize β-galactosidase expression in the 3D context of the developing heart in order to get an accurate view of the development of the DMP . At E11 . 5 analysis of both strains shows there is an expansion of Isl1-expressing mesenchymal tissue from within the right pulmonary ridge , which forms the DMP and has started to protrude ventrally into the cavity of the right atrium ( Figure 4 ) . One day later at E12 . 5 , the DMP has protruded more ventrally and it is now in contact with the inferior atrioventricular cushion . At these stages both reporters give a similar picture , as the DMP is actively expressing Isl1 . Later in development ( E13 . 5 ) the expression of Isl1 diminishes and totally disappears by E14 . 5 ( Figure 4; Snarr et al . , 2007b ) . However development of the DMP can still be followed using the Isl1Cre/Rosa26RLacZ fate reporter strain . This revealed that by E14 . 5 the DMP forms the ventro-caudal buttress at the core of the AV junction sandwiched between the atrial septum and the endocardial cushions that have now developed into the tricuspid and mitral valves ( Figure 4 ) . Overall , using two different genetic lineage markers of the SHF , these data show a detailed 3D view of the spatio-temporal development of the DMP . To test whether the AVSD seen in Dp1Tyb mice are caused by defects in development of the DMP , as has been proposed , we imaged the DMP in these mice using the Isl1nLacZ reporter strain . We found that at both E11 . 5 and E12 . 5 the DMP was present in Dp1Tyb embryos at a similar location to that seen in Wt mice ( Figure 5a ) . Volumetric analysis showed that at E11 . 5 in both Wt and Dp1Tyb embryos , the DMP appears rounded in shape and attached to the dorsal extracardiac mesenchyme . Furthermore , the size and shape factor of the DMP in Dp1Tyb mice was similar to that in Wt mice . At E12 . 5 the size of the DMP was reduced in the mutant mice , but in both strains the DMP was more elongated and showed a similar shape factor ( Figure 5a ) . At E11 . 5 and E12 . 5 it is not possible to identify which of the hearts would have developed AVSD , and thus we could not tell if mice with largely normal DMP development would have gone on to show defects . To evaluate this issue directly we examined E14 . 5 Dp1Tyb hearts with AVSD for the presence of the DMP . Once again we used HREM with 3D modeling and examined the same Wt and Dp1Tyb hearts using three different views . In a short axis view across the AV junction we saw an AVSD in the Dp1Tyb heart with the superior and inferior bridging leaflets across the single AV junction while the Wt heart showed a normal AV junction ( Figure 5b , left panels ) . A long axis four-chamber view of the same hearts showed an intact ventricular septum in the Wt heart but a VSD in the Dp1Tyb heart ( Figure 5b , middle panels ) . Finally , a more dorsal plane of the long axis view of the same hearts showed that the DMP was located in the correct position at the AV junction in both the Wt and the Dp1Tyb hearts ( Figure 5b , right panels , and Video 6 ) . We performed the same analysis on the Dp3Tyb mice and once again found that the DMP was present in embryos with AVSD ( not shown ) . Taken together these data show that the AVSD found in these new mouse models of DS are not caused by perturbations of the growth of the DMP . Importantly , we note that defects in the DMP have been previously shown to result in atrial septal defects or AVSD with atrial shunting ( Briggs et al . , 2012 ) , neither of which are seen in Dp1Tyb or Dp3Tyb mice . In contrast these strains show AVSD with exclusive ventricular shunting , implying that this subtype of AVSD must be due to defects in tissues other than the DMP .
Cardiac abnormalities are very common in DS . Approximately half of all babies born with DS have a heart defect , many of which are serious and need to be surgically repaired . In order to understand the genetic and molecular mechanisms that lead to CHD in DS it is essential to establish mouse models for DS that accurately recapitulate them . Here we report the generation of a new mouse strain Dp1Tyb with a 23 Mb duplication of the entire region of Mmu16 orthologous to Hsa21 , and show that it recapitulates the main types of CHD seen in DS . The Dp ( 16 ) 1Yey mouse strain carries a similar duplication to Dp1Tyb encompassing the same set of genes and was also reported to have heart defects during embryogenesis ( Li et al . , 2007 ) . Interestingly , the cardiac phenotypes in Dp1Tyb are very similar to the ones we previously observed in the transchromosomic strain Tc1 ( Dunlevy et al . , 2010; O'Doherty et al . , 2005 ) , despite differences in the Hsa21 orthologous gene content in the two strains and the different species origin for the duplicated genes: mouse in Dp1Tyb mice , human in Tc1 mice . This phenotypic similarity supports the conclusion that the CHD seen in these strains are caused by the same pathological mechanisms as those in people with DS . As with CHD in humans with DS we observed incomplete penetrance of the phenotype in Dp1Tyb mice . In humans this has been ascribed to genetic modifiers ( Li et al . , 2012 ) . However the Dp1Tyb mouse strain was analyzed on an inbred background , suggesting that incomplete penetrance may be caused by stochastic effects during embryonic cardiogenesis . While the recovery of E14 . 5 Dp1Tyb embryos was consistent with expected Mendelian numbers ( not shown ) , the recovery of Dp1Tyb mice at weaning was reduced to 50% of the expected numbers ( Table 1 ) . Some of this loss could be due to the most severe CHDs , since AVSDs were found in ~25% of E14 . 5 Dp1Tyb embryos , but it is likely that there are other unknown causes for this perinatal lethality . The other strain to show AVSDs , Dp3Tyb , showed a 25% reduction in recovery of mice at weaning , similar to the observed frequency of AVSDs at E14 . 5 , making this a likely cause of the reduced yield of mutant mice . Together with the Dp1Tyb strain we generated a collection of 7 new mouse strains with segmental duplications ranging from 1 . 5 Mb to 23 Mb that provide a genetic resource for mapping dosage-sensitive genes required in 3 copies to cause DS phenotypes . We used this mapping panel to identify a 4 . 9 Mb genomic region ( from Mir802 to Zbtb21 ) that when present in 3 copies is sufficient to cause CHD . Furthermore , from analysis of shorter duplications , we show that the phenotype is caused by at least two distinct dosage-sensitive loci . By combining this data with analysis of the Ts1Rhr strain , we have determined that , minimally , one of the two loci lies either in an interval at the centromeric end of the duplication in Dp3Tyb ( from Mir802 to Setd4 ) or at the telomeric end ( from Mx2 to Zbtb21 ) ( Figure 6a ) . These two segments between them contain only 7 coding genes ( Setd4 , Mx2 , Tmprss2 , Ripk4 , Prdm15 , C2cd2 and Zbtb21 ) and one microRNA gene ( Mir802 ) , none of which had been previously implicated in causing CHD in DS . A previous study proposed that a 3 . 7 Mb genomic region of Mmu16 ( Ifnar1 – Kcnj6 ) in Dp ( 16 ) 4Yey was sufficient to cause CHD ( Liu et al . , 2013 ) ; this region overlaps the centromeric end of the duplicated region of Dp3Tyb by 1 . 9 Mb from Mir802 to Kcnj6 , which may help to further narrow down the search for causative genes ( Figure 6b ) . We note that while the Dp5Tyb and Ts1Rhr strains did not show a statistically significant increase in CHDs compared to their wild-type littermate controls , in both strains we detected 2 mutant embryos with AVSDs ( out of 20 or 21 embryos analyzed respectively ) . Thus it is possible that these strains have a weak phenotype – a larger number of embryos from these strains would need to be analyzed to establish whether this is significant . If true , it would suggest that one or more of the causative genes may reside within the Dp5Tyb interval ( which is entirely contained within the Ts1Rhr interval ) . In contrast , analysis of humans with partial trisomies of Hsa21 identified distinct regions that contribute to cardiac abnormalities ( Barlow et al . , 2001; Korbel et al . , 2009 ) . An individual ( PM ) trisomic for the 7 . 7 Mb region from PSMG1 to PRMT2 had a VSD , whereas the shortest partial trisomy ( individual BA ) that gave rise to AVSD extended 10 . 0Mb from HLCS to PRMT2 ( Figure 6b ) . We note that that while both of these intervals overlap substantially with the minimal Dp3Tyb region identified in our studies , the duplicated region in individual PM does not overlap at all with the Ifnar1 – Kcnj6 region identified by Liu et al as being sufficient to cause CHDs ( Liu et al . , 2013 ) . In particular the human studies suggested that increased dosage of the DSCAM gene may be responsible for the CHDs ( Korbel et al . , 2009 ) . In contrast the mouse studies show that duplication of Dscam is neither necessary or sufficient to cause CHDs , since the gene is not within the minimal Ifnar1 – Kcnj6 region identified by Liu et al as being sufficient for CHDs ( Liu et al . , 2013 ) but is duplicated in the Dp6Tyb strain which we show here does not have CHDs . Taking the human and mouse data together , this suggests that several dosage-sensitive genes may contribute to CHDs , with potentially no single gene being absolutely required . Dp1Tyb mice show AVSD with ventricular but not atrial shunting . Analysis of AVSD in people with DS showed that the most common subtype was a complete AVSD with both atrial and ventricular shunting and the next most common was AVSD with exclusive ventricular shunting similar to that seen in the Dp1Tyb mouse ( Freeman et al . , 2008 ) . Thus , although Dp1Tyb mice show AVSD , they only model one subtype of AVSD seen in people with DS – AVSD with ventricular shunting . It is not known why some cases of DS have complete AVSD whereas others have subtypes with only atrial or ventricular shunting , but it is tempting to speculate that there are distinct mechanisms driving defects in the atrial and ventricular septa . AVSD have been shown to result from defects in development of the DMP ( Briggs et al . , 2013; Goddeeris et al . , 2008; Rana et al . , 2014; Tian et al . , 2010; Webb et al . , 1999; Xie et al . , 2012 ) . Importantly , a variety of different genetic manipulations that affect the DMP all resulted in AVSD with exclusive atrial shunting , demonstrating the importance of the DMP for the correct formation of the atrial side of the AV junction ( Briggs et al . , 2012 ) . In contrast , in Dp1Tyb mice we found normal DMP development and no AVSD with atrial shunting , but instead saw exclusively AVSD with ventricular shunting ( Figure 6c ) . Thus we propose that defects in the atrial septum causing the ostium primum type of AVSD are caused by malformation of the DMP , whereas ventricular defects are due to perturbations in other tissues , such as endocardial cushions or myocardium . Interestingly , Gata4-deficient mice display AVSD and have a normal DMP , but show perturbations in myocardial signaling to the adjacent atrioventricular cushion mesenchyme , suggesting that defects in myocardial signaling could lead to AVSD ( Misra et al . , 2014; Rajagopal et al . , 2007 ) . To our knowledge the Dp1Tyb and Dp3Tyb mouse strains are the first to have been shown to have AVSD with ventricular shunting and thus will allow direct studies of the pathological mechanisms underlying this subtype of AVSD . In conclusion , we have generated a mapping panel of 7 mouse strains that can be used to identify dosage-sensitive genes underlying the broad range of DS phenotypes , and we have exploited this panel to map the location of genes causing CHD in DS .
We generated ES cells with a duplication from Lipi to Zbtb21 using Cre/loxP-mediated chromosome engineering ( Yu and Bradley , 2001 ) following strategies similar to those previously used to make mice with duplications of Hsa21-orthologous regions ( Brault et al . , 2015; Li et al . , 2007; Liu et al . , 2013; Liu et al . , 2011; Olson et al . , 2004; Pereira et al . , 2009; Yu et al . , 2010 ) . MICER vectors ( Adams et al . , 2004 ) MHPP352i17 ( coordinates of homology region 16:74930370–16:74937378 Mb , mouse assembly GRCm38/mm10 ) and MHPN352i16 ( 16:97977263–16:97982380 Mb ) were used sequentially as targeting vectors to insert loxP sites proximal to Lipi and distal to Zbtb21 respectively in HM-1 ES cells ( Magin et al . , 1992 ) ; the cells had been tested to be negative for mycoplasma contamination . Targeting was carried out by standard procedures . Cre recombinase was transiently expressed in double-targeted ES cells to induce recombination between the loxP sites . Segmental duplication was confirmed by Southern blot analysis , targeted clones were injected into blastocysts to generate chimeric mice and these were bred to establish the C57BL/6J . 129P2-Dp ( 16Lipi-Zbtb21 ) 1TybEmcf/Nimr ( Dp1Tyb ) mouse strain by standard methods . ES cells with a duplication from Mis18a to Runx1 were generated in the same way using MICERs MHPP323h04 ( 16:90563769 – 16:90577148 Mb ) and MHPN219i02 ( 16:93054020 – 16:93062456 Mb ) to target loxP sites proximal to Mis18a and distal to Runx1 respectively , followed by Cre-mediated recombination , and correctly targeted ES cells were used to establish the C57BL/6J . 129P2-Dp ( 16Mis18a-Runx1 ) 2TybEmcf/Nimr ( Dp2Tyb ) mouse strain . For the remaining duplication strains C57BL/6J . 129P2-Dp ( 16Mir802-Zbtb21 ) 3TybEmcf/Nimr ( Dp3Tyb ) , C57BL/6J . 129P2-Dp ( 16Mir802-Dscr3 ) 4TybEmcf/Nimr ( Dp4Tyb ) , C57BL/6J . 129P2-Dp ( 16Dyrk1a-B3galt5 ) 5TybEmcf/Nimr ( Dp5Tyb ) , C57BL/6J . 129P2-Dp ( 16Igsf5-Zbtb21 ) 6TybEmcf/Nimr ( Dp6Tyb ) , and C57BL/6J . 129P2-Dp ( 16Lipi-Hunk ) 9TybEmcf/Nimr ( Dp9Tyb ) we used an in vivo Cre-mediated recombination strategy whereby we bred female mice containing the Hprttm1 ( cre ) Mnn allele ( Tang et al . , 2002 ) and two loxP sites located in trans configuration on Mmu16 at the boundaries of the desired duplication , to C57BL/6JNimr males and Cre activity in the female germline from the Hprttm1 ( cre ) Mnn allele resulted in occasional pups ( 0 . 7–6% ) with recombination between the loxP sites generating the duplication . The loxP sites were derived from targeting with the 4 MICERs described above as well as MHPP432c09 ( 16:94538615–16:94546849 Mb ) located between Dscr3 and Dyrk1a and MHPN235b18 ( 16:96327324–16:96331804 Mb ) located between B3Galt5 and Igsf5 . Dp ( 16Cbr1-Fam3b ) 1Rhr ( Ts1Rhr ) mice ( Olson et al . , 2004 ) , Isl1tm3Sev ( Isl1nLacZ ) ( Sun et al . , 2007 ) , Isl1tm1 ( cre ) Sev ( Isl1Cre ) ( Cai et al . , 2008 ) , Gt ( ROSA ) 26Sortm1Sor ( ROSA26RLacZ ) ( Soriano , 1999 ) , Hprttm1 ( cre ) Mnn and all duplication mouse strains generated above were maintained by backcrossing to C57BL/6JNimr . All mice used for experiments had been backcrossed for at least 5 generations . Specifically , Dp1Tyb was analyzed after backcrossing to C57BL/6JNimr for 6–9 generations ( N6-N9 ) , Dp2Tyb at N5 , Dp3Tyb at N8 , Dp4Tyb at N5-N7 , Dp5Tyb at N5 , Dp6Tyb at N5-N11 , Dp9Tyb at N11 and Ts1Rhr at N11-N12 . Genotyping was carried out using custom probes ( Transnetyx ) . All animal work was carried out under a Project Licence granted by the UK Home Office . To count numbers of genes in the duplication intervals we used the Biomart feature of Ensembl ( mouse genome assembly GRCm38 . p4 ) to count numbers of genes within a given interval , filtering either for protein-coding genes or for protein-coding genes with orthology to human . In addition we included the Mx2 gene as a protein-coding gene , since Ensembl did not automatically classify this gene as coding . Genomic DNA was prepared from the tail of each of the duplication strains and from C57BL/6JNimr mice to be used as a reference control using either phenol-chloroform extraction or DNeasy Blood and Tissue Kit ( Qiagen , UK ) . DNA ( 1 μg ) was analyzed by Roche Diagnostics Limited using a mouse 3 × 720 K array ( Roche NimbleGen ) or by Oxford Gene Technology using a mouse 1 × 1 M array ( Agilent Technologies ) . 50–75 mers probes were used and the design was based on the genome assembly mm9 . The hybridized aCGH slides were scanned for Cy3 ( test ) and Cy5 ( control ) channels . The Log2 ratios of the test/control were calculated and plotted as graphs using Prism 7 . E14 . 5 embryonic hearts were dissected and fixed for 30 min in 4% paraformaldehyde followed by a 1 hr wash in distilled water and secondary fixation overnight . Fixed samples were dehydrated and embedded in modified JB4 methacrylate resin ( Weninger et al . , 2006 ) and sectioned at 2 μm . HREM imaging ( isometric resolution of 2 μm ) used a Hamamatsu Orca-HR camera . Data sets were normalized and subsampled prior to 3D volume rendering using Osirix v5 . 6 ( Rosset et al . , 2004 ) . As the expected CHD were not fully penetrant , a minimum of 17 E14 . 5 mutant embryos were compared to littermate controls . Phenotype analysis was performed blind for genotype , and classification of type of CHD was carried out as previously described ( Dunlevy et al . , 2010 ) . For dual-wavelength HREM , a conventional Xgal reaction was performed followed by 4% paraformaldehyde fixation , dehydration and embedding , and imaging was carried out using a Jenoptik ProgRes C14 camera with dual filter ( 59022bs , Chroma Technology Corp ) . Image analysis of the DMP was done using ITKsnap 2 . 4 . 0 and Volocity 6 . 2 . 1 software packages . Volocity was used to calculate the volume of the DMP and the shape factor ( shape factor is 1 for a perfect sphere and <1 for more irregular shapes ) . A minimum of 9 biological replicates per group was analyzed and analysis of the DMP was performed blind for genotype . | Down syndrome is a condition caused by having an extra copy of one of the 46 chromosomes found inside human cells . Specifically , instead of two copies , people with Down syndrome are born with three copies of chromosome 21 . This results in many different effects , including learning and memory problems , heart defects and Alzheimer’s disease . Each of these different effects is caused by having a third copy of one or more of the approximately 230 genes found on chromosome 21 . However , it is not known which of these genes cause any of these effects , and how an extra copy of the genes results in such changes . Now , Lana-Elola et al . have investigated which genes on chromosome 21 cause the heart defects seen in Down syndrome , and how those heart defects come about . This involved engineering a new strain of mouse that has an extra copy of 148 mouse genes that are very similar to 148 genes found on chromosome 21 in humans . Like people with Down syndrome , this mouse strain developed heart defects when it was an embryo . Using a series of six further mouse strains , Lana-Elola et al . then narrowed down the potential genes that , when in three copies , are needed to cause the heart defects , to a list of just 39 genes . Further experiments then showed that at least two genes within these 39 genes were required in three copies to cause the heart defects . The next step will be to identify the specific genes that actually cause the heart defects , and then work out how a third copy of these genes causes the developmental problems . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"developmental",
"biology"
] | 2016 | Genetic dissection of Down syndrome-associated congenital heart defects using a new mouse mapping panel |
Serotonin is implicated in many aspects of behavioral regulation . Theoretical attempts to unify the multiple roles assigned to serotonin proposed that it regulates the impact of costs , such as delay or punishment , on action selection . Here , we show that serotonin also regulates other types of action costs such as effort . We compared behavioral performance in 58 healthy humans treated during 8 weeks with either placebo or the selective serotonin reuptake inhibitor escitalopram . The task involved trading handgrip force production against monetary benefits . Participants in the escitalopram group produced more effort and thereby achieved a higher payoff . Crucially , our computational analysis showed that this effect was underpinned by a specific reduction of effort cost , and not by any change in the weight of monetary incentives . This specific computational effect sheds new light on the physiological role of serotonin in behavioral regulation and on the clinical effect of drugs for depression . Clinical trial Registration: ISRCTN75872983 DOI: http://dx . doi . org/10 . 7554/eLife . 17282 . 001
Selective serotonin reuptake inhibitors ( SSRI ) are the most prescribed medications for major depressive episodes ( Bauer et al . , 2008 ) . The effects of SSRI on improving mood and reducing anxiety have been well documented ( Trivedi et al . , 2006; Stahl , 2008; Cipriani et al . , 2009 ) and many experimental studies showed that serotonin modulates emotional processing even in healthy volunteers ( Harmer et al . , 2009; Serretti et al . , 2010 ) . Loss of interest or pleasure in daily activities is also a key symptom of depression ( American Psychiatric Association , 2013 ) . However , the impact of SSRI on the motivation deficit ( apathy ) remains rather controversial ( Papakostas et al . , 2006; Weber et al . , 2009 ) . Indeed , several studies reported that SSRI treatments do not reduce apathy as much as other symptoms ( Fava et al . , 2014 ) or can even induce apathy in patients ( Barnhart et al . , 2004; Sansone and Sansone , 2010; Padala et al . , 2012 ) . However , other studies also reported that SSRI treatments increase motivation or at least the sensitivity to reward ( Tang et al . , 2009; Stoy et al . , 2012; Yuen et al . , 2014 ) . The related animal literature is also contradictory: some studies on cost/benefit trade-off showed reduced effort expenditure with SSRIs ( Yohn et al . , 2016 ) or no effect with serotonin blockers ( Denk et al . , 2005 ) , while other studies reported increased motor activity induced by SSRI ( Weber et al . , 2009 ) or optogenetic stimulation of the dorsal raphe nucleus ( Warden et al . , 2012 ) . One issue with clinical studies is that SSRI effects may be confounded by an interaction with the pathological state: an SSRI treatment was reported to decrease apathy in late-life depression ( Yuen et al . , 2014 ) and to increase it in Parkinson’s disease ( Zahodne et al . , 2012 ) . Another issue is that apathy may not be a simple construct . It is typically captured as a loss of behavioral activation ( Cléry-Melin et al . , 2011; Treadway et al . , 2012 ) , which could arise from different causes at the cognitive level: for instance a diminished sensitivity to reward attractiveness , or alternatively an exacerbated sensitivity to action cost . The ambition of the present study therefore was two-fold: we aimed ( 1 ) to clarify the effects of an SSRI on the motivation of effortful behavior in healthy volunteers , unconfounded by psychopathology , and ( 2 ) to reconcile these effects with a more general role of serotonin in the sensitivity to action costs vs . benefits . Although experimental findings seem diverse at first sight , they might suggest a general role for serotonin in the behavioral adaptation to action costs . Previous studies suggested that serotonin is implicated in the motor aspects of action production and also the valuation processes that motivate the action . Notably , serotonin was found to condition impulse control and the capacity for behavioral inhibition ( Cools et al . , 2005; Crockett et al . , 2009; Warden et al . , 2012; Guitart-Masip et al . , 2014 ) and to promote patience for delayed rewards ( Schweighofer et al . , 2008; Miyazaki et al . , 2014; Worbe et al . , 2014 ) . Serotonin also impacts the determinants of actions , for instance , how positive and negative outcomes guide learning in humans ( Chamberlain et al . , 2006; Crockett et al . , 2009; Faulkner and Deakin , 2014 ) , monkeys ( Clarke et al . , 2004 ) and rodents ( Bari et al . , 2010 ) . Several attempts have been made previously to capture serotonin function in a coherent computational theory . Key ideas are that serotonin ( 1 ) regulates the impact of action costs such as punishments ( Daw et al . , 2002; Niv et al . , 2007 ) or delay ( Niv et al . , 2007; Cools et al . , 2011; Miyazaki et al . , 2014 ) and ( 2 ) adjusts the propensity to activate versus inhibit the action ( Boureau and Dayan , 2011; Cools et al . , 2011 ) . Borrowing from these models , we retain the general working hypothesis that serotonin regulates behavioral activation by modulating the weight of action costs rather than benefits . We further suggest that the notion of action cost could be more general than initially envisaged . Indeed , in the literature quoted above , actions are mostly implemented as binary responses ( e . g . , approach vs . avoidance ) and their costs or benefits are manipulated through the valence of their outcome ( e . g . , monetary gain or loss ) . Such tasks therefore over-simplify the real-life situation , where actions may often require more or less effort , depending on their intensity and duration . Here , we tested the specific hypothesis that serotonin regulates the weight of effort cost in action production , as opposed to the weight of expected benefit . To assess the impact of changing cerebral serotonin levels and how it unfolds over time , we compared different groups of healthy subjects treated under double blind conditions with escitalopram ( an SSRI ) or placebo during eight weeks . Participants performed a previously published task ( Meyniel et al . , 2013 ) , during which they allocated physical effort over time in order to maximize a monetary payoff that increased with effort duration ( see Figure 1A ) . The task thus involves trading a physical effort cost against distinct levels of monetary incentives . Although changes in cost and benefit were independently manipulated in the task , they may result in intricate effects at the behavioral level . To disentangle between potential effects of serotonin on cost or benefit estimates , or both , we used a formal model of how decisions are generated by a hidden level of computations in our task , as previously described ( Meyniel et al . , 2013 , 2014 ) . Based on this computational analysis of the behavior , we could pinpoint the specific effect of serotonin on cost-related parameters , and track the predicted effects of such a computational change onto the experimental measures . 10 . 7554/eLife . 17282 . 003Figure 1 . Task design and behavioral performance . ( A ) The screenshots depict a trial as it was presented to subjects . Subjects were free to allocate their effort as they wished over the 30s corresponding to the trial duration . They were instructed that their monetary payoff would be proportional to both the monetary incentive and the effort duration , i . e . the time spent squeezing a handgrip harder than a target force level , which varied with task difficulty . Subjects were provided with on-line feedback on the payoff accumulated in the trial ( score on the top ) and on the instantaneous pressure exerted on the grip ( fluid level in the thermometer ) . The force time series of an example trial is shown below the screenshots , revealing 3 effort periods , with rewarded effort ( force above target ) plotted in black ( not gray ) . Two factors were manipulated across trials: ( i ) the incentive level , shown as a coin image ( 1 , 2 or 5p ) and ( ii ) the difficulty level , corresponding to the same white bar in the thermometer reached with different target force levels ( 70% , 80% or 90% of the maximal force ) . The last screen summarized the payoff cumulated over preceding trials . ( B ) Using a double-blind procedure , healthy subjects were assigned to one of the two treatment groups , corresponding to a daily intake of either placebo or escitalopram ( 10 mg during the initial phase , 20 mg during the intermediate and late phase ) during 9 weeks . Each subject completed the effort allocation task three times at distinct treatment phases ( initial , intermediate and late ) . Numbers of subjects and visits correspond to data sets included in the analysis after compliance and quality checks . ( C ) The three left-most graphs show task performance ( as reflected in monetary payoff ) sorted by treatment group ( black: placebo; gray: escitalopram ) and time since treatment onset . Statistical significance was assessed with two-sample , two-sided t-tests . On the right-most plot , payoff was averaged over visits at the subject level . Statistical significance was assessed with ANOVAs including treatments as between-subject factors and test phase ( initial , intermediate or late ) as a within-subject factor . *p<0 . 05; **p<0 . 005 . Error bars indicate Student's 95% confidence intervals . DOI: http://dx . doi . org/10 . 7554/eLife . 17282 . 00310 . 7554/eLife . 17282 . 004Figure 1—source data 1 . The MATLAB data file contains the payoff earned by each participant at each visit , in the placebo and escitalopram groups . DOI: http://dx . doi . org/10 . 7554/eLife . 17282 . 004
58 subjects were included in the analysis , see Table 1 . We took the cumulated payoff as the primary measure of performance in the task and compared it between groups and visits ( see Figure 1C ) . We found a significant effect of treatment group: performance was significantly improved in the escitalopram group ( F1 , 58 . 4=9 . 37 , p=0 . 003 ) as compared to the placebo group . This difference was stable over time: there was no significant interaction between groups and visits ( F2 , 91=0 . 22 , p=0 . 8 ) . The average payoff per visit was £35 . 9±1 . 21 s . e . m . and £30 . 1±1 . 44 s . e . m . in the escitalopram and placebo groups respectively . The better performance observed in the escitalopram group could be underpinned by different mechanisms: alleviation of effort costs or inflation of incentive values , or both . To disentangle between the two mechanisms , we relied on a computational model of effort allocation that was previously proposed ( Meyniel et al . , 2013; Meyniel et al . , 2014 ) . This model assumes that a single computational variable , termed cost evidence , accounts for the decision to stop and resume effort exertion . Cost evidence waxes during effort ( with slope Se ) until reaching an upper bound where effort is stopped , and it wanes during rest ( with slope Sr ) until reaching a lower bound where effort is resumed . The distance between bounds is the cost-evidence amplitude ( denoted A ) . Effort and rest durations are determined by the ratios of amplitude and slopes ( see Figure 2A ) so that performance depends on the value of latent parameters ( A , Se , Sr ) and its potential modulation by task factors ( monetary incentive and effort difficulty ) . We used a Bayesian Model Selection procedure previously validated ( Meyniel et al . , 2013 , 2014 ) to pinpoint the effect of the task factors onto these latent parameters ( see Materials and methods ) . 10 . 7554/eLife . 17282 . 005Figure 2 . Computational results . ( A ) The cost-evidence accumulation model assumes that effort and rest durations are respectively determined by the accumulation ( mean slope Sem ) and dissipation ( mean slope Srm ) of cost evidence between bounds ( mean amplitude Am ) . Possible modulations of these parameters by incentive and difficulty levels were implemented in 20 distinct models . In the best model identified ( #20 ) by Bayesian selection , increasing effort difficulty shortens effort duration by steepening the accumulation slope ( a parametric effect controlled by parameter Sed and illustrated with colors from yellow to red ) . Increasing the incentive level has two effects: first , it shortens rest duration by speeding up the dissipation ( parametric effect of Sri , illustrated by colors from dark to light blue ) ; second , it lengthens effort duration by pushing back the bounds ( parametric effect of Ai , illustrated by green scaling ) . ( B ) Plots show inter-subject means and Student's 95% confidence intervals obtained for the fitted values of model parameters ( which were averaged over visits at the subject level ) . To facilitate visual comparison , scales and offsets were adjusted so that mean and error bars are visually equal across plots in the placebo group . Statistical significance corresponds to ANOVAs including treatment group ( escitalopram vs . placebo ) as a between-subject factor and treatment phase as a within-subject factor ( initial , intermediate or late ) ; **p<0 . 005 . ( C ) Data in the placebo group served as a baseline to simulate effort and rest durations after imposing a 20% increase in computational parameters . In the table , each row corresponds to a simulated change in one single parameter . Colors denote the effect sizes recovered by model fitting for each parameter , as percent of change compared to baseline . Numbers indicate the percentage of 'hit' ( on the diagonal ) and 'false alarm' ( off-diagonal ) in detecting a significant change in parameter values with a paired t-test thresholded at p<0 . 01 . ( D ) The graph illustrates why the effect of escitalopram , characterized at the computational level as a reduced accumulation slope of cost-evidence during effort ( Sem ) , should translate at the behavioral level into both a longer effort duration and an increased sensitivity of effort duration to incentive level . DOI: http://dx . doi . org/10 . 7554/eLife . 17282 . 00510 . 7554/eLife . 17282 . 006Figure 2—source data 1 . The MATLAB data file contains the fitted value of parameters Ai , Sem , Sed , Srm , Sri ( see Materials and methods , Equation 2 ) , for each participant at each visit , in the placebo and escitalopram groups . DOI: http://dx . doi . org/10 . 7554/eLife . 17282 . 006 Replicating our previous studies , the best model showed that incentives impacted the amplitude ( A ) and dissipation slope ( Sr ) , whereas effort difficulty impacted the accumulation slope ( Se ) , as illustrated in Figure 2A , right . To capture all the effects , the cost-evidence accumulation model therefore necessitates five free parameters: the mean slope of cost-evidence accumulation ( Sem ) and its steepening for higher difficulty levels ( Sed ) ; the mean slope of cost-evidence dissipation during rest ( Srm ) and its steepening for higher incentives ( Sri ) ; and the expansion of the cost-evidence amplitude as higher incentives push the bounds back ( Ai ) . The fact that the same best model was found independently in each treatment group with high confidence levels ( exceedance probabilities xp>98% in each group ) indicates that all subjects can be characterized within this common computational framework . We computed for each subject the best-fitting values of the model parameters ( Ai , Sem , Sed , Srm , Sri ) . Since the interaction between treatments ( placebo vs . escitalopram ) and treatment phase was not significant for any parameter ( all F2 , 91<2 . 03 , p>0 . 14 ) , we provide fitted values pooled over treatment phases in Figure 2B . Only one model parameter was significantly different between the placebo and escitalopram groups: the cost-evidence accumulation slope ( Sem , F1 , 58 . 0=9 . 88 , p=0 . 003 ) . This parameter captures the average value of effort cost across conditions: there was a 31% decrease in the escitalopram group compared to the placebo group ( on average with ± s . e . m . , placebo: 0 . 16±0 . 012 , escitalopram: 0 . 11±0 . 008 ) ; the difference was actually significant at each visit ( all p<0 . 015; two-sided t-test ) . The difference between escitalopram and placebo never reached significance for the other model parameters ( all p>0 . 08 , see Table 2; on average with ± s . e . m . for placebo vs . escitalopram , Ai: 0 . 15±0 . 028 vs . 0 . 19±0 . 022 , Sed: 0 . 021±0 . 004 vs . 0 . 013±0 . 002 , Srm: 0 . 37±0 . 022 vs . 0 . 42±0 . 029 , Sri: 0 . 086±0 . 018 vs . 0 . 126±0 . 016 ) . We checked the sensitivity and specificity of our model fitting procedure in detecting a treatment effect through simulations . A 20% change in any given parameter was reliably detected and the difference recovered by model fitting was significant in a 96% of simulations at least for Ai , Sem , Srm and Sri and in a 77% of simulations for Sed . Importantly , a change in one single parameter was recovered without propagating to other parameters and the false alarm rate was below 5% for all parameters ( Figure 2C ) . 10 . 7554/eLife . 17282 . 007Table 1 . Details on participants N corresponds to the number of subjects per treatment type and phase . A few datasets were not available due technical problems and late withdrawals . Based on criteria specific to the present task ( and not to the clinical trial ) , some subjects were excluded from the analysis ( 'excluded' ) . We report the age and sex of participants included in the analysis and the exact time of their test since the treatment onset . DOI: http://dx . doi . org/10 . 7554/eLife . 17282 . 007Treatment typeTreatment phaseN not availableN excludedN after exclusionSex ( Male / female ) Age ( years ) ± SD of included subjectsTime since treatment onset ( days ) ± SD for included subjectsPlaceboInitial052714/13 23 . 4 ± 4 . 35 3 . 0 ± 0 . 68 PlaceboIntermediate062614/12 23 . 2 ± 4 . 33 13 . 8 ± 1 . 13 PlaceboLate042815/13 23 . 4 ± 4 . 27 54 . 7 ± 4 . 98 EscitalopramInitial182311/12 24 . 5 ± 4 . 71 3 . 1 ± 0 . 63 EscitalopramIntermediate162512/13 24 . 5 ± 4 . 51 14 . 0 ± 0 . 87 EscitalopramLate262410/14 24 . 6 ± 4 . 61 55 . 3 ± 4 . 69 10 . 7554/eLife . 17282 . 008Table 2 . Treatment effect on computational parameters and behavioral measures . All numbers are p-values obtained from ANOVAs . p-values lower than 0 . 05/5=0 . 01 ( computational parameters ) and 0 . 05/6=0 . 008 ( behavioral measures ) appears in bold to show significant effects that survive correction for multiple comparisons . DOI: http://dx . doi . org/10 . 7554/eLife . 17282 . 008VariableTreatmentVisitInteractionAi0 . 3880 . 1970 . 406Sem0 . 003 0 . 0390 . 260Sed0 . 07970 . 5430 . 137Srm0 . 1860 . 7780 . 612Sri0 . 1300 . 1960 . 557Effort duration – mean0 . 002 0 . 1990 . 875Effort duration – sensitivity to incentive0 . 0230 . 1870 . 115Effort duration – sensitivity to difficulty0 . 2470 . 8130 . 318Rest duration – mean0 . 2130 . 4820 . 531Rest duration – sensitivity to incentive0 . 1620 . 9370 . 807Rest duration – sensitivity to difficulty0 . 1150 . 4230 . 681 To further test the specificity of SSRI effect on cost-evidence accumulation , we performed another Bayesian model selection that contrasted the two groups of subjects . Parameters of the cost-evidence accumulation model were fitted on the placebo group data to serve as a reference . In the escitalopram group , they were fixed to these reference values and only one or two modulations were permitted to capture the treatment effects . Models including a modulation of Sem outperformed the others ( log Bayes Factor , all Δ>147 . 9 ) , see Table 3 . A version with modulations of both Sem and Sed was slightly more likely than a version with only a modulation of Sem ( Δ=3 ) and much more likely than any other combination ( Δ>5 . 9 ) . 10 . 7554/eLife . 17282 . 009Table 3 . Model comparison assessing the specificity of treatment effect . Data in the escitalopram group were fitted with the cost-evidence accumulation model . The parameters were fixed to the values fitted onto the placebo group , excepted when a modulation was permitted . The first row contains models that permit the modulation of one single parameter , whereas the remaining rows correspond to models that permit a combination of two modulations . Each cell gives log Bayes Factor ( i . e . log model evidence ) relative to the null model . Higher values denote better models . DOI: http://dx . doi . org/10 . 7554/eLife . 17282 . 009SemSedSrmSriAiOnly one modulation173 . 619 . 9−4 . 2 −3 . 8 3 . 7Also includes Sem176 . 6169 . 6169 . 7170 . 7Also includes Sed15 . 716 . 121 . 7Also includes Srm−7 . 3 −1 . 1 Also includes Sri1 . 2 Thus , the specific effect of SSRI on cost-related parameters supports our hypothesis that serotonin is involved in the estimation of action cost , rather than benefits . Reducing the effort cost ( Sem ) only should have two effects at the behavioral level ( see Figure 2D ) . The first is straightforward: the duration of effort epochs should be longer . The second is less trivial: the sensitivity of effort duration to incentive level should be increased . This is because in our model , incentive level modulates the amplitude between bounds ( as captured by parameter Ai ) . Thus , the effect of incentive level on effort duration is proportional to the accumulation slope Sem: if the accumulation is slower , then a given displacement of the upper bound will have a larger effect on effort duration . Fulfilling these predictions , we found both a longer effort duration per se ( F1 , 58 . 1=10 . 72 , p=0 . 002 , on average placebo: 7 . 63s±0 . 56 , escitalopram: 10 . 2s±0 . 55 , with s . e . m . ) and a higher sensitivity of effort duration to incentive level ( F1 , 58 . 5=5 . 46 , p=0 . 023 , on average with ± s . e . m . , placebo: 1 . 11±0 . 22 , escitalopram: 1 . 95±0 . 26 ) in the escitalopram group compared to placebo ( see Figure 3A ) . The difference between the two treatment groups never reached significance for the other behavioral variables ( all p>0 . 11 , on average with ± s . e . m . , for placebo vs . escitalopram , effort sensitivity to difficulty: −0 . 84±0 . 13 vs . −1 . 06±0 . 14 , rest duration: 3 . 17s±0 . 19 vs 2 . 85s±0 . 19 , rest sensitivity to incentives: −0 . 19±0 . 04 vs . −0 . 30±0 . 05 , rest sensitivity to difficulty: 0 . 077±0 . 018 vs . 0 . 025±0 . 03 ) . 10 . 7554/eLife . 17282 . 010Figure 3 . Behavioral results . ( A ) Plots show inter-subject means and Student's 95% confidence intervals obtained from linear regression . Regression coefficients were averaged over visits at the subject level . To facilitate visual comparison , scales and offsets were adjusted so that mean and s . e . m . are visually equal across plots in the placebo group . Statistical significance corresponds to ANOVAs including treatment group ( escitalopram vs . placebo ) as a between-subject factor and treatment phase as a within-subject factor ( initial , intermediate or late ) ; *p<0 . 05 , **p<0 . 005 . ( B ) As predicted by the cost-evidence accumulation model , effort duration and its sensitivity to incentive level are correlated across subjects ( one dot corresponds to one subject; values were averaged across visits for each subject ) . The line shows the linear regression fit obtained when pooling the two treatment groups ( ρ56=0 . 55 , p<10–5 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 17282 . 01010 . 7554/eLife . 17282 . 011Figure 3—source data 1 . The MATLAB data file contains a description of the behavior obtained by linear regressions for each participant at each visit , in the placebo and escitalopram groups . The regression weights correspond to the mean effort duration and mean rest duration , and their modulation by incentive levels and difficulty levels . DOI: http://dx . doi . org/10 . 7554/eLife . 17282 . 011 Crucially , because they arise from a common cause , the model also predicts that the two effects should be correlated: the more Sem is reduced , the more effort duration and its sensitivity to incentive level should be increased . The Pearson correlation over subjects was indeed significantly positive in both groups ( placebo: ρ27=0 . 43 , p=0 . 02; escitalopram: ρ27=0 . 54 , p=0 . 002 , see Figure 3B ) .
Our behavioral results in healthy volunteers show that the SSRI escitalopram improves global performance and hence payoff in a task that involves trading effort cost against monetary benefit . Taking advantage of the independent manipulation of cost and benefit , our computational analysis characterized the effect of escitalopram as a specific diminution of effort cost . Together with previous findings , our results support the hypothesis that cost may be a general functional domain for serotonin ( Boureau and Dayan , 2011; Dayan , 2012 ) . For instance , interventions targeting serotonergic transmission during probabilistic and reversal learning paradigms in rodents , monkeys and humans suggested that serotonin impacts sensitivity to negative rather than positive feedback ( Clarke et al . , 2004; Chamberlain et al . , 2006; Crockett et al . , 2009; Bari et al . , 2010; Cohen et al . , 2015 ) . The reduction of serotonergic transmission in humans following acute tryptophan depletion ( for a review , see Faulkner and Deakin , 2014 ) reduced information sampling in a decision-making task , but crucially , only when sampling had a financial cost ( Crockett et al . , 2012 ) . Serotonin may also impact moral costs , such as unfairness in social decision-making ( Crockett et al . , 2008 ) . Furthermore , serotonin may control the cost of delay in reward delivery in rodents , such that a higher firing rate of serotonergic neurons correlates with an increased ability to wait for bigger rewards ( Miyazaki et al . , 2014 ) , without affecting the sensitivity to the reward itself ( Fonseca et al . , 2015 ) . Conversely , low serotonin levels , e . g . after acute tryptophan depletion , were suggested to exacerbate sensitivity to the cost of waiting , which could result in impulse control disorders ( Crockett et al . , 2009; Cools et al . , 2011; Dayan , 2012 ) . Thus , it is tempting to build a parsimonious computational theory on the idea that serotonin is involved in processing all kinds of action costs . This is compatible with the notion that different types of costs are processed by distinct neural circuits , since neuromodulators such as serotonin can affect many brain regions . However , such a generalization would necessitate assessing the putative role of each mono-amine and their interactions with different types of costs within the same study . Previous attempts in rodents have observed effects that are inconsistent with the present results: in cost/benefit trade-off tasks , serotonin blockers impacted choices only when costs were delays , and not physical efforts ( Denk et al . , 2005 ) , whereas dopamine blockers affected both types of costs . Such discrepancies may call for caution when translating the results of pharmacological studies in rodents into medication effects in humans . A general role for serotonin in cost processing also seems compatible with the effects on apathy and impulsivity reported in healthy and pathological conditions ( Cools et al . , 2005; Papakostas et al . , 2006; Schweighofer et al . , 2008; Crockett et al . , 2009; Weber et al . , 2009; Warden et al . , 2012; Guitart-Masip et al . , 2014; Miyazaki et al . , 2014; Worbe et al . , 2014 ) . Indeed , according to our interpretation , SSRIs might reduce the perceived cost of performing actions , which would promote behavioral activation , hence the alleviation of apathy . As SSRIs might also reduce the perceived cost of delaying action , they would improve response control and reduce impulsivity ( Cools et al . , 2005; Cools et al . , 2011; Crockett et al . , 2009; Dayan , 2012; Bari and Robbins , 2013; Miyazaki et al . , 2014; Fonseca et al . , 2015 ) . However , it is important to note that the SSRI effects obtained here relate to effort cost and not to time discounting . Although time is central in our task , it departs from delay and waiting paradigms in several respects . First , the overall task duration was fixed and its pace was independent from how subjects allocated their effort within trials . Second , the reward was gained instantaneously , concomitantly to effort production , without delay . Therefore , the SSRI effect on effort duration ( how long subjects sustain an effort ) cannot be explained by a change in temporal discounting ( how long subjects wait for a reward ) . The direction of the effect obtained in the present study suggests that higher serotonin level alleviates the effort cost . This interpretation is based on the assumption that the primary effect of SSRI is to increase serotonin level in the synaptic cleft ( Nutt et al . , 1999; Stahl , 2008 ) . However , the net effect of SSRI might not be so straightforward to interpret at the molecular level for several reasons . First , a high tonic concentration could have the paradoxical effect of reducing the sensitivity to phasic serotonergic signals ( Faulkner and Deakin , 2014 ) . Second , the ubiquitous negative feedback regulation by auto-receptors can initially revert the response expected from high extracellular levels of serotonin ( Stahl , 2008; Fischer et al . , 2014 ) and also produce non-linear dose-response functions for cognitive performance ( Bari et al . , 2010 ) . Third , much evidence suggests that different serotonin projections to different forebrain systems mediate varying acute and chronic adaptive responses to aversive events ( Hale et al . , 2013; Faulkner and Deakin , 2014 ) . The effect of serotonin also depends on the location in the brain and on the type of receptors , for instance , 5-HT1A and 5-HT2A receptors respectively inhibit and excite motoneurons ( Jacobs and Azmitia , 1992 ) . Thus , the superficial interpretation that serotonin helps overcoming effort costs will need to be refined by addressing the complexity of SSRI effects at the molecular level . We nevertheless note that the superficial interpretation is consistent with demonstrations that serotonin also helps overcoming other costs such as delay in humans ( Schweighofer et al . , 2008 ) . Refinement is also needed at the computational level . Our model does not specify how exactly serotonin could attenuate the impact of effort cost on action production . An indirect effect through an increase in muscular capacity can be excluded since there was no difference in maximal force between the placebo and SSRI groups at any visit . The computational analysis simply suggests that a slower accumulation of effort cost ( lower Sem ) under SSRI prolonged effort duration . It does not distinguish between down-regulation of the cost signal itself , or down-regulation of the weight this cost has in the decision to produce an effort . The former ( perceptual ) view would be in line with an analgesic effect of escitalopram , which is consistent with the findings that nociception and/or somatosensory perception are both modulated by serotonin at central levels ( Jacobs and Azmitia , 1992 ) , that serotonin is targeted by common drugs modulating pain like acetaminophen ( Smith , 2009 ) and that serotonin modulates the inhibitory feedback loop that allows muscular fatigue to down-regulate the motor command ( Gandevia , 2001; Cotel et al . , 2013 ) . The latter ( decisional ) view would be more generalizable to the capacity of overcoming other types of costs . As the serotonin and dopamine systems strongly interact ( Dremencov et al . , 2009; Cools et al . , 2011; Schilström et al . , 2011; Fischer et al . , 2014 ) , the SSRI effect observed here might also be at least in part mediated by dopamine . Indeed , the beneficial effect of SSRI shown here is reminiscent of previous reports about dopaminergic manipulations in mesolimbic structures , which affected the propensity to choose high reward – high effort over low reward – low effort options in rodents ( Cousins et al . , 1996; Salamone et al . , 2007 ) . However , if SSRIs antagonize dopamine release ( Dremencov et al . , 2009; Cools et al . , 2011 ) , escitalopram should have reduced effort production in our task ( but see Schilström et al . , 2011 ) . Moreover , the fact that escitalopram did not modulate incentive effects on parameters such as the amplitude or dissipation slope argues against a participation of dopamine . Indeed dopaminergic manipulations in humans have been repeatedly shown to affect reward processing , not only in learning contexts ( Frank et al . , 2004; Pessiglione et al . , 2006; Palminteri et al . , 2009 ) but also in reward/effort trade-off paradigms ( Wardle et al . , 2011; Treadway et al . , 2012; Le Bouc et al . , 2016 ) . It might be that serotonin and dopamine have complementary roles in promoting action production ( and alleviating apathy ) : the former by reducing effort cost and the latter by enhancing the incentive value of potential rewards . Previous studies suggest that we should distinguish between acute and chronic effects of SSRIs ( Fischer et al . , 2014 ) . We did not find an interaction with time in our results but only a main effect of treatment group . Both parametric and non-parametric statistics showed that such a group difference is very unlikely to arise from chance in sampling the population ( under the null hypothesis ) . General psychological assessment at baseline was also similar in the two groups . However , we acknowledge that our shortest time since treatment onset ( 3 days ) may already depart from the acute regime , that ceiling effects in the SSRI group could in principle mask an interaction with time , and that our study may lack the statistical power to reveal such an effect . This absence of an interaction contrasts with the apparently delayed clinical effect of SSRIs in depressed patients , which usually take weeks to improve mood ( Stahl , 2008 ) . However , at the molecular level , serotonin release is boosted by SSRIs from treatment onset ( Kobayashi et al . , 2008 ) and SSRI effects on emotional processing also occur with little or no delay in healthy subjects ( Harmer et al . , 2003 ) . In pathological conditions , time may be needed to adjust to the new , less negative perception of costs , as well as the reduced emotional bias , and convert these implicit changes into a conscious subjective improvement that can be reported to the practitioner ( Harmer et al . , 2009; Cools et al . , 2011 ) . This idea has been formalized in a recent study showing how positive and negative outcomes can shape mood on the long run ( Eldar and Niv , 2015 ) . Further studies in depressed patients are needed to assess whether an early detection of effort cost attenuation could be used to predict long-term treatment effects on clinical symptoms . Finally , our results also indicate that non-trivial behavioral effects can be accounted for by a change in a single computational parameter ( Sem ) : a specific modulation of cost can result in both longer efforts and an increased sensitivity of effort duration to potential benefits in our results . Intuitively , this effect on reward sensitivity corresponds to the idea that when perceived costs are too high , a change in reward prospect will have little effect . Thus , our analysis shows that the behavioral consequences of cost and benefit estimates are intricate but computationally tractable . Our paradigm and model could provide experimental and conceptual tools to refine the description of motivational disorders . Distinct dysfunctions , such as amplification of effort cost vs . flattening of reward prospects , might call for different treatments: if drugs modulating serotonin only affect cost estimates , then other drugs ( possibly dopaminergic ) should aim to impact the valuation of potential benefits .
Healthy volunteers ( 18–45 years old ) were recruited to the study by public advertisement after approval by the Ethics Committee of Berkshire , UK ( protocol CL1-20098-81 , on 28th Sept . 2011 ) and registration as a clinical trial , ISRCTN75872983 . Participants gave their written informed consent prior to participation . Normal health was checked by clinical and psychiatric examinations including laboratory tests . Exclusions and withdrawals from the study were adjusted to include 64 participants ( 2 treatment groups of 32 participants , 16 men per group ) , tested at 3 separate visits , resulting in 192 completions of the task . Given that the clinical trial was exploratory and also included tests from other research groups , the sample size was not selected specifically for our study; however it appeared reasonable given typical studies in the field . Indeed , published between-subject comparisons of placebo and anti-depressant treatments were made on a lower sample size per group ( e . g . N=14 in Harmer et al . , 2004 , N=20 in Chamberlain et al . , 2006 and N=30 in Guitart-Masip et al . , 2014 ) , and a single visit per subject ( while we have three ) . Due to technical problems or late withdrawals , 4 task completions were not fully acquired and therefore unusable . The remaining data were checked for quality by F . M . prior to unblinding: 11 were excluded due to mis-calibration of task difficulty; 9 due to hardware default or signal quality; 15 for non-compliance with task instructions . Subjects were asked to produce an effort in every trial . As there were eight trials per condition in the task , a given visit was excluded for non-compliance when the total number of effort ( or rest ) was lower than three in at least one condition . As a result , the number of participants per visit varied between 23 and 28 in each treatment group , for a total of 153 task completions . Note that a given participant may produce an interpretable dataset at a given visit and not at another . The total number of participants with at least one interpretable dataset across the three visits happened to be 29 in each treatment group . The finally included data set is summarized in Table 1 . Sex ratio ( 14/29 vs . 16/29 , Z test for proportion: z= 0 . 53 , p=0 . 6 ) , ratio of excluded data sets ( 15/96 vs . 20/92 , z=1 . 08 , p=0 . 28 ) and age ( mean with ± s . e . m . , 23 . 2 ± 0 . 8 vs . 24 . 2 ± 0 . 8 , p=0 . 38 ) were similar in the two groups . We also used psychological tests to assess differences between groups of subjects at baseline , before treatment . T-test comparison showed no significant difference in Mood Visual Analog Scales for the items 'happy' ( p=0 . 75 ) , 'sad' ( p=0 . 96 ) , 'hostile' ( p=0 . 48 ) , 'alert' ( p=0 . 73 ) , 'anxious' ( p=0 . 51 ) , 'calm' ( p=0 . 14 ) , nor in the Hospital Anxiety Depression score for anxiety ( p=0 . 68 ) and depression ( p=0 . 35 ) . The difference in State-Trait Anxiety Inventory score was at p=0 . 04 at baseline but p=0 . 45 and p=0 . 83 after 7 and 55 days of treatment respectively . Therefore , psychological variables were not significantly different at baseline when correcting for multiple comparisons ( at p=0 . 05/9 ) and the one passing the uncorrected threshold p=0 . 05 was not significantly different during the testing phase . Two centers participated in the study ( Oxford and Manchester , UK ) . Data were collected between January 2012 and July 2013 . Participants were randomly assigned to one of the two parallel treatment groups following a double-blind procedure: placebo or escitalopram ( 10 mg during week #1 and #9; 20 mg from week #2 to #8 ) . The randomisation list was constructed in blind , by the Institut de Recherches Internationales Servier , with stratification by gender and center ( Oxford , Manchester ) . Treatments were conditioned by Les Laboratoires Servier Industrie so as to be visually indistinguishable and shipped in numbered containers to the investigators . The randomisation list was not made available to the investigators until the final data set had been checked for quality and locked by transfer to a Contract Research Organization ( Biotrial ) . Participants took a daily oral capsule around 8 P . M . for 9 weeks and performed the Effort Allocation Task three times at distinct latencies ( Figure 1B ) : initial ( 2–4 days after treatment onset ) , intermediate ( 12–17 days ) and late ( 52–60 days for all subjects but 3 , who were tested between the 33rd and 44th days before withdrawing from the study ) . The task was not performed on week #9 . The clinical trial included other tests , not presented here , to assess emotional processing , sexual acceptability and learning abilities . Safety evaluations ( adverse events collection , blood pressure/heart rate and laboratory test ) were performed along with the study . The Effort Allocation Task is schematized in Figure 1A and detailed in a previous publication ( Meyniel et al . , 2013 ) . The only change was the adoption of the local currency ( British pounds ) . On each trial , participants had 30 s that could be spent either resting or squeezing a handgrip . They were instructed that the payoff would be proportional to both the monetary incentive and the time spent above a force target corresponding to effort difficulty . The task lasted approximately one hour and was split into 8 blocks . The factor levels ( monetary incentives: 1 , 2 or 5p and effort difficulty: 70 , 80 , 90% of the maximal force ) were manipulated independently and crossed , resulting in 9 conditions , each corresponding to a trial , presented in a randomized order in each block . Left and right hands were used alternatively over blocks . Participants were encouraged to maximize their payoff at each trial , and told that the cumulated payoff would be added to their financial compensation for participating in the study . Unbeknown to them , this payoff was rounded up to a fixed amount after the last visit such that all participants eventually received the same total . The task difficulty was adjusted to the subject's maximal force , which was measured at each visit before the test . The procedure for measuring the maximal force ( sustained handgrip squeezing ) is detailed in ( Meyniel et al . , 2013 ) and follows published guidelines ( Gandevia , 2001 ) . The maximal force was not affected by treatments ( F3 , 113 . 7=0 . 93 , p=0 . 42 ) , visits ( F2 , 180=1 . 34 , p=0 . 26 ) or by the interaction of these two factors ( F6 , 180=0 . 27 , p=0 . 95 ) . The cumulated effort duration determined the payoff obtained at each trial . Yet there are many ways of chunking this cumulated duration depending on the duration of each effort and rest epochs . Effort and rest epochs were determined based on the force time series ( Figure 1A provides an example ) using an off-line algorithm ( Meyniel et al . , 2014 ) . The offline detection algorithm was based on both the force signal normalized by the calibration maximal force , and its temporal derivative . Samples with positive derivative , exceeding one standard deviation of the derivative time series , and force level higher than 0 . 5 ( half the maximum ) were tentatively marked as effort onsets . Effort offsets were defined similarly for negative derivative values and force levels below 0 . 5 . When multiple offsets were detected between two onsets , all but the last one were discarded . If multiple onsets were detected before an offset , the one with minimum force was kept . An offset was marked at the trial end if the effort was still sustained at that moment . Elapsed time between effort onsets and offsets determined effort and rest durations . The first rest duration was the elapsed time between trial onset and the first effort onset . We performed a model-free analysis of the behavior , for each subject and visit , with multiple linear regressions done separately for effort and rest durations . The linear models included the factors of interest ( incentive and difficulty ) and temporal factors ( block number; trial number within a block , effort or rest epoch number within a trial ) . Regressors ( excepted the constant ) were z-scored so that regression coefficients ( beta estimates ) correspond to standardized effect sizes . These beta values were then compared between treatments using ANOVA ( see statistical analysis below ) . We characterized the effect of treatments on ( 1 ) the monetary payoff , ( 2 ) the fitted parameters of the best computational model ( Equation 2 ) , ( 3 ) the regression coefficients of the model-free analysis . We provide source data files for each of these three sets of variables . We used a repeated-measure ANOVA with subjects as a random factor , treatment group as a between-subject factor and time since treatment onset ( initial , intermediate , late ) as a within-subject factor . Interaction between the within-subject factors was included in all ANOVAs . We followed up the ANOVA results with two-sample two-sided t-tests at a given time since treatment onset . Significance levels corrected for comparing multiple variables are included in Table 2 . Given our large sample size ( N=58 included subjects ) , these classical parametric statistics reliably quantify the likelihood that the observed differences between treatment groups reported in our study may be due to chance in sampling the population . Indeed , we confirmed these significance levels with non-parametric permutation tests , using 10 , 000 permutations of treatment labels between subjects to estimate the probability that an equal or more extreme statistic ( F or T depending on the test ) could occur by chance . | Neuromodulators are chemicals released in the brain that affect the activity of brain cells . Serotonin is a neuromodulator with the most complicated role: it is released in most brain regions and affects behavior in diverse ways . Serotonin is implicated in the regulation of mood , anxiety , impulsivity and learning . Moreover , most medications for depression target serotonin . A lack of motivation is an important symptom of depression , but exactly how serotonin affects motivation still remains unclear . Meyniel et al . studied how increasing the amount of serotonin in the brain affects motivation in healthy people . The volunteers in the experiments squeezed a handgrip: the longer they squeezed , the more money they got as a reward . Before the experiment , some of the volunteers received an antidepressant drug that increases the amount of serotonin surrounding their brain cells , while others received a placebo . The experiments revealed that , compared to the people who had the placebo , those who received the drug put in more effort to get a reward . More serotonin could increase motivation by reducing the perceived cost of putting in more effort , or by making people value the reward more . A mathematical model of the results showed that the increased motivation in the antidepressant group was more consistent with serotonin reducing the cost of putting in an effort , rather than increasing how much the reward was valued . Combined with previous findings , these results suggest that serotonin affects the processing of cost associated with tasks – be that the amount of effort required , delays in getting a reward , or a punishment . Further experiments are now required to understand if the same mechanism operates in people with depression , and if so , whether it can be altered to promote recovery . It will also be important to better understand the interaction between serotonin and other neuromodulators such as dopamine . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"computational",
"and",
"systems",
"biology",
"neuroscience"
] | 2016 | A specific role for serotonin in overcoming effort cost |
SARS-CoV-2 is notable both for its rapid spread , and for the heterogeneity of its patterns of transmission , with multiple published incidences of superspreading behaviour . Here , we applied a novel network reconstruction algorithm to infer patterns of viral transmission occurring between patients and health care workers ( HCWs ) in the largest clusters of COVID-19 infection identified during the first wave of the epidemic at Cambridge University Hospitals NHS Foundation Trust , UK . Based upon dates of individuals reporting symptoms , recorded individual locations , and viral genome sequence data , we show an uneven pattern of transmission between individuals , with patients being much more likely to be infected by other patients than by HCWs . Further , the data were consistent with a pattern of superspreading , whereby 21% of individuals caused 80% of transmission events . Our study provides a detailed retrospective analysis of nosocomial SARS-CoV-2 transmission , and sheds light on the need for intensive and pervasive infection control procedures .
Reducing the spread of SARS-CoV-2 is a crucial priority for controlling and limiting the impact of the COVID-19 pandemic . Key metrics in assessing transmission are the basic and effective reproduction R numbers , which describe the mean number of infections caused by a typical infected individual in totally and partially susceptible populations , respectively ( Anderson and May , 1992 ) . However , individual variations from this mean can be of vital importance ( Lloyd-Smith et al . , 2005 ) ; a study of SARS-CoV-2 in Hong Kong suggested that 80% of transmission events resulted from only 19% of cases ( Adam et al . , 2020 ) . Superspreader events are widely reported to play a key role in the spread of the virus in community settings ( Shen et al . , 2004; Kucharski and Althaus , 2015; Hamner , 2020; Ebrahim and Memish , 2020; Lemieux et al . , 2020 ) . Transmission within hospitals has been identified as a critical concern in managing the COVID-19 pandemic ( Iacobucci , 2020 ) . Studying transmission in the hospital environment requires care in distinguishing cases acquired in the community from cases of nosocomial infection ( Sikkema et al . , 2020 ) . The identification of outbreaks may be complicated by the potential for asymptomatic carriage of the virus ( Rivett et al . , 2020 ) . As such , testing of asymptomatic health care workers ( HCW ) has been proposed as a means to reduce viral spread and protect the workforce and patients ( Black et al . , 2020; Jones et al . , 2020 ) . Evaluating SARS-CoV-2 transmission in a hospital context is not a trivial task . Factors such as the date of symptom onset relative to the date of admission can be used to identify cases of likely nosocomial transmission ( Rickman et al . , 2021; Price et al . , 2021 ) . Phylogenetic methods can be used to identify putative clusters of infection occurring within a single ward or other location within a hospital setting ( Meredith et al . , 2020 ) . Epidemiological methods can be used to look at potential contacts and opportunities for transmission between cases of infection ( Cluster Track , Camart Ltd , Cambridge , UK ) . However , these approaches do not always provide the detail of who infected whom within a single outbreak or cluster , lacking the resolution to resolve the fine detail of transmission clusters . Viral genome sequences provide a valuable resource for evaluating nosocomial transmission . At the most basic level , highly distinct sequences are unlikely to be related via transmission . Multiple approaches have been proposed to infer transmission patterns from genome sequences . Typically , phylogenetic reconstruction is used to infer relationships between sequences , an evolutionary model being combined with epidemiological data to infer a network of transmission events ( Volz and Frost , 2013; Ypma et al . , 2012; Didelot et al . , 2014; Hall et al . , 2015 ) . Modelling approaches have been extended to include factors such as unsampled hosts ( De Maio et al . , 2016 ) , the availability of multiple samples per patient ( Wymant et al . , 2018; Worby et al . , 2016 ) , incomplete epidemics ( Didelot et al . , 2017 ) , and deep sequence data ( Ratmann et al . , 2019 ) . Here , we evaluated patterns of viral transmission occurring in epidemiological clusters in Cambridge University Hospitals NHS Foundation Trust ( CUH ) , United Kingdom , where multiple patients with suspected hospital-acquired COVID-19 infections and/or HCW working on the same wards tested positive for SARS-CoV-2 within a 2-week period . Using a novel approach to combine genetic and epidemiological data , we inferred networks of SARS-CoV-2 transmission between these individuals . The tight clustering of genome sequences collected within a single ward places an imperative on the exploitation of non-genetic information to identify potential transmission events . We did this by combining an evolutionary model with symptom and location data for individuals considered , and knowledge of SARS-CoV-2 infection dynamics . Examining data from the largest clusters of infection identified within the hospital , we showed that the spread of infection in these clusters was driven by a small set of superspreader individuals .
We developed a method to infer networks of transmission events between individuals within CUH . Our method combines knowledge of SARS-CoV-2 infection dynamics with viral genome sequences and data describing the movements of patients and HCWs within the hospital . Applying our method to these data , we generated maximum likelihood reconstructions of the pattern of transmission events occurring within five infection clusters , each of which was centred on a ward at CUH . For reasons of data protection we term these wards A to E . These five wards were chosen as they contained the largest number of patients with hospital-onset infections and/or healthcare worker infections in CUH up to the end of the study period . Of the wards analyzed , wards A to D were ‘green’ wards ( designated for patients who had not tested positive for SARS-CoV-2 ) , while ward E was a ‘red’ ward ( designated for known COVID-19 patients ) . Although referred to here as a ‘ward’ for simplicity , one of the green wards was a number of neighbouring clinical areas within the hospital . A preliminary analysis of the data , treating individuals in a pairwise manner , suggested that transmission events between the identified wards was unlikely ( Figure 1 ) . Reconstructed transmission networks for the four green wards are shown in Figure 2 . Our method requires each transmission in a network to be consistent with a statistical model of pairwise viral transmission ( Illingworth , 2020 ) . As such , a broad range of possibilities could in theory be inferred . At one extreme , the infections on a ward could all arise from a single introduction of the virus , with all cases arising via transmission from a single individual . At the other extreme , the infections could all be entirely independent of one another , with no transmission between cases at all . Our approach uses sequence data and epidemiological information to identify cases that are plausibly linked by direct transmission , before inferring the maximum likelihood network reconstruction of events . Across the green wards , the majority of cases were inferred to be connected to at least one other via transmission , with 42 out of 54 cases being joined into networks . These networks involved between 2 and 11 cases each ( mean 5 . 9 cases ) . This contrasts with results from the single red ward ( Figure 3 ) , in which only 9 of 19 cases were inferred to be linked to others via transmission ( mean inferred network size 2 . 3 cases ) . Our result corresponds to the nature of the wards studied; the repeated transfer of infected patients onto a red ward leads to an increased number of independent introductions . Individuals in our study were divided into patients and HCWs allowing for the estimation of rates of transmission between these categories . Of the 38 transmission events in the maximum likelihood networks , 20 were patient-to-patient , 8 were from patient to HCW , 8 were HCW-to-HCW , and just 2 were from HCW to patient ( Figure 2—figure supplement 1 ) These results suggest that patients were significantly more likely to be infected by other patients than by health care workers ( p-value 6 . 1 x 10−5 , one-tailed binomial test ) . By contrast , HCWs were at approximately equal risk of being infected by patients and other HCWs . Some of the wards analysed appeared to show uneven patterns of viral transmission , with a small number of individuals responsible for most of the infections observed . For example , in the maximum likelihood reconstruction derived for ward A , the majority of individuals infected did not pass on the virus , while individuals A6 and A10 were the sources , respectively , of four and five transmission events ( Figure 2A ) . A statistical analysis of the inferred networks provided evidence for a role for superspreading behaviour during transmission . As a first step in evaluating this , we calculated the level of uncertainty in each inferred network , combining data from the maximum likelihood network with that from other plausible , but lower-likelihood networks . Figure 2—figure supplement 2 shows statistical ensembles of networks inferred for the green wards . In this figure , the width of an arrow is proportional to the probability that transmission occurred between each pair of individuals . The maximum likelihood network inferred for Ward E was the only plausible solution given by our reconstruction method . In a second step , we fitted models to data from these ensembles , calculating probability distributions of the number of individuals infected by each person in the dataset . A negative binomial model , in which the extent of viral spreading was overdispersed , gave a better explanation of the data , measured using the Bayesian Information Criterion , than a simpler model in which all individuals transmitted equally ( Figure 4 ) . In the best-performing model of viral spreading , 87% of individuals either did not transmit the virus , or transmitted only to one other . Taken across all individuals , 21% of individuals were responsible for 80% of viral transmission , a result very similar to that found among the general population ( Adam et al . , 2020 ) . A repeat of this calculation for the green wards alone gave similar statistics ( Figure 4—figure supplement 1 ) , with 23% of individuals in these wards being responsible for 80% of transmission ( Figure 4—figure supplement 1 ) . In our maximum likelihood reconstructions , a total of five individuals infected three or more others , including one HCW and four patients . Clinical characteristics of these individuals were explored , but are not described in detail or assigned to their anonymised ward clusters to preserve patient anonymity . Of the four patients , all had suspected hospital-acquired COVID-19 and significant comorbidities: two had a history of chronic liver disease , and two had previous haematological malignancies , one of whom was still on immunosuppressive treatment . Immunosuppression has been associated with prolonged viral shedding ( Italiano et al . , 2020; Avanzato et al . , 2020 ) . One superspreader was confused and mobile on the ward . Another had a fever for several days before being tested for SARS-CoV-2 , which had been attributed to a pre-existing community-acquired bacterial infection . The only HCW superspreader exclusively infected other HCWs , and shared accommodation with several of these individuals . Cycle threshold ( Ct ) values of samples collected from identified superspreader individuals were not statistically distinct from those from individuals in the study in general ( Figure 4—figure supplement 2 ) . Inferred timings of transmission events caused by superspreaders showed a variety of patterns ( Figure 4—figure supplements 3–5 ) . In ward B , the initial three infections of HCWs by the individual B0 are likely to have occurred within a short period of the SARS-CoV-2 virus entering the ward ( within 4 days with 95% certainty ) , suggesting that an outbreak caused by superspreading may spread rapidly to multiple individuals . However , the inferred timings on other wards were less conclusive; in wards A and C the inferred distributions of infection times were more diffuse . We note simply that where superspreader events occur , the potential exists for multiple transmissions to occur within a short space of time .
We have here outlined a novel approach for the inference of transmission networks . Our approach combines an evolutionary model with specific information about SARS-CoV-2 transmission dynamics to identify the most probable set of transmission events linking a set of cases of infection . Our approach builds upon previous approaches to analysing SARS-CoV-2 data from hospital settings , going beyond the identification of clusters to infer directional networks of viral transmission . The multiple forms of data used by our method each play a critical role in network inference . Where individuals in a ward become symptomatic at similar times , sequence data provide a strong indication of whether these infections are linked via transmission or arise from completely independent events . However , the potentially short time spanned by a local outbreak may be insufficient for substantial genetic variation to accumulate in the virtual population; in such cases , other information , such as dates of symptom onset , become critical for network reconstruction . The inference of networks allows for detailed study of how SARS-CoV-2 can spread within a clinical environment . Contacts between patients appear to be crucial , as they are primarily infected by other patients rather than through transmission from HCWs . This finding has potential implications for the application of protective measures within a hospital environment , wherease face mask usage was enforced for individuals in outpatients and for HCWs in all areas of the hospital , inpatients were not at the time of data collection subject to the same precautions . A recent study has suggested SARS-CoV-2 aerosolisation to be high in areas where patients with COVID-19 are coughing ( Hamilton , 2021 ) . Our study is biased in its consideration of the largest clusters of infection identified in CUH wards during the first wave of the pandemic . Examining data from these clusters , we identified a pattern of superspreading , in which a small proportion of individuals were responsible for the majority of nosocomial transmission events . Our result is interesting in the context of previous studies of superspreading ( Shen et al . , 2004; Kucharski and Althaus , 2015 ) , providing an example of this in a hospital context , and a case in which a small proportion of ‘superspreader individuals’ drive ‘superspreader events’ . We note that , while prolonged or increased viral shedding would increase the chance of an individual becoming a superspreader , behavioural and environmental factors may also be influential . A key feature of SARS-CoV-2 that makes infection prevention and control ( IPC ) particularly challenging is its significant infectivity prior to the onset of symptoms . This means that isolating staff or patients once symptoms are recognised is not sufficient to prevent transmission . The superspreaders identified here illustrate several principles for limiting the spread of SARS-CoV-2 in hospitals . First , scrupulous adherence to infection control practices including use of appropriate personal protective equipment ( PPE ) at all times , even on green wards and in non-clinical hospital areas , is required to limit transmission between asymptomatic patients and staff in which COVID-19 is not suspected . Use of masks by patients , including on green wards , should be instituted if tolerated , particularly when staff are present in patient bed spaces . Second , healthcare professionals must be vigilant to the possibility of hospital-onset COVID-19 and have a low threshold for testing inpatients , even where an alternative differential diagnosis for the patients' symptoms exists . Third , as soon as positive cases are confirmed , appropriate isolation and PPE precautions should be used , along with contact tracing , testing and isolation . Patients who have been in direct contact with confirmed cases on green wards should be isolated . Fourth , regular screening of asymptomatic individuals can help to identify patients and staff that may be unsuspectingly infected with COVID-19 and infectious , either pre-symptomatic , pauci-symptomatic or asymptomatic , prompting isolation and contact tracing ( Jones et al . , 2020 ) . Fifth , ventilation should be improved to reduce the risk of aerosol dispersal . Of note , our recommendations concur with the current UK guidance for COVID-19 infection prevention and control ( Public Health England , 2020 ) , which have evolved during the course of the pandemic . Finally , our identification of transmission from patients to HCWs highlights the use of higher grade respiratory precaution to protect HCWs ( such as FFP3 respirators ) as an important topic for future research . The potential for superspreading enhances the difficulty of controlling hospital-acquired infection , particularly as most transmission events from superspreaders to other people inferred in this study occurred within a relatively short time period . By the time a second linked case in a ward is identified , the potential exists for an index case to have infected multiple individuals , making it too late to prevent a broader outbreak . While this study does not allow for a complete characterisation of superspreading individuals , it may suggest possible risk factors in these instances such as immunosuppression ( associated with prolonged shedding ) , more mobile behaviour that may have contributed to increased risk of transmission , and extended symptoms ( fever ) prior to testing and isolation ( due to fever being attributed to an alternative cause ) . Our inferences of transmission were performed on the basis of a largely complete dataset . Sampling of infections within wards was likely very close to complete , with sample collection from symptomatic patients and health care workers being conducted in parallel to asymptomatic screening of hospital staff . A screening programme for asymptomatic HCW was set up at CUH in April 2020 ( Rivett et al . , 2020 ) and voluntary weekly screening is currently offered to all HCW . SARS-CoV-2 seroprevalence among staff tested in CUH between 10th June and 7th August 2020 was 7 . 2% ( Cooper , 2020 ) . The five outbreaks described here occurred earlier in the pandemic ( March to June 2020 ) , when staff seroprevalence would have been lower . The proportion of staff with neutralising antibodies during the outbreaks was therefore low , and likely played a minor role in transmission dynamics . Sequencing was attempted for all positive samples; across the green wards data was of high quality for 55 out of 71 individuals for whom data was collected ( >80% unambiguous nucleotides with no more than one ambiguous nucleotide at at a variant site ) consensus viral genome in addition to data describing their location during the period of the study . We acknowledge several limitations to our study . There is potential for missing or incomplete data , with some aspects of the data more vulnerable than others to omission . Asymptomatic screening was offered to all staff working on the five wards analysed in the study during the outbreaks . It is theoretically possible that HCW could have caught COVID-19 early on in the outbreaks and cleared the virus quickly , becoming negative at time of testing , or caught the virus asymptomatically after the screening test , or had levels of virus below the detection limit of the assay ( and thus have been false negatives ) . However , levels of SARS-CoV-2 RNA below the assay detection limit are unlikely to be infectious ( and thus not significant for the inferred transmission networks ) , and overall HCW testing coverage was high . Testing of asymptomatic patients varied by ward . Asymptomatic screening was done for all patients on Wards A and B during the outbreaks , and all patients entering Ward E ( the only ‘red’ ward included in the study ) were known SARS-CoV-2 positives . However , for Wards C and D , systematic asymptomatic screening of all patients on the ward during the outbreaks was not performed , and it is possible some asymptomatic infections ( that could have contributed to the transmission networks ) were missed . Data describing the wards on which patients were treated is likely to be complete , but the same ward data for HCWs may miss the potential for interactions between workers outside of their base wards for example in communal non-clinical areas within the hospital . Missing location data would lead to the non-identification of genuine contacts; our approach may therefore underestimate the number of infections caused by transmission between health care workers . Where data were missing our method does not attempt to identify cases of indirect transmission , for example invoking the presence of unobserved individuals . Only potential cases of transmission that were compatible with a model of direct transmission were included in our networks . The number of superspreader individuals identified in this study ( five ) is too small to draw general conclusions on superspreader characteristics . Moreover , it is not possible to disentangle whether superspreading was driven mainly by individual factors ( such as infectivity or behaviour ) or environmental factors ( such as patient placement and ventilation at time of peak infectivity ) , or a combination of these . Ct values can vary for many reasons including the timing of sampling during COVID-19 infection , sampling type and technique , viral transport , sample preparation and variability between PCR runs . The finding that Ct values did not vary significantly between superspreader and non-superspreader individuals should therefore be interpreted with caution . In conclusion , we have here applied a combined statistical approach to infer and examine SARS-CoV-2 transmission networks within a hospital environment during the first wave of the pandemic in the United Kingdom . For the largest ward outbreaks of hospital-onset COVID-19 , the majority of transmission was driven by a small proportion of individuals . Future developments could include exploring the impact of variables that may be associated with an increased transmission risk . Examples would include novel SARS-CoV-2 variants such as B1 . 1 . 7 and B1 . 617 . 2 , which appear to be more readily transmissible ( Rambaut et al . , 2020 ) , patient characteristics such as immunosuppression which are associated with prolonged viral shedding ( Avanzato et al . , 2020 ) and environmental factors such as patient placement and room ventilation . Nevertheless , this unusually comprehensive dataset has provided detailed insights into the processes of hospital-based transmission . Combining data from multiple sources into a single analysis provides increased resolution and insight into the pervasive problem of nosocomial viral transmission .
Prospective surveillance studies of COVID-19 infection in patients and healthcare workers ( HCW ) were conducted at Cambridge University Hospitals NHS Foundation Trust ( CUH ) , as previously described ( Rivett et al . , 2020; Meredith et al . , 2020 ) . Nasopharyngeal swab samples were collected and submitted to the Public Health England ( PHE ) Clinical Microbiology and Public Health Laboratory ( CMPHL ) or the Department of Medicine , University of Cambridge for SARS-CoV-2 diagnostic testing , as detailed below . Samples included in this study were collected during the first epidemic wave , between 22nd March and 14th June 2020 . Clinical , laboratory , and patient location data were extracted from the hospital information system ( EPIC Systems Corporation , Verona , USA ) . HCW ward location data were collected by members of the HCW screening team . PHE recommendations for COVID-19 infection prevention and control , including PPE use , were followed throughout the course of the study . Patients and HCW had separate testing criteria and sample workflows . HCW were tested in the CUH HCW screening programme , which included both asymptomatic screening and symptomatic testing arms . Asymptomatic screening at the time of this study was focused on staff working on COVID-19 ‘red’ wards ( designated for patients with confirmed COVID-19 infection ) , wards with hospital-acquired infection outbreaks , and wards with high rates of staff positivity . Suggested symptomatology to prompt staff testing are described in Rivett et al . , 2020 , divided into ‘major’ criteria ( e . g . fever and/or new persistent cough ) and ‘minor’ criteria ( e . g . coryzal symptoms , headache , myalgia ) . For all five of the outbreaks described in this study , all staff working on the outbreak wards were invited for screening by the CUH HCW screening team ( i . e . tested regardless of any symptoms or if asymptomatic ) during the outbreak periods ( prompted by the outbreak investigations and/or high rates of staff positivity on the wards ) . There was no systematic asymptomatic screening for inpatients in CUH during the study period , but targeted patient screening on wards with hospital-onset COVID-19 outbreaks was performed . Ward E was a COVID-19 ‘red’ ward; all patients on this ward had tested positive prior to placement there . Wards A to D were all ‘green’ wards ( designated for non-COVID-19 patients ) at the time the hospital-onset COVID-19 outbreaks started . For Ward A , symptomatic contacts of confirmed cases were tested initially , and as the outbreak grew and more cases were confirmed , ultimately all patients on the ward were screened ( including asymptomatics ) . For Ward C , contacts of confirmed positives and/or patients who developed symptoms were screened , and for Ward D , symptomatic contacts of the index case were tested . Thus , for Wards C and D , systematic asymptomatic screening of all patients on the ward during the outbreaks was not performed . For the Ward B outbreak , when the index patient ( case B0 ) tested positive , all staff members who had worked on the cluster of clinical areas referred to as ‘Ward B’ within the preceding 2 weeks plus all patients on those wards were screened ( regardless of any symptoms ) . Thus , there is high confidence for Wards A , B and E that all infections among both staff and patients were detected ( providing the amount of SARS-CoV-2 RNA was sufficient for detection ) . Visitor restrictions were introduced on 25th March 2020 and so were present for almost all of this study ( first positive swab was for Ward E , collected a few days before this ) . After 25th March , visitors to adult patients were only permitted in exceptional circumstances: for patients at the end of life or visitors with a direct care role for the patient . Samples underwent nucleic acid extraction and were tested for presence of SARS-CoV-2 using a validated in-house RT qPCR assay , as previously described ( Price et al . , 2021 ) . The test was reported as SARS-CoV-2 PCR positive if the cycle threshold ( Ct ) value was less than or equal 36 . A 15 microlitre aliquot of the RNA extract of each positive sample was transferred to the Department of Pathology , University of Cambridge , for sequencing . Samples were assigned COG-UK sequencing codes and sequenced using a multiplex PCR based approach according to the modified ARTIC v2 protocol with v3 primer set ( artic-ncov , 2019; Quick , 2020 ) . Amplicon libraries were sequenced using MinION flow cells v9 . 4 . 1 ( Oxford Nanopore Technologies , Oxford , UK ) . Genomes were assembled using reference-based assembly and a bioinformatic pipeline ( Artic Network , 2021 ) . All sequences underwent quality control ( QC ) filtering , including a 20x minimum coverage cut-off for any region of the genome and 50 . 1% cut-off for calling single nucleotide polymorphisms ( SNP ) . Patients from CUH were determined to have indeterminate , suspected or definite hospital-acquired infections ( HAI ) on the basis of days from admission to first positive SARS-CoV-2 test , using the same definitions as in Meredith et al . , 2020; Price et al . , 2021: indeterminate = positive test after 48 hr and less than 7 days post admission; suspected = positive test 7 to 14 days post admission; definite = positive test greater than 14 days post admission . Wards were ranked by their combined number of indeterminate , suspected and definite HAI cases plus HCW cases ( taking the HCW ward to be any ward each HCW had worked on within the preceding 14 days before testing positive ) . 15 wards at CUH had two or more HAI and HCW cases occur within 14 days of each-other . The five wards with the largest number of combined HAI and HCW cases were chosen for this study and named wards A to E ( Supplementary file 1 ) . Each of the selected wards had 10 or more HAI plus HCW cases , therefore representing the largest HAI ward-based clusters in CUH during the study period , which encompasses the ‘first wave’ of the pandemic in the East of England region . Four were ‘green’ wards ( A to D ) , intended to house patients who did not have COVID-19 , and one was a ‘red’ ward ( E ) , intended to house confirmed COVID-19 cases . In four out of five wards , the majority of cases were HCW . Ward characteristics cannot be described in detail in order to preserve confidentiality . They were typically organised into bays , with four to six patients per bay , and a limited number of side rooms ( which are critical for infection control purposes ) . In summary , Ward A had 30 beds with three side rooms; Ward C had 27 beds with three side rooms; Ward D had 30 beds with four side rooms; Ward E had 26 beds with three side rooms . The Ward B outbreak was focused around a ward with 32 beds of which 12 were side rooms , although several adjacent clinical areas were screened as part of the outbreak investigation ( as staff were shared between these wards ) . An initial network analysis of patient bed movements was conducted to add patients to each ward cluster that may have been in contact with the HAI cases , either with community onset infections on the same ward or while they were co-located on other wards outside of the ‘outbreak wards’ themselves . This analysis was undertaken in two steps using SQL v18 . 5 . 1 and FoodChain-Lab ( an extension of the Knime Analytics Platform v3 . 6 . 1 ) . The first step involved utilising SQL to process case and ward movements data from CUH patients , creating a list of ward-based case co-locations that were within the set parameters of the network analysis model . These parameters were ( 1 ) an infectious period that included the 4 days prior to symptom onset up until 7 days after symptom onset , and ( 2 ) a susceptibility period of 14 days prior to symptom onset . Where symptom onset date was not available , the case positive specimen date was used instead . The second step was to import the case co-locations data into FoodChain-Lab , which was used to draw a social network diagram of cases and their ward co-locations with one another that met the set parameters . This network diagram was used to identify any clustering of cases by ward that had not met the original criteria of HAI and HCW cases but could have been involved in shared transmission with those individuals based on their co-location within the infection period of the virus . This yielded the final case set for each ward cluster taken forward for analysis using the transmission reconstruction model . Data were collected on COVID-19 symptom onset dates for all included individuals from the five ward clusters . These were collected separately for HCW and patients . HCW testing was part of the CUH HCW screening programme , and testing criteria are described in Rivett et al ( Sikkema et al . , 2020 ) . ( Table 1 ) . Staff who tested positive were then contacted by members of the HCW screening team and asked retrospectively about their symptoms , and onset dates were recorded by the HCW screening team and used in this analysis . For patients , symptom onset dates were collected by retrospective review of patients’ electronic hospital records ( EPIC Systems ) , usually noted at presentation by the clerking doctor but all records were examined for any suggestion of symptoms . Symptom definitions followed the standard national recommendations at the time: initially fever , breathlessness , and new continuous cough; muscle aches were added in early May and loss of taste or smell was included from mid-May . If the infection was asymptomatic , or symptom onset dates could not be determined , then the date of first positive SARS-CoV-2 test was used instead . Patient ward movement data through the hospital was downloaded from the hospital electronic records system ( EPIC ) . Data on HCW shift patterns were collected manually using hospital shift rostering information and , in some cases , directly contacting the HCW . HCW were defined as being either present or absent on each ‘outbreak’ ward within each 24 hr period ( i . e . time of day or shift length was not taken into account ) . No HCW worked on multiple outbreak wards , although staff contact outside of wards ( during lunch , outside of work etc ) cannot be excluded . Each anonymised patient code was linked to COG-UK sequencing codes , with sample collection date and laboratory receive dates recorded for each sample that was sequenced . Dates of symptom onset and sample collection for each individual are shown in Figure 5 . Given a set of data from infected individuals , comprising viral genome sequences collected from each infection , dates on which individuals became symptomatic , and when individuals were co-located , we sought to identify whether and how these cases are linked by transmission . We noted that , prior to analysis , we did not know whether the cases for which we had data were connected via transmission; a set of cases may be anything from completely unlinked to a single connected transmission network . As a first stage in our method , we partitioned our data into subsets of cases that might plausibly be linked by transmission events . This step was conducted using a pairwise measure of the consistency of the data observed from two individuals with the hypothesis of direct transmission having occurred between these two individuals; calculation of this measure was previously implemented in the A2B-COVID software package ( Illingworth , 2020 ) . The mathematical principles of this pairwise measure of consistency are explained fully elsewhere ( Illingworth , 2020 ) ; below we provide a summary . Our pairwise likelihood function combines information from a broad variety of sources . Firstly , dates on which individuals first reported symptoms give an indication of the likelihood of transmission between individuals . Previous publications have characterised the infectivity profile of SARS-CoV-2 ( i . e . the time between becoming symptomatic to transmitting the virus to another person conditional on causing infection ) , by fitting a shifted gamma distribution to data from confirmed infection , inferring the distribution parameters α=97 . 185 , β=0 . 2689 , and shift s=25 . 625 ( Ashcroft et al . , 2020; He et al . , 2020 ) . Further , the time between an individual being infected with SARS-CoV-2 and developing symptoms has been described as a lognormal distribution , with inferred parameters μ=1 . 434 and σ=0 . 6612 ( He et al . , 2020; Li et al . , 2020 ) . We use these distributions to assess the extent to which the observed symptom-onset times are compatible with transmission . Specifically , we denote by XT the event that transmission from i to j occurred on day T , and by X the event that transmission from i to j occurred at all . For convenience we collect together parameters , writing θ1={α , β , s , μ , σ} . Supposing that individuals i and j became symptomatic on the days Si and Sj respectively , we then derive the expression:P ( T|θ1 , Si , X ) PSj|θ1 , XT=∫T-Si-0 . 5T-Si+0 . 5e- ( x+s ) /β ( x+s ) α-1β-αΓαdx∫Sj-T-0 . 5Sj-T+0 . 5e-logx-μ22σ2xσ2πdxwhere a time of becoming symptomatic was unknown , the date at which that individual tested positive was used , corrected by a constant term estimated from the distribution of known times between positive test dates and dates of becoming symptomatic ( Illingworth , 2020 ) . We elaborate on this model by incorporating data describing the location of individuals , noting that transmission can only occur when people are co-located , and viral genome sequences collected from individuals in the study . We have , for individuals in our study , information describing the wards on which patients were hosted , and the wards on which HCWs worked shifts . Given individuals i and j we defined a contact metric wij ( T ) equal to the probability that i and j were co-located on any given day T . Ward data describing patient locations and health care worker shift dates were used to calculate values wij for each pair of individuals . For any given ward , we set the value wi ( W , T ) to equal one if the individual i was known to be on ward W on day T . If a healthcare worker worked a shift on ward W on day T , we further assigned a minimum value of 0 . 5 to the values wi ( W , T-1 ) and wi ( W , T+1 ) , accounting for night shifts overlapping days , and the potential for fomite transmission . Removing these additional days led to some small alterations in the inferred networks ( Figure 5—figure supplement 1 ) , but did not substantially affect our primary result , with again 80% of cases being caused by 21% of individuals ( Figure 5—figure supplement 2 ) . For all other dates , we set wi ( W , T ) to be zero . In the event that location data was unknown , we assigned a value for wi ( W , T ) of 1 for patients for the most common ward on which individuals in a cluster were located , and a value for wi ( W , T ) of 4/7 for health care workers , reflecting common shift patterns of work . The value wij ( T ) for any two individuals i and j was then defined as the maximum of the product of wi ( X , T ) and wj ( X , T ) calculated across all wards X for that value of T . Our pairwise likelihood accounted for data from viral genome sequencing using a separate likelihood function . For each pair of individuals , sequences were assessed according to the number of nucleotides by which they differed from a pairwise consensus , calculated from the two sequences , with reference to the broader set of sequences collected from a ward . At a given genome position , the consensus was defined as the nucleotide shared by the two sequences if they were identical , or by the most common nucleotide in the broader set if the two sequences differed . The Hamming distances between each individual sequence and the pairwise consensus were measured . Each distance reflects both the potential evolution of the virus and the extent of measurement error inherent to the sequencing process . Using data from CUH , an estimate of the measurement error was previously calculated as E=0 . 414 nucleotides per pair of sequences , or 0 . 212 nucleotides per sequence ( Illingworth , 2020 ) . To model viral evolution we adopted the rate γG=0 . 0655 nucleotide substitutions per day , equal to the global rate of viral adaptation ( Hadfield et al . , 2018 ) . We then derived a Poisson model for the observed number of nucleotide changes resulting from evolution and error . We denote the days on which viral sequence data was collected from the individuals i and j as D = {Di , Dj} , and the Hamming distances of each sequence from their consensus as Hi and Hj , and again use a shorthand expression θ2={E , γG} . We then have the result thatPHi , Hj|θ2 , D , XT=E/2+γGPiHie-E/2+γGPiHi ! E/2+γGDj-QiHje-E/2+γGDj-QiHj ! We now combine the above terms , writing y = {Si , Sj , Hi , Hj} , θ = {θ1 , θ2} , and denoting by X the event that transmission occurred from i to j at any time to obtainP ( y|θ , D , X ) =∑TP ( T|Si , θ1 ) PSj|θ1 , XTwij ( T ) PHi , Hj|θ2 , D , XT This expression was used to assess the consistency of the data y from each pair of individuals with the hypothesis of transmission from i to j . The value p ( y|θ , D , X ) is defined over a discrete space of symptom times and sequence measurements . Calculating this over all feasible terms within this discrete space , we derived threshold values , conditional upon D , for which 95% and 99% of genuine transmission events would obtain values greater than the threshold . In this way , potential transmission events were classified into ‘consistent’ events ( probability greater than the 95% threshold ) , ‘borderline’ events ( probability between the 95% and 99% thresholds ) and ‘unlikely’ events ( probability below the 99% threshold ) . The pairwise analysis , described to this point , was used to assess the data for potential between-ward transmission events . As a first step in identifying transmission networks in our data , we clustered individuals using the pairwise thresholds . By way of notation we denote the transmission event from individual i to individual j as i→j . Beginning with a single subset containing an arbitrary individual , we added an individual j to the subset S if for some individual i in S , either i→j or j→i has a likelihood that was ‘consistent’ or ‘borderline’ under the pairwise likelihood . If no such subset existed , j was placed in a new subset , repeating the process until all individuals were clustered into sets between which transmission events were unlikely . If clustering identified individuals that were alone within a subset , these individuals were removed from the analysis at this point , giving further attention to clusters that contained at least two individuals . Building upon our pairwise method , we calculated the likelihood of a transmission network . A transmission network must contain at least one transmission event . We consider the network N , comprising transmission events i1→j1 , i2→j2 , etc , in which the transmission even ik→jk occurs on day Tk . We here write T* = {Tk} to represent the set of all transmission times . Similar to the calculation above , we consider symptom onset times , the locations of individuals , and sequence components , using these to derive an expression for a particular network of transmission events . Considering the non-genetic data , we make the assumption that the intrinsic dynamics of each transmission are independent of one another , so that the joint distribution of the observed symptom times Si given a particular network and set ot transmission times XTk , isPSi|θ1 , XTk=∏kwikjkTkP ( Tk|θ1 , Sik ) PSjk|θ1 , XTk For the sequence component of the likelihood , we consider the transmission tree defined by our network . Adopting an infinite sites model ( Kimura , 1969 ) , we assume that each observed nucleotide substitution occurred only once , with the reversion of substitutions being impossible . Further , we assumed that , while a substitution that was observed in only one sequence could have arisen from measurement error , substitutions that were observed in two or more sequences could not have resulted from error . Under these assumptions , we classified potential substitutions by the sequences they would be observed in , then adopted a Poisson likelihood model , comparing the periods of time in which sets of substitutions could have occurred to the numbers of substitutions that were observed in the data ( Felsenstein , 1981 ) . We first describe this likelihood for a transmission network in which all times of transmission are known . From the sequence data , we identified nucleotide substitutions occurring in viral sequences relative to the consensus sequence . We denote the number of substitutions observed in a sequence or group of sequences I by MI . Windows of time were then assigned to substitutions in reverse order . Diagrams illustrating this process are shown in Figure 5—figure supplement 3 . We first consider the final transmission event to occur ( Figure 5—figure supplement 3A ) . We suppose that this transmission is A→B , occurring at time tAB . It is clear that the time of sample collection DB occurs no earlier than tAB; we divide the scenario into two possibilities . Firstly , if DA ≥ tAB , we have that substitutions observed only in individual B must have occurred in the DB - tAB days between tAB and DB , while substitutions observed only in A must have occurred in the DA - tAB days between tAB and DA . We note that substitutions observed only in single individuals are treated as potentially arising from either evolution or noise . We therefore calculated the likelihood of having observed MA substitutions from a Poisson distributed variable with expectation E/2 + γG ( DA - tAB ) , and MB substitutions from a Poisson distributed variable with expectation E/2 + γG ( DB - tAB ) . Alternatively , if DA < tAB , then there exists a time P , equal to the latest of the set of times including DA and all other times of transmission tAX from A to an individual X other than B . We note that mutations observed only in B can then have occurred in the DB - P days of this interval , while mutations that were observed only in A can only have arisen through sequencing error . We therefore calculated the likelihood of having observed MA substitutions from a Poisson distributed variable with expectation E/2 + γG ( DA - P ) , and MB substitutions from a Poisson distributed variable with expectation E/2 . We note that the inclusion of a non-zero rate of sequencing error allows for A to transmit to B at a time after the observation of a viral consensus sequence from A that contains a substitution not observed in the viral consensus from B . A similar logical process was carried for each transmission event working backwards in time . A more general case , representing a joining of two branches of the tree , is shown in Figure 5—figure supplement 1B . We here consider the individuals A , B , C , and D , noting that these may not be terminal branches in the tree . That is , if a mutation occurs early in the infection of individual C , it will be observed both in the sequence collected from C and in sequences collected from individuals to which C later transmits the virus . In this case , there are a variety of possibilities here to enumerate . If DA ≥ tAD , then mutations which are observed precisely in A , D , and individuals downstream of A and D , must occur in the window between tAB and tAD . If DA < tAD but DA ≥ tAB , then the same mutations must occur in the window between tAB and DA . Finally , if DA < tAB , it is impossible for mutations to be observed in precisely A , D , and individuals downstream of A and D . However , there exists a point P , equal to the latest of a set of times including DA and all other times of transmission from A to another individual prior to tAB . In this scenario , substitutions occurring between P and tAB will be observed in B , C , D , and all individuals of these . In all the above scenarios , mutations which are observed in B and C , plus individuals downstream of B and C , must occur in the window between tAB and the latter of DB and tBC . In each case , we calculate the likelihood of having observed the given number of substitutions from a Poisson distributed variable with expectation given by the rate of evolution γG multiplied by the time available for substitutions to occur , plus the measurement error E/2 where a variant was observed in only one sequence . In this manner , we derived a phylogenetic hierarchy , dividing the transmission tree into sets of individuals that would be expected to share genetic substitutions; we denote these sets Ia , for a=0 , 1 , 2 , etc . For each such set , we calculate the length of time in which these substitutions would have to occur; we denote these times ta , counting the number of substitutions in each set , denoted Ma . We then have thatPIa , ta , Ma|θ2=∏aδaE/2+γGtaMae-δaE/2+γGtaMa ! where the term δa is equal to one if Ia contains a single individual and 0 otherwise . We now construct our final expression . We write G = {{Ia} , {ta} , {Ma}} , and note that G , through the values ta , is dependent on the times of transmission T* and the times of sample collection , which we denote more generally by D = {Dk} . Denoting by y the set of all data collected from the individuals in the network , we then have the resultPy|θ , D , XN=∑T*=Tk∏k=1nwikjkTkP ( Tk|θ1 , Sik ) PSjk|θ1 , XTkPG|θ2 , D , XT*where XN is the event that the transmission events in the network occurred at some set of times . We convert this into a log likelihood , calculating the likelihood of a given network N:logLN|y=logPy|θ , D , XN We find the maximum likelihood network N , and using likelihoods calculated for multiple networks generate statistical ensembles of networks , to estimate , for example , the number of individuals infected by a specific person on a ward . The likelihood calculation above presupposes the existence of an ordered transmission network with times assigned to each transmission event . Calculations were performed in order to derive these networks . Plausible networks describe sets of transmission events that could potentially describe a pattern of individuals within a subset . Four criteria were used to identify such networks . Firstly and secondly , networks had to be acyclic , and span all individuals in a subset . Thirdly , networks had to be likely , with each transmission event in the network having a pairwise likelihood that was classified as ‘consistent’ with the pairwise likelihood model . Where computational time allowed it , that is , for all but ward A , this criterion was relaxed to also consider transmission events that were classified as ‘borderline’; details on computability are given later in the Methods . Fourthly , networks had to be consistent with the observed sequence data . Consistency with sequence data can be described in terms of patterns of shared substitutions observed in sequences from different individuals . We consider a set Ia of all individuals in the subset with viral sequences that share a substitution or set of substitutions . Under a maximum parsimony assumption that variants are gained only once and cannot revert , consistency requires that ( i ) there can only be one transmission event i→j with i ∉ Ia and j ∈ Ia and ( ii ) with the exception of this specific individual j in Ia , there can be no transmission events j→k with j ∈ Ia and k ∉ Ia ( Figure 5—figure supplement 4 ) . This has to apply for all such sets Ia . For each subset of individuals created using the pairwise likelihood an exhaustive search of plausible networks was conducted . In the event that no such network was identified , the assumption that the network must span all individuals in the subset was gradually relaxed , reducing the number of individuals required successively by one until at least one plausible network was identified . In some cases , this relaxation led to the identification of multiple plausible networks involving non-overlapping sets of individuals; we return to this later in the Methods section . Given a plausible network , we next considered orderings of the transmission events that it comprised . To achieve this , transmission events were given a nominal index . In an ordering the placement of the transmission ia→ja before ib→jb implied that ia→ ja occurred before or at an identical time to ib→jb if it was true of the indices that a < b , but implied that ia→ja occurred strictly before ib→jb if a > b . Plausible orderings were required to fulfil three criteria . Firstly , on a logical principle , ia→ja must occur before ib→jb if ja = ib; an individual must be infected before transmitting the virus . We insisted that an individual who had received the virus via transmission could not themselves transmit until at least one day after they had been infected . Secondly , orderings had to fulfil characteristics imposed upon them by the likelihood function . If the first day on which the pairwise likelihood LP describing transmission between ib and jb was greater than zero was after the last day on which the equivalent likelihood for the transmission between ia and ja was greater than zero , it followed that ia→ja must occur before ib→jb . A third criterion was imposed by the existence of shared substitutions in the sequence data . Returning to the criteria imposed for network plausibility we suppose that for the set of sequences Ia that contain shared substitutions we have identified the transmission event i→j for which i ∉ Ia but j ∈ Ia , and that there exist a series of transmission events in the network j→k for this j . In this case , the last transmission for which k ∉ Ia must occur before the first transmission for which k ∈ Ia , the substitution event necessarily occurring in individual j between these two times ( Figure 5—figure supplement 5 ) . All plausible orderings were stored for each plausible network; we note that the existence of a plausible network did not imply the existence of a plausible ordering of transmission events . Potential times for transmission events were determined by the first and last time points for which each event had a non-zero pairwise symptomatic likelihood . These times , in addition to the ordering constraints , impose a combinatorial set of possible times at which transmission occurred . Finally , the likelihood of a transmission network was calculated as the sum over all orderings of the sum over all sets of times of the timings-dependent network likelihood . Denoting the set of all orderings of transmission events as O , and the set of possible sets T* as T** , we havelogLn=log∑O∑T**LnT** Calculating likelihoods for plausible transmission networks , the maximum likelihood network was identified . Uncertainty about properties of the network , for example whether a network includes a particular transmission event , and the number of individuals infected by each individual , was quantified by Bayesian methods . Assuming a uniform prior over the space of possible networks , the posterior probability that each potential network is the true network is obtained as the likelihood of this specific network divided by the sum of the likelihoods of all possible networks . Further statistics were calculated using these network probabilities . For example , the probability that individual A infected individual B was calculated as the sum of the probabilities of networks for which A infected B , while the probability that A infected a total of n other individuals was calculated as the sum of the probabilities of the networks in which A infected n other individuals . Where in the initial subsetting of individuals disjoint subsets were identified , independent network reconstructions were calculated for each subset . Where the largest complete network in a subset omitted more than one individual from the subset , repeat calculations , removing those individuals inferred to be in that network , were performed , aiming to identify transmission events between other individuals in the subset . Our algorithm runs at a combinatorial level of complexity dependent upon the number of individuals and the number of plausible paths through the network . We adopted different computational methods to identify maximum likelihood networks and statistical properties . For wards other than A , a rapid and comprehensive calculation of likelihoods for all plausible networks was possible . Uncertainty about properties of the network was then computed over the complete space of possible networks . For ward A , a first calculation was performed , generating likelihoods for a systematically chosen set of 0 . 2% of the plausible networks . From these networks , a heuristic likelihood threshold was chosen , identifying 10 networks with the highest likelihood values . These networks were used as starting points for independent downhill optimisation calculations . Given each network , the set of adjacent networks was identified , each constructed by the breaking and replacement of a single transmission event . Likelihoods were calculated for these networks , choosing the maximum likelihood network among these before repeating the calculation . Across these independent calculations , convergence to the reported maximum likelihood network was observed . Uncertainty calculations for ward A were performed initially over the systematic sample of network space , plus networks sampled in the downhill optimisation , plus all networks one or two steps adjacent from the maximum likelihood network; this included a total of 14 , 040 networks . A second calculation of statistics describing network uncertainty was performed across all these networks but also including any remaining networks three steps adjacent from the maximum likelihood network; this included a total of 19 , 238 networks . Statistics derived from the smaller and larger ensembles of networks were then compared . The high degree of similarity between these statistics suggests convergence towards the true statistical average ( Figure 5—figure supplement 5 ) . Estimates of the distribution of the number of individuals infected by each person in each ward were used to assess the existence of superspreader individuals , comparing the fit of two models to the data . Our network inference provided for each individual a distribution of the number of people infected , so we could say that each individual i infected j others with probability pij . We used these values to define ‘latent data’ to which we can fit Poisson models describing different hypotheses about heterogeneity in transmission rates . Under a model with no superspreading , the number of individuals infected by each person would be expected to follow a Poisson distribution with some rate R of transmission . The marginal likelihood of this model , defined as the Poisson likelihood integrated over the distribution of the latent number of individuals infected by each person , is derived aslogL1=∑ilog∑jpijrje-rj ! This was maximised to estimate the common transmission rate r . We compared this hypothesis to an alternative , negative binomial model . This is equivalent to a model where the transmission rate is different for each person , so that each person infects a number of people described by a Poisson with a random rate drawn from a gamma distribution . We found parameters p and r maximising the marginal likelihoodlogL2=∑ilog∑jpijr+j-1r-11-pjpr The marginal likelihoods L1 and L2 were then compared using the Bayesian Information Criterion ( Schwarz , 1978 ) to account for the additional parameters in the second model . In the maximum likelihood networks , of 22 cases of patients being infected , two were inferred to originate from HCWs , with 20 of these being infected by other patients . We compared these values using a one-tailed test , calculating the cumulative density function of a binomial distribution with N=22 and probability 0 . 5 for the observed sample . This study utilised 129 SARS-CoV-2 genomes from 98 individuals across the five outbreak wards . The COG-UK and GISAID sequence IDs for these samples are shown in Supplementary files 1 and 2 . Genomic data are publicly accessible through the COG-UK website data section ( https://www . cogconsortium . uk/data/ ) and GISAID ( https://www . gisaid . org/ ) . Sequences generated through the COG-UK consortium have associated public metadata , including age , sex , collection date ( if available ) , and location to the level of UK county . COG-UK samples are sequenced under statutory powers granted to the UK Public Health Agencies . Matched patient data is securely released to the COG-UK consortium under a data sharing framework which strictly controls the handling of patient data . Information on whether individuals are healthcare workers or patients , and groupings of patients into their shared ward locations in hospital , are not for public release linked to their sequencing identifiers ( eg . COG-UK sequence codes ) . This is because of the risk of deductive disclosure , potentially compromising study participant anonymity . However , code to fully reproduce the transmission network analysis using anonymised metadata and altered SARS-CoV-2 sequences is available via GitHub at https://github . com/cjri/a2bnetwork ( Illingworth , 2021; copy archived at swh:1:rev:2c08d1a789b7f1a9ce758a86db27fc3d78b9d003 ) If a researcher requires access to restricted metadata ( including healthcare worker status and patient ward locations ) linked to the COG-UK sequence codes , then this will require a formal data sharing agreement with the COG-UK Consortium and Cambridge University Hospitals NHS Foundation Trust ( CUH ) . Data will only be shared for public health and research purposes , not for commercial enterprise , and only to individuals working at reputable research and public health institutions for which data security can be assured . Should this be required researchers should contact the study corresponding authors in the first instance . | The COVID-19 pandemic , caused by the SARS-CoV-2 virus , presents a global public health challenge . Hospitals have been at the forefront of this battle , treating large numbers of sick patients over several waves of infection . Finding ways to manage the spread of the virus in hospitals is key to protecting vulnerable patients and workers , while keeping hospitals running , but to generate effective infection control , researchers must understand how SARS-CoV-2 spreads . A range of factors make studying the transmission of SARS-CoV-2 in hospitals tricky . For instance , some people do not present any symptoms , and , amongst those who do , it can be difficult to determine whether they caught the virus in the hospital or somewhere else . However , comparing the genetic information of the SARS-CoV-2 virus from different people in a hospital could allow scientists to understand how it spreads . Samples of the genetic material of SARS-CoV-2 can be obtained by swabbing infected individuals . If the genetic sequences of two samples are very different , it is unlikely that the individuals who provided the samples transmitted the virus to one another . Illingworth , Hamilton et al . used this information , along with other data about how SARS-CoV-2 is transmitted , to develop an algorithm that can determine how the virus spreads from person to person in different hospital wards . To build their algorithm , Illingworth , Hamilton et al . collected SARS-CoV-2 genetic data from patients and staff in a hospital , and combined it with information about how SARS-CoV-2 spreads and how these people moved in the hospital . The algorithm showed that , for the most part , patients were infected by other patients ( 20 out of 22 cases ) , while staff were infected equally by patients and staff . By further probing these data , Illingworth , Hamilton et al . revealed that 80% of hospital-acquired infections were caused by a group of just 21% of individuals in the study , identifying a ‘superspreader’ pattern . These findings may help to inform SARS-CoV-2 infection control measures to reduce spread within hospitals , and could potentially be used to improve infection control in other contexts . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"evolutionary",
"biology",
"microbiology",
"and",
"infectious",
"disease"
] | 2021 | Superspreaders drive the largest outbreaks of hospital onset COVID-19 infections |
Genetic effects on gene expression and splicing can be modulated by cellular and environmental factors; yet interactions between genotypes , cell type , and treatment have not been comprehensively studied together . We used an induced pluripotent stem cell system to study multiple cell types derived from the same individuals and exposed them to a large panel of treatments . Cellular responses involved different genes and pathways for gene expression and splicing and were highly variable across contexts . For thousands of genes , we identified variable allelic expression across contexts and characterized different types of gene-environment interactions , many of which are associated with complex traits . Promoter functional and evolutionary features distinguished genes with elevated allelic imbalance mean and variance . On average , half of the genes with dynamic regulatory interactions were missed by large eQTL mapping studies , indicating the importance of exploring multiple treatments to reveal previously unrecognized regulatory loci that may be important for disease .
Cells exist within complex environments , where levels of metabolites and signaling molecules can change rapidly . In order to thrive under such conditions , cells have evolved mechanisms to control gene expression in response to environmental perturbation . Variation in environmental exposure can explain variation in gene expression in human population samples ( Gibson , 2008; Idaghdour et al . , 2010; Favé et al . , 2018; Aguirre-Gamboa et al . , 2016; Horst et al . , 2016; Maghbooli et al . , 2018; Wang et al . , 2015 ) . For example , a highly correlated cluster of surfactant-related genes was found to be highly expressed in lung tissue from GTEx donors who died while on a ventilator ( McCall et al . , 2016 ) . Similarly , thousands of gene expression differences were identified between sun-exposed and non-sun-exposed skin ( Kita and Fraser , 2016 ) . In addition to changing gene expression , environmental exposures are also able to alter splicing processes ( Pai and Luca , 2019 ) . This may be a direct effect of the environment altering normal splicing patterns , as well as the result of a compensatory effect initiated by the cell to overcome a stressful situation . Splicing is an important co-transcriptional and post-transcriptional process taking place within the nucleus , in which portions of pre-mRNAs called introns are removed , and others called exons are joined together to form a mature transcript ( Ule and Blencowe , 2019 ) . All forms of splicing require a variable set of proteins and small nuclear RNAs to form small-nuclear ribonucleoproteins ( snRNPs ) , ultimately giving rise to the spliceosome ( Wahl et al . , 2009 ) which , together with a variable set of accessory proteins ( Han et al . , 2017; Gonatopoulos-Pournatzis et al . , 2018 ) finely tune splicing . Expression of both spliceosomal and accessory factors may be influenced by the environment , which means the environment is eventually able to alter splicing processes ( Richards et al . , 2017 ) . It is widely accepted that gene expression varies across individuals and that such variation is under genetic control . Expression quantitative trait loci ( eQTL ) mapping in samples from a variety of tissues has elucidated the tissue specificity of the genetic control of gene expression ( van der Wijst et al . , 2018; Dimas et al . , 2009; Flutre et al . , 2013 ) . Specifically , the GTEx consortium observed a U-shaped pattern for eQTL tissue specificity , with eQTLs tending to be either highly shared amongst tissues or highly tissue-specific ( The Gtex Consortium , 2020 ) . These findings underscore the importance of evaluating the genetic control of gene expression across cell and tissue types . To address the environment-specific control of gene expression , eQTL studies performed on in vitro treated cells from many donors are commonly used . The resulting genetic variants which influence gene expression response are known as response eQTL ( reQTL ) and represent gene-environment interactions ( G×E ) for molecular phenotypes . reQTL have been identified for infectious agents , drugs , and hormones , among other stimuli/perturbations ( Knowles et al . , 2018; Manry et al . , 2017; Nédélec et al . , 2016; Alasoo et al . , 2018; Kim-Hellmuth et al . , 2017; Quach et al . , 2016; Çalışkan et al . , 2015; Lee et al . , 2014; Fairfax et al . , 2014; Maranville et al . , 2011; Mangravite et al . , 2013; Barreiro et al . , 2012; Alasoo et al . , 2019; Huang et al . , 2021 ) . These studies have consistently shown that genes with G×E on gene expression are enriched for association with complex traits . However , given the large sample sizes needed to detect reQTLs , only a small number of conditions in a limited number of cell types can be tested at one time . Allele-specific expression ( ASE ) is an alternative to eQTL mapping to identify the genetic control of gene expression . Rather than associating a genetic variant with gene expression , ASE uses heterozygous sites within coding regions to measure allelic imbalance in RNA-seq reads . A significant allelic imbalance within an individual implies cis-regulatory variation , as the trans-environment is constant . Similar to reQTL mapping , differences in ASE between environmental conditions is indicative of G×E , and we refer to such SNPs as conditional ASE ( cASE ) ( Moyerbrailean et al . , 2016 ) . cASE has important implications for human health . For example , nearly 50% of genes with cASE were involved in complex traits by GWAS , which is significantly greater than ASE or eQTL genes ( Moyerbrailean et al . , 2016 ) . Unlike reQTL mapping , ASE can identify G×E in small sample sizes , thereby allowing for interrogation of a broader spectrum of environmental exposures ( Moyerbrailean et al . , 2016 ) . While gene expression is governed by both genetic and environmental factors , it is subject to fluctuations due to stochastic factors ( Raser and O’Shea , 2004 ) . This noise in gene expression is gene-specific , dependent on promoter elements , and can be affected by genetic mutation ( Raser and O’Shea , 2004; Mogno et al . , 2010 ) , indicating its importance in gene regulation . Indeed , there has been selection to minimize gene expression noise in yeast ( Lehner , 2008 ) , and gene expression variation has been linked to differential gene expression in response to perturbation in flies and humans ( Sigalova et al . , 2020 ) . While the importance of variation in gene expression has been clearly established , much less is known on variation in the genetic control of gene expression . Taking advantage of the reduced sample size needed to measure genetic effects by ASE , we have established a system to measure the genetic effects on gene expression in three cell types from six individuals across a large number of environmental conditions . This study design allows us to simultaneously quantify context ( cell type and environment ) dependent and genetic effects on gene expression . In addition to identifying thousands of instances of cASE , we have partitioned the variance in gene and allelic expression into individual , treatment , and cell type components , illustrating how each of these components can influence genetic control .
We have reprogrammed lymphoblastoid cell lines ( LCLs ) from six individuals into induced pluripotent stem cells ( IPSCs ) , which were further differentiated into cardiomyocytes ( CMs ) ( Figure 1A ) . Our study was performed in batches on 96-well plates . Each batch consisted of the same cell type from three individuals , with 28 treatments and two controls . Each experiment was performed in duplicate for a total of 12 batches . ( Figure 1 ) . In order to shift CM cellular metabolism from fetal-associated glycolysis to adult-associated aerobic respiration , we replaced the cell culture media with galactose-containing media on day 20 ( Rana et al . , 2012; Ward and Gilad , 2019 ) . We assessed the purity of CM cultures on days 25 and 27 using flow cytometry to measure the percentage of cells expressing cardiac troponin 2 ( TNNT2 ) . Purities ranged from 44 . 9 to 95% ( LABEL:SuppTable14 ) . CMs were derived in a single differentiation experiment per individual , so all experiments from CMs in a single individual have identical purities . Principal components analysis on gene expression reveals three distinct clusters corresponding to the three cell types ( Figure 1—figure supplement 1 ) . Additionally , to further verify the identity of the three cell types , we evaluated the expression of LCL , IPSC , and CM marker genes ( Figure 1—figure supplement 2 ) , including TNNT2 , which , as expected , showed high expression in CM samples and low expression in IPSC and LCL samples ( Figure 1—figure supplement 2C ) . We exposed all cell lines to 28 different treatments ( LABEL:SuppTable1 ) , resulting in 84 cell type/treatment combinations of cellular contexts . Treatments included hormones , common drugs , vitamins , and environmental contaminants , among others . We used a two-step sequencing approach ( Moyerbrailean et al . , 2015 ) to identify changes in gene expression , splicing , and allelic expression across cellular contexts . First , we performed an initial shallow RNA-sequencing step ( median depth of 9 . 5M reads ) to identify treatments which caused a significant response , indicated by changes in gene expression when comparing each treatment to its vehicle control with DESeq2 ( Love et al . , 2014 ) ( Figure 1B , LABEL:SuppTable7 ) . To characterize changes in gene expression , splicing , and allelic expression , we selected 12 of the 28 treatments which induced a substantial change in global gene expression for a total of 36 different contexts ( in bold in Figure 1B , see methods ) . From the 12 deep-sequenced treatments across all cell types ( 36 contexts and six controls , median depth/sample: 146M for LCLs , 148M for IPSCs , 273M for CMs , see LABEL:SuppTable13 ) , we identified differentially expressed genes between each treatment and its vehicle control in each cell type using DESeq2 ( Figure 1C , LABEL:SuppTable8 ) . We found between 53 ( insulin in IPSCs ) and 21 , 593 ( copper in LCLs ) differentially expressed genes ( DEGs; FDR < 10% ) out of 58 , 300 genes . We confirmed that the treatments were having physiological effects by examining gene ontology enrichment of differentially expressed genes . For example , as expected glucose catabolic pathways were enriched ( FDR = 0 . 40% ) in genes upregulated in response to insulin in CMs relative to all genes expressed in CMs ( LABEL:SuppTable2 ) . Also , divalent metal ion transport ( FDR = 1 . 8% ) and transition metal ion transport ( FDR = 2 . 8% ) were enriched in genes upregulated in response to copper in LCLs . When we considered a model with all cell types together , we identified 4835 genes that show evidence of cell type × treatment interaction effects on gene expression as measured by a likelihood ratio test in DESeq2 ( FDR < 10% , LABEL:SuppTable17 , see Materials and methods Equation 2 ) . These cell type × treatment interactions indicate that for these genes , the transcriptional response to the environment varies by cell type . Inter-individual variation within the regulatory sequences governing these cell-type-specific responses could lead to G×E in a cell-type-dependent manner . Interestingly , 96% of differentially expressed genes in CMs in response to treatment with insulin are not differentially expressed in LCLs or IPSCs . These unique insulin-CM genes are enriched for cholesterol and ADP metabolic processes . Together these results highlight the importance of studying environmental effects on gene expression in different cell types to avoid missing physiologically important cell type × treatment effects . To comprehensively identify global shifts in splicing patterns across cell types , individuals , and environments we employed LeafCutter ( Li et al . , 2018 ) , an annotation-free splicing detection tool based on intron excision from pre-mRNAs . We identified 22 , 334 , 519 unique introns , corresponding to 101 , 450 different transcripts and 14 , 066 different genes among all cell types and treatment exposures . We found between 9 ( insulin in IPSCs ) and 4106 ( copper in LCLs ) differentially spliced genes ( DSGs; FDR < 10% ) between treatment and control in each cell type . As one would expect , we identified treatments triggering consistent splicing alterations across all three cell types , as well as treatments predominantly having an effect in just one or two of them ( Figure 1C ) . For instance , selenium exposure leads to splicing alterations in all three cell types , particularly prominent in IPSCs ( 3223 DSGs ) , while in LCLs and CMs the number of differentially spliced genes is much lower ( 1006 and 130 DSGs , respectively ) . Intron excision can be directional , meaning a given intron can be more or less excised in any given condition , with respect to baseline levels ( Richards et al . , 2017 ) . This may result from a treatment inducing a concerted regulation of splicing towards , for example , intron retention across genes . To investigate this phenomenon , in each condition ( i . e . treatment-cell type combination ) , we extracted the percent spliced in ( Ψ ) values and performed a two-sided binomial test to determine whether observed excision direction significantly shifts from a random 50:50 distribution . We identified 10 out of 36 environments ( ∼29% ) that show a consistent shift ( FDR < 10% ) in one direction of intron splicing ( Figure 1—figure supplement 3 ) . More specifically , four of them were identified in LCLs ( insulin , selenium , zinc , copper ) , two in IPSCs ( caffeine , zinc ) , and four in CMs ( aldosterone , nicotine , dexamethasone , insulin ) . When considering the same treatment across different cell types , we found varying consistency in splicing direction . For instance , zinc treatment is associated with an increase in intron excision events in all the three cell types considered , although being significant only in LCLs and IPSCs . Response to insulin , instead , is characterized by an increase in intron retention in CMs and intron excision in LCLs . This would suggest that interaction between the environment and cellular contexts eventually determines whether cells tend to overall retain or excise introns . Cells are continuously exposed to different stimuli and environments , even in physiological conditions . Responses to those exposures are multi-layered , involving changes at both the transcriptional and post-transcriptional level . Because of that , differentially expressed and differentially spliced genes may be involved in distinct processes , but ultimately contributing to restoring cellular homeostasis . To understand how gene expression changes and alternative splicing contribute to the cellular response , we investigated whether the same genes are DEG and DSG in a certain context . Overall , the number of DSGs is significantly correlated with the number of DEGs ( Figure 1—figure supplement 4 , ρ=0 . 91 , p=1 . 4×10-14 ) , thus indicating that the extent of the environmental effects on the transcriptome is similar for gene expression and splicing . DSGs are also generally enriched for DEGs , except for notable examples , such as insulin in CMs ( Figure 1—figure supplement 5 ) . On average 12 . 6% of the DSGs were found to be also DEGs in a given condition ( Figure 1D ) . Overall , this suggests that there are both co-transcriptional and also independent mechanisms controlling pre and post-transcriptional responses . Furthermore , gene regulatory mechanisms involved in different biological processes responding to the environment likely modify transcription by splicing or changes in expression independently . We identified different biological processes affected by differential gene expression and splicing in response to the treatments in the three cell types ( Figure 1E , Figure 1—figure supplement 6 and Figure 1—figure supplement 7 ) . DEGs in CMs are enriched for biological processes related to ion channel activity and transmembrane signaling across seven treatments ( vitamin A , dexamethasone , caffeine , nicotine , copper , insulin , and acetaminophen ) , whereas DSGs were enriched for cytoskeletal protein binding across seven treatments ( aldosterone , dexamethasone , caffeine , nicotine , copper , zinc , and insulin ) . Similarly , IPSCs showed a difference , with DEGs being involved in ion channel activity in seven treatments ( vitamin A , caffeine , nicotine , copper , selenium , zinc , and acetaminophen ) , whereas DSGs were enriched for DNA binding and RNA biology terms across all treatments . On the other hand , both DEGs and DSGs in LCLs were highly enriched for cancer and viral-related processes , pathways and diseases across most treatments , without showing any specific difference between DEGs and DSGs . We also considered disease-gene network annotations and found that DEGs and DSGs in CMs are both enriched for genes linked to cardiovascular diseases , including different forms of cardiomyopathies and vascular disorders across all treatments except insulin for DSGs ( Figure 1—figure supplement 8 ) . In order to identify the determinants of gene expression and splicing variation , we partitioned the variance in both gene expression and intron excision ( Hoffman and Schadt , 2016 ) . In particular , we were interested in uncovering the extent of the contribution of genetics , environmental and cell type effects to gene expression and intron excision . Variance which cannot be attributed to any of these defined components is counted as residual , which can be generally interpreted as due to stochasticity in the transcription process , technical variation , and/or unknown variables ( e . g . cell cycle stage ) . First , we considered the relative contribution of cell type , treatment , individual and batch effects ( i . e . plate ) to the overall variance . To this end , we partitioned the variance of all deep-sequenced samples . As expected , cell type identity has the strongest effect on both gene expression and splicing variance across samples for most genes ( variance explained median value 74% and 70% , respectively , which is consistently bigger than the batch effect ) , with a relatively small contribution from the individual and treatment ( Figure 2 ) . Within each cell type , we considered treatment , individual , batch effects ( i . e . plate ) , and we also considered the effect of any interaction between individual and treatment . These interactions should capture potential G×E effects for a specific treatment , but may also include epigenetic interactions . As expected , once we removed cell type effects , we observed a larger contribution from the other factors to both gene expression and splicing variation across samples . Similar patterns were observed for both gene expression and splicing . For example the contribution of the individual is largest in CMs ( 55% gene expression , 45% splicing ) , followed by LCLs ( 36% gene expression , 28% splicing ) and IPSCs ( 28% gene expression , 20% splicing ) , similar to what was shown previously for gene expression ( Banovich et al . , 2018 ) . To investigate whether this result may reflect variation in the purity of the CMs , we considered the expression of the gene TNNT2 which encodes for the Cardiac muscle troponin T . The expression of this gene is used as a marker of differentiation of CMs and a surrogate of CM purity ( Ward and Gilad , 2019 ) . We repeated the analysis of variance for the CMs by including TNNT2 expression as an additional factor . The results show that the proportion of variance explained by the individual component does not change and that the median percent variance explained by TNNT2 expression is 6% ( Figure 2—figure supplement 1 ) . Treatment explained the greatest percentage of variance in LCLs ( 22% gene expression , 12% splicing ) , followed by IPSCs ( 11% gene expression , 6% splicing ) , then CMs ( 4% gene expression , 2% splicing ) . When we focused on the variance explained by measured variables ( plate , treatment , individual and their interaction ) , we found that treatment ( marginal and interaction with individual ) explained at least half of the variance for 64% of the genes . To investigate the genetic control of gene expression across cell types and treatments , we identified SNPs exhibiting allele-specific expression ( ASE ) . ASE occurs when there is a transcriptional imbalance between the maternal and paternal copies of an allele . Because ASE is measured within each sample , trans factors are kept constant , so any differences in allelic expression is most likely due to cis-regulatory variants . We used QuASAR ( Harvey et al . , 2015 ) to identify heterozygous genotypes and to provide an initial estimate of the ASE ( LABEL:SuppTable3 ) . In total , we quantified ASE at 282 , 278 unique SNPs in 22 , 397 genes . The number of SNPs with ASE across treatments within each cell type ranges from 612 to 1052 for LCLs , 607-1342 for IPSCs , and 812-1310 for CMs ( Figure 3A ) . For each individual , we measured ASE in up to 84 experimental conditions ( 3 cell types , 14 treatments , 2 technical replicates ) . This gives us an unprecedented opportunity to determine the factors that modify allelic expression in an individual . To avoid excluding conditions where an allele may be lowly expressed , we tested for ASE using a new linear model ( see Materials and methods ) , which incorporated ASE measurements across all conditions for a SNP in a given individual , for a total of 69 , 205 SNPs ( 10 , 142 genes ) that can be tested . By doing so , we directly infer the noise inherent in ASE measurements using the linear model ( ANOVA ) . We first identified 15 , 497 ( ∼23% ) SNPs that show evidence of ASE in any condition ( ANOVA test Equations 3 and 4 , FDR < 10% , LABEL:SuppTable9 ) , corresponding to 5640 genes . Reassuringly , we observed high correlation of ASE between replicates ( i . e . same individual , treatment , and cell type in different batches; Spearman’s correlation median=0 . 66 ) , which confirms that technical effects have a limited impact and that ASE measurements are reproducible ( Figure 3B ) . When we consider control conditions on the same plate , the median correlation is identical to the one between replicates , thus confirming that the two control conditions can be considered technical replicates of each other . Finally , we observe a significant decrease in the correlation between ASE measurements when we consider different treatments or different cell types , which indicates G×E effects ( p§lt;10-16 , Kolmogorov-Smirnov test ) . We then used the linear model to identify significant cell-type-specific effects , treatment-specific effects , and cell type × treatment effects on ASE , which we refer to as conditional ASE ( cASE ) ( ref . Materials and methods , Equation 5 , LABEL:SuppTable10 ) . We identified 7866 instances of cell type cASE ( 5452 unique SNPs , FDR < 10% ) , corresponding to gene × cell type interactions; 1409 instances of treatment cASE ( 1102 unique SNPs , FDR 10% ) , corresponding to gene × treatment interactions; and 929 instances of cell type × treatment cASE ( 715 unique SNPs , FDR < 10% ) , corresponding to gene × cell type × treatment interactions ( Figure 3C and D ) . When we considered genes with at least one SNP with ASE , we found 2822 unique genes with gene × cell type interactions , 979 unique genes with gene × treatment interactions , and 689 unique genes with gene × cell type × treatment interactions . We next investigated whether these genetic effects on gene expression have been previously observed in large scale eQTL mapping studies that largely ignored dynamic regulatory interactions . For this analysis , we considered the CMs and their tissue counterparts in GTEx ( left ventricle and atrial appendage ) , the LCLs and the GEUVADIS dataset ( Lappalainen , 2013 ) , and IPSCs and the I2QTL consortium data ( Bonder et al . , 2021 ) . In CMs , we identified 3033 genes with ASE . Of these genes , 50% ( 1519 ) and 52% ( 1619 ) were eGenes in left ventricle and atrial appendage from GTEx , respectively . This translates to a 1 . 52- and 1 . 46-fold enrichment of ASE genes in GTEx genes in the left ventricle and atrial appendage , respectively ( p§lt;10-16 , Fisher’s exact test , Figure 3E ) . Of the 338 genes with treatment cASE in CMs , 170 and 178 are eGenes in left ventricle and atrial appendage in GTEx . Interestingly , treatment cASE genes in CMs were not significantly enriched in either tissue ( odds ratio = 1 . 02 , 0 . 95 ) . In addition to eQTL mapping , GTEx also used ASE to measure cis-regulatory effects ( Castel et al . , 2020 ) . As with GTEx eGenes , GTEx left ventricle and atrial appendage genes with ASE were enriched for CM ASE genes ( odds ratio = 1 . 75 , 1 . 66 for each tissue , respectively , p§lt;10-16 ) , but not CM cASE genes . A similar pattern was seen in LCLs and IPSCs . In LCLs , there were 3237 genes with ASE , and 24% ( 764 ) were eGenes in Geuvadis ( 1 . 55-fold enrichment , p=1 . 4×10-13 ) . Of the 167 genes with cASE in LCLs , 26 were eGenes in Geuvadis . This represented a significant depletion of cASE genes in Geuvadis eGenes ( odds ratio = 0 . 55 , p=7 . 7×10−3 ) . In IPSCs , there were 3113 genes with ASE , and 80% were eGenes in i2QTL ( 1 . 49-fold enrichment , p=1 . 0×10-10 ) . Of the 352 genes with cASE in IPSCs , 284 were eGenes in i2QTL . As with the CMs , these cASE genes were not significantly enriched ( odds ratio = 1 . 03 ) . These results indicate that investigating the control of gene expression across different environmental contexts , even in a small number of individuals , can identify new instances of genetic regulation that are missed in large eQTL mapping studies that do not explicitly sample different environmental contexts . To investigate whether our G×E results replicate in other environments previously investigated , we calculated the overlap between genes with gene × treatment interactions in our dataset and genes with G×E identified in fourteen previous ASE and eQTL mapping studies spanning a range of cell types and treatments . Of 979 genes with treatment cASE , 850 ( 87% ) replicated in at least one of these datasets ( with p<0 . 05 in the original study , LABEL:SuppTable4 ) . Knowles et al . , 2018 exposed IPSC-derived CMs to doxorubicin , a chemotherapeutic agent , and performed response eQTL ( reQTL ) and response splicing QTL ( rsQTL ) mapping . When considering genes with a reQTL , 105 could be tested for cASE in CMs and 79 of these genes had nominally significant cASE ( p < 0 . 05 ) , while 12 displayed cASE after multiple test correction ( FDR < 10% ) spanning nine different treatments in our study ( LABEL:SuppTable5 ) . For example , for three of those genes ( FRAS1 , PDGFC , and MPHOSPH6 ) we can now annotate a G×E interaction with caffeine in addition to doxorubicin . Out of 740 genes with one or more sQTLs , 182 and 88 have ASE and cASE in our datasets ( p < 0 . 05 ) , respectively , whereas of the 62 genes with response sQTLs to doxorubicin 16 have ASE and 11 have cASE ( p < 0 . 05 ) . Overall , these results point to different regulatory mechanisms that can lead to G×E and that are context-specific . The number of genes with gene × treatment interactions varies with a low of 96 ( 93 SNPs ) for aldosterone and a high of 176 ( 159 SNPs ) for copper ( 1 . 8-fold range between minimum and maximum number of cASE ) . This is in contrast with the spread in gene expression and splicing responses across treatments ( two orders of magnitude lower than the range between minimum and maximum number of DEGs ) . When we considered cASE in each cell type separately , indeed we did not find a significant enrichment or depletion for cASE genes being differentially expressed or differentially spliced between treatment and control ( Figure 3—figure supplement 1 ) . However , despite the number of cASE being similar across treatments , the genes with G×E are largely different . Most treatment cASE SNPs were specific to one treatment ( 80 . 5% , Figure 3—figure supplement 2 ) , so there was little sharing between treatments ( Figure 3 ) . While this could be due to a lack of power in detecting cASE with small effect sizes , it also suggests that genes with the strongest cASE are different across conditions . SNPs with ASE and cASE may have a direct effect on gene expression or more often may indirectly reflect the effect of a regulatory variant located in non-coding regions . Regulatory sites that could be located within the gene transcript itself include those in the 3’ and 5’ UTRs , in splicing junctions , premature stop codons , and possibly exonic enhancers . Thus , SNPs with ASE and cASE located within these regulatory sites could be putative causal sites for the observed allelic imbalance ( Mohammadi et al . , 2019 ) . We retrieved the genomic location of the SNPs associated with ASE and cASE , as well as their predicted functional impact from dbSNP ( https://www . ncbi . nlm . nih . gov/snp , Sherry et al . , 2001 ) ( Figure 3—figure supplement 3 ) . When we considered all variants that could be tested for ASE , both ASE and cASE SNPs were found to be significantly enriched in 5’ and 3’ UTRs , and among variants classified as missense or altering a stop codon ( FDR < 10% , Fisher’s exact test; Figure 3G ) . On the contrary , SNPs located within upstream/downstream transcript and genic regions , within introns and in splicing sites were found to be significantly depleted in ASE/cASE SNPs . Taken together , these data suggest that functional SNPs with ASE or cASE most likely exert their function by changing transcriptional and/or RNA stability regulatory sequences because the 5’ and 3’ regions tend to affect these processes , rather than affecting splicing processes . To investigate the role of genes with cASE in human phenotypes , we considered genes associated with complex traits . To this end , we used the results from Probabilistic Transcriptome Wide Association Studies ( PTWAS ) ( Zhang et al . , 2020 ) , which combined eQTL data from GTEx and GWAS data from several large scale studies to identify which genes are most likely to be in the causal pathway for complex traits . First , we considered eight cardiovascular traits , including blood lipids , hypertension , coronary artery disease , and heart rate . Heart rate was the only trait with a significant overlap with CM-specific cASE , compared to the average overlap with other cardiovascular traits ( FDR < 10% ) . Interestingly , heart rate is the trait among those tested where cardiomyocytes have a direct physiological role ( Figure 4A ) . We then investigated the role of G×E in a larger panel of 45 complex traits ( Figure 4B ) . Five traits were enriched for treatment cASE genes in at least one condition ( p < 0 . 05 ) compared to the other treatments . For example , genes associated with low birthweight are enriched for genes that have caffeine cASE ( Figure 4B ) . For the 428 genes associated with coronary artery disease ( CAD ) , cASE for metal treatments have the greatest overlap , with cadmium and copper each having five cASE genes associated with CAD , and zinc having three ( Figure 4C ) . Interestingly , cadmium causes endothelial dysfunction and promotes atherosclerosis ( Messner et al . , 2009 ) , while copper deficiency has been linked to the development of CAD ( DiNicolantonio et al . , 2018 ) . FAM213A ( also known as PRXL2A ) , which is involved in redox regulation ( Guo et al . , 2015; Xu et al . , 2010 ) is shared as treatment cASE for cadmium and copper , in addition to being treatment cASE for nicotine . We also considered an atrial fibrillation GWAS which identified 151 genes near associated loci ( Nielsen et al . , 2018 ) . Ten of these genes displayed treatment cASE in at least one condition , for a total of 21 gene-cASE condition pairs ( Figure 4—figure supplement 1 , LABEL:SuppTable12 ) . The greatest overlap was with dexamethasone cASE . Corticosteroids are antiinflammatory drugs which share molecular pathways with inflammatory disease . Inflammation is a risk factor for atrial fibrillation ( Hu et al . , 2015; Harada et al . , 2015 ) . Additionally , in multiple population-based , case-control studies , corticosteroid use was associated with new-onset atrial fibrillation ( Van Der Hooft et al . , 2006; Christiansen et al . , 2009 ) . Unlike atrial fibrillation , GWAS for heart failure have not identified many association signals , likely due to the highly heterogeneous nature of the disease . Two recent meta-analyses identified 3 and 13 genes , respectively ( Arvanitis et al . , 2020; Shah et al . , 2020 ) . Three of the genes associated with heart failure had treatment cASE: FAM241A in five conditions ( copper , dexamethasone , insulin , caffeine , and vitamin A ) ; BAG3 in five conditions ( dexamethasone , caffeine , vitamin A , nicotine , and aldosterone ) ; and KLHL3 in triclosan . While most causes of heart disease are not due to a mutation in a single gene , regulatory variation controlling the expression of Mendelian disease genes can affect complex trait risk ( Freund et al . , 2018 ) . In CMs , we found three genes in OMIM ( Online Mendelian Inheritance in Man ) ( Hamosh et al . , 2005 ) with gene × treatment interactions that are known to cause Mendelian forms of heart disease: PSMA6 , AARS2 , DSC2 . AARS2 displayed cASE in response to nicotine and mutations in this gene cause Combined Oxidative Phosphorylation Deficiency 8 , which manifests as fatal infantile hypertrophic mitochondrial cardiomyopathy ( Figure 4D ) . DSC2 , instead , displayed cASE in response to copper and mutations in this gene cause Arrhythmogenic right ventricular dysplasia 11 ( Figure 4E ) . Much of complex trait heritability is not explained by genome-wide significant variants , but rather results from the contribution of many variants with smaller effect sizes ( Manolio et al . , 2009; Yang et al . , 2010; Boyle et al . , 2017 ) . Methods to partition the heritability of complex traits can be used to determine the proportion of heritability attributable to genomic annotations ( Yang et al . , 2011; Finucane et al . , 2015 ) . We used RHE-mc ( Pazokitoroudi et al . , 2020 ) to partition the heritability of complex traits from UK Biobank based on cASE annotations . First , we partitioned the heritability of 17 traits , using an annotation of genes with cell type cASE for CMs . CM cell type cASE was significantly enriched for diastolic blood pressure , autoimmune disease , and height . We then considered 22 traits , using treatment cASE as an annotation ( Figure 4—figure supplement 2 ) . Diastolic blood pressure was enriched for vitamin A , systolic blood pressure was enriched for vitamin A , copper , and zinc . Respiratory disease was enriched for caffeine , and smoking status for selenium . The presence of cASE indicates that the genetic control of gene expression varies significantly with changes in the environment . To further investigate the role of the environment and cell type in controlling ASE , we used a mixed-effects model to partition the variance in ASE for each SNP-individual pair across all experimental conditions , similar to our analysis of gene expression and splicing variance ( see Materials and methods ) . When considering all cell types together , the greatest amount of explainable variance in ASE was captured by the cell type ( Figure 5A , LABEL:SuppTableMixed_All ) , similar to the results explaining variability in gene expression ( Figure 3A ) . However , when considering variance in ASE within each cell type , a much larger proportion of the explainable variance was attributed to the effect of a treatment ( Figure 5B , LABEL:SuppTableMixed_Sep ) . Specifically 17 . 5% , 16 . 6% , and 16 . 8% of variance ( median ) in ASE was attributed to treatment effects in CMs , IPSCs , and LCLs , respectively , compared to 3 . 6% , 11 . 2% , and 21 . 9% variance ( median ) in total gene expression . There has been extensive investigation into gene expression stochasticity ( Sigalova et al . , 2020 ) . However , the role of stochasticity in the genetic control of gene expression has proven more difficult to study ( Sarkar et al . , 2019 ) . Because our experimental design measures ASE across many conditions , we can analyze the residual variance as a measure of genetic control stochasticity . We found that residual variance at a particular SNP is correlated ( r∈[0 . 55-0 . 59] , p§lt;10-16 , 95% confidence intervals = [0 . 55 , 0 . 63] for LCL vs IPSC , [0 . 50 , 0 . 59] for LCL vs CM , and [0 . 55 , 0 . 61] for IPSC vs CM ) across cell types , indicating that unknown factors that contribute to the genetic control of gene expression are conserved across tissues ( Figure 5C ) . In LCLs , genes with high residual variance were enriched in 22 GO terms , with the greatest enrichments being related to tissue morphogenesis and developmental processes , while genes with low residual variance were enriched in 82 GO terms , including RNA processing and immune activation . In IPSCs , low residual variance genes were enriched in 128 GO terms , including peptide metabolic pathways , which was also seen in CMs . We found that low residual variance genes in CMs were enriched in 123 GO terms , including processes related to cell signalling ( LABEL:SuppTable6 , Figure 5—figure supplement 1 ) . To investigate the biological significance of ASE changes in magnitude and variance , we considered nine genomic features: genes differentially expressed in response to any treatment , genes differentially spliced in response to a treatment , genes with a TATA-box promoter , CpG percentage in the promoter , GC percentage in the promoter , genes which cause mendelian cardiomyopathies , genes associated with complex traits by TWAS , gene tolerance to loss of function mutations , and dN/dS ratio ( the ratio of nonsynonymous to synonymous mutations in a gene , which is a measure of selective pressure ) ( Figure 5D ) . We first considered the distribution of allelic expression for each of these categories . DEGs were characterized by high ASE compared to genes that do not respond to the treatments ( Figure 5E ) . Allelic expression was also elevated for genes tolerant to loss of function mutations , while genes associated with complex traits tend to have lower allelic expression ( Figure 5F and G ) , which reflects the difference in phenotypic relevance for these gene categories . Genes with lower ratio of nonsynonymous to synonymous mutations ( dN/dS , i . e . genes under negative selection ) are associated with lower mean ASE . When we considered regulatory features of the genes with ASE , we found lower ASE for genes with a TATA box , and for genes with higher promoter CpG and GC content . When focusing on variance in allelic expression , we found differential gene expression to be associated with increased residual variance , suggesting that for genes with more dramatic responses to the environment , gene expression is under a less stringent genetic control ( Figure 5E ) . Differentially spliced genes , however , are associated with increased treatment variance , and not increased residual variance ( Figure 5—figure supplement 2 ) . We then considered promoter architecture . SNPs in genes with a TATA box had increased residual variance compared to SNPs in genes without a TATA box ( Figure 5 ) . We found no significant relationship between treatment variance and whether the gene contains a TATA box . In addition , GC content and CpG islands in a gene’s promoter was associated with increased treatment variance and decreased residual variance . Genes which are known to cause cardiomyopathies when mutated had a small increase in residual variance ( p = 0 . 0032 ) , while genes implicated in any of the 103 TWAS in Zhang et al . , 2020 had decreased residual variance . Finally , we considered if intolerance to loss of function and dN/dS ratio could be related to ASE variance . As expected , genes that are less tolerant to loss-of-function mutations and genes under negative selection ( i . e . low dN/dS ) have significantly lower residual variance .
A fundamental open question in the field of genomics is understanding the sources of variation in gene expression across different individuals , cell types , and environmental contexts . Each of these components has been investigated separately in previous studies ( Gibson , 2008; Idaghdour et al . , 2010; Favé et al . , 2018; Aguirre-Gamboa et al . , 2016; Horst et al . , 2016; Maghbooli et al . , 2018; Wang et al . , 2015; van der Wijst et al . , 2018; Dimas et al . , 2009; Flutre et al . , 2013; Knowles et al . , 2018; Manry et al . , 2017; Nédélec et al . , 2016; Alasoo et al . , 2018; Kim-Hellmuth et al . , 2017; Quach et al . , 2016; Çalışkan et al . , 2015; Lee et al . , 2014; Fairfax et al . , 2014; Maranville et al . , 2011; Mangravite et al . , 2013; Barreiro et al . , 2012; Alasoo et al . , 2019; Huang et al . , 2021; Moyerbrailean et al . , 2016; Findley et al . , 2019 ) . Here , in order to investigate the transcriptome along the three different axes , we used a study design that combines IPSC technology , high-throughput screening and allele-specific expression analyses . The cell type context has the overall strongest effect on gene expression , splicing , and allelic expression . This is observed both in terms of differences of the mean gene expression levels across conditions using fixed effect models , or in analyzing the variance on gene expression using random effect models . Importantly , we demonstrate that a large number of context-specific genetic effects on gene expression are not captured by existing large cohort eQTL studies ( e . g . GTEx , GEUVADIS , i2QTL ) , but can be discovered even with limited sample sizes , when using an allele-specific expression study design . Our unique approach , accounting for cell-type but also genetic and environmental influences has also revealed that environmental impact on gene expression is substantial and particularly important for genes that influence complex traits . IPSCs are a valuable model system for studying primary cell types which are difficult to obtain and/or culture ( Sterneckert et al . , 2014 ) . Extensive work by our group and others have shown that despite the potential introduction of reprogramming effects , IPSC-derived CMs mimic primary CMs ( Pavlovic et al . , 2018 ) , recapitulate in vivo CM phenotypes ( Matsa et al . , 2016; Burridge et al . , 2016; Carvajal-Vergara et al . , 2010 ) , retain donor-specific expression signatures ( Panopoulos et al . , 2017; Burrows et al . , 2016; DeBoever et al . , 2017; Rouhani et al . , 2014 ) , and can be used to assess the effects of environmental exposures on CM gene expression and cellular physiology ( Ward and Gilad , 2019; Knowles et al . , 2018; Sharma et al . , 2017; Kitani et al . , 2019; Matsa et al . , 2016; Sharma et al . , 2014; Burridge et al . , 2016 ) . This ability to study CMs themselves is especially important given that GWAS studies for heart failure have revealed little of the genetic architecture of the disease . While pedigree analysis suggests that the heritability of heart failure is 26 - 34% ( Lindgren et al . , 2018 ) , two recent , large-scale GWAS identified only eleven ( Shah et al . , 2020 ) and three ( Arvanitis et al . , 2020 ) genomic loci associated with heart failure . This could be due to a lack of quantitative measurements of heart failure , complex etiologies , or the significant environmental contributions to developing heart failure ( Rau et al . , 2015 ) . Our approach allows for interrogation of the effects of environmental exposures in a controlled environment on CM gene expression . Some treatments had very dramatic effects on gene expression , such as copper and selenium in LCLs with more than 10 , 000 DEGs each . However , these large changes in gene expression were seen in our previous work ( Moyerbrailean et al . , 2016 ) ( 7869 DEGs in copper , 14 , 057 DEGs in selenium ) , and the increased sequencing depth and greater number of individuals in this study increased our power to detect DEGs . On average , we show the number of DSGs is directly correlated with the number of DEGs , therefore suggesting that environmental perturbations have a similar degree of effect on both transcriptional and post-transcriptional processes . However , only a minor fraction of genes were found to be both differentially expressed and spliced , which suggests that changes in gene expression and splicing represent distinct regulatory mechanisms by which cells can respond to their environment . Indeed , this is supported by GO and KEGG enrichment analyses , showing DSGs in CMs are involved in cytoskeletal activities , whereas DEGs were found to play a role in ion channel processes . Partitioning the variance in gene expression and intron excision across all cell types gives a broad overview of the relative contribution of cell type and treatment components to transcription and splicing processes . There were similar patterns for gene expression and splicing . Similar to previous work on partitioning the variance in gene expression across a panel of LCLs , IPSCs , and CMs by Banovich et al . , 2018 , variance due to the individual component was greatest in CMs , followed by LCLs and IPSCs . Yet herein , we were additionally able to consider the treatment component , which was greatest in LCLs and least in CMs . One possible explanation is that the chosen treatments may have more effect in LCLs; alternatively , this may be a consequence of greater inter-individual heterogeneity in CMs compared to LCLs . Regardless of cell type , residual variance was greater for splicing than gene expression . This is consistent with previous work showing that splicing is an error-prone process with a high degree of noise ( Melamud and Moult , 2009; Pickrell et al . , 2010; Wan and Larson , 2018 ) . A key new feature of our study design is that it allowed us to analyze allelic expression across 12 environments ( and two controls ) , three cell types , six individuals and two technical replicates each , for a total of 504 experimental samples . This is the largest single study of allelic expression comparing in parallel cell type effects , treatment conditions , and their interactions while controlling for technical variation . We used a new linear model to precisely estimate allelic expression and its variance directly from all the measurements of ASE for each individual ( up to 84 ) . As a result , we have a more complete view of the pervasiveness of G×E and we were able to directly investigate whether environmental effects on genetic regulation of gene expression differ across cell types . Our study design allowed us to specifically investigate interactions between genotype , cell type , and treatment . Interestingly , many treatment effects on allelic expression vary across cell types , therefore it is important to consider jointly cellular and environmental contexts . As the IPSC technology allows investigators to study multiple cell types from the same individuals , future study designs should consider the importance of studying environmental effects across cell types to learn about pleiotropic effects of potential biomedical relevance . Both eQTL and ASE studies have demonstrated that genetic effects on gene expression vary significantly across cellular ( The Gtex Consortium , 2020 ) and environmental contexts ( Moyerbrailean et al . , 2016; Knowles et al . , 2018; Manry et al . , 2017; Nédélec et al . , 2016; Alasoo et al . , 2018; Kim-Hellmuth et al . , 2017; Quach et al . , 2016; Çalışkan et al . , 2015; Lee et al . , 2014; Fairfax et al . , 2014; Maranville et al . , 2011; Mangravite et al . , 2013; Barreiro et al . , 2012; Alasoo et al . , 2019; Huang et al . , 2021 ) . However , we do not know how stable the genetic control of gene expression is across technical and environmental contexts . This cannot be investigated with eQTL mapping because the genetic effects on gene expression are estimated across individuals , but can be explored with multiple measurements of allelic expression from the same individual . We observed that residual variance of allelic expression is conserved across cell types , which indicates that the intrinsic properties controlling variation in allelic expression are consistent in different tissues and have biological significance . For example , genes which had a strong expression response to at least one treatment had increased residual variance over genes which had no response or a smaller response , while genes with a splicing response had greater treatment variance only . We hypothesize that greater residual variance indicates a less stringent control of allelic expression . This may depend on the regulatory architecture of these genes and may similarly enable large fluctuations in gene expression across contexts . Prior work on gene expression variance has demonstrated that genes with a TATA box have increased noise ( Mogno et al . , 2010 ) . This was also shown by Sigalova et al . , 2020 as well as in our data here ( Figure 2—figure supplement 2 ) . However , this finding has not been evaluated for the genetic control of allelic expression . Our results indicate that genes with ASE which are under the control of a TATA box promoter have increased residual variance over ASE genes without a TATA box , and thus greater noise . TATA box promoters , as opposed to CpG island-associated promoters , have been associated with tissue-specific genes and high conservation across species ( Carninci et al . , 2006 ) . Genes which are more tolerant to loss of function mutations showed greater allelic expression residual variance , indicating that redundancy in gene function allows for less stringent genetic control . This could allow for the evolution of new regulatory elements , resulting in new patterns of gene expression . Conversely , genes which are under negative selection ( i . e . low dN/dS ) have low residual variance , underscoring the importance of preserving stable expression of these genes . Tolerance to ASE variation could result in robustness against regulatory decoherence ( Lea et al . , 2019 ) , which could be further explored in future studies . Cell type and treatment-specific effects are the largest identifiable sources of variation in allelic expression for hundreds of genes , as we demonstrated using both variance decomposition and fixed effects as statistical models . Splicing sites have very conserved sequences , which are recognized by spliceosomal and accessory proteins which ultimately determine splicing patterns . Moreover , splicing is finely tuned by regulatory cis RNA sequences within both exons and introns , which are recognized by several RNA-binding proteins ( RBPs ) . Mutations within these consensus sequences therefore have catastrophic consequences in pre-mRNA splicing , eventually being associated with a plethora of different pathological conditions ( Faustino and Cooper , 2003; Lukong et al . , 2008; Anna and Monika , 2018 ) , including several forms of cystic fibrosis ( Friedman et al . , 1999; Bobadilla et al . , 2002 ) and in both Becker and Duchenne muscular dystrophies ( Habara et al . , 2009 ) . In addition to deleterious mutations , naturally occurring polymorphisms such as SNPs may contribute to alter the strength of splicing signals , eventually changing splicing outcomes . However , SNPs linked either to ASE or cASE were found to be mainly depleted in both donor and acceptor splicing sites , eventually suggesting that differential intron splicing is not a major mechanism underlying ASE/cASE . In fact , those SNPs were found to be enriched within several genic regions , in particular within UTRs and , interestingly , causing either the loss or gain of stop codons . These events have dramatic effects on RNA stability , as improper stop codon localization can activate the nonsense mRNA decay ( NMD ) pathway , resulting in transcript degradation ( Brogna and Wen , 2009 ) . Moreover , synonymous polymorphisms may alter the codon optimization of a given mRNA , a process which has been linked to reduced half-life ( Presnyak et al . , 2015 ) . All these events may eventually lead to reduced abundance of the transcript carrying a particular allele , compared to the other copy , thus resulting in ASE . We identified 3198 genes with cASE , including 2822 genes with cell type cASE , 979 genes with treatment cASE and 689 genes with treatment × cell type cASE . Many genes with treatment cASE overlap with genes that have context-specific genetic effects discovered by other studies; yet there is little sharing for any specific pair of conditions . In our study , since we compare ASE within the same genetic background , this limited sharing of ASE between treatments strongly suggests the existence of independent regulatory variants in context specific cis-regulatory modules . eQTL mapping consortia , like GTEx , identify eQTLs under one arbitrary condition , so genetic effects on gene expression which occur only transiently during development or under certain environmental conditions will be missed ( Moyerbrailean et al . , 2016; Strober et al . , 2019; Findley et al . , 2019; Resztak et al . , 2021; Umans et al . , 2021; Cuomo et al . , 2020 ) . A common approach to increasing identification of eQTLs is to increase the sample size . However , interrogating additional environmental conditions will be required to gain a more complete understanding of the genetic control of gene expression . This is reflected in the lack of enrichment for cASE genes in eQTLs from large studies in three tissues/cell-types . Even with a sample size of only six individuals , we identified many new genes which are genetically regulated under specific environmental conditions . For example , in CMs approximately half ( ∼1500 ) of the ASE genes were not eGenes in GTEx heart tissues . They were identified as ASE genes in our study due to the wide range of compounds we exposed them to . More than 160 of these ASE genes unidentified by GTEx showed evidence of cASE , thus pinpointing the environmental condition which was responsible for altering regulation of gene expression . Indeed a large fraction of CM cASE genes ( >47% ) are not eGenes in GTEx as detected by eQTL mapping or ASE . Because the CM samples were sequenced to very high depth , and because the power to detect allelic bias is based on read depth , this result could be partially due to having high statistical power to detect ASE and cASE . However , this does not seem to be the case , as we obtained similar results for LCLs and IPSCs , compared to the eQTL results in the Geuvadis and i2QTL datasets , respectively , with ASE genes being enriched in eGenes , but not cASE . While comparisons across studies may be complicated by several factors including differences in haplotype structures , study populations , and sequencing depth , the results are highly concordant and support the same conclusion . Our results show the importance of considering the relevant tissue type , as CM-specific cASE are enriched for heart rate , which is a phenotype with a direct physiological role for the heart muscle . The importance of the identification of G×E in a range of environments is illustrated by our findings that Mendelian cardiovascular disease genes and TWAS genes in a wide variety of traits display G×E . In total , across all cardiovascular related traits we identified 169 genes with cASE , spanning the 12 conditions tested . In cardiomyocytes , 52 genes with cASE were associated with cardiovascular disease traits . Among the treatment effects , metal ions ( cadmium and copper ) and corticosteroids ( dexamethasone ) are the most common environments interacting with genetic risk for CVD and atrial fibrillation , respectively . For example , the aryl hydrocarbon receptor nuclear translocator two gene ( ARNT2 ) was identified by GWAS for atrial fibrillation ( Nielsen et al . , 2018 ) and is a treatment cASE gene in CMs for insulin . ARNT2 is part of the hypoxia-inducible factor ( HIF ) pathway ( Mandl et al . , 2016 ) , which has a role in the progression of obesity and metabolic disease ( Gaspar and Velloso , 2018 ) . Indeed , insulin itself is known to induce the expression of HIF1A ( Treins et al . , 2002 ) , which we confirm in our differential gene expression data . In turn , the HIF pathway is especially important for cardiovascular disease and the response to cardiac ischemia ( Semenza , 2014 ) . This approach could be further expanded by targeting individuals with large polygenic risk scores to discover additional G×E effects of clinical relevance , in future studies . The polygenic nature of human complex traits provides a formidable challenge to tackle the genetic and molecular basis of interindividual variation . Our study demonstrates that interactions between genetic and environmental factors are common , but require specifically designed studies to be discovered . These results also have direct implications on issues related to the portability and interpretation of polygenic risk scores across individuals exposed to different environments . While gene-environment interactions further complicate the overall picture of human complex trait variation , they may represent an important contribution to the overall missing heritability that requires further study and careful consideration .
Experiments were conducted using three cell types derived from the same six individuals: Lymphoblastoid Cell Lines ( LCL ) , Induced Pluripotent Stem Cells , ( IPSC ) and IPSC-derived Cardiomyocytes ( CM ) . All cells were cultured at 37°C with 5% CO2 . Each batch consisted of the same cell type from three individuals , with 28 treatments and two controls . Each experiment was performed in duplicate for a total of 12 batches ( Figure 1 ) . LCLs from six Yoruba individuals were purchased from Coriell: GM18858 , GM18855 , GM18505 , GM18912 , GM18520 , and GM19209 . LCLs were maintained at a density of 200 , 000 to 1 million cells/ml in supplemented RPMI media ( 500 mL RPMI-1640 with glutamine [Fisher Scientific , 15-040-CM] , 75 mL FBS [Genemate S1200-500] , 5 mL GlutaMAX [35050-061 , ThermoFisher Scientific] and 5 mL penicillin/streptomycin ) . A total of 50 , 000 cells were plated per well of a round-bottom 96-well plate in 100 ul supplemented RPMI media 48 hr before treatment . Cells from each individual were plated in 32 wells representing 28 treatments and two controls . Each of the six LCLs were reprogrammed into iPSCs using episomal reprogramming vectors and expanded on a layer of MEF prior to conversion to feeder-independent growth as previously described ( Banovich et al . , 2018 ) . iPSCs were seeded on plates coated with a 1:100 dilution of Matrigel hESC-qualified Matrix ( 354277 , Corning , Bedford , MA , USA ) and cultured in iE8 media ( Invitrogen A1517001 ) supplemented with penicillin/streptomycin . Cells were passaged every 3–5 days using dissociation reagent ( 0 . 5 mM EDTA , 300 mM NaCl in PBS ) . During plating of cells , media was supplemented with 10 uM ROCK inhibitor ( stemolecule Y27632 , stemgent 04-0012 ) to aid in cell adherence . Media was changed every day thereafter . 50 , 000 cells were plated per well of a Matrigel-coated flat-bottom 96-well plate in supplemented E8 media 48 hours before treatment . Cells from each individual were plated in 32 wells representing 28 treatments and two controls performed in duplicate . Additional information on IPSC reprogramming , including number of passages and differentiation batch can be found in LABEL:SuppTable15 . CMs were differentiated from iPSCs using small molecules as previously described ( Banovich et al . , 2018 ) . Briefly , Wnt signaling was modulated by treating iPSCs with 12 µM of the GSK3 inhibitor CHIR99021 ( 4953 , Tocris Bioscience , Bristol , UK ) followed by 2 µM of the Wnt signaling inhibitor Wnt-C59 ( 5148 , Tocris Bioscience ) . Cells start to spontaneously beat between days 7 and 10 . The cardiomyocyte population was enriched through metabolic purification by culturing the cells in glucose-free , lactate-containing media from days 14 to 20 . In order to promote aerobic respiration following purification , on Day 20 , the cell culture media was replaced with galactose-containing media ( 500mL DMEM [without glucose , Life Tech A11430-01] , 50 mL FBS [GeneMate S1200-500] , 5 mL sodium pyruvate [Gibco 11360-070] , 2 . 5 mL HEPES [Fisher SH3023701] , 5 mL GlutaMAX [35050-061 , ThermoFisher] , 5 mL penicillin/streptomycin and 990 mg galactose [Sigma G5388] ) . On Day 25 , 5 days before treatment , cells were trypsinized and split to 114 , 000 cells/well in 100 μL of galactose-containing media on a Matrigel-coated flat-bottom 96-well plate . Cells from each individual were plated in 32 wells representing 28 treatments and 2 controls . Purity of the cell cultures was determined by measuring the expression of the cardiomyocyte-specific marker cardiac troponin T ( 564767 , BD Biosciences ) by flow cytometry . Flow cytometry was performed on Day 25 for 18912 , 18520 , and 19209 and on Day 27 for 18858 , 18855 , and 18505 . Purities ranged from 44% to 95% ( LABEL:SuppTable14 ) . Additional information on CM differentiation , including cell counts and the day at which the cells spontaneously started beating can be found in LABEL:SuppTable16 . All lines tested negative for mycoplasma contamination . All cell lines were authenticated by genotyping . Samples were treated on a total of twelve plates , and each plate was processed as a batch . Each plate contained samples from three individuals from a single cell type , exposed to 28 treatments plus two vehicle controls ( water and ethanol , referred as Control 1 and Control 2 ) . Additionally , each plate had a technical replicate performed the following day . Importantly , all three cell types for the same group of individuals were treated and harvested in parallel on the same day , to avoid that cell-type effects are confounded with batch effects . Cells were treated for six hours at concentrations listed in LABEL:SuppTable1 . As in Moyerbrailean et al . , 2016 , concentrations were chosen based on the highest physiological concentrations as reported by the Mayo Clinic ( http://www . mayomedicallaboratories . com ) or the CDC ( http://www . cdc . gov/biomonitoring/ ) , as available . For RNA library preparation , each plate was prepared as a batch . Treated cells were collected by centrifugation at 2000 rpm and washed 2x using ice cold PBS . Collected pellets were lysed on the plate , using Lysis/Binding Buffer ( Invitrogen ) , and frozen at −80°C . Poly-adenylated mRNAs were subsequently isolated from thawed lysates using the Dynabeads mRNA Direct Kit ( Ambion ) and following the manufacturer instructions . RNA-seq libraries were prepared using a protocol modified from the NEBNext Ultra II Directional RNA library preparation protocol to use 96 Barcodes from BIOOScientific added by ligation , as described in Moyerbrailean et al . , 2015 . Libraries from the same plate were pooled together and quantified using the KAPA Library Quantification Kit , following the manufacturer instructions and using a custom-made series of standards obtained from serial dilutions of the phi-X DNA ( Illumina ) . Library pools were sequenced to an average of 9 . 5M 75bp PE reads . Within each pool , individual library concentrations were normalized and repooled to achieve comparable sequencing depths . Twelve treatments were selected for deep sequencing on the basis of the strong transcriptional response they provoked in at least one cell type , in addition to both controls . Samples selected for deep sequencing were pooled within each plate and further sequenced on the NovaSeq 6000 using 300bp PE reads . Each plate of LCLs and IPSCs were sequenced once on one lane for an average of 147M reads , and each plate of CMs was sequenced twice on one lane for an average of 273M reads . The number of reads per library can be seen in LABEL:SuppTable13 . RNA-sequencing reads were aligned to the human reference genome using HISAT2 ( Kim et al . , 2015 ) ( https://ccb . jhu . edu/software/hisat2/index . shtml , version hisat2-2 . 0 . 4 ) with the following options: HISAT2 -x <genome> −1 <fastq_R1 . gz> −2 <fastq_R2 . gz> where <genome> represents the location of the genome file ( genome_snp_tran ) , and <fastqs_R1 . gz> and <fastqs_R2 . gz> represent that sample’s fastq files . Multiple sequencing runs were merged for each sample using samtools ( version 2 . 25 . 0 ) . We removed PCR duplicates and further removed reads with a quality score of <10 ( equating to reads mapped to multiple locations ) . Shallow and deep RNA-sequencing reads were aligned in an identical manner except for the reference used . GRCh38 was used for the deeply sequenced data and GRCh37 for the low coverage data which is not used after the initial screening . To identify differentially expressed ( DE ) genes in shallow and deep sequencing data , we used DESeq2 ( Love et al . , 2014 ) ( R version 3 . 5 . 2 , DESeq2 version 1 . 28 . 1 ) . coverageBed was utilized to count reads in transcripts from the Ensembl gene annotation with -s to account for strandedness and -split for BED12 input . The counts were then utilized in DESeq2 to determine changes in gene expression under the different treatment conditions . ( 1 ) Model1:CellType+Treatment . ID Each treatment was compared to its relevant vehicle control , except for plate CM1R2 , where we identified a technical problem with control 1 , so we only used control two for this plate ( LABEL:SuppTable13 ) . Note that comparison between the two controls yielded the expected finding of fewer than six differentially expressed genes for all other plates , thus confirming that the two controls are essentially technical replicates of the untreated condition . Multiple test correction was performed using the Benjamini-Hochberg procedure ( Benjamini and Hochberg , 1995 ) with a significance threshold of 10% . A gene was considered a DEG if at least one of its transcripts was differentially expressed and had an absolute log2 fold change >0 . 25 . Each cell type was run independently , and the model corrected for a composite variable of library preparation batch and individual . Full DESeq2 results from the shallow sequencing can be found in LABEL:SuppTable7 . After identifying DEGs from all 28 treatments in the initial shallow sequencing , we selected 12 to sequence more deeply . Nine of these twelve treatments were selected because they resulted in at least 60 DEGs in CMs . Selenium and Zinc were added because they induced a strong response in both IPSCs and LCLs , and Cadmium was added to complete the set of metal treatments . Full DESeq2 results from the deep sequencing can be found in LABEL:SuppTable8 . To identify genes that show evidence of treatment × cell type interactions , we analyzed the deep sequencing data across all cell types and treatments using the following likelihood ratio test to compare Model 1 and Model 2: ( 2 ) Model2:CellType+Treatment . ID+CellType:Treatment . ID To detect shifts in splicing patterns across cell types and environments we used LeafCutter ( Li et al . , 2018 ) , an intron-based splicing analysis tool . Briefly , LeafCutter uses short RNA-seq reads spanning exon-exon junctions ( i . e . split reads ) to estimate internal introns usage , hence being able to ultimately infer all splicing events which can be summarized with differential introns excisions . Overlapping introns sharing a splice sites are then identified , which are subsequently used to construct a graph where the connected components represent clusters . Lastly , for each cluster , the counts of the composing introns are jointly modeled with a Dirichlet-multinomial generalized linear model . We used the provided bam2junc . sh script to convert HISAT2-generated . bam files from the deep sequencing data to . junc files , as well as leafcutter_cluster . py script to perform intron clustering with the -s yes -m 50 -l 500000 options ( keep strand information , >50 split reads supporting each cluster and accepting introns with length up to 500kb ) . We removed clusters localizing on sex , mitochondrial and scaffold chromosomes , and used leafcutter_ds . R to perform differential intron excision analysis by contrasting each treatment with the appropriate control , with individuals ( i . e . cell lines ) as confounders . LeafCutter produces an output file in which , for every intron within a successfully tested cluster it computes the percentage spliced in index ( Ψ ) in the control and treatment condition , as well as the corresponding ΔΨ . LeafCutter exploits that retention and alternative excision of introns ( Braunschweig et al . , 2014; Boutz et al . , 2015 ) act as proxies of several subtypes of alternative splicing events , including exon skipping ( ES ) , intron retention ( IR ) , mutually exclusive exons ( MxE ) , alternative first and last exons ( AFE/ALE ) , alternative 5’ and 3’ splice sites ( A5SS/A3SS ) , among others . For every condition , we extracted introns belonging to clusters considered to be differentially spliced ( FDR < 10% ) , and we performed a two-sided binomial test defining n as the number of introns with ΔΨ§gt;0 , and p equal to the average proportion of positive events among all conditions . We considered a shift to be significant if the corresponding FDR < 10% . We used ClusterProfiler ( Yu et al . , 2012 ) R package to perform Gene Ontology ( Ashburner et al . , 2000; The Gene Ontology Consortium , 2019 ) , KEGG ( Wixon and Kell , 2000 ) , and Disease-Gene Network ( Piñero et al . , 2020 ) enrichment analyses . We considered two separate approaches for enrichment analysis . First , we calculated an enrichment of DEGs within each treatment per cell type against a background of all expressed genes in that treatment . This identified the pathways that were most enriched in DEGs for each treatment ( LABEL:SuppTable2 ) . Second , we calculated an enrichment within each cell type across all environments to compare GO , KEGG and DGN enrichments using the compareCluster function by submitting the lists of differentially spliced or expressed ( FDR < 10% ) genes in each condition , as well as one background gene list comprising all tested genes in that given cell type . We considered a process to be significantly enriched if FDR < 10% . For performing GO analysis on ASE variance , we first identified the top 20% and bottom 20% of genes based on their variance ( high variance or low variance genes , respectively ) . We tested for pathway enrichments using Gene Ontology annotations comparing the high variance against the low variance genes using ClusterProfiler with the same thresholds as described above ( LABEL:SuppTable6 ) . We used the Variance Partition package ( Hoffman and Schadt , 2016 ) to partition the total variance in gene expression . For each gene , the tool determines the fraction of the observed variance explained by each variable by implementing a linear-mixed model , ultimately allowing multiple dimensions of variation to be analyzed simultaneously . Variance which cannot be attributed to any of the provided covariates is counted as residuals , which can be generally interpreted as due to noise that can have technical or biological origin . We included all transcripts with at least one count per million reads ( cpm ) in 50% of the libraries , and the resulting data was voom-normalized ( Ritchie et al . , 2015; Law et al . , 2014 ) . All variables included in the model were categorical and were therefore modeled as random effects . For the model including all cell types , we used the fitExtractVarPartModel function with the following formula: ( 1|Individual ) + ( 1|CellType ) + ( 1|Plate ) + ( 1|Treatment ) . When each cell type was considered independently , we dropped CellType and added a Individual:Treatment variable , accounting for the interaction between these two covariates . We restricted this analysis to genes tested for both splicing and gene expression so we can compare the origins of variation in these two transcriptional processes . For splicing , we merged all LeafCutter . junc files across every cell type and condition , where each column represents a sample and each row a different intron , in order to recover the full list of detected introns . We applied the suggested cutoff to extract introns considered to be expressed ( on average , at least one cpm in half of the libraries ) , and used the cell type covariate to design the matrix eventually used by voom to normalize expression counts . Finally , we used the same methods as for gene expression to partition the variance . Using the common SNPs from dbSNP Build 144 , we calculated the number of RNA-seq reads mapping to each allele using samtools mpileup . We inferred genotypes from these pileups using all samples from a given individual across all treatments and each plate together using QuASAR ( Harvey et al . , 2015 ) . Counts for each pair of controls on the same plate were combined . For all subsequent analyses , we focused on heterozygous SNPs with a read coverage greater than five and located on autosomes . Initial testing for ASE was performed for each sample separately using QuASAR . The amount of ASE identified via QuASAR aggregated by treatment is shown in Figure 3A , and the full output from QuASAR for tested SNPs can be found in LABEL:SuppTable3 . Previous efforts for detecting ASE have commonly used binomial or beta-binomial tests ( Harvey et al . , 2015; Knowles et al . , 2017; Degner et al . , 2012 ) . Given that we have a large number of replicates and experimental conditions and to avoid excluding conditions where an allele may be lowly expressed , we tested for ASE using a linear model which incorporated ASE measurements across all conditions for a SNP in a given individual . To this end , we re-calculated ASE by adding a pseudocount to both reference and alternate read counts , and then computed the natural logarithm of the number of reference reads divided by the number of alternate reads . By doing so , we directly infer the noise inherent in ASE measurements using the linear model , obviating the need for binomial models to quantify the technical variance . To detect ASE in the model with all cell types combined , we selected 69 , 683 SNPs for which ASE was measured in at least five conditions per cell type , including at least one measurement for each control . We used an ANOVA test to compare a full model including the control ID , cell type , treatment , and cell type × treatment interaction term against a reduced model with the intercept set to 0: ( 3 ) Full:ASE∼Control+CellType+Treatment+CellType:Treatment ( 4 ) Reduced:ASE∼0 We only consider this test for SNPs with a sufficient number of observations given the number of parameters being estimated ( i . e . , at least 5 degrees of freedom ) . Full results for the ANOVA model can be found in LABEL:SuppTable9 . For SNPs with ASE detected by the ANOVA model , we performed spearman correlations for every pair of libraries using all SNPs shared between them . Pairs of libraries were then placed into four categories based on their differences . These were: ‘Across Cell’ ( n = 968 pairs ) , which included all pairs of libraries from the same individual and treatment , but different cell types; ‘Across Individual’ ( n = 490 pairs ) , which included all pairs of libraries from the same treatment , cell type , and plate , but different individuals; ‘Across Replicate’ ( n = 242 pairs ) , which included all pairs of libraries from the same individual , cell type , and treatment , but different plate; and ‘Across Treatment’ ( n = 3146 pairs ) , which included all pairs of libraries from the same individual , cell type , and plate , but different treatment . Finally , the category for ‘Control 1 vs Control 2’ ( n = 33 pairs ) includes correlations between the two control libraries from the same individual , cell type , and plate . To identify SNPs displaying cASE due to cell type , treatment , or the interaction between cell type and treatment , we used a fixed effect linear model . All SNPs with significant ASE detected via the ANOVA model were tested for cASE . Additionally , SNPs must have ASE measured in at least one of each control for each cell type to be tested . We modeled the ASE values for each SNP in an individual across all conditions as a function of cell type , treatment , cell type-treatment interaction , and a variable representing the vehicle used for that treatment ( Control ) : ( 5 ) ASE∼Control+CellType+Treatment+CellType:Treatment Results can be found in LABEL:SuppTable10 , and are summarized in LABEL:SuppTable_SigTestAll . We also investigate ASE , and cASE for each cell type separately which may recover situations in which a gene is only expressed in one cell type . First we detect ASE for each cell type across any condition: ( 6 ) Full:ASE∼Control+Plate+Treatment ( 7 ) Reduced:ASE∼0 Then , we test for each treatment component in the Full model to detect cASE . For the model with the cell types analyzed separately , ANOVA results can be found in LABEL:SuppTable_ANOVAsep , and results from the Full , fixed effect model ( Equation 6 ) , can be found in LABEL:SuppTable_FixedSep and are summarized in LABEL:SuppTable_SigTestSep . Using the genomic annotations from dbSNP build 153 , we assigned SNPs tested for ASE to one of 18 categories on the basis of their location in relation to a gene and predicted function . For each category , we calculated the enrichment of ASE and cASE SNPs separately relative to all SNPs tested for ASE and performed Fisher’s exact test to test for significance . Multiple test correction of p-values was performed using the Benjamini-Hochberg procedure . To identify traits which are enriched for CM cell type cASE , we focused on 8 TWAS for cardiovascular traits from Zhang et al . , 2020 . For each trait , we intersected all cell type cASE with TWAS genes , and calculated the proportion of overlap which is attributable to CM cell type cASE only . We used the one-sample proportion test to determine if the CM cell type cASE overlap for a given trait was significantly different from the average trait overlap for CM cell type cASE . To identify genes with treatment cASE that could influence complex traits , we intersected our cASE genes from the model with all cell types combined with genes whose expression has been implicated by transcriptome-wide association study ( TWAS ) to influence complex traits in 103 TWAS ( Zhang et al . , 2020 ) . Forty-five TWAS had at least 20 cASE gene overlaps across all treatments , and those were selected for further analysis . As above for CM cell type cASE , we used the one-sample proportion test to determine if the treatment cASE overlap for a given trait was significantly different from the average overlap for that trait across all treatments . P ( t , c ) =O ( t , c ) N ( t ) Z ( t , c ) =P ( t , c ) -P0 ( c ) P0 ( c ) * ( 1-P0 ( c ) ) /N ( t ) where t is the trait , c is the treatment , O ( t , c ) is the number of genes which are significant for both the trait and the treatment , N ( t ) is the sum of all genes for the trait which are treatment cASE in any treatment , and P0 ( c ) is the average P ( t , c ) across all traits . For plotting purposes , we abbreviated the TWAS study names from Zhang et al . , 2020 according to LABEL:SuppTable11 . To identify genes responsible for Mendelian traits , we downloaded OMIM’s Synopsis of the Human Gene Map ( morbidmap . txt ) on December 19 , 2019 and intersected treatment cASE genes from the model containing only the CM data with OMIM genes . This resulted in 95 OMIM traits , which were manually curated to identify traits relevant to heart disease . To investigate whether our G×E results replicate in other environments previously investigated , we calculated the overlap between genes with gene × treatment interactions in our dataset and genes with G×E ( P < 0 . 05 ) identified in fourteen previous ASE and eQTL mapping studies ( Maranville et al . , 2011; Idaghdour et al . , 2012; Mangravite et al . , 2013; Çalışkan et al . , 2015; Moyerbrailean et al . , 2016; Quach et al . , 2016; Zhernakova et al . , 2017; Knowles et al . , 2017; Leland Taylor et al . , 2018; Huang et al . , 2021; Knowles et al . , 2018; Lee et al . , 2014; Nédélec et al . , 2016; Barreiro et al . , 2012 ) . The full table of treatment cASE genes which replicated in these other studies can be found in LABEL:SuppTable4 . We used Fisher’s exact test to test for an enrichment in ASE and cASE genes in eGenes from three large eQTL studies for CMs , IPSCs , and LCLs , respectively: GTEx left ventricle and atrial appendage ( The Gtex Consortium , 2020 ) , i2QTL ( Bonder et al . , 2021 ) , and Geuvadis ( Lappalainen , 2013; Wen et al . , 2015 ) . ASE and cASE genes for each cell type were detected using the model that examined each cell type separately . For each cell type , we restricted our analysis to include genes which had been tested for being eGenes in the relevant eQTL study and had also been tested for ASE or cASE in our study . GTEx v8 eQTL data was downloaded from https://storage . googleapis . com/gtex_analysis_v8/single_tissue_qtl_data/GTEx_Analysis_v8_eQTL . tar . GTEx ASE data were downloaded from https://github . com/secastel/phaser/blob/master/gtex_v8_analyses/gtex_v8_tissue_by_gene_imbalance . tar . gz . i2QTL eGenes and tested genes were found in Supplemental Tables 3 and 7 in Bonder et al . , 2021 . Geuvadis eGenes were downloaded from https://www . ebi . ac . uk/arrayexpress/experiments/E-GEUV-3/files/analysis_results/EUR373 . gene . cis . FDR5 . best . rs137 . txt . gz . The list of tested Geuvadis genes were downloaded from http://www-personal . umich . edu/~xwen/geuvadis/geuv . fm . tar . gz . To estimate the heritability explained by G×E , we used RHE-mc ( Pazokitoroudi et al . , 2020 ) . We performed two separate analyses using the UK Biobank data . First , we quantified the heritability of complex traits using an annotation of CM gene × cell type genes . Second , we quantified the heritability of complex traits using an annotation of gene × treatment genes . In both cases , SNPs were annotated to G×E genes within 100 Kb . For the gene × treatment analysis , we included both gene × treatment and gene × treatment × cell type genes , with each treatment forming a separate annotation . The heritability partitioning was performed as described in Pazokitoroudi et al . , 2020 . Briefly , we excluded SNPs with greater than 1% missingness and minor allele frequency smaller than 0 . 1% . Further , we excluded SNPs that fail the Hardy-Weinberg test at significance threshold 10-7 as well as SNPs that lie within the MHC region ( Chr6: 25–35 Mb ) to obtain 7 , 774 , 235 SNPs . We included age , sex , and the top 20 genetic principal components ( PCs ) as covariates in our analysis for all traits . We used PCs precomputed by the UK Biobank from a superset of 488 , 295 individuals . Additional covariates were used for waist-to-hip ratio ( adjusted for BMI ) and diastolic/systolic blood pressure ( adjusted for cholesterol-lowering medication , blood pressure medication , insulin , hormone replacement therapy , and oral contraceptives ) . For the CM gene × cell type annotation , we partitioned the heritability of 17 complex traits , while for the gene × treatment analysis , we expanded to 22 traits in the same sample from the UK Biobank . For each annotation , we computed the heritability enrichment as the ratio of the percentage of heritability explained to the percentage of SNPs in that annotation . To quantify ASE variance , we used the same linear model as for identifying cASE , with the exception of modeling the cell type , treatment , cell type-treatment interaction , and batch variables as random effects: ( 8 ) ASE∼ ( 1|Control ) + ( 1|CellType ) + ( 1|Treatment ) + ( 1|CellType:Treatment ) As before , we only tested SNPs with significant ASE as determined by the ANOVA . To analyze each cell type independently , we used the following model: ( 9 ) ASE∼ ( 1|Control ) + ( 1|Batch ) + ( 1|Treatment ) Residual variance is partially a function of sequencing depth , so for all analyses of variance components for ASE , at each SNP , we have adjusted for the total number of reads covering that position . To determine the biological significance of ASE variance , we identified biological features which contributed to differences in treatment or residual variance . Given the overwhelming contribution of cell type to ASE variance , we used the ASE variance calculated within each cell type . We tested for a relationship between residual and treatment variance and six annotations using the following model: ( 10 ) Variance∼Annotation+SNPexpression We assigned SNPs tested for ASE to the genes in which they reside , and considered the following sevengene annotations: Sequencing files have been uploaded to the Sequence Read Archive ( SRA ) under Bioproject PRJNA694697 . Code is available at https://github . com/piquelab/GxExC ( Findley , 2021; copy archived at swh:1:rev:15df015227a05ce566fff158d312bd1a666e1235 ) . | The activity of the genes in a cell depends on the type of cell they are in , the interactions with other genes , the environment and genetics . Active genes produce a greater number of mRNA molecules , which act as messenger molecules to instruct the cell to produce proteins . The amount of mRNA molecules in cells can be measured to assess the levels of gene activity . Genes produce mRNAs through a process called transcription , and the collection of all the mRNA molecules in a cell is called the transcriptome . Cells obtained from human samples can be grown in the lab under different conditions , and this can be used to transform them into different types of cells . These cells can then be exposed to different treatments – such as specific chemicals – to understand how the environment affects them . Cells derived from different people may respond differently to the same treatment based on their unique genetics . Exposing different types of cells from many people to different treatments can help explain how genetics , the environment and cell type affect gene activity . Findley et al . grew three different types of cells from six different people in the lab . The cells were exposed to 28 different treatments , which reflect different environmental changes . Studying all these different factors together allowed Findley et al . to understand how genetics , cell type and environment affect the activity of over 53 , 000 genes . Around half of the effects due to an interaction between genetics and the environment and had not been seen in other larger studies of the transcriptome . Many of these newly observed changes are in genes that have connections to different diseases , including heart disease . The results of Findley et al . provide evidence indicating to which extent lifestyle and the environment can interact with an individual’s genetic makeup to impact gene activity and long-term health . The more researchers can understand these factors , the more useful they can be in helping to predict , detect and treat illnesses . The findings also show how genes and the environment interact , which may be relevant to understanding disease development . There is more work to be done to understand a wider range of environmental factors across more cell types . It will also be important to establish how this work on cells grown in the lab translates to human health . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"chromosomes",
"and",
"gene",
"expression",
"genetics",
"and",
"genomics"
] | 2021 | Functional dynamic genetic effects on gene regulation are specific to particular cell types and environmental conditions |
Spore killers in fungi are selfish genetic elements that distort Mendelian segregation in their favor . It remains unclear how many species harbor them and how diverse their mechanisms are . Here , we discover two spore killers from a natural isolate of the fission yeast Schizosaccharomyces pombe . Both killers belong to the previously uncharacterized wtf gene family with 25 members in the reference genome . These two killers act in strain-background-independent and genome-location-independent manners to perturb the maturation of spores not inheriting them . Spores carrying one killer are protected from its killing effect but not that of the other killer . The killing and protecting activities can be uncoupled by mutation . The numbers and sequences of wtf genes vary considerably between S . pombe isolates , indicating rapid divergence . We propose that wtf genes contribute to the extensive intraspecific reproductive isolation in S . pombe , and represent ideal models for understanding how segregation-distorting elements act and evolve .
Mendel’s law of equal segregation stipulates that paternal and maternal alleles of a gene should have an equal chance of being transmitted to progenies . This law guarantees a fair competition between different alleles and enables beneficial ones to prevail during natural selection . Meiotic drivers , a type of selfish genetic element , break Mendel’s law by skewing transmission ratios to their advantage , and thus can spread in a population even when having a deleterious effect on organismal fitness ( Lindholm et al . , 2016; Werren , 2011 ) . The term ‘meiotic drive’ was initially coined to describe segregation distortion resulting from preferential inclusion in the gamate during asymmetric female meiosis but has now been used more broadly to include biased transmission caused by postmeiotic mechanisms . In fact , some of the best-known meiotic drivers , such as Segregation Distorter in Drosophila and the t haplotype in mouse ( Lyon , 2003; Larracuente and Presgraves , 2012 ) , act postmeiotically to disable male gametes ( sperms ) that do not inherit them . This type of meiotic driver , called gamete killer , exists in animals , plants , and fungi ( Burt and Trivers , 2006 ) . Fungal gamete killers , or spore killers , have been found in several filamentous ascomycetes , most notably Neurospora and Podospora anserina ( Dalstra et al . , 2003; Grognet et al . , 2014; Hammond et al . , 2012; Turner and Perkins , 1979 ) . It is unclear how widespread spore killers are among fungal species . The fission yeast Schizosaccharomyces pombe is a prominent model organism for molecular and cell biology and has been increasingly used to study natural variation and genome evolution ( Brown et al . , 2011; Hu et al . , 2015; Jeffares et al . , 2015; Rhind et al . , 2011 ) . S . pombe natural isolates , which are nearly all haploids and have pair-wise nucleotide differences of less than 1% ( Jeffares et al . , 2015; Rhind et al . , 2011 ) , can readily mate with each other to form hybrid diploids , but the viability of spores derived from inter-isolate crosses is often below 5% and in many instances under 1% ( Gutz and Doe , 1975; Kondrat’'eva and Naumov , 2001; Naumov et al . , 2015 ) . This is in stark contrast to Saccharomyces cerevisiae , whose natural isolates have a similar level of nucleotide diversity but much better spore viability when inter-crossed ( Hou et al . , 2014 ) . One explanation for the within-species hybrid sterility in S . pombe is chromosomal rearrangement ( Avelar et al . , 2013; Brown et al . , 2011; Zanders et al . , 2014 ) . However , because one rearrangement reduces spore viability at most by half ( Avelar et al . , 2013; Hou et al . , 2014 ) , other factors are likely in play . It was shown recently that when the S . pombe laboratory strain , which was isolated from French grape juice ( Hu et al . , 2015; Osterwalder , 1924 ) , was crossed to a strain isolated from fermented tea ( initially called Schizosaccharomyces kambucha [Singh and Klar , 2003] , later renamed S . pombe var . kambucha [Rhind et al . , 2011] ) , at least three spore killers contributed to hybrid sterility ( Bomblies , 2014; Zanders et al . , 2014 ) , but the identities of the killer genes were unknown . In this study , through investigating the causes of intraspecific hybrid sterility of fission yeast , we uncovered the molecular identities of two active spore killers , which both belong to the wtf gene family . We analyzed their killing behaviors in both native and non-native genomic contexts . Furthermore , we performed comparative genomic analysis of wtf genes and revealed interesting patterns of divergence among them .
We have previously studied CBS5557 , an S . pombe strain isolated from grapes in Spain ( Hu et al . , 2015; Rankine and Fornachon , 1964 ) . Like most S . pombe isolates , CBS5557 differs from the laboratory strain , whose genome is the S . pombe reference genome , in a 2 . 23 Mb pericentric inversion on chromosome I ( Brown et al . , 2011 ) . After introducing the same inversion into the laboratory strain ( Hu et al . , 2015 ) , the viability of spores derived from the cross between the laboratory strain and CBS5557 increased from 14% to 22% ( Figure 1—figure supplement 1 and Supplementary file 1 ) . To uncover the causes of the remaining sterility , we performed next-generation sequencing ( NGS ) -assisted bulk segregant analysis ( Hu et al . , 2015 ) and found that among the viable progenies from the cross , SNP alleles of the laboratory strain ( reference alleles ) on chromosome III were markedly under-represented ( Figure 1A ) . Two broad transmission distortion peaks , one on the left arm ( hereafter referred to as the left peak ) and one on the right arm ( hereafter referred to as the right peak ) , were observed , suggesting that there are at least two spore killer genes on chromosome III of CBS5557 . 10 . 7554/eLife . 26057 . 003Figure 1 . The identification of cw9 and cw27 genes as spore killers . ( A ) NGS-assisted bulk segregant analysis of viable progenies from a cross between DY9974 , a heterothallic derivative of CBS5557 , and DY8531 , a laboratory strain harboring the 2 . 23 Mb chromosome I inversion ( Hu et al . , 2015 ) . Reference allele frequencies at SNP positions are plotted as dots . Red trend lines are based on a rolling median calculation . Two vertical dashed lines denote the positions of cw9 and cw27 genes on chromosome III . Black triangles mark the positions of centromeres . ( B and C ) Bulk segregant analysis of viable progenies from crosses between a laboratory strain and the backcrossed-1 strain , and between a laboratory strain and the backcrossed-2 strain , respectively . Blue and orange colored segments in the diagrams and on the X-axes of the plots denote chromosome III regions with reference genome sequence and those with CBS5557 genome sequence , respectively , in the two backcrossed strains . ( D and E ) Schematics of the regions surrounding cw9 and cw27 in the CBS5557 genome and the corresponding regions in the reference genome . Gene structures of cw9 and cw27 were predicted using the AUGUSTUS web server ( Stanke et al . , 2008 ) . Solo LTRs were annotated based on BLAST analysis . DOI: http://dx . doi . org/10 . 7554/eLife . 26057 . 00310 . 7554/eLife . 26057 . 004Figure 1—figure supplement 1 . Hybrid sterility between CBS5557 and the laboratory strain can be partially rescued by eliminating the difference in the 2 . 23 Mb chromosome I inversion . ( A ) Representative tetrads from a cross between LD775 , an h- laboratory strain , and DY9974 , an h+ derivative of CBS5557 . ( B ) Representative tetrads from a cross between DY8531 , an h- laboratory strain harboring the 2 . 23 Mb chromosome I inversion ( Hu et al . , 2015 ) , and DY9974 . ( C ) Quantitation of spore viability assessed by tetrad analysis . p-value was calculated using Fisher’s exact test . Numerical data are provided in Supplementary file 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 26057 . 00410 . 7554/eLife . 26057 . 005Figure 1—figure supplement 2 . Alignment of the 5’ portions of the cw9 gene of CBS5557 and the wtf9 gene of the reference genome . The sequences were aligned using MAFFT via Jalview ( Katoh et al . , 2002; Waterhouse et al . , 2009 ) . Identical bases are indicated by blue background . Red and green boxes denote the first and second predicted protein coding exons , respectively . Predicted start codons are highlighted with red letters . Gene structure of cw9 was predicted using the AUGUSTUS web server ( Stanke et al . , 2008 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 26057 . 00510 . 7554/eLife . 26057 . 006Figure 1—figure supplement 3 . Alignment of the two directly oriented LTRs flanking cw27 in the CBS5557 genome ( top and bottom sequences , respectively ) and the LTR at the corresponding location in the reference genome ( middle sequence ) . The reference genome LTR was referred to as PCRW_055 or C3_1895301 in Bowen et al . ( 2003 ) . According to the classification proposed in Bowen et al . , these three LTR sequences belong to the ζ clade . The sequences were aligned using MAFFT via Jalview . Identical bases are indicated by grey background . 7-nt-long homology at the breakpoint is highlighted in red . Matched sequences upstream and downstream of the breakpoint are colored in green . DOI: http://dx . doi . org/10 . 7554/eLife . 26057 . 006 To narrow down the genomic regions containing the spore killer genes , we performed multiple rounds of backcross and obtained strains with only a portion of chromosome III originating from CBS5557 and the rest of the genome coming from the laboratory strain ( see 'Materials and methods' ) . We selected two backcrossed strains that retained the ability to skew inheritance of markers inserted in the left and right peak regions , respectively . These two strains , referred to as backcrossed-1 and backcrossed-2 , respectively , were crossed to the laboratory strain and viable progenies were subjected to NGS-assisted bulk segregant analysis , which revealed the locations of the CBS5557 genomic segments present in the two backcrossed strains and confirmed the allele transmission bias ( left-side plots in Figure 1B and C ) . Through inspecting the Illumina and PacBio sequencing data of the CBS5557 genome , we discovered that , within the CBS5557 genomic segments in the two backcrossed strains , sequences most divergent from the reference genome are those containing wtf family genes ( Figure 1D and E ) . wtf ( for with Tf LTRs ) genes were so named because they are often in close proximity to solo long terminal repeats ( LTRs ) derived from Tf retrotransposons ( Bowen et al . , 2003; Wood et al . , 2002 ) . wtf genes are predicted to contain multiple introns and encode multi-transmembrane proteins and have been shown to be highly up-regulated transcriptionally during meiosis and sporulation by large-scale analyses ( Bowen et al . , 2003 ) . No homologs of wtf genes have been found in any other species , including other Schizosaccharomyces species ( Rhind et al . , 2011 ) . We deleted from the backcrossed-1 strain a wtf gene locating at the position of wtf9 in the reference genome but exhibiting a low sequence identity ( 42% ) to wtf9 in the 5’ portion ( Figure 1D and Figure 1—figure supplement 2 ) and found that the deletion notably reduced the segregation distortion ( right-side plot in Figure 1B ) , suggesting that this wtf gene , which we named cw9 ( the nomenclature of wtf genes in the CBS5557 genome will be explained later in this paper ) , is a spore killer . The residual allele frequency bias is probably due to additional spore killer gene ( s ) in the CBS5557 genomic segments present in the backcrossed-1 strain . Within the CBS5557 genomic segment in the backcrossed-2 strain , we found a wtf gene absent in the reference genome , possibly owing to a deletion event mediated by homologous recombination between LTRs ( Figure 1E and Figure 1—figure supplement 3 ) . Removing this wtf gene , which we named cw27 , from the backcrossed-2 strain completely abolished segregation distortion ( right-side plot in Figure 1C ) . To determine how cw9 and cw27 behave in their native strain background , we generated h+/h- CBS5557 diploid strains containing homozygous or heterozygous deletion of cw9 or cw27 , and performed tetrad analysis to evaluate spore viability ( Figure 2A and Supplementary file 1 ) . Homozygous deletion of cw9 or cw27 had no effect on spore viability ( Figure 2A and Figure 2—figure supplement 1 ) . In contrast , heterozygous deletion of either cw9 or cw27 caused a significant reduction of spore viability , which resulted mainly from the death of spores carrying the deletion ( Figure 2B ) . When cw9 and cw27 were both in the heterozygous state , a more severe loss of spore viability was observed , with the double deletion mutant spores suffering a particularly high level of death ( Figure 2A and B , and Supplementary file 1 ) . These data indicate that cw9 and cw27 can act as spore killers in the self-cross of CBS5557 when either of them is in the heterozygous state , and inheriting one killer can prevent its killing effect but not that of the other killer . 10 . 7554/eLife . 26057 . 007Figure 2 . cw9 and cw27 act as spore killers when in the heterozygous state in the CBS5557 background . ( A ) Heterozygous deletion but not homozygous deletion of cw9 or cw27 in CBS5557 h+/h- diploid caused spore viability loss . Spore viability was measured using tetrad analysis . Representative tetrads are shown in Figure 2—figure supplement 1 and in panel B . p-values were calculated using Fisher’s exact test . Numerical data are provided in Supplementary file 1 . ( B ) Among the spores derived from CBS5557 diploids with heterozygous deletion , loss of viability mainly occurred to spores with deletion . Asterisks indicate significant deviation from 50% ( p=1 . 94E-7 and 1 . 42E-11 for cw9∆ spores from cw9∆/cw9+ diploid and cw27∆ spores from cw27∆/cw27+ diploid , respectively , exact binomial test ) . Numerical data are provided in Supplementary file 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 26057 . 00710 . 7554/eLife . 26057 . 008Figure 2—figure supplement 1 . Representative tetrads from a wild type CBS5557 h+/h- diploid strain ( DY21782 ) , a CBS5557 h+/h- diploid strain with homozygous cw9 deletion ( DY21838 ) , and a CBS5557 h+/h- diploid strain with homozygous cw27 deletion ( DY21842 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 26057 . 008 To determine whether cw9 and cw27 can cause spore killing in a non-native genomic context , we constructed integrating plasmids carrying cw9 or cw27 , integrated the plasmids at various genomic locations in the laboratory strain , and subjected the resulting strains to crosses . In crosses where only one of the parental strains harbored cw9 or cw27 , spores inheriting the killer had normal viability , whereas spores not inheriting the killer suffered significant loss of viability , regardless of the genomic location of the killer ( Figure 3A and Figure 3—figure supplement 1 ) . We then performed crosses in which both parental strains harbored a killer at the leu1 locus ( Figure 3B and Figure 3—figure supplement 2 ) . When both parental strains contained the same killer , spore viability was normal . In contrast , when one parental strain had cw9 and the other parental strain had cw27 , spores inheriting either cw9 or cw27 experienced a dramatic loss of viability . These results demonstrate that cw9 and cw27 can act independently of genomic context and confirm that they do not exhibit mutual resistance . 10 . 7554/eLife . 26057 . 009Figure 3 . cw9 and cw27 act as spore killers when inserted into the genome of the laboratory strain . ( A ) In crosses of laboratory strains , when only one parental strain had an insertion of cw9 or cw27 at his3 , lys1 , or leu1 locus , spores without the insertion suffered viability loss . Representative tetrads are shown in Figure 3—figure supplement 1 . p-values were calculated using Fisher’s exact test . Numerical data are provided in Supplementary file 1 . ( B ) In crosses of laboratory strains , when both parental strains had an insertion of cw9 or cw27 at the leu1 locus , spore viability was normal when parents had the same killer , but was severely low when parents had different killers . The two parental alleles were distinguished by leu1-linked antibiotic resistance markers . Representative tetrads are shown in Figure 3—figure supplement 2 . Numerical data are provided in Supplementary file 1 . ( C ) Electron microscopy analysis of laboratory-background h+/h- diploid cells undergoing synchronous meiosis and sporulation . DOI: http://dx . doi . org/10 . 7554/eLife . 26057 . 00910 . 7554/eLife . 26057 . 010Figure 3—figure supplement 1 . Representative tetrads from laboratory-background crosses in which only one of the parental haploid strains had a vector or a killer-containing plasmid integrated at the his3 , lys1 , or leu1 locus . DOI: http://dx . doi . org/10 . 7554/eLife . 26057 . 01010 . 7554/eLife . 26057 . 011Figure 3—figure supplement 2 . Representative tetrads from three laboratory-background h+/h- diploid strains homozygous for plasmid integration at the leu1 locus and a laboratory-background h+/h- diploid strain heterozygous for plasmid integration at the leu1 locus , with one allele containing cw9 and the other allele containing cw27 . kanMX and hphMX markers were inserted between the coordinates 1963090 and 1963096 on chromosome II so that they are tightly linked to leu1 . DOI: http://dx . doi . org/10 . 7554/eLife . 26057 . 011 To observe the ultrastructural details of the spore killing process , we performed electron microscopy analysis on laboratory-background h+/h- diploid cells undergoing synchronous meiosis and sporulation induced by a shift to a nitrogen-free synthetic sporulation medium ( SSL-N medium ) ( Figure 3C ) . Two diploid strains were used: a strain heterozygous for a cw27-containing plasmid integrated at the his3 locus and a control strain heterozygous for an empty vector integrated at the same locus . 12 hr after the shift , cw27-containing cells formed four-spored asci indistinguishable from those formed by the control cells . At this time point , organelles inside spores were clearly visible . 18 hr after the shift , the cytoplasm of spores derived from the control cells became electron-dense and organelles were no longer discernable , owing to a little-understood spore maturation process ( Yoo et al . , 1973 ) . Interestingly , at the same time point , asci derived from the cw27-containing cells often exhibited a 2:2 pattern of spore morphology: two spores in an ascus looked like the matured spores in the control sample , whereas the other two spores exhibited cytoplasm with a lower electron density and visible organelles , an electron-lucent spore wall , and frequently a non-spherical shape . These data suggest that killer-affected spores fail to undergo a proper maturation process . cw9 and cw27 are both flanked by solo LTRs , and upstream of their predicted start codons they share a nearly identical 288-bp-long sequence ( hereafter referred to as the conserved_up sequence ) ( Figure 4—figure supplement 1 ) , which is also conserved upstream of the coding sequences of many other wtf genes ( Bowen et al . , 2003 ) . To determine whether these and other sequence features are important for spore killing , we performed truncation analysis ( Figure 4A and Figure 4—figure supplement 2 ) . The truncated versions were named Ta , Tb , Tc , Td , and Te . Their effects on spore viability in a heterozygous cross of laboratory strains were examined . The Ta versions , which lack the LTRs , caused spore killing phenotypes similar to those caused by longer constructs containing the flanking LTRs , indicating that the LTRs are not important . Interestingly , the Tb versions missing the sequence between the upstream LTR and the conserved_up sequence caused spores containing the killer to suffer a moderate viability loss , suggesting that these truncated killer genes partially lost the ability to protect killer-containing spores from being harmed . Removing from the Ta versions , the sequence downstream of the predicted stop codon ( Tc versions ) did not have a notable effect . Removing from the Tc versions , the last 10 amino-acid-coding codons ( Td versions ) greatly diminished spore killing , whereas removing from the Tc versions the sequences upstream of the start codons ( Te versions ) resulted in more indiscriminate killing than that caused by the Tb versions . These results suggest that intact C termini are important for the killing activity , and the upstream sequences are important for the protecting activity . 10 . 7554/eLife . 26057 . 012Figure 4 . Sequence requirement for the killing and the protecting activities of cw9 and cw27 . ( A ) Truncation analysis to assess the involvement of the 5’ and 3’ sequences of cw9 and cw27 in spore killing . In the diagrams on top , blue arrows represent LTRs and yellow bars represent the conserved_up sequence . Representative tetrads are shown in Figure 4—figure supplement 1 . p-values were calculated using Fisher’s exact test . Numerical data are provided in Supplementary file 1 . ( B ) The Td versions of cw9 and cw27 are able to effectively protect against killing despite their lack of killing activity . Representative tetrads are shown in Figure 4—figure supplement 2 . p-values were calculated using Fisher’s exact test . Numerical data are provided in Supplementary file 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 26057 . 01210 . 7554/eLife . 26057 . 013Figure 4—figure supplement 1 . Alignment of the 5’ portions of cw9 and cw27 . The starting positions of these two sequences correspond to those of the Ta truncations . The sequences were aligned using MAFFT via Jalview . Identical bases are indicated by blue background . Yellow and red boxes denote the conserved_up sequence and the first predicted protein coding exon , respectively . Predicted start codons are highlighted with red letters . DOI: http://dx . doi . org/10 . 7554/eLife . 26057 . 01310 . 7554/eLife . 26057 . 014Figure 4—figure supplement 2 . Representative tetrads from laboratory-background h+/h- diploid strains heterozygous for plasmid integration at the leu1 locus . DOI: http://dx . doi . org/10 . 7554/eLife . 26057 . 01410 . 7554/eLife . 26057 . 015Figure 4—figure supplement 3 . Representative tetrads from laboratory-background h+/h- diploid strains heterozygous for a Tc-version killer-containing plasmid integrated at the leu1 locus on chromosome II , and also heterozygous for a vector or a Td-version killer-containing plasmid integrated at the ars1 replication origin region upstream of the hus5 gene on chromosome I . DOI: http://dx . doi . org/10 . 7554/eLife . 26057 . 015 To determine whether the Td versions , which are largely devoid of the killing activity , can still protect , we performed crosses in which a strain with a Td-version-containing plasmid integrated at the ars1 locus was crossed to a strain with a Tc-version-containing plasmid integrated at the leu1 locus , and found that the Td versions of both cw9 and cw27 can effectively protect spores harboring them from being killed by the respective Tc-version killers ( Figure 4B and Figure 4—figure supplement 3 ) . Thus , perturbing the C termini resulted in separation-of-function mutants that retain the protecting activity but can no longer kill . Such kind of mutants , if present in natural populations , would behave similarly to the resistant alleles of other previously studied gamete killers , and may also represent an intermediate state toward the extinction of a killer ( Burt and Trivers , 2006 ) . In the reference genome , there are 25 wtf genes at 20 locations ( 15 singletons and five tandem pairs ) ( Bowen et al . , 2003 ) ( Figure 5A and B ) . Using PacBio sequencing , we found that in the CBS5557 genome , these 20 locations are occupied by 29 wtf genes ( 12 singletons , seven pairs , and one triplet ) , with four locations containing an extra wtf gene compared to the reference genome ( Figure 5A and B ) . We systematically named these 29 genes according to their syntenic relationship with the 25 wtf genes in the reference genome , with each name consisting of the prefix cw , a number ( from 1 to 25 ) , and a suffix for genes at locations where extra genes exist in the CBS5557 genome ( Figure 5A and B ) . In addition , there are three CBS5557 singleton wtf genes without syntenic counterparts in the reference genome and we named them cw26 , cw27 , and cw28 according to their order in the genome ( Figure 5A ) . Like cw27 , cw26 and cw28 also appeared to have been lost in the laboratory strain through LTR-mediated recombination ( Figure 5—figure supplement 1 ) , suggesting that this is a common mode of wtf gene turnover . 10 . 7554/eLife . 26057 . 016Figure 5 . wtf genes vary both in numbers and sequences between the reference genome and the CBS5557 genome . ( A ) Genomic locations of wtf genes in the reference genome and in the CBS5557 genome . The 25 wtf genes in the reference genome have been named according to their order in the genome ( Bowen et al . , 2003 ) . Their locations are depicted as 20 blue vertical bars , including five thick bars denoting five tandem pairs . Our PacBio sequencing analysis revealed that in the CBS5557 genome , there are 32 wtf genes , whose locations are depicted as 23 vertical bars , including three thin red bars at locations where no wtf genes exist in the reference genome and four thick red bars at locations where compared to the reference genome one extra wtf gene is found . The wtf genes in the CBS5557 genome are named with the prefix cw and a number from the name of the syntenic gene in the reference genome . Two genes of a tandem pair corresponding to a singleton in the reference genome are distinguished using the suffixes a and b . Among the three genes of the triplet , two are named cw11 and cw12 based on their homology to wtf11 and wtf12 , respectively , and the gene situated between cw11 and cw12 is named cw11x . Genes at new locations are named cw26 , cw27 , and cw28 . Chromosome lengths are not drawn to scale . ( B ) Diagrams depicting the eight genomic locations with more than one wtf gene in at least one of the two genomes . Genes are shown as arrows . Introns are not shown . wtf17 is depicted not according to its annotation at PomBase , with its 5’ boundary revised based on sequence alignment . ( C ) Maximum likelihood phylogenetic tree of 57 wtf genes of the reference genome and the CBS5557 genome . DNA sequences including the conserved_up regions , predicted coding sequences , and associated introns were aligned using the L-INS-i iterative refinement algorithm of MAFFT ( Katoh and Standley , 2014 ) ( Figure 5—source data 1 ) . Maximum likelihood analysis was performed using IQ-TREE ( Nguyen et al . , 2015 ) . The tree was rooted by midpoint rooting ( Hess and De Moraes Russo , 2007 ) . Red dots on nodes indicate IQ-tree-calculated ultrafast bootstrap ( UFBoot ) support values < 95% . Colored rectangles highlight phylogenetic neighbors . Two genes are considered phylogenetic neighbors if they are separated by a single internal node with a support value >= 95% . Green rectangles indicate the 11 pairs of neighbors each composed of a reference wtf gene and a syntenic CBS5557 wtf gene . Yellow rectangles indicate the five pairs each composed of two wtf genes locating at different genomic positions . Magenta brackets denote the 12 genes that share a 150 bp sequence within the predicted intron 1 ( see Figure 5—figure supplement 3 ) . These 12 genes are divided into two subtypes , Intron-1-ATG genes and Exon-2-ATG genes , based on the locations of the first ATG codons downstream of the 150-bp-long sequence . The brown dashed box indicates genes with six exons . Scale bar , 0 . 1 nucleotide substitutions per nucleotide site . DOI: http://dx . doi . org/10 . 7554/eLife . 26057 . 01610 . 7554/eLife . 26057 . 017Figure 5—source data 1 . MAFFT-aligned DNA sequences of 57 wtf genes of the reference genome and the CBS5557 genome . The sequences include the conserved_up regions , predicted coding sequences , and associated introns . DOI: http://dx . doi . org/10 . 7554/eLife . 26057 . 01710 . 7554/eLife . 26057 . 018Figure 5—source data 2 . Gene structure predictions for 57 wtf genes of the reference genome and the CBS5557 genome . DOI: http://dx . doi . org/10 . 7554/eLife . 26057 . 01810 . 7554/eLife . 26057 . 019Figure 5—source data 3 . MAFFT-aligned amino acid sequences of the predicted protein products of 57 wtf genes of the reference genome and the CBS5557 genome . DOI: http://dx . doi . org/10 . 7554/eLife . 26057 . 01910 . 7554/eLife . 26057 . 020Figure 5—figure supplement 1 . Schematics of how the three CBS5557-only singleton wtf genes may have been lost in the laboratory strain through LTR-mediated recombination . Shown at the bottom of each diagram is the breakpoint junction sequence with 100% identity between the two directly oriented LTRs flanking the CBS5557-only singleton wtf gene and its position in the reference genome . DOI: http://dx . doi . org/10 . 7554/eLife . 26057 . 02010 . 7554/eLife . 26057 . 021Figure 5—figure supplement 2 . Schematics depicting the high levels of diversity between five singleton CBS5557 wtf genes and their counterparts in the reference genome . Genes are shown as maroon arrows . Introns are not shown . Sequence alignment was performed using MAFFT via Jalview . Pair-wise identity was calculated by dividing the numbers of identical bases by the alignment length . DOI: http://dx . doi . org/10 . 7554/eLife . 26057 . 02110 . 7554/eLife . 26057 . 022Figure 5—figure supplement 3 . cw9 , cw27 , and 10 other wtf genes in the reference genome and the CBS5557 genome share a 150 bp conserved sequence in the predicted intron 1 . Red boxes denote the 150 bp sequence . The transcription start site of a short wtf23 isoform identified by the Iso-Seq method ( Kuang et al . , 2017 ) is situated closely downstream of this 150 bp sequence , and at an optimal distance from a canonical TATA box ( TATAAA ) ( Li et al . , 2015 ) . The first ATG codons downstream of the 150 bp sequence are located within the predicted intron 1 of the top five genes , which we call Intron-1-ATG genes , and within the predicted exon 2 of the bottom 7 genes , which we call Exon-2-ATG genes . These ATG codons are all in frame with the predicted coding sequences . Proteins initiated from these ATG codons correspond to the poison isoforms in the model proposed by Nuckolls et al . , 2017 ) . Sequence coordinates and alignment are based on the MAFFT-aligned DNA sequences in Figure 5—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 26057 . 02210 . 7554/eLife . 26057 . 023Figure 5—figure supplement 4 . The predicted protein products of 6-exon-containing wtf genes in the reference genome and the CBS5557 genome . Alignment is based on the MAFFT-aligned protein sequences in Figure 5—source data 3 . Protein products of wtf8 , wtf22 , wtf24 , and cw16 , which belong to the 6-exon group but appear to have suffered pseudogenizing mutations , are not shown . Green brackets denote the eight transmembrane helices predicted by PolyPhobius ( http://phobius . sbc . su . se/poly . html ) ( Käll et al . , 2005 ) . Helices 3 and 8 are indicated with light green brackets , because the former is absent in some of the aligned sequences , and the latter has low residue-wise posterior probabilities in the PolyPhobius prediction . DOI: http://dx . doi . org/10 . 7554/eLife . 26057 . 02310 . 7554/eLife . 26057 . 024Figure 5—figure supplement 5 . The predicted protein products of 5-exon-containing wtf genes in the reference genome and the CBS5557 genome . Alignment is based on the MAFFT-aligned protein sequences in Figure 5—source data 3 . Protein products of the eight most divergent genes ( wtf7 , wtf11 , wtf14 , wtf15 , cw7 , cw11 , cw14 , and cw15 ) and eight genes that appear to have suffered pseudogenizing mutations ( wtf1 , wtf2 , wtf3 , wtf6 , wtf12 , wtf17 , cw3 , and cw12 ) are not shown . Green brackets denote the five transmembrane helices predicted by PolyPhobius . Helix three is indicated with a light green bracket , because it is absent in some of the aligned sequences . DOI: http://dx . doi . org/10 . 7554/eLife . 26057 . 024 Among the 12 CBS5557 singleton wtf genes with syntenic counterparts in the reference genome , cw9 and five others exhibit exceptionally high DNA sequence diversity ( <85% identity ) from their counterparts in the reference genome ( Figure 1D , Figure 1—figure supplement 2 , and Figure 5—figure supplement 2 ) . Given that the genome-wide nucleotide difference between CBS5557 and the laboratory strain is only 3 . 1 SNPs/kb ( Hu et al . , 2015 ) , wtf genes must have diverged at a much faster pace than average genes . To examine the phylogenetic relationship among all 57 wtf genes of the two genomes , we performed a maximum likelihood analysis ( Figure 5C ) . In the resulting phylogenetic tree , we used colored rectangles to highlight 16 pairs of genes , with each pair consisting of phylogenetic neighbors , defined as two genes separated by a single internal node that has an ultrafast bootstrap support value >= 95% ( Figure 5C ) . Interestingly , only 11 of these 16 pairs conform to typical orthologous relationship where two members of a pair are syntenic ( highlighted by green rectangles in Figure 5C ) , whereas the five other pairs each consist of two genes that are non-syntenic ( highlighted by yellow rectangles in Figure 5C ) . This pattern suggests that ectopic gene conversion ( also called non-allelic or interlocus gene conversion ) may contribute to the inter-isolate difference of wtf genes ( Chen et al . , 2007; Petes and Hill , 1988 ) . Interestingly , genes belonging to the same gene cluster ( pair or triplet ) , despite being subject more strongly to the homogenizing effect of gene conversion due to their physical proximity , are without exception very distinct from each other . One possible explanation is that diversifying selection may have prevented sequence homogenization within gene clusters . The substantial difference in the numbers and sequences of wtf gene between the laboratory strain and CBS5557 suggests that rapid sequence divergence and independent loss/gain of wtf genes have resulted in different S . pombe natural isolates harboring distinct sets of active spore killers , which become reproductive barriers during inter-isolate crosses and contribute to the extensive and severe hybrid sterility within this species .
Only a small number of gamete killers have been characterized at the molecular level . Two well-studied gamete killers in animals , Segregation Distorter in Drosophila and the t haplotype in mouse , both contain at least two key loci , a killer locus and a target locus ( Lyon , 2003; Burt and Trivers , 2006; Larracuente and Presgraves , 2012; Lindholm et al . , 2016; Bauer et al . , 2012 ) . In fungi , spore killers of Neurospora also appear to comprise of two genes , rsk ( resistant to Spore killer ) and a yet-to-be-identified killer gene , which are proposed to act together in a fashion similar to the toxin-antitoxin ( TA ) systems of bacterial plasmids , in which a stable toxin and a labile antitoxin together ensure that bacterial cells cannot lose the plasmid expressing them ( Hammond et al . , 2012; McLaughlin and Malik , 2017; Saupe , 2012 ) . The involvement of more than one gene renders these gamete killer systems vulnerable to recombination that may separate the component genes , and as a consequence , they have all evolved recombination-suppressing features such as chromosomal inversions ( Burt and Trivers , 2006; Harvey et al . , 2014 ) . On the other hand , two types of single-gene spore killers , [Het-s] prion and the Spok genes , have been found in the filamentous fungus Podospora anserina ( Dalstra et al . , 2003; Grognet et al . , 2014 ) . The killing activity of the [Het-s] prion requires the presence of the HET-S protein , whereas the Spok genes can act autonomously . Fission yeast wtf killers share two prominent features with the Spok killers: being composed of a single predicted coding sequence and acting in a self-sufficient manner . These features arguably make wtf and spok killers the most compact and efficient forms of gamete killers . Despite the similarities they share , wtf and spok killers are very different entities . Firstly , wtf genes only exist in a single fission yeast species S . pombe , indicating that they probably have arisen after the divergence of S . pombe from its closest relatives S . octosporus and S . cryophilus around 120 million years ago ( Rhind et al . , 2011 ) , and have not jumped across the species barriers . In contrast , Spok family genes have a patchy but widespread distribution among Pezizomycotina species ( Grognet et al . , 2014 ) , whose last common ancestor dates to 400 million years ago ( Prieto and Wedin , 2013 ) , suggesting that Spok killers have originated much earlier than wtf killers or have been especially prone to transfer horizontally across species . Secondly , spok family genes encode soluble proteins with a C-terminal kinase-like domain ( InterPro:IPR011009 ) , whereas wtf genes are predicted to encode multi-transmembrane proteins , suggesting that the detailed molecular mechanisms of these two types of killers are likely to be quite distinct . The single-gene nature of wtf killers begs the question: how does a single gene encode both the killing and the protecting activities ? Our findings on the mutations that can disrupt one but not the other activity suggest the possibility that more than one type of expression products can be generated from a wtf killer gene . Consistent with this idea , Nuckolls et al . showed that the wtf4 gene from the S . pombe var . kambucha strain ( Sk wtf4 ) acts as a spore killer by expressing two overlapping transcripts , with the shorter transcript encoding a spore-killing poison and the longer transcript encoding an antidote that protects the spores from the killing effect of the poison ( Nuckolls et al . , 2017 ) . This dual-transcript model can satisfactorily explain why the Tb and Te forms of cw9 and cw27 lost protecting activities but not killing activities ( Figure 4A ) . Presumably , the transcription of the longer isoforms , which encode antidote proteins initiated from a start codon in the predicted exon 1 , partially relies on the sequence between the upstream LTR and the conserved_up sequence and absolutely requires the conserved_up sequence; in contrast , the transcription of the shorter isoforms , which encode poison proteins initiated from a downstream alternative start codon , is not affected by removing the upstream sequences . We hypothesized that the transcription of the poison isoforms may depend on a special promoter residing within the predicted intron 1 of cw9 and cw27 . Indeed , through inspecting the DNA sequence alignment ( Figure 5—source data 1 ) , we found that cw9 , cw27 , and 10 other wtf genes in the reference genome and the CBS5557 genome share a 150 bp conserved sequence in the predicted intron 1 , and this sequence is closely upstream of the transcription start site of a shorter-than-predicted wtf23 transcript isoform uncovered by the Iso-Seq long-read RNA sequencing method ( Figure 5—figure supplement 3 ) ( Kuang et al . , 2017 ) . These 12 genes can be classified into two subtypes , which we call Intron-1-ATG genes ( wtf4 , wtf13 , cw4 , cw18 , and cw23 ) and Exon-2-ATG genes ( wtf19 , wtf23 , cw8a , cw9 , cw21 , cw22 , and cw27 ) , based on the locations of the first ATG codons downstream of the 150 bp sequence ( Figure 5C and Figure 5—figure supplement 3 ) . These ATG codons are all in frame with the predicted coding sequences . The within-intron-1 ATG codons in the 5 Intron-1-ATG genes are only 7 bp away from the predicted exon 2 , and if used as start codons , would result in proteins beginning with three intron-1-coded amino acids ( Met , Leu , and Ser ) . The two active spore killer genes found in the S . pombe var . kambucha strain by Nuckolls et al . , Sk wtf4 and Sk wtf28 , also possess this 150 bp sequence and can be classified as an Intron-1-ATG gene and an Exon-2-ATG gene , respectively . We propose that this 150 bp sequence is a hallmark of the ‘poison-and-antidote’ genes that can , at least potentially , generate both poison and antidote isoforms according to the dual-transcript model proposed by Nuckolls et al . The other 45 wtf genes in the reference genome and the CBS5557 genome lack this 150 bp sequence . In addition , except for the eight most divergent genes ( wtf7 , cw7 , wtf11 , cw11 , wtf14 , cw14 , wtf15 , and cw15 ) , they do not have an in-frame ATG codon in intron one or in the 5’-section of exon 2 , suggesting that most of these genes are not able to generate a shorter isoform according to the dual-transcript model , and are thus ‘antidote-only’ genes . Alternatively , it is possible that they may generate two different protein products using other mechanisms , such as alternative splicing ( Kuang et al . , 2017 ) , overlapping proteins translated in different frames ( Duncan and Mata , 2014 ) , and post-translational processing and modification . Five wtf genes in the reference genome , wtf1 , wtf2 , wtf3 , wtf22 , and wtf24 , are annotated by PomBase as pseudogenes ( McDowall et al . , 2015 ) . Our analysis of the gene structures of the 57 wtf genes in the reference genome and the CBS5557 genome showed that in addition to these five genes , four other wtf genes in the reference genome ( wtf6 , wtf8 , wtf12 , and wtf17 ) , and three wtf genes in the CBS5557 genome ( cw3 , cw12 , and cw16 ) , appear to have suffered pseudogenizing mutations ( see Materials and methods and Figure 5—source data 2 ) , which include loss of start codons ( wtf3 and cw3 ) , premature stop codons ( wtf1 , wtf12 , cw12 , and cw16 ) , frameshifting indels ( wtf6 , wtf8 , wtf22 , and wtf24 ) , and segmental deletions ( wtf2 and wtf17 ) . These ‘dead’ wtf genes presumably are not able to generate either antidote or poison proteins . Most of the wtf genes in the reference genome and the CBS5557 genome are predicted to contain either 5 or 6 exons . Interestingly , the 6-exon genes , including cw9 , cw27 , and all but one of the other ten 150-bp-sequence-containing genes , clustered together in the DNA sequence-based phylogenetic tree ( Figure 5C ) , and their protein products are predicted to be longer and contain two or three more transmembrane helices than the protein products of 5-exon genes ( Figure 5—source data 3 , Figure 5—figure supplement 4 , and Figure 5—figure supplement 5 ) . For each of the 12 150-bp-sequence-containing genes , the antidote protein product is approximately 50 amino acids longer than the poison protein product at the N-terminal soluble region . The 10 C-terminal amino acids missing from the protein products of the Td forms of cw9 and cw27 are downstream of the last transmembrane helix , and are strongly conserved among the predicted protein products of 6-exon genes ( Figure 5—figure supplement 4 ) , indicating that Td-like mutations do not occur naturally to a significant extent . How do transmenbrane proteins act as killers ? In the prokaryotic toxin-antitoxin ( TA ) systems , many toxins are transmembrane proteins and kill cells by damaging cellular membranes ( Schuster and Bertram , 2013 ) . The best-known examples include the type I TA toxins Hok and TisB , and the type V TA toxin GhoT ( Cheng et al . , 2014; Gerdes et al . , 1986; Unoson and Wagner , 2008 ) . We speculate that Wtf proteins may also possess abilities to disrupt the integrity of certain cellular membranes . wtf genes were named after their association with the solo LTRs of the Tf retrotransposons ( Wood et al . , 2002 ) . It was initially hypothesized that retrotransposition may aid the expansion of the wtf gene family ( Wood et al . , 2002 ) . However , the presence of introns in wtf genes and the lack of sequence similarity between wtf-flanking LTRs indicate that wtf genes have not been retrotransposed ( Bowen et al . , 2003 ) . Bowen et al . have put forward an explanation for the enrichment of LTRs near wtf genes: the regions flanking wtf genes may have been favored insertion targets of Tf retrotransposons , perhaps owing to the high transcription levels of wtf genes during meiosis and sporulation ( Bowen et al . , 2003 ) . Our findings that LTRs flanking cw9 and cw27 are dispensable for spore killing support the idea that the wtf-LTR association does not benefit wtf genes but may rather reflect an exploitation of wtf genes by Tf retrotransposons for their own expression and/or inheritance benefit . Furthermore , we show that three singleton wtf genes ( cw26 , cw27 , and cw28 ) appear to have been lost in the laboratory strain due to LTR-mediated recombination , suggesting that the wtf-LTR association promotes wtf gene loss and is detrimental to the long-term persistence of wtf genes . In addition , it may not be coincidental that both cw9 and cw27 require an approximately 90-bp-long sequence between the upstream LTR and the conserved_up sequence for the full protecting activity , raising the possibility that some wtf genes that lack such an intervening sequence ( e . g . wtf4 and cw4 ) may have suffered functional loss due to a Tf insertion too close to the conserved_up sequence or may have become functionally dependent on the upstream LTR . In terms of the number of genes per genome , wtf genes are probably the most evolutionarily successful gamete killers known to date . During a relatively short evolutionary time span , wtf gene family has emerged and become the largest species-specific gene family in S . pombe ( Lespinet et al . , 2002 ) . The compactness ( minimal sizes of fully functional cw9 and cw27 are both around 1 . 9 kb ) and the self-sufficiency of the wtf killers likely have been advantageous to its family size expansion . Furthermore , given the dramatic divergence of wtf genes between the reference genome and the CBS5557 genome , and the mutual-killing behaviors of cw9 and cw27 , it is possible that two identical copies of a wtf gene generated by a gene duplication event can undergo relatively rapid sequence diversification to become mutually non-resistant , so that their selective advantage is not diminished by resistance . Ectopic gene conversion is known to play a causal role in the sequence diversification of animal major histocompatibility complex ( MHC ) genes ( Ohta , 1991; Takuno et al . , 2008 ) , plant disease-resistance genes ( R genes ) ( Mondragon-Palomino and Gaut , 2005 ) , and Plasmodium falciparum var genes ( Claessens et al . , 2014 ) . Our phylogenetic analysis suggests that ectopic gene conversion may also contribute to the historical expansion and the ongoing dynamic change of the wtf gene family . It is intriguing that most of the wtf genes locate on chromosome III , the smallest of the three fission yeast chromosomes . To explain this phenomenon , Bowen et al . proposed that the laboratory strain may have inherited only chromosome III but not the other two chromosomes from an ancestor strain in which wtf genes were abundant on all three chromosomes ( Bowen et al . , 2003 ) . Our finding here of the same kind of chromosomal distribution bias in CBS5557 , whose nucleotide diversity from the laboratory strain is higher on chromosome III than on the other two chromosomes ( Hu et al . , 2015 ) , makes this explanation less likely . The only type of viable aneuploidy in fission yeast is chromosome III disomy ( Niwa et al . , 2006; Niwa and Yanagida , 1985 ) , which occurs at a frequency of >1% during normal meiosis ( Molnar et al . , 1995 ) . We speculate that the tolerance of chromosome III disomy may in some way be advantageous to the expansion of the wtf gene family on this chromosome . Spore killers and other meiotic drivers , despite having only been found in a small number of taxa , are believed to exist in a wide range of species , where they are an important force influencing genome organization , gametogenesis , and speciation ( Burt and Trivers , 2006; Lindholm et al . , 2016; Werren , 2011 ) . The molecular basis of their action and the evolutionary mechanisms behind their birth , spread , and extinction are scarcely known . The unveiling of the molecular identities of spore killers in S . pombe by this study and that of Nuckolls et al . ( Nuckolls et al . , 2017 ) , promises to bring the power of a highly tractable model organism to the pursuit of the many mysteries of these evolutionary wonders .
Fission yeast strains used in this study are listed in Supplementary file 2 in Microsoft Word format . Genetic methods and composition of media are as described ( Forsburg and Rhind , 2006 ) . Gene deletion was constructed by PCR-based gene targeting . Integrating plasmids were based on pDUAL and related vectors ( Matsuyama et al . , 2004 , Matsuyama et al . , 2008 ) . A pDUAL-based plasmid was linearized with NotI digestion and integrated at the leu1 locus , or linearized with MluI digestion and integrated at the ars1 replication origin region upstream of the hus5 gene . To convert homothallic strains to heterothallic and to mark the mating type locus , we created the mat1Δ17 mutation ( Arcangioli and Klar , 1991 ) , by replacing a 140 bp sequence between the H1 homology box at the mat1 locus and the nearby SspI restriction site with an antibiotic resistance marker using PCR-based gene targeting . h+/h- diploid strains were constructed by mating on SPAS medium h+ and h- strains with different antibiotic resistance markers inserted at the mat1 locus , and a few hours later restreaking the mating mixture onto YES medium containing two types of antibiotics . To construct strains with only a portion of chromosome III originating from CBS5557 and the rest of the genome coming from the laboratory strain , we first introduced rec12∆ mutation , which blocks meiotic recombination , into the laboratory strain and CBS5557 , respectively . Through crossing the two resultant strains with each other , we obtained a strain whose chromosome I and II are from the laboratory strain and whose chromosome III is from CBS5557 . This strain was then crossed to a rec12+ laboratory-background strain carrying an antibiotic resistance marker at a locus on the left arm of chromosome III ( mug123:hphMX ) and another antibiotic resistance marker at a locus on the right arm of chromosome III ( rps2802:natMX ) ( first backcross ) . Progeny strains with only one of the loci marked were kept . They were then crossed to laboratory-background strains with the other locus marked by kanMX ( second backcross ) . By scoring the marker segregation ratios of the second backcross , we selected those first-backcross progenies that conferred strong segregation distortion at the kanMX-marked locus . Their second-backcross progenies that harbored the kanMX marker were tested by crossing , and those that conferred segregation distortion at the kanMX-marked locus were selected . One of the second-backcross progenies that conferred segregation distortion at the mug123 locus is the backcrossed-1 strain used in the bulk segregant analysis shown in Figure 1B . Two chromosome III genomic regions from CBS5557 , one between coordinates 247743 and 568805 , and the other between coordinates 814348 and 1711635 , are present in the backcrossed-1 strain . Another second-backcross progeny that conferred segregation distortion at the rps2802 locus underwent additional rounds of backcross to yield the backcrossed-2 strain used in the bulk segregant analysis shown in Figure 1C . The CBS5557 chromosome III genomic region present in the backcrossed-2 strain is between coordinates 1839246 and 1916857 . For Illumina sequencing-assisted bulk segregant analysis , >150 viable progeny colonies derived from random spore analysis were pooled together . Genomic DNA extraction , sequencing libraries construction , and Illumina sequencing were performed as described ( Hu et al . , 2015 ) . Sequencing data were deposited at NCBI SRA under the following accession numbers: viable progenies from CBS5557 × laboratory strain ( DY9974 × DY8531 ) , SRR5131575; viable progenies from backcrossed-1 × laboratory strain ( cross-derived diploid strain was named DY26097 ) , SRR5131578; viable progenies from backcrossed-1 cw9∆ × laboratory strain ( cross-derived diploid strain was named DY26100 ) , SRR5131579; viable progenies from backcrossed-2 strain × laboratory strain ( cross-derived diploid strain was named DY26095 ) , SRR5131577; viable progenies from backcrossed-2 cw27∆ strain × laboratory strain ( cross-derived diploid strain was named DY26092 ) , SRR5131576 . Sequencing reads were mapped to the reference genome using BWA-MEM ( version 0 . 7 . 15-r1140 , RRID:SCR_010910 ) . SAM files were converted to BAM files and duplicates were removed using SAMtools ( version 0 . 1 . 19–44428 cd , RRID:SCR_002105 ) . The average read depths were 14 . 9× for DY9974 × DY8531 , 10 . 4× for DY26097 , 18 . 4× for DY26100 , 16 . 2× for DY26095 , and 23 . 1× for DY26092 . The software bam-readcount version 0 . 7 . 4 was run with the option -q 60 -b 35 ( -q 60 -b 27 for DY26097 , which had lower depth coverage and lower sequencing quality ) to obtain the read counts of four different bases at each of the 38783 SNP positions identified previously ( Hu et al . , 2015 ) . Reference allele frequency at each SNP position was calculated by dividing the read count of the reference base by the sum of the read counts of the reference base and the variant base . Only SNP positions with the sum of the read counts of the reference base and the variant base >= 10 ( > = 7 for DY26097 ) were used in the plots . A trend line was drawn in each plot by calculating a rolling median of reference allele frequencies at 45 consecutive SNP positions using the rollapply function of R’s zoo package ( version 1 . 7–13 ) . For PacBio sequencing of CBS5557 genome , we prepared genomic DNA from DY9971 , an h- derivative of CBS5557 , by first grinding cells in liquid nitrogen and then extracting DNA using the Maxi Column Fungal DNAOUT kit ( Tiandz , Beijing , China ) . Sequencing library construction and sequencing on the PacBio RS II platform using P6C4 chemistry were performed by BGI ( Shenzhen , China ) . After filtering out low-quality reads and removing adaptor sequences , we obtained 91 , 333 subreads no shorter than 1000 bp ( mean length 7370 bp ) . These reads were used for de novo assembly by the SMRT Analysis software . The assembly was polished using Pilon version 1 . 21 ( RRID:SCR_014731 ) together with our previously published CBS5557 Illumina sequencing data ( Hu et al . , 2015; Walker et al . , 2014 ) . PacBio sequencing data were deposited at NCBI SRA under the accession number SRR5133273 . For the induction of synchronous meiosis and sporulation , h+/h- diploid strains were pre-grown in liquid YES medium to log phase , cultured in SSL+N synthetic liquid medium for 10 hr , and then shifted to nitrogen-free SSL-N medium ( Egel and Egel-Mitani , 1974 ) . Four-spored asci represented about half of the cells 10 hr after shifting to SSL-N ( Olson et al . , 1978 ) . Electron microscopy was performed as described ( Sun et al . , 2013 ) . Spore viability was assessed by tetrad analysis using a TDM50 tetrad dissection microscope ( Micro Video Instruments , Avon , USA ) . At least 40 tetrads ( 160 spores ) were analyzed for each cross . Based on empirical evidence , such a sample size is large enough for reliably assessing spore viability . Numerical data of the tetrad analysis in Excel format are provided as Supplementary file 1 . For statistical analysis of the spore viability data , Fisher’s exact test and exact binomial test of goodness-of-fit were performed using Excel spreadsheets downloaded from http://www . biostathandbook . com/fishers . html and http://www . biostathandbook . com/exactgof . html , respectively ( McDonald , 2014 ) . Except for the phylogenetic analysis , DNA sequences were aligned using MAFFT via Jalview ( Katoh et al . , 2002; Waterhouse et al . , 2009 ) . Pair-wise identity was calculated by dividing the numbers of identical bases by the alignment length using the Sequence Manipulation Suite web server ( http://www . bioinformatics . org/sms2/ident_sim . html ) ( Stothard , 2000 ) . For the phylogenetic analysis shown in Figure 5C , an alignment of DNA sequences containing the conserved_up regions , the coding sequences , and the associated introns was generated using the L-INS-i algorithm of MAFFT version 7 . 310 for Mac OS X ( RRID:SCR_011811 ) with the command line option --localpair --maxiterate 16 --reorder ( Katoh and Standley , 2014 ) , and manually adjusted to correct an obvious misalignment of the wtf2 sequence . The final alignment of 57 wtf genes of the reference and CBS5557 genome in FASTA format is provided as Figure 5—source data 1 . Maximum likelihood analysis was performed using IQ-TREE version 1 . 5 . 3 for Mac OS X with the command line option -m TEST -alrt 1000 -bb 1000 ( Nguyen et al . , 2015 ) . The tree was rooted by midpoint rooting using FigTree version 1 . 4 . 2 ( http://tree . bio . ed . ac . uk/software/figtree/ ) and was visualized using Phylo . io ( Robinson et al . , 2016 ) . Gene structures of the 57 wtf genes were predicted using the AUGUSTUS web server ( version 3 . 2 . 3 , RRID:SCR_008417 ) with the parameter settings --genemodel=exactlyone --sample=100 --keep_viterbi=true --alternatives-from-sampling=true --minexonintronprob=0 . 08 --minmeanexonintronprob=0 . 3 --maxtracks=20 ( Stanke et al . , 2008 ) . When multiple predications were made for a gene , we manually inspected the predictions and selected the one that best conforms to conserved patterns of intron positions . For the majority of genes , the selected prediction is the top-scoring prediction , and for 10 genes ( wtf6 , wtf7 , wtf14 , wtf24 , cw6a , cw7 , cw12 , cw14 , cw16 , cw25 ) , the selected prediction is the one with the second best score . The final predictions for all 57 genes in plain text format are provided as Figure 5—source data 2 , which can be loaded into Jalview as an annotation file for the DNA sequence alignment in Figure 5—source data 1 . The failure to predict an exon flanked by conserved intron-exon junction sequences is an indication that pseudogenizing mutation ( s ) exist . Among the predicted protein sequences obtained based on these gene structure predictions , the sequences of Wtf4 , Wtf6 , Wtf7 , Wtf12 , Wtf13 , Wtf15 , Wtf17 , and Wtf18 are different from those annotated at PomBase . A MAFFT ( L-INS-i algorithm ) -generated alignment of the 57 predicted protein sequences in FASTA format is provided as Figure 5—source data 3 . Illumina and PacBio sequencing data reported in this paper are available at the Sequence Read Archive ( SRA ) under the accession number SRP095878 ( BioProject PRJNA358837 ) ( Hu et al . , 2017 ) . The DNA sequences and annotations of the 32 wtf genes of the CBS5557 genome have been deposited at GenBank under the accession numbers KY926712-KY926743 . | During evolution , new species emerge when individuals from different populations of similar organisms no longer breed with each other , or when the offspring produced if they do breed are sterile . This process is known as “reproductive isolation” and , for over 100 years , evolutionary biologists have tried to better understand how this process happens . Animals , plants and fungi produce sex cells – known as gametes – when they are preparing to reproduce . These cells are made when cells containing two copies of every gene in the organism divide to produce new cells that each only have one copy of each gene . Therefore , a particular gene copy usually has a 50% chance of being carried by an individual gamete . There are genes that selfishly increase their chances of being transmitted to the next generation by destroying the gametes that do not carry them . These “gamete killer” genes reduce the fertility of the organism and lead to reproductive isolation . Fission yeast is a fungus that is widely used in research . There are different strains of fission yeast that are reproductively isolated from each other , but it is not known whether gamete killers are responsible for this isolation . To address this question , Hu et al . investigated the causes of reproductive isolation in fission yeast . The experiments identified two gamete killers , referred to as cw9 and cw27 . Both genes belong to the wtf gene family . Each gene is believed to encode two different proteins , one that acts as a poison and one that acts as an antidote . The poison is capable of killing all gametes , but the antidote protects the cells that contain the gamete killer gene . Further experiments show that the antidote produced by one of the gamete killer genes cannot protect cells against the poison produced by the other gene . A separate study by Nuckolls et al . found that another member of the wtf family also acts as a gamete killer in fission yeast . Together , these findings shed new light on the causes of reproductive isolation , and will contribute to deeper understanding of speciation and evolution in general . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"genetics",
"and",
"genomics"
] | 2017 | A large gene family in fission yeast encodes spore killers that subvert Mendel’s law |
Optogenetics enables genome manipulations with high spatiotemporal resolution , opening exciting possibilities for fundamental and applied biological research . Here , we report the development of LiCre , a novel light-inducible Cre recombinase . LiCre is made of a single flavin-containing protein comprising the AsLOV2 photoreceptor domain of Avena sativa fused to a Cre variant carrying destabilizing mutations in its N-terminal and C-terminal domains . LiCre can be activated within minutes of illumination with blue light without the need of additional chemicals . When compared to existing photoactivatable Cre recombinases based on two split units , LiCre displayed faster and stronger activation by light as well as a lower residual activity in the dark . LiCre was efficient both in yeast , where it allowed us to control the production of β-carotene with light , and human cells . Given its simplicity and performances , LiCre is particularly suited for fundamental and biomedical research , as well as for controlling industrial bioprocesses .
The wealth of knowledge currently available on the molecular regulations of living systems – including humans – largely results from our ability to introduce genetic changes in model organisms . Such manipulations have been extremely informative because they can unambiguously demonstrate causal effects of molecules on phenotypes . The vast majority of these manipulations were made by first establishing a mutant individual – or line of individuals – and then studying it . This classic approach has two limitations . First , the mutation is present in all cells of the individual . This complicates the analysis of the contribution of specific cells or cell types to the phenotypic alterations that are observed at the whole-organism level . Second , when a mutation is introduced long before the phenotypic analysis , it is possible that the organism has ‘adapted’ to it either via compensatory regulations or , in case of mutant lines maintained over multiple generations , by compensatory mutations . For these reasons , other approaches relying on site-specific recombinases were developed to introduce specific mutations in a restricted number of cells of the organism and at a specific time . For instance , the Cre/LoxP system ( Van Duyne , 2015; Rajewsky et al . , 1996 ) consists of two manipulations: a stable insertion , in all cells , of foreign 34 bp DNA sequences called LoxP and the expression of the Cre recombinase in some cells only , where it modifies the DNA by catalyzing recombination between the LoxP sites . The result is a mosaic animal – or plant , or colony of cells – where chromosomal DNA has been rearranged in some cells only . Cre is usually introduced via a transgene that is only expressed in the cells to be mutated . The location and orientation of LoxP sites can be chosen so that recombination generates either a deletion , an inversion , or a translocation . Similar systems were developed based on other recombinases/recognition targets , such as Flp/FRT ( Lee and Luo , 2001 ) or Dre-rox ( Anastassiadis et al . , 2009 ) . To control the timing of recombination , several systems were made inducible . Tight control was obtained using recombinases that are inactive unless a chemical ligand is provided to the cells . For example , the widely used Cre-ERT chimeric protein can be activated by 4-hydroxy-tamoxifen ( Feil et al . , 1996 ) . Other inducible systems rely on chemical-induced dimerization of two halves of the recombinase . For instance , the FKBP–FRB split Cre system consists of two inactive proteins that can assemble in the presence of rapamycin to form a functional recombinase complex ( Jullien et al . , 2003 ) . Similar systems were reported that rendered dimerization of the split Cre fragments dependent on phytohormones ( Weinberg et al . , 2019 ) . Although powerful , these systems present some caveats: ligands are not always neutral to cells and can therefore perturb the biological process under investigation; since they diffuse in tissues , the control of activation is sometimes not precise enough in space and/or time; and the cost or side effects of chemical inducers can be prohibitive for industrial or biomedical applications . More recently , several authors modified these dimerizing split recombinases to make them inducible by light instead of chemicals . This presents several advantages because ( i ) light can be used with extreme spatiotemporal precision and high reproducibility; ( ii ) when applied at low energy , it is neutral to many cell types; and ( iii ) it is very cheap and therefore scalable to industrial processes . The dimerization systems that were used come from developments made in optogenetics , where various light , oxygen , or voltage ( LOV ) protein domains have been used as photosensory modules to control transcription ( de Mena et al . , 2018 ) , protein degradation ( Renicke et al . , 2013 ) , dimerization ( Kennedy et al . , 2010; Strickland et al . , 2012; Nihongaki et al . , 2014 ) , or subcellular relocalization ( Niopek et al . , 2014; Witte et al . , 2017 ) . LOV domains belong to the Per-Arnt-Sim ( PAS ) superfamily found in many sensors . They respond to light via a flavin cofactor located at their center . In the A . sativa phototropin 1 LOV2 ( AsLOV2 ) domain , blue light generates a covalent bond between a carbon atom of a flavin mononucleotide ( FMN ) cofactor and a cystein side chain of the PAS fold ( Crosson and Moffat , 2001; Swartz et al . , 2001 ) , resulting in a conformational change including the unfolding of a large C-terminal α-helical region called the Jα helix ( Swartz et al . , 2002; Harper et al . , 2003 ) . Diverse optogenetics tools have been developed by fusing LOV domains to functional proteins in ways that made the Jα folding/unfolding critical for activity ( Pudasaini et al . , 2015 ) . Among these tools are several photodimerizers that proved useful to control the activity of recombinases . One study reported blue-light-dependent heterodimerization of a split Cre recombinase using the CIB1-CRY2 dimerizers from the plant Arabidopsis thaliana ( Taslimi et al . , 2016 ) and others successfully used the nMag/pMag dimerizers derived from Vivid ( VVD ) , a protein of the fungus Neurospora crassa ( Kawano et al . , 2016; Sheets et al . , 2020 ) . A third system was based on dimerizers derived from the chromophore-binding photoreceptor phytochrome B ( PhyB ) of A . thaliana and its interacting factor PIF3 . In this case , red light was used for stimulation instead of blue light , but the system required the addition of an expensive chemical , the chromophore phycocyanobilin ( Hochrein et al . , 2018 ) . An ideal inducible recombinase is one that ensures both low basal activity and high induced activity , that is simple to implement , cheap to use , and fast to induce . All dimerizing split Cre systems have in common that two protein units must be assembled in order to form one functional Cre . Thus , the probability of forming a functional recombination synapse , which normally requires four Cre molecules , is proportional to the product of the two units' cellular concentrations to the power of four . Split systems therefore strongly depend on the efficient expression of their two different coding sequences , as previously reported ( Meador et al . , 2019 ) . An inducible system based on a single protein may avoid this limitation . Its implementation by transgenesis would also be simpler , especially in vertebrates . We report here the development of light-inducible Cre ( LiCre ) , a novel light-inducible Cre recombinase that is made of a single flavin-containing protein . LiCre can be activated within minutes of illumination with blue light , without the need of additional chemicals , and it shows extremely low-background activity in the absence of stimulation as well as high induced activity . Using the production of carotenoids by yeast as a case example , we show that LiCre and blue light can be combined to control metabolic switches that are relevant to the problem of metabolic burden in bioprocesses . We also report that LiCre can be used efficiently in human cells , making it suitable for biomedical research . Since LiCre offers cheap and precise spatiotemporal control of a genetic switch , it is amenable to numerous biotechnological applications , even at industrial scales .
A variety of optogenetic tools have been successfully developed based on specific LOV domain proteins possessing α-helices that change conformation in response to light ( Weitzman and Hahn , 2014 ) . We reasoned that fusing a LOV domain to a helical domain of Cre that is critical for its function could generate a single protein with light-dependent recombinase activity . We searched for candidate α-helices by inspecting the structure of the four Cre units complexed with two LoxP DNA targets ( Ennifar et al . , 2003; Guo et al . , 1997; Figure 1a , b ) . Each subunit folds in two domains that bind to DNA as a clamp . It was initially reported that helices αA and αE of the amino-terminal domain , as well as helix αN of the C-terminal domain , participate to inter-units contacts ( Guo et al . , 1997 ) ; and this role of helix αN was later confirmed ( Ennifar et al . , 2003 ) . Contacts between αA and αE associate all four amino-terminal domains ( Figure 1a ) , and contacts involving αN lock the four carboxy-terminal domains in a cyclic manner ( Figure 1b ) . These helices were therefore good candidates for manipulating Cre activity . We focused on αA and αN because their location at protein extremities was convenient to design chimeric fusions . We tested the functional importance of helices αA and αN by gradually eroding them . We evaluated the corresponding mutants by expressing them in yeast cells , where an active Cre can excise a repressive DNA element flanked by LoxP sites and thereby switch ON the expression of a green fluorescent protein ( GFP ) ( Figure 1c ) . After inducing the expression of Cre mutants with galactose , we counted by flow cytometry the proportion of cells that expressed GFP and used this measure to compare recombinase activities of the different mutants ( Figure 1d ) . As a control , we observed that the wild-type Cre protein activated GFP expression in all cells under these conditions . Mutants lacking the last two or the last three carboxy-terminal residues displayed full activity . In contrast , mutants lacking four or more of the C-ter residues were totally inactive . This was consistent with a previous observation that deletion of the last 12 residues completely suppressed activity ( Rongrong et al . , 2005 ) . Our series of mutants showed that helix αN is needed for activity and that its residue E340 ( located at position −4 from the protein end ) is implicated . The role of this glutamic acid is most likely to stabilize the complex: the tetramer structure indicates that it interacts with residue R192 of the adjacent unit ( Figure 1e ) . Interestingly , although D341 was not essential for activity , the salt bridges of this residue with R139 probably also reinforce attraction between the two protein units . Biomolecular simulations using a simplistic force-field model showed that the free-energy barrier for displacing the αN helix was much lower if E340 and D341 were replaced by alanines ( Figure 1f ) . Consistent with this prediction , we observed that a double mutant E340A D341A lost ~10% of activity ( Figure 1g ) . This mild ( but reproducible ) reduction of activity suggested that the double mutation E340A D341A led to a variant of Cre where multimerization was suboptimal . We also tested the functional importance of α-helix A either in a normal context where the C-terminal part of Cre was intact or where it carried the destabilizing E340A D341A mutation ( Figure 1g ) . Deletion of residues 2–37 , which entirely ablated helix A , eliminated enzymatic activity ( Figure 1g ) . Very interestingly , the effect of shorter deletions depended on the C-terminal context . When the C-terminus was wild-type , removing residues 2–21 ( immediately upstream of helix A ) had no effect and removing residues 2–28 ( partial truncation of αA ) decreased the activity by ~10% . When the C-terminus contained the E340A D341A mutation , deletions 2–21 and 2–28 were much more severe , reducing the activity by 12% and 80% , respectively . This revealed genetic interactions between the extremities of the protein , which is fully consistent with a cooperative role of helices αA and αN in stabilizing an active tetramer complex . From these observations , we considered that photo-control of Cre activity might be possible by fusing αA and αN helices to LOV domain photoreceptors . We obtained successful fusions ( Figure 2 ) after screening several types of candidate constructs . Our first strategy was to fuse the αN carboxy-terminal helix of Cre to the amino-terminal cap of the LOV domain of protein VVD , a well-characterized photosensor from N . crassa ( Nihongaki et al . , 2014; Zoltowski et al . , 2009; Vaidya et al . , 2011 ) . The resulting chimeric protein , which contained the full-length Cre connected to VVD via four amino acids , did not display light-dependent recombinase activity ( Figure 2—figure supplement 1 ) . Introducing the ( E340A D341A ) double mutation in this chimeric protein reduced the overall activity – as expected from above – but did not cause light dependency . Changing the size of the linker or introducing different mutations at positions 340 and 341 also failed to generate light-dependency ( Figure 2—figure supplement 1 ) . We noted , however , that the activity of mutants ( E340R D341A ) and ( E340R D341R ) was markedly reduced , probably because attraction between E340 and R192 was changed into repulsion between two positively charged arginines . Our next strategy was based on a modified version of the AsLOV2 domain from Avena sativa , which had been optimized and used to build an optogenetic dimerizer by fusing its Jα C-ter helix to the bacterial SsrA peptide ( Guntas et al . , 2015 ) . Instead , we fused Jα to the αA amino-terminal helix of Cre . Using the same GFP reporter system as described above for detecting in vivo recombination in yeast , we built a panel of constructs with various fusion positions and directly quantified their activity with and without blue-light illumination . All fusions displayed reduced activity in both dark and light conditions compared to wild-type Cre . Although variability was high between independent transformants , we identified four constructs – corresponding to fusions of AsLOV2 to residues 17 , 19 , 27 , and 32 of Cre , respectively – where the assay indicated a higher activity after light stimulation ( Figure 2—figure supplement 2 ) . To confirm this , we recovered the corresponding plasmids from individual yeast transformants , amplified them in bacteria to verify their sequence , and re-transformed them in yeast . This validated a differential activity between dark and light conditions for the constructs corresponding to fusion positions 19 , 27 , and 32 ( Figure 2a ) . Fusion at position 32 ( named LOV_Cre32 ) displayed the highest induction by light , with activity increasing from 15% in dark condition to 50% after 30 min of illumination . Although this induction was significant , a 15% activity of the non-induced form remained too high for most applications . We therefore sought to reduce this residual activity , which we did in two ways . First , we randomized the residues located at the junction between AsLOV2 and Cre . We used degenerate primers and in vivo recombination ( see Materials and methods ) to mutagenize LOV_Cre32 at these positions and directly tested the activity of about 90 random clones . Six of them showed evidence of low residual activity in the dark , and we characterized them further by sequencing and re-transformation . For one clone , photo-induction was not confirmed after re-transformation ( Figure 2b , clone ii ) . For the other five clones , residual activity was indeed reduced compared to LOV_Cre32 , with the strongest reduction being achieved by an isoleucine insertion at the junction position ( Figure 2b , clone v ) . However , this improvement was also accompanied by a weaker induced activity and a larger variability between independent assays . As a complementary approach to reduce residual activity , we took advantage of the above-described genetic interaction between N-ter truncations and C-ter mutations targeting residues 340 and 341 . We built another series of constructs where AsLOV2 fusions to αA helix were combined with the A340A341 double mutation . Results differed from those observed in the context of a wild-type αN ( Figure 2—figure supplement 3 ) , confirming genetic interaction between N-ter and C-ter mutations . In this assay , two constructs – corresponding to fusion positions 20 and 22 – showed higher recombination activity upon blue-light illumination . We extracted , sequence-verified , and re-tested them together with the one corresponding to fusion position 21 . This revealed that fusion at position 20 , but not 21 or 22 , conferred light-dependent activity ( Figure 2c ) . The construct corresponding to fusion of AsLOV2 at position 20 of the Cre A340A341 double mutant displayed a residual activity that was indistinguishable from the negative control and a highly reproducible induced activity of ~25% . Sequencing revealed that it also included a A->D mutation at the fusion junction ( Supplementary file 1 – Supplementary Text S2 ) . We called this construct LiCre and characterized it further . We placed LiCre under the expression of the PMET17 promoter and tested various lighting conditions . We first used the same illumination system as above , an LED box apparatus commercialized under the name ‘PAUL’ , which was originally designed for viability-PCR assay . This device is suitable for continuous illumination at room temperature . We applied varying intensities and durations on cells that were cultured to stationary phase in the absence of methionine ( full LiCre expression ) . Activity was very low without illumination and increased with both the intensity and duration of light stimulation . The minimal intensity required for stimulation comprised between 0 . 057 and 1 . 815 mW/cm2 . The highest activity ( ~65% of switched cells ) was obtained with 90 min illumination at 36 . 3 mW/cm2 ( Figure 3a ) . At comparable duration and intensity ( 30 min and 36 . 3 mW/cm2 ) , LiCre generated more recombination events in this experiment than in the previous one ( Figure 3a vs . Figure 2c ) . This improvement likely results from the different expression systems and culture media that were used ( induction of the PMET17vs . PGAL1 promoter ) . Extending illumination to 180 min did not further increase the fraction of switched cells . Remarkably , we observed that 2 min of illumination was enough to switch 5% of cells , and 5 min illumination generated 10% of switched cells ( Figure 3b ) . We compared these performances with those of two previous systems that were both based on light-dependent complementation of a split Cre enzyme . We constructed plasmids coding for proteins CreN59-nMag and pMag-CreC60 described earlier ( Kawano et al . , 2016 ) and transformed them in our yeast reporter strain . Similarly , we constructed and tested plasmids coding for the previously described ( Taslimi et al . , 2016 ) proteins CRY2L348F-CreN and CIB1-CreC . All four coding sequences were placed under the control of the yeast PMET17 promoter . We analyzed the resulting strains as above after adapting light to match the intensity recommended by the authors ( 1 . 815 mW/cm2 for nMag/pMag and 5 . 45 mW/cm2 for CRY2L348F/CIB1 ) . As shown in Figure 3c , we validated the photoactivation of nMag/pMag split Cre in yeast , where activity increased about fourfold following 90 min of illumination , but we were not able to observe photoactivation of the CRY2L348F/CIB1 split Cre system ( Figure 3d ) . In addition , the photoactivation of nMag/pMag split Cre was not as fast as the one of LiCre since 30 min of illumination was needed to observe a significant increase of activity . This observation is consistent with the fact that dimerization of split Cre , which is not required for LiCre , probably limits the rate of formation of an active recombination synapse . Another difference was that , unlike LiCre , nMag/pMag split Cre displayed a mild but significant background activity in the absence of illumination ( ~6% of switched cells ) ( Figure 3c ) . We then used another lighting device , a DMX-controlled LED spot , which provided more precise control on both the intensity and dynamics of illumination . We applied varying intensities on reporter cells expressing either LiCre , nMag/pMag split Cre , or CRY2L348F/CIB1 split Cre ( Figure 3e ) . Absolute efficiencies were weaker than those observed on the PAUL device . This difference is likely due to two factors: the presence of mirrors in the PAUL box ( better distribution of light to the cells ) and a difference of temperature , which was controlled at 30°C under the DMX spot but not in the PAUL box . Nonetheless , the relative efficiencies were very concordant between the two devices . The DMX spot stimulated an activity of nMag/pMag split Cre that was above background but much weaker than for LiCre , and it did not activate CRY2L348F/CIB1 split Cre . This was true at all intensities tested . We observed the highest activation of nMag/pMag split Cre at 1 . 5 mW/cm2 intensity , which matched the recommendation of its authors ( Kawano et al . , 2016 ) . This calibration also revealed that 0 . 1 mW/cm2 was enough – although clearly suboptimal – to stimulate LiCre above its residual activity . Intriguingly , we observed a weaker activity at 35 . 6 mW/cm2 than at 17 . 9 mW/cm2 . However , this difference could be artifactual because it was within the range of variability observed in one of the conditions ( 3 . 7 mW/cm2 ) . In parallel , we inspected if blue light at maximal intensity was toxic to yeast cells . We scored colony-forming units ( CFUs ) from cells exposed or not to 35 . 6 mW/cm2 blue light ( DMX spot ) during 1 hr ( Figure 3—figure supplement 1 ) . We observed no significant differences in the number of CFUs generated by illuminated and non-illuminated samples , nor between LiCre-expressing and non-expressing cells . We conclude that this light exposure is safe regarding yeast viability . We then interrogated the response of LiCre and nMag/pMag split Cre to periodic stimulations . We expected different response dynamics from these systems because AsLOV2 and VVD have different photocycle lifetimes , with a fast ( ~80 s ) and slow ( ~18 , 000 s ) decay rate of their adduct ( activated ) form , respectively ( Pudasaini et al . , 2015 ) . At the optimal energy for nMag/pMag ( 1 . 5 mW/cm2 ) , LiCre outperformed nMap/pMag split Cre when light episodes were prolonged but not when they were short ( Figure 3f ) . Brief light pulses occurring 2 min apart were insufficient to activate LiCre but fully stimulated nMag/pMag split Cre ( Figure 3f , 5% of lighting time ) . This is entirely consistent with a faster decay rate for the AsLOV2 domain of LiCre compared to the VVD domain of nMag/nMag . To specifically and more precisely characterize the response dynamics of LiCre , we applied periodic stimulations using a higher energy ( 35 . 6 mW/cm2 ) and a slow ( 2 min ) or a fast ( 10 s ) period ( Figure 3g ) . As expected , the fast regime better stimulated recombination . Remarkably , light pulses of 0 . 5 s occurring 10 s apart were sufficient to fully stimulate LiCre . Altogether , these results show that , at least in the yeast cellular context , LiCre outperforms the two previous split systems in terms of efficiency , rapidity , and residual background activity . To demonstrate the control of a biological activity by light , we built a reporter where Cre-mediated excision enabled the expression of the HIS3 gene necessary for growth in the absence of histidine . We cultured cells carrying this construct and expressing LiCre and spotted them at various densities on two HIS- selective plates . One plate was illuminated during 90 min while the other one was kept in the dark , and both plates were then incubated for growth . After 3 days , colonies were abundant on the plate that had been illuminated and very rare on the control plate ( Figure 4a ) . LiCre can therefore be used to trigger cell growth with light . We then sought to observe the switch in individual cells . To do so , we replaced GFP by mCherry in our reporter system so that the excitation wavelength of the reporter did not overlap with stimulation of LiCre . We expressed and stimulated LiCre ( 90 min at 3 . 63 mW/cm2 ) in cells carrying this reporter and subsequently imaged them over time . As expected , we observed the progressive apparition of mCherry signal in a fraction of cells ( Figure 4b , c ) . Although convenient for high-throughput quantifications , reporter systems based on the de novo production and maturation of fluorescent proteins require a delay between the time of DNA excision and the time of acquisition . We wished to bypass this limitation and directly quantify DNA recombination . For this , we designed oligonucleotides outside of the region flanked by LoxP sites . The hybridization sites of these primers are too distant for efficient amplification of the non-edited DNA template , but , after Cre-mediated excision of the internal region , these sites become proximal and PCR amplification is efficient ( Figure 4d ) . We mixed known amounts of edited and non-edited genomic DNA and performed real-time qPCR to build a standard curve that could be used to infer the proportion of edited DNA from qPCR signals . After this calibration , we applied this qPCR assay on genomic DNA extracted from cells collected immediately after different durations of illumination at moderate intensity ( 3 . 63 mW/cm2 , PAUL device ) . Results were in full agreement with GFP-based quantifications ( Figure 4e ) . Excision of the target DNA occurred in a significant fraction of cells after only 2 min of illumination , and we estimated that excision occurred in about 30% and 40% of cells after 20 and 40 min of illumination , respectively . To determine if DNA excision continued to occur after switching off the light , we re-incubated half of the cells for 90 min in the dark prior to harvest and genomic DNA extraction . The estimated frequency of DNA excision was strikingly similar to the one measured immediately after illumination ( Figure 4f ) . This was fully consistent with the fast cycling properties of LiCre described above , with a rapid ( tens of seconds ) reversal of activated LiCre to its inactive state . The qPCR assay also allowed us to compare the efficiency of light-induced recombination between cell populations in exponential growth or stationary phase . This revealed that LiCre photoactivation was about fourfold more efficient in non-dividing cells ( Figure 4g ) . Although the reasons for this difference remain to be determined , this increase of LiCre photoactivation at stationary phase makes it particularly suitable for bioproduction applications , where metabolic switching is often desired after the growth phase . LiCre offers a way to change the activities of cells without adding any chemical to their environment . This potentially makes it an interesting tool to address the limitations caused by metabolic burden , the natural trade-off between the fitness of host cells , and their efficiency at producing exogenous compounds ( Wu et al . , 2016 ) . We therefore tested the possibility to use LiCre to control the production of a commercial compound with light . Carotenoids are pigments that can be used as vitamin A precursors , antioxidants , or coloring agents , making them valuable for the food , agriculture , and cosmetics industries ( Mata-Gómez et al . , 2014 ) . Commercial carotenoids are generally produced by chemical synthesis or extraction from vegetables , but alternative productions based on microbial fermentations offer remarkable advantages , including the use of low-cost substrates and therefore a high potential for financial gains . Bioproduction of carotenoids can be achieved by introducing recombinant biosynthesis pathways in host microorganisms , which offers the advantage of a well-known physiology of the host and optimizations by genetic engineering . For these reasons , strategies were previously developed to produce carotenoids in the yeast Saccharomyces cerevisiae . Expressing three enzymes ( crtE , crtI , and crtYB ) from Xanthophyllomyces dendrorhous enabled S . cerevisiae to efficiently convert farnesyl pyrophosphate ( FPP ) into β-carotene ( Verwaal et al . , 2007 ) . FPP is naturally produced by S . cerevisiae from acetyl-CoA and serves as an intermediate metabolite , particularly for the production of ergosterol that is essential for cellular viability ( Figure 5 ) . This production is associated with a trade-off: redirecting FPP to β-carotene limits its availability for ergosterol biosynthesis and therefore impairs growth; and its consumption by the host cell can limit the flux toward the recombinant pathway . Consistently , metabolic burden associated with carotenoids production was shown to be substantial ( Verwaal et al . , 2010 ) . A promising way to deal with this trade-off would be to favor the flux toward ergosterol during biomass expansion and , after enough producer cells are obtained , switch the demand in FPP toward β-carotene . We therefore explored if LiCre could offer this possibility . First , we tested if LiCre could allow us to switch ON the exogenous production of carotenoids with light . If so , one could use it to trigger production at the desired time of a bioprocess . We constructed a S . cerevisiae strain expressing only two of the three enzymes required for β-carotene production . Expression of the third enzyme , a bifunctional phytoene synthase and lycopene cyclase , was blocked by the presence of a floxed terminator upstream of the coding sequence of the crtYB gene ( Figure 5b ) . Excision of this terminator should restore a fully functional biosynthetic pathway . As expected , this strain formed white colonies on agar plates , but it formed orange colonies after transformation with an expression plasmid coding for Cre , indicating that β-carotene production was triggered ( Figure 5c ) . To test the possible triggering by light , we transformed this strain with a plasmid encoding LiCre and selected several transformants , which we cultured and exposed – or not – to blue light before spotting them on agar plates . The illuminated cultures became orange while the non-illuminated ones remained white . Plating a dilution of the illuminated cell suspension yielded a majority of orange colonies , indicating that LiCre triggered crtYB expression and β-carotene production in a high proportion of plated cells ( Figure 5c ) . We quantified bioproduction by dosing total carotenoids in cultures that had been illuminated or not . This revealed that 72 hr after the light switch the intracellular concentration of carotenoids had jumped from background levels to nearly 200 μg/g ( Figure 5d ) . Thus , LiCre allowed us to switch ON the production of carotenoids by yeast using blue light . We then tested if LiCre could allow us to switch OFF with light the endogenous ergosterol pathway that competes with carotenoid production for FPP consumption . The first step of this pathway is catalyzed by the Erg9p squalene synthase . Given the importance of FPP availability for the production of various compounds , strategies have been reported to control the activity of this enzyme during bioprocesses , especially in order to reduce it after biomass expansion ( Asadollahi et al . , 2008; Xie et al . , 2015; Tippmann et al . , 2016 ) . These strategies were not based on light but derived from transcriptional switches that naturally occur upon addition of inhibitors or when specific nutrients are exhausted from the culture medium . To test if LiCre could offer a way to switch ERG9 activity with light , we modified the ERG9 chromosomal locus and replaced the coding sequence by a synthetic construct comprising a floxed sequence coding for Erg9p and containing a transcriptional terminator , followed by a sequence coding for the catalytic domain of the 3-hydroxy3-methylglutaryl coenzyme A reductase ( tHMG1 ) ( Figure 5e ) . This design prepares ERG9 for a Cre-mediated switch: before recombination , Erg9p is normally expressed; after recombination , ERG9 is deleted and the tHMG1 sequence is expressed to foster the mevalonate pathway . Given that ERG9 is essential for yeast viability in the absence of ergosterol supplementation ( Fegueur et al . , 1991 ) , occurrence of the switch can be evaluated by measuring the fraction of viable yeast cells prior to and after the induction of recombination . When doing so , we observed that expression of Cre completely abolished viability , regardless of illumination . In contrast , cultures expressing LiCre were highly susceptible to light: they were fully viable in the absence of illumination and lost ~23% of viable cells after light exposure ( Figure 5f ) . Thus , LiCre offers the possibility to abolish the activity of the yeast squalene synthase by exposing cells to light . Beyond yeast , LiCre may also have a large spectrum of applications on multicellular organisms . Therefore , we tested its efficiency in human cells . For this , we constructed a lentiviral vector derived from the simian immunodeficiency virus ( SIV ) and encoded a human-optimized version of LiCre with a nuclear localization signal fused to its N-terminus . To quantify the efficiency of this vector , we also constructed a stable reporter cell line where the expression of a membrane-located mCherry fluorescent protein could be switched ON by Cre/Lox recombination . We obtained this line by Flp-mediated insertion of a single copy of the reporter construct into the genome of Flp-In 293 cells ( Figure 6a , see Materials and methods ) . Our assay consisted of producing LiCre-encoding lentiviral particle , depositing them on reporter cells for 24 hr , illuminating the infected cultures with blue light , and , 28 hr later , observing cells by fluorescence microscopy . As shown in Figure 6b , mCherry expression was not detected in non-infected reporter cells . In cultures that were infected but not illuminated , a few positive cells were observed . In contrast , infected cultures that had been exposed to blue light contained numerous positive cells . We sought to quantify this photoactivation and compare it to the one conferred by nMag/pMag split Cre . For this , we produced two additional sets of lentiviral particles: one encoding the originally described nMag/pMag split Cre ( Kawano et al . , 2016 ) , and one encoding the 'Magnets-opti' coding sequence of ‘PA-Cre3 . 0’ , where codons had been optimized to avoid intramolecular sequence homology ( Morikawa et al . , 2020 ) . For consistency , all three constructs shared the same vector backbone and CMV promoter . We tested these systems in parallel as above using light energy that was either suited to LiCre ( 3 . 6 mW/cm2 ) or nMag/pMag split Cre ( 1 . 8 mW/cm2 ) and quantified recombination efficiencies by flow cytometry . As shown in Figure 6c , the residual activity of Magnets-opti in the dark was very high ( ~50% of switched cells ) , which was surprising . LiCre and the original nMag/pMag split Cre system displayed similar levels of induced activity at both low and high light energies; they also showed significant residual activity with 5–10% positive cells for LiCre in the dark and 10–15% for the original nMag/pMag . The highest fold induction was obtained with LiCre at 3 . 6 mW/cm2 , with 5% residual and 31% induced activities , respectively . LiCre can therefore be used to switch genetic activities in human cells with blue light .
We built a structural model of LiCre to conceptualize its mode of activation ( Figure 7a ) . We based this model on ( i ) the available structure of the Cre tetramer complexed with its target DNA ( Guo et al . , 1997 ) , ( ii ) the available structure of AsLOV2 in its dark state ( Guntas et al . , 2015 ) , and ( iii ) knowledge that the Jα helix of AsLOV2 domains unfolds after light activation ( Swartz et al . , 2002; Harper et al . , 2003 ) . From this model , we hypothesize that LiCre photoactivation may occur via two synergistic effects . First , the domain AsLOV2 likely prevents Cre tetramerization in the dark state simply because of its steric occupancy . The unfolding of the Jα helix in the light state may allow AsLOV2 to liberate the multimerizing interface . Second , because the Jα helix of AsLOV2 and the αA helix of Cre are immediately adjacent , they may disturb each other in their proper folding so that they cannot exist both simultaneously in their native conformation . The unfolding of Jα in the light state may therefore stimulate proper folding of αA , and thereby allow αA to bind to the adjacent Cre unit . This predicts two possible models of activation of recombinase activity , as depicted in Figure 7b . If LiCre is not bound to DNA in its dark state ( Model 1 ) , then photoactivation allows it to form a protein dimer bound to each LoxP sites . The two sites can then associate to form the synaptic complex . Alternatively , LiCre may already bind to DNA in the dark state ( Model 2 ) . In this case , photoactivation likely reinforces protein–protein interaction of the two bound units and it also allows the two dimers to assemble into a functional recombination synapse . Experiments interrogating LiCre:DNA interactions in dark and light conditions are now needed to test these models . Several tools already exist for inducing site-specific recombination with light . They fall in two groups: those that require the addition of a chemical and those that are fully genetically encoded . The first group includes the utilization of photocaged ligands instead of 4-hydroxy-tamoxifen to induce the activity of Cre-ERT . This pioneering approach was successful in cultured human cells ( Link et al . , 2005 ) as well as fish ( Sinha et al . , 2010 ) and mouse ( Lu et al . , 2012 ) . Later , a more complex strategy was developed that directly rendered the active site of Cre photoactivatable via the incorporation of photocaged amino acids ( Luo et al . , 2016 ) . In this case , cells were provided with non-natural amino acids , such as the photocaged tyrosine ONBY , and were genetically modified in order to express three macromolecules: a specifically evolved pyrrolysyl tRNA synthetase , a pyrrolysine tRNACUA , and a mutant version of Cre where a critical amino acid such as Y324 was replaced by a TAG stop codon . The tRNA synthetase/tRNACUA pair allowed the incorporation of the synthetic amino acid in place of the nonsense mutation , and the resulting enzyme was inactive unless it was irradiated with violet or ultraviolet light . This strategy successfully controlled recombination in cultured human cells ( Luo et al . , 2016 ) and zebrafish embryos ( Brown et al . , 2018 ) . We note that it presents several caveats: its combination of chemistry and transgenes is complex to implement , the presence of the tRNA synthetase/tRNACUA pair can generate off-target artificial C-terminal tails in other proteins by bypassing natural stop codons , and violet/ultraviolet light can be harmful to cells . More recently , a radically different chemical approach was proposed , which consisted of tethering an active TAT-Cre recombinase to hollow gold nanoshells ( Morales et al . , 2018 ) . When delivered to cells in culture , these particles remained trapped in intracellular endosomes . Near-infrared photostimulation triggered activity by releasing the recombinase via nanobubble generation occurring on the particle surface . A fourth system is based on the chromophore phycocyanobilin , which binds to the PhyB receptor of A . thaliana and makes its interaction with PIF3 dependent on red light . Photostimulation of this interaction was used to assemble split Cre units into a functional complex in yeast ( Hochrein et al . , 2018 ) . A major interest of these last two systems is to offer the possibility to use red light , which is less harmful to cells than blue or violet light and better penetrates tissues . However , all these strategies require to efficiently deliver chemicals to the target cells at the appropriate time before illumination , and their underlying chemistry can be expensive , especially for applications in the context of large volumes such as industrial bioprocesses . Other systems , such as LiCre , are based on flavin chromophores that are readily available across a wide variety of organisms . This avoids the need of chemical additives . To our knowledge , there are currently three such systems . One is based on the sequestration of Cre between two large photo-cleavable domains ( Zhang et al . , 2017 ) . The principle of light-induced protein cleavage is very interesting , but its application to Cre showed important limitations: a moderate efficiency ( ~30% of ON cells after the switch ) , the dependence on a cellular inhibitory chaperone , and the need of violet light . The two other systems are the CRY2/CIB1 and nMag/pMag split Cre ( Taslimi et al . , 2016; Kawano et al . , 2016 ) that we evaluated here in comparison to LiCre . An important advantage of LiCre over these systems is that it is made of a single protein . The first benefit of this is simplicity . More efforts are needed to establish transgenic organisms expressing two open reading frames ( ORFs ) compared to a single one . This is particularly true for vertebrate systems , where inserting several constructs requires additional efforts for characterizing transgene insertion sites and conducting genetic crosses . For this reason , the two ORFs of the split Cre system were previously combined in a single construct , where they were separated either by an internal ribosomal entry site ( IRES ) or by a sequence coding a self-cleaving peptide ( Taslimi et al . , 2016; Kawano et al . , 2016; Morikawa et al . , 2020 ) . Although helpful , these solutions have important limits: with an IRES , the two ORFs are not expressed at the same level; with a self-cleaving peptide , cleavage of the precursor protein can be incomplete , generating uncleaved products with unknown activity . This was the case for nMag/pMag split Cre in mammalian cells , where a non-cleaved form at ~72 kDa was reported and where targeted modifications of the cleavage sequence increased both the abundance of this non-cleaved form and the non-induced activity of the system ( Morikawa et al . , 2020 ) . The second benefit of LiCre being a single protein is to avoid problems of suboptimal stoichiometry between the two protein units , which was reported as a possible issue for CRY2/CIB1 split Cre ( Meador et al . , 2019 ) . A third benefit is to avoid possible intramolecular recombination between the homologous parts of the two coding sequences . Although not demonstrated , this undesired possibility was suspected for nMag/pMag split Cre because its two dimerizers derive from the same sequence ( Morikawa et al . , 2020 ) . The other advantages of LiCre are its performances . In the present study , we used a yeast-based assay to compare LiCre with split Cre systems . Unexpectedly , although we used the improved version of the CRY2/CIB1 split Cre containing the CRY2-L348F mutation ( Taslimi et al . , 2016 ) , it did not generate photo-inducible recombination in our assay . This is unlikely due to specificities of the budding yeast , such as improper protein expression or maturation , because the original authors reported activity in this organism ( Taslimi et al . , 2016 ) . We do not explain this result , but it is consistent with previous studies reporting that photoactivation of the original version of the CRY2/CIB1 split Cre can be extremely low ( Kawano et al . , 2016 ) , and that the induced activity of the CRY2-L348F/CIB1 system can also be low and highly variable ( Morikawa et al . , 2020 ) . In contrast , we validated the efficiency of nMag/pMag split Cre and so did other independent laboratories ( Morikawa et al . , 2020; Takao et al . , 2020; Allen et al . , 2019; Weinberg et al . , 2019 ) . We observed a significant photoactivation of this system both in yeast and human cells . Recently , the nMag/pMag split Cre system was expressed in mice as a transgene – dubbed PA-Cre3 . 0 – which comprised the promoter sequence of the chicken beta actin gene ( CAG ) and the 'Magnets-opti' modified version of the original self-cleaving coding sequence ( Morikawa et al . , 2020 ) . The authors reported that this strategy abolished residual activity and attributed this improvement to a reduction of the expression level of the transgene ( Morikawa et al . , 2020 ) . In our human cell assay , where we used the CMV promoter , the 'Magnets-opti' system instead showed a much higher residual activity than the original nMag/pMag split Cre . This was surprising; it is probably very important to carefully calibrate the expression of Magnets-opti when using it . In contrast , LiCre displayed weaker residual activity than the original nMag/pMag split Cre in the dark . Importantly , LiCre also displayed higher induced activity – especially in yeast – and a faster response to light compared to nMag/pMag split Cre . This strong response probably results from its simplicity since the activation of a single protein involves fewer steps than the activation of two units that must then dimerize to become functional . Finally , we showed that the photocycle of active LiCre is faster than the one of active nMag/pMag split Cre . This offers higher temporal precision for the induction of recombination . In conclusion , LiCre provides a cheap , simple , low-background , highly efficient and fast-responding way to induce site-specific recombination with light . Given that it works in both yeast and mammalian cells , it opens many perspectives from fundamental and biomedical research to industrial applications .
Plasmids , strains , and oligonucleotides used in this study are listed in Supplementary file 1 ( Supplementary Tables S1 , S2 , and S3 , respectively ) . LiCre plasmids are available from Addgene ( http://www . addgene . org/ ) under accession numbers 166660 , 166661 and 166663 . We ordered the synthesis of sequence LoxLEULoxHIS ( Supplementary file 1 – Supplementary Text S1 ) from GeneCust , who cloned the corresponding BamHI fragment in plasmid pHO-poly-HO to produce plasmid pGY262 . The PTEF-loxP-KlLEU2-STOP-loxP-spHIS5 construct can be excised from pGY262 by NotI digestion for integration at the yeast HO locus . This way , we integrated it in a leu2 his3 strain , which could then switch from LEU+ his- to leu- HIS+ after Cre-mediated recombination ( Figure 4e ) . To construct a GFP-based reporter , we ordered the synthesis of sequence LEULoxGreen ( Supplementary file 1 – Supplementary Text S1 ) from GeneCust , who cloned the corresponding NheI-SacI fragment into pGY262 to obtain pGY407 . We generated strain GY984 by crossing BY4726 with FYC2-6B . We transformed GY984 with the 4 kb NotI insert of pGY407 and obtained strain GY1752 . To remove the ade2 marker , we crossed GY1752 with FYC2-6A and obtained strain GY1761 . Plasmid pGY537 targeting integration at the LYS2 locus was obtained by cloning the BamHI-EcoRI fragment of pGY407 into the BamHI , EcoRI sites of pIS385 . Plasmid pGY472 was produced by GeneCust , who synthesized sequence LEULoxmCherry ( Supplementary file 1 – Supplementary Text S1 ) and cloned the corresponding AgeI-EcoRI insert into the AgeI , EcoRI sites of pGY407 . We generated GY983 by crossing BY4725 with FYC2-6A . We obtained GY2033 by transformation of FYC2-6B with a 4 kb NotI fragment of pGY472 . We obtained GY2207 by transformation of GY983 with the same 4 kb NotI fragment of pGY472 . To generate GY2206 , we linearized pGY537 with NruI digestion , transformed in strain GY855 , and selected a LEU+ Lys- colony ( pop-in ) , which we re-streaked on 5-FoA plates for vector excision by counter-selection of URA3 ( pop-out ) ( Sadowski et al . , 2007 ) . Strain GY2214 was a diploid that we obtained by mating GY2206 with GY2207 . Mutations E340A D341A were introduced by GeneCust by site-directed mutagenesis of pSH63 , yielding plasmid pGY372 . We generated the N-terΔ21 mutant of Cre by PCR amplification of the PGAL1 promoter of pSH63 using primer 1L80 ( forward ) and mutagenic primer 1L71 ( reverse ) , digestion of pSH63 by AgeI and co-transformation of this truncated plasmid and amplicon in a trp1Δ63 yeast strain for homologous recombination and plasmid rescue . We combined the N-terΔ21 and the C-ter E340A D341A mutations similarly , but with pGY372 instead of pSH63 . We generated N-terΔ28 and N-terΔ37 mutants , combined or not with C-ter E340A D341A mutations , by the same procedure where we changed 1L71 by mutagenic primers 1L72 and 1L73 , respectively . To generate a Cre-VVD fusion , we designed sequence CreCVII ( Supplementary file 1 – Supplementary Text S1 ) where the Cre sequence from GENBANK AAG34515 . 1 was fused to the VVD-M135IM165I sequence from Zoltowski et al . , 2009 via four additional residues ( GGSG ) . We ordered its synthesis from GeneCust and co-transformed it in yeast with pSH63 ( previously digested by NdeI and SalI ) for homologous recombination and plasmid rescue . This generated pGY286 . We then noticed an unfortunate error in AAG34515 . 1 , which reads a threonine instead of an asparagine at position 327 . We cured this mutation from pGY286 by site-directed mutagenesis using primers 1J47 and 1J48 , which generated pGY339 that codes for Cre-4-VVD described in Figure 2—figure supplement 1 . We constructed mutant C-terΔ14 of Cre by site-directed mutagenesis of pGY286 using primers 1J49 and 1J50 , which simultaneously cured the N327T mutation and introduced an early stop codon . Mutants C-terΔ2 , C-terΔ4 , C-terΔ6 , C-terΔ8 , C-terΔ10 , C-terΔ12 of Cre were constructed by GeneCust , who introduced early stop codons in pGY339 by site-directed mutagenesis . We generated Cre-VVD fusions carrying point mutations at amino acid positions 340 and 341 , with or without truncation of the GGSG linker ( Figure 2—figure supplement 1 ) , by site-directed mutagenesis of pGY339 . To test LOV_Cre fusions , we first designed sequence EcoRI-LovCre_chimJa-BstBI ( Supplementary file 1 – Supplementary Text S1 ) corresponding to the fusion of AsLOV2 with Cre via an artificial α-helix . This helix was partly identical to the Jα helix of AsLOV2 and partly identical to the αA helix of Cre . This sequence was synthesized and cloned in the EcoRI and BstBI sites of pSH63 by GeneCust , yielding pGY408 . We then generated and directly tested a variety of LOV_Cre fusions . To do so , we digested pGY408 with BsiWI and MfeI and used this fragment as a recipient vector; we amplified the Cre sequence from pSH63 using primer 1G42 as the reverse primer , and one of primers 1M42 to 1M53 as the forward primer ( each primer corresponding to a different fusion position ) ; we co-transformed the resulting amplicon and the recipient vector in strain GY1761 , isolated independent transformants , and assayed them with the protocol of photoactivation and flow cytometry described below . We generated and tested a variety of LOV_CreAA fusions by following the same procedure where plasmid pGY372 was used as the PCR template instead of pSH63 . A transformant corresponding to LOV_Cre32 and showing light-dependent activity was chosen for plasmid rescue , yielding plasmid pGY415 . A transformant corresponding to LOV_CreAA20 was chosen for plasmid rescue , yielding plasmid pGY416 . Sanger sequencing revealed that the fusion sequence present in pGY416 was QID instead of QIA at the peptide junction ( position 149 on LiCre sequence of Supplementary file 1 – Supplementary Text S2 ) . All further experiments on LiCre were derived from the fusion protein coded by pGY416 . To introduce random residues at the peptide junction of LOV_Cre32 ( Figure 2b ) , we first generated pGY417 using the same procedure as for the generation of pGY415 but with pSH47 instead of pSH63 as the PCR template so that pGY417 has a URA3 marker instead of TRP1 . We then ordered primers 1N24 , 1N25 , and 1N26 containing degenerate sequences , used them with primer 1F14 to amplify the Cre sequence of pSH63 , co-transformed in strain GY1761 the resulting amplicons together with a recipient vector made by digesting plasmid pGY417 with NcoI and BsiWI , and isolated and directly tested individual transformants with the protocol of photoactivation and flow cytometry described below . Plasmids from transformants showing evidence of reduced background were rescued from yeast and sequenced , yielding pGY459 to pGY464 . To replace the PGAL1 promoter of pGY416 by the PMET17 promoter , we digested it with SacI and SpeI , PCR-amplified the PMET17 promoter of plasmid pGY8 with primers 1N95 and 1N96 , and co-transformed the two products in yeast for homologous recombination , yielding plasmid pGY466 . We changed the promoter of pGY415 using exactly the same procedure , yielding plasmid pGY465 . We changed the promoter of pSH63 similarly using primer 1O83 instead of 1N96 , yielding plasmid pGY502 . To express the nMag/pMag split Cre system in yeast , we designed sequence CreN-nMag-NLS-T2A-NLS-pMag-CreC partly ( Supplementary file 1 – Supplementary Text S1 ) and ordered its synthesis from GeneCust . The corresponding BglII fragment was co-transformed in yeast for homologous recombination with pGY465 previously digested with BamHI ( to remove AsLOV2 and part of Cre ) , yielding plasmid pGY488 that contained the full system . We then derived two plasmids from pGY488 , each one containing one half of the split system under the control of the Met17 promoter . We obtained the first plasmid ( pGY491 , carrying the TRP1 selection marker ) by digestion of pGY488 with SfoI and SacII and co-transformation of the resulting recipient vector with a PCR product amplified from pGY465 using primers 1O80 and 1O82 . We obtained the second plasmid ( pGY501 , carrying the URA3 selection marker ) in two steps . We first removed the pMag-CreC part of pGY488 by digestion with NdeI and SacII followed by Klenow fill-in and religation . We then changed the selection marker by digestion with PfoI and KpnI and co-transformation in yeast with a PCR product amplified from pSH47 with primers 1O77 and 1O89 . To express the CRY2L348F/CIB1 split Cre system in yeast , we designed sequences CIB1CreCter and CRY2CreNter and ordered their synthesis from GeneCust , obtaining plasmids pGY526 and pGY527 , respectively . To obtain pGY531 , we extracted the synthetic insert of pGY527 by digestion with BglII and co-transformed it in yeast with the NdeI-BamHI fragment of pGY466 for homologous recombination . To obtain pGY532 , we extracted the synthetic insert of pGY526 by digestion with BglII and co-transformed it in yeast with the SacI-BamHI fragment of pSH47 for homologous recombination . To build a switchable strain for carotene production , we modified EUROSCARF strain Y41388 by integrating a LoxP-KlLEU2-TADH1-LoxP cassette immediately upstream the CrtYB coding sequence of the chromosomally integrated expression cassette described by Verwaal et al . , 2007 . This insertion was obtained by transforming Y41388 with a 6 . 6 kb BstBI fragment from plasmid pGY559 and selecting a Leu+ transformant , yielding strain GY2247 . To obtain pGY559 , we first deleted the crtE and crtI genes from YEplac195-YB_E_I ( Verwaal et al . , 2007 ) by MluI digestion and religation . We then linearized the resulting plasmid with SpeI and co-transformed it for recombination in a leu2Δ yeast strain with a PCR amplicon obtained with primers 1P74 and 1P75 and template pGY407 . After Leu+ selection , the plasmid was recovered from yeast , amplified in bacteria , and verified by restriction digestion and sequencing . We used CRISPR/Cas9 to build a switchable strain for squalene synthase . We cloned the synthetic sequence gERG9 ( Supplementary file 1 – Supplementary Text S1 ) in the BamHI-NheI sites of the pML104 plasmid ( Laughery et al . , 2015 ) so that the resulting plasmid ( pGY553 ) coded for a gRNA sequence targeting ERG9 . This plasmid was transformed in GY2226 together with a repair-template corresponding to a 4 . 2 kb EcoRI fragment of pGY547 that contained LoxP-synERG9-TADH1-LoxP with homologous flanking sequences . The resulting strain was then crossed with Y41388 to obtain GY2236 . We used synthetic ( S ) media made of 6 . 7 g/l Difco Yeast Nitrogen Base without amino acids and 2 g/l of a powder that was previously prepared by mixing the following amino acids and nucleotides: 1 g of adenine , 2 g of uracil , 2 g of alanine , 2 g of arginine , 2 g of aspartate , 2 g of asparagine , 2 g of cysteine , 2 g of glutamate , 2 g of glutamine , 2 g of glycine , 2 g of histidine , 2 g of isoleucine , 4 g of leucine , 2 g of lysine , 2 g of methionine , 2 g of phenylalanine , 2 g of proline , 2 g of serine , 2 g of threonine , 2 g of tryptophane , 2 g of tyrosine , and 2 g of valine . For growth in glucose condition , the medium ( SD ) also contained 20 g/l of D-glucose . For growth in galactose condition ( induction of PGAL1 promoter ) , we added 2% final ( 20 g/l ) raffinose and 2% final ( 20 g/l ) galactose ( SGalRaff medium ) . Media were adjusted to pH = 5 . 8 by addition of NaOH 1 N before autoclaving at 0 . 5 bar . For auxotrophic selections or PMET17 induction , we used media where one or more of the amino acids or nucleotides were omitted when preparing S . For example , SD-W-M was made as SD but without any tryptophane or methionine in the mix powder . Unless mentioned otherwise , quantitative tests were done by flow cytometry using yeast reporter strain GY1761 . We used two devices for photoactivation . The first device was a PAUL apparatus from GenIUL: a box with mirrors on its walls and floor , and equipped with 460 nm blue LEDs on its ceiling . Using a NovaII photometer ( Ophir Photonics ) , we measured that a 100% intensity on this apparatus corresponded to an energy of 36 . 3 mW/cm2 . We used Zomei ND filters when we needed to obtain intensities that were not tunable on this device . The second device was a computer-controlled DMX LED spot ( 450 nm ) purchased from Neptune-LED ( Grenoble , France ) , which we placed above a thermostated platform . The yeast reporter strain was transformed with the plasmid of interest , pre-cultured overnight in selective medium corresponding to conditions of transcriptional activation of the plasmid-borne Cre construct ( SGalRaff-W for PGAL1 plasmids , SD-W-M for PMET17 plasmids , SD-W-U-M for split Cre systems ) with no particular protection against ambient light . The saturated culture was transferred to two 96-well polystyrene flat-bottom Falcon sterile plates ( 100 μl per well ) and one plate was illuminated at the indicated intensities while the other plate was kept in the dark . After the indicated duration of illumination , cells from the two plates were transferred to a fresh medium allowing expression of GFP but not cell division ( SD-W-H or SD-W-U-H , strain GY1761 being auxotroph for histidine ) , and these cultures were incubated at 30°C for 90 min . Cells were then either analyzed immediately by flow cytometry or blocked in phosphate-buffered saline ( PBS + 1 mM sodium azide ) and analyzed the following day . We acquired data for 10 , 000 events per sample using a FACSCalibur ( BD Biosciences ) or a MACSQuant VYB ( Miltenyi Biotech ) cytometer after adjusting the concentration of cells in PBS . We analyzed raw data files in the R statistical environment ( http://www . r-project . org ) using custom-made scripts based on the flowCore package ( Hahne et al . , 2009 ) from bioconductor ( http://www . bioconductor . org ) . We gated cells automatically by computing a perimeter of ( FSC-H , SSC-H ) values that contained 40% of events ( using 2D-kernel density distributions ) . We plotted the gates of all samples belonging to a single experiment ( usually corresponding to the same day ) . This occasionally highlighted samples for which the gates were clearly shifted ( outliers ) , indicating cell populations of unexpected size , and we therefore discarded these rare samples for further analysis . A threshold of fluorescent intensity ( GFP or mCherry ) was set to distinguish ON and OFF cells ( i . e . , expressing or not the reporter ) . To do this , we included in every experiment a negative control made of the reporter strain transformed with an empty vector and chose the 99 . 9th percentile of the corresponding 4000 fluorescent values ( gated cells ) as the threshold . No statistical power analysis was done prior to the experiment , no masking of samples IDs was applied during the experiments , and no data randomization was used . For Figure 3g , we segmented individual cells on bright-field images using the ImageJ Lasso plugin . Then , we measured on the fluorescence images the mean gray value of pixels in each segmented area , providing single-cell measures of fluorescence . For each image , the background fluorescence level was quantified from eight random regions outside of cells and with areas similar to single cells . This background level was subtracted from the fluorescence level of each cell . We strictly followed the procedure described in Verwaal et al . , 2007 , which consists of mechanical cell lysis using glass beads , addition of pyrogallol , KOH-based saponification , and extraction of carotenoids in hexane . Quantification was estimated by optical absorption at 449 nm using a Biowave spectrophotometer . No statistical power analysis was done prior to the experiment , no masking of samples IDs was applied during the experiments , and no data randomization was used . We built a reporter construct for Cre-mediated recombination in human cells based on Addgene's plasmids 55779 , containing a membrane-addressed mCherry sequence ( Yost et al . , 2007 ) ( mCherry-Mem ) , and 51269 , containing a zsGreen-based reporter of Cre recombination ( Hermann et al . , 2014 ) . Re-sequencing revealed that 51269 did not contain three terminator sequences but only one between the LoxP sites . We applied a multi-steps procedure to ( i ) restore three terminators , ( ii ) replace zsGreen with mCherry-Mem , and ( iii ) have the final reporter in a vector suitable for targeted single-site insertion . First , we inserted a LoxP site between restriction sites NheI and HindIII of pCDNA5/FRT ( Invitrogen ) by annealing oligonucleotides 1O98 and 1O99 , digesting and cloning this adaptor with NheI and HindIII , which yielded plasmid pGY519 . Second , we replaced in two different ways the zsGreen sequence of 51269 by the mCherry-Mem sequence of 55779: either by cloning a SmaI-NotI insert from 55779 into EcoRV-NotI of 51269 , yielding plasmid pGY520 , or cloning a EcoRI-NotI from 55779 into EcoRI-NotI of 51269 , yielding plasmid pGY521 . Third , we inserted the HindIII-NotI cassette of pGY520 into the HindIII-NotI sites of pGY519 , yielding plasmid pGY523 . Fourth , we inserted the HindIII-NotI cassette of pGY521 into the HindIII-NotI sites of pGY519 , yielding plasmid pGY524 . Fifth , a HindIII-BamHI fragment of pGY523 containing one terminator and a BglII-EcoRI fragment of pGY524 containing another terminator were simultaneously cloned as consecutive inserts in the BglII-EcoRI sites of 51269 . Finally , the resulting plasmid was digested with HindIII and BamHI to produce a fragment that was cloned into the HindIII-BglII sites of pGY524 to produce pGY525 . To establish stable cell lines , Flp-In T-REx 293 cells were purchased from Invitrogen ( ThermoFisher ) and transfected with both the Flp recombinase vector ( pOG44 , Invitrogen ) and pGY525 . Selection of clonal cells was first performed in medium containing 300 μg hygromycin ( Sigma ) . After 2 weeks , we identified foci of cell clusters , which we individualized by transferring them to fresh wells . One of these clones was cultured for three additional weeks with high concentrations of hygromycin ( up to 400 μg ) to remove potentially contaminating negative cells . The resulting cell line was named T4-2PURE . A synthetic sequence was ordered from GeneCust and cloned in the HindIII-NotI sites of pCDNA3 . 1 ( Invitrogen V79020 ) . This insert contained an unrelated additional sequence that we removed by digestion with BamHI and XbaI , followed by blunt-ending with Klenow fill-in . The resulting plasmid ( pGY561 ) encoded LiCre optimized for mammalian codon usage , in-frame with a N-ter located SV40-NLS signal . This NLS-LiCre sequence was amplified from pGY561 using primers Sauci and Flard ( Table S3 ) , and the resulting amplicon was cloned in the AgeI-HindIII sites of the GAE0 Self-Inactivating Vector ( Mangeot et al . , 2004 ) , yielding pGY577 . To obtain a similar vector expressing nMag/pMag split Cre , we ordered its synthetic sequence from GeneCust , who cloned it into the AgeI-HindIII sites of pGY577 , yielding pGY625 . This plasmid was further modified by replacing the AgeI-EcoRV portion with a synthetic fragment corresponding to the ‘Magnets-opti’ coding sequence of ‘PA-Cre3 . 0" ( Morikawa et al . , 2020 ) , yielding pGY626 . Sequences of pGY577 , pGY625 , and pGY626 are provided in Supplementary file 1 – Supplementary Text S1 . Lentiviral particles were produced in Gesicle Producer 293 T cells ( TAKARA ref 632617 ) transiently transfected by an HIV-1 helper plasmid ( 45% of total DNA ) , a plasmid encoding the VSV-g envelope ( 15% of total DNA ) , and either pGY577 , pGY625 , or pGY626 ( 40% of total DNA ) as previously described ( Mangeot et al . , 2004 ) . Particle-containing supernatants were clarified , filtered through a 0 . 45 μm membrane , and concentrated by ultracentrifugation at 40 , 000 g before resuspension in 1× PBS ( 100-fold concentration ) . To quantify the particle load of these preparations , we designed a qPCR assay ( primers 1Q74 , 1Q75 ) and calibrated it on a standard curve obtained by serial dilutions of plasmid DNA . These measures ( as mean ± s . e . m . million particles per μl , n = 3 replicate measures ) on the preparations used in Figure 6c were the following: 16 ± 0 . 7 for LiCre , 8 . 6 ± 0 . 4 for nMag/pMag split Cre , and 5 . 2 ± 0 . 04 for Magnets-opti . About 3 × 105 cells of cell line T4-2PURE were plated in two 6-well plates . After 24 hr , 100 μl of viral particles suspension were added to each well . After another 24 hr , one plate was illuminated with blue light ( 460 nm ) using the PAUL apparatus installed in a 37°C incubator while the other plate was kept in the dark . For illumination , we applied a sequence of 20 min ON , 20 min OFF under CO2 atmosphere , 20 min ON , where ON corresponded to 3 . 63 mW/cm2 illumination ( or 1 . 8 mW/cm2 when indicated in Figure 6c ) . Plates were then returned to the incubator and , after 28 hr , were either imaged on an Axiovert135 inverted fluorescent microscope or treated with trypsin , resuspended in 1× PBS , and fixed with paraformaldehyde ( PFA ) prior to acquisitions on a MACSQuant flow cytometer . Flow-cytometry data was then processed as described above for yeast cells . Four samples were discarded from further analysis because the culture medium had become yellow and because flow cytometry revealed aberrant cell sizes ( lower forward scatter [FSC] and very high side scatter [SSC] ) . Samples IDs were not masked during the experiments . We did not retype the T4-2PURE reporter cell line or the Gesicle Producer cell line , nor did we control them for mycoplasma contamination prior to the experiments because we used them in all the reporter assays that we applied to various constructs and it is extremely unlikely that a contamination would bias results for some constructs and not others . We calculated the free-energy profile ( reported in Figure 1f ) for the unbinding of the C-terminal α-helix in the tetrameric Cre-recombinase complex ( Ennifar et al . , 2003 ) ( PDB entry 1NZB ) as follows . The software we used were the CHARMM-GUI server ( Lee et al . , 2016 ) to generate initial input files; CHARMM version c39b1 ( Brooks et al . , 2009 ) to setup the structural models and subsequent umbrella sampling by molecular dynamics; WHAM version 2 . 0 . 9 ( http://membrane . urmc . rochester . edu/content/wham/ ) to extract the PMF; and VMD version 1 . 9 . 2 ( Humphrey et al . , 1996 ) to visualize structures . To achieve sufficient sampling by molecular dynamics , we worked with a structurally reduced model system . We focused thereby only on the unbinding of the C-terminal α-helix of subunit A ( residues 334:340 ) from subunit F . Residues that did not have at least one atom within 25 Å from residues 333 to 343 of subunit A were removed including the DNA fragments . Residues with at least one atom within 10 Å were allowed to move freely in the following simulations; the remaining residues were fixed to their positions in the crystal structure . For the calculation of the double mutant A340A341 , the corresponding residues were replaced by alanine residues . The systems were simulated with the CHARMM22 force field ( GBSW and CMAP parameter file ) and the implicit solvation model FACTS ( Haberthür and Caflisch , 2008 ) with recommended settings for param22 ( i . e . , cutoff of 12 Å for non-bonded interactions ) . Langevin dynamics were carried out with an integration time step of 2 fs and a friction coefficient of 4 ps−1 for non-hydrogen atoms . The temperature of the heat bath was set to 310 K . The hydrogen bonds were constrained to their parameter values with SHAKE ( Ryckaert et al . , 1977 ) . The PMF was calculated for the distance between the center of mass of the α-helix ( residues 334:340 of subunit A ) and the center of mass of its environment ( all residues that have at least one atom within 5 Å of this helix ) . Umbrella sampling ( Torrie and Valleau , 1977 ) was performed with 13 independent molecular dynamics simulations where the system was restrained to different values of the reaction coordinate ( equally spaced from 4 to 10 Å ) using a harmonic biasing potential with a spring constant of 20 kcal mol−1 Å−1 ( GEO/MMFP module of CHARMM ) . Note that this module uses a pre-factor of ½ for the harmonic potential ( as in the case of the program WHAM ) . For each simulation , the value of the reaction coordinate was saved at every time step for 30 ns . After an equilibration phase of 5 ns , we calculated for blocks of 5 ns the PMF and the probability distribution function along the reaction coordinate using the weighted histogram analysis method ( Kumar et al . , 1992 ) . A total of 13 bins were used with lower and upper boundaries at 3 . 75 and 10 . 25 Å , respectively , and a convergence tolerance of 0 . 01 kcal mol−1 . Finally , we determined for each bin its relative free energy Fi=−kTln ( p−i ) , where k is the Boltzmann constant , T the temperature ( 310 K ) , and p-i the mean value of the probability of bin i when averaged over the five blocks . The error in the Fi estimate was calculated with σFi=kT σp-i/p-i , where σp-i is twice the standard error of the mean of the probability . An offset was applied to the final PMF so that its lowest value was located at zero . We grew 10 colonies of strain GY1761 carrying plasmid pGY466 overnight at 30°C in SD-L-W-M liquid cultures . The following day , we used these starter cultures to inoculate 12 ml of SD-W-M medium at OD600 = 0 . 2 . When monitoring growth by optical density measurements , we observed that it was fully exponential after 4 hr and until at least 8 . 5 hr . At 6 . 5 hr of growth , for each culture , we dispatched 0 . 1 ml in 96-well plate duplicates using one column ( 8 wells ) per colony , stored aliquots by centrifuging 1 ml of the cell suspension at 3300 g and freezing the cell pellet at −20°C ( ‘exponential’ negative control ) , and re-incubated the remaining of the culture at 30°C for later analysis at stationary phase . We exposed one plate ( Figure 4j ‘exponential’ cyan samples ) to blue light ( PAUL apparatus , 460 nm , 3 . 63 mW/cm2 intensity ) for 40 min while the replicate plate was kept in the dark ( Figure 4j ‘exponential’ gray samples ) . We pooled cells of the same column and stored them by centrifugation and freezing as above . The following day , we collected 1 ml of each saturated , frozen , and stored cells as above ( ‘stationary’ negative control ) . We dispatched the remaining of the cultures in a series of 96-well plates ( 0 . 1 ml/well , two columns per colony ) and exposed these plates to blue light ( PAUL apparatus , 460 nm , 3 . 63 mW/cm2 intensity ) for the indicated time ( 0 , 2 , 5 , 10 , 20 , or 40 min ) . For each plate , following illumination , we collected and froze cells from six columns ( Figure 4h samples ) and re-incubated the plate in the dark for 90 min before collecting and freezing the remaining six columns ( Figure 4i , x-axis samples ) . For genomic DNA extraction , we pooled cells from six wells of the same colony ( one column ) , centrifuged and resuspended them in 280 μl in 50 mM EDTA , and added 20 μl of a 2 mg/ml Zymolyase stock solution ( SEIKAGAKU , 20 U/mg ) to the cell suspension and incubated it for 1 hr at 37°C for cell wall digestion . We then processed the digested cells with the Wizard Genomic DNA Purification Kit from Promega . We quantified DNA on a Nanodrop spectrophotometer and used ~100 , 000 copies of genomic DNA as template for qPCR , with primers 1P57 and 1P58 to amplify the edited target and primers 1B12 and 1C22 to amplify a control HMLalpha region that we used for normalization . We ran these reactions on a Rotorgene thermocycler ( Qiagen ) . This allowed us to quantify the rate of excision of the floxed region as NLox/NTotal , where NLoxis the number of edited molecules and NTotal the total number of DNA template molecules . To estimate NLox , we prepared mixtures of edited and non-edited genomic DNAs at known ratios of 0% , 0 . 5% , 1% , 5% , 10% , 50% , 70% , 90% , and 100% and applied ( 1P57 , 1P58 ) qPCR using these mixtures as templates . This provided us with a standard curve that we then used to convert Ct values of the samples of interest into NLox values . To estimate NTotal , we qPCR-amplified the HMLalpha control region from templates made of increasing concentrations of genomic DNA . We then used the corresponding standard curve to convert the Ct value of HMLalpha amplification obtained from the samples of interest into NTotal values . No statistical power analysis was done prior to the experiment , no masking of samples IDs was applied during the experiments , and no data randomization was used . | In a biologist’s toolkit , the Cre protein holds a special place . Naturally found in certain viruses , this enzyme recognises and modifies specific genetic sequences , creating changes that switch on or off whatever gene is close by . Genetically engineering cells or organisms so that they carry Cre and its target sequences allows scientists to control the activation of a given gene , often in a single tissue or organ . However , this relies on the ability to activate the Cre protein ‘on demand’ once it is in the cells of interest . One way to do so is to split the enzyme into two pieces , which can then reassemble when exposed to blue light . Yet , this involves the challenging step of introducing both parts separately into a tissue . Instead , Duplus-Bottin et al . engineered LiCre , a new system where a large section of the Cre protein is fused to a light sensor used by oats to detect their environment . LiCre is off in the dark , but it starts to recognize and modify Cre target sequences when exposed to blue light . Duplus-Bottin et al . then assessed how LiCre compares to the two-part Cre system in baker's yeast and human kidney cells . This showed that the new protein is less ‘incorrectly’ active in the dark , and can switch on faster under blue light . The improved approach could give scientists a better tool to study the role of certain genes at precise locations and time points , but also help them to harness genetic sequences for industry or during gene therapy . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"biochemistry",
"and",
"chemical",
"biology",
"tools",
"and",
"resources",
"genetics",
"and",
"genomics"
] | 2021 | A single-chain and fast-responding light-inducible Cre recombinase as a novel optogenetic switch |
Visual perception and behavior are mediated by cortical areas that have been distinguished using architectonic and retinotopic criteria . We employed fluorescence imaging and GCaMP6 reporter mice to generate retinotopic maps , revealing additional regions of retinotopic organization that extend into barrel and retrosplenial cortices . Aligning retinotopic maps to architectonic borders , we found a mismatch in border location , indicating that architectonic borders are not aligned with the retinotopic transition at the vertical meridian . We also assessed the representation of visual space within each region , finding that four visual areas bordering V1 ( LM , P , PM and RL ) display complementary representations , with overlap primarily at the central hemifield . Our results extend our understanding of the organization of mouse cortex to include up to 16 distinct retinotopically organized regions .
Mammalian visual cortex consists of a series of interconnected areas , which correspond to topographically-organized collections of neurons with similar functional properties and patterns of connectivity ( Van Essen , 2003 ) . In mouse neocortex , 12 visual areas have been identified , each corresponding to a retinotopically-organized map of the visual field ( Dräger , 1975; Wagor et al . , 1980; Olavarria et al . , 1982; Olavarria and Montero , 1989; Schuett et al . , 2002; Kalatsky and Stryker , 2003; Wang and Burkhalter , 2007; Marshel et al . , 2011; Wang and Burkhalter , 2013; Garrett et al . , 2014; Tohmi et al . , 2014 ) . The borders between visual areas have been located using architectonic and retinotopic criteria . Architectonic borders are associated with changes in staining ( chemoarchitectonics ) , cell density ( cytoarchitectonics ) , or myelination patterns ( myeloarchitechtonics ) . Retinotopic borders are associated with a change in chirality of the retinotopic map and have been identified anatomically by labeling projections between visual areas , and functionally by recording the responses of neurons with electrodes or imaging techniques . The architectonic and retinotopic borders of visual areas in the mouse are generally thought to be co-aligned , but the shapes and locations of visual areas can differ across studies . We generated retinotopic maps from GCaMP6 transgenic mice ( Chen et al . , 2013; Madisen et al . , 2015 ) and compared functional retinotopic , anatomical retinotopic , chemoarchitectonic , cytoarchitectonic and myeloarchitechtonic maps . Our results revealed new regions of retinotopic organization in mouse neocortex , some within and some outside the accepted boundaries of mouse visual cortex , and a mismatch between retinotopic and architectonic maps that may help reconcile the differences between visual area maps generated by these complementary approaches .
We mapped visual areas in three GCaMP6 reporter lines: Ai95 ( RGL-GCaMP6f ) , Ai96 ( RGL-GCaMP6s ) and Ai93 ( TITL-GCaMP6f ) . Each reporter line was crossed with the Emx1-IRES-Cre line , driving GCaMP6 expression in pyramidal neurons in all layers of neocortex ( Figure 1—figure supplement 1 ) . We imaged through a 5 mm diameter glass window implanted over visual areas ( Figure 1A , B ) . The time-averaged fluorescence , measured in primary visual cortex in the absence of visual stimuli , was greater in all three reporter lines than in wild-type mice ( Figure 1H; fluorescence normalized to wild-type , Emx1-Ai95 10 . 3 ± 1 . 3 , Emx1-Ai96 4 . 7 ± 0 . 5 , Emx1-Ai93 49 . 7 ± 5 . 9 ) . 10 . 7554/eLife . 18372 . 003Figure 1 . Amplitude and kinetics of fluorescence transients to brief visual stimuli . ( A ) Fluorescence image focused on the cortical surface of an Emx1-Ai95 mouse . Approximate edge of 5 mm diameter craniotomy is marked with a dashed line . ( B and C ) Baseline ( B ) and peak ( C ) change in fluorescence for the field of view shown in A . In C , dashed line and circle indicate the regions used to extract values for D and E , respectively . ( D ) Spatial extent of the fluorescence change along line marked in panel C . Width of response at half height ( dashed line ) is 190 µm . ( E ) Fluorescence time course for a single trial , from the region marked in panel C . D and E are in arbitrary fluorescence units . ( F ) Mean ± SEM fractional fluorescence changes for each mouse line . Four wild-type mice ( black ) , 6 Emx1-Ai95 mice ( red ) , 4 Emx1-Ai96 mice ( green ) , 4 Emx1-Ai93 mice ( blue ) . Stimulus indicated with black bar . ( G ) Mean change in fluorescence ( △F ) for each mouse line , normalized to the peak amplitude of the fluorescence change in wild-type mice ( △FWT ) . Shaded areas denote ± SEM . ( H ) Time-averaged baseline fluorescence , normalized to wild-type mice . Bears denote mean ± SEM from four wild-type mice , 6 Emx1-Ai95 mice , 4 Emx1-Ai96 mice , 4 Emx1-Ai93 mice . ( I ) Peak fractional fluorescence change . ( J ) Time to peak fluorescence , measured from the onset of the stimulus . DOI: http://dx . doi . org/10 . 7554/eLife . 18372 . 00310 . 7554/eLife . 18372 . 004Figure 1—figure supplement 1 . Expression of GCaMP6 in Emx1-Ai95 , Emx1-Ai96 and Emx1-Ai93 mice under the Emx1-IRES-Cre driver . Images of GCaMP6 fluorescence in 100 µm-thick coronal sections . ( A ) Two-photon image of GCaMP6 fluorescence throughout the depth of cortex in an Emx1-Ai93 mice . White matter is to the left of the image , pia to the right . Three stitched images in a single optical plane . Note the absence of GCaMP6 from the nuclei; we observed nuclear exclusion in all neurons examined ( 271 neurons in layer 2/3 of visual cortex , 1 Emx1-Ai93 mouse ) . ( B ) Widefield images of visual cortices of an Emx1-Ai95 , an Emx1-Ai96 and a wild-type mouse . Same spatial and intensity scale for all images . ( C ) Widefield image from an Emx1-Ai93 mouse . Same spatial scale , but a different intensity scale from the images in B . ( D ) Summary of laminar variations in fluorescence intensity in visual cortex . Bars represent mean ± SEM fluorescence ( arbitrary units ) . n = 4 ( two hemispheres from two mice for each line ) . Emx1-Ai93: asterisks denote significant difference in intensity compared to layer 4 ( p<0 . 05 , two-tailed t-test ) . ( E ) Widefield fluorescence images of three 100 µm-thick coronal sections from an Emx1-Ai95 mouse . ( F ) Widefield fluorescence images of three coronal sections from an Emx1-Ai93 mouse . Approximate of cortical areas ( arrow heads ) were derived from Franklin and Paxinos ( 2007 ) . M motor cortex , S1a anterior primary somatosensory cortex ( likely jaw and forepaw representations ) , S1p posterior primary somatosensory cortex ( likely trunk and hindpaw representations ) , S1v vibrissal primary somatosensory cortex , S2 secondary somatosensory cortex , V visual cortex , A auditory cortex . ( G ) Summary of areal variations in fluorescence intensity . Bars represent mean ± SEM fluorescence ( arbitrary units ) . n = 4 ( two hemispheres from two mice for each line ) . In Emx1-Ai95 mice , GCaMP6 was evenly distributed across the layers of neocortex ( Figure 1B , D ) and across neocortical areas ( Figure 1E , G ) . Fluorescence displayed laminar and areal variations in Emx1-Ai93 mice ( Figure 1C , D , F , G ) , likely due to use of the CaMK2a-tTA line to enhance expression of GCaMP6 ( Mayford et al . , 1996; Krestel et al . , 2001 ) ; Allen Mouse Brain Connectivity Atlas , Transgenic Characterization http://connectivity . brain-map . org/transgenic ) . DOI: http://dx . doi . org/10 . 7554/eLife . 18372 . 004 To measure response amplitudes and kinetics , we presented mice with a brief visual stimulus consisting of a white circle on a black background ( diameter 20 degrees; center location 60° azimuth , 0° altitude; 50 ms duration ) . In wild-type mice , brief stimuli evoked a decrease in fractional fluorescence of −2 . 3 ± 0 . 3% △F/F ( four mice , Figure 1F ) . In all three mouse lines we observed transient activation of visual areas ( Figure 1C-E ) that was larger and faster than autofluorescence signals in wild-type mice ( Figure 1F , G , I , J ) , followed by a decline in fluorescence that may result from vasodilation ( Pisauro et al . , 2013 ) . In Emx1-Ai93 mice , brief stimuli evoked a peak fractional fluorescence change of 10 ± 2 . 4% △F/F ( range 5 . 7–15 . 0% △F/F , 4 mice , Figure 1F ) and a mean peak change in absolute GCaMP6 fluorescence that was 216 times the peak fluorescence change observed in wild-type mice ( Figure 1G ) . All GCaMP6 mice displayed kinetics that were faster than the autofluorescence signal in wild-type mice ( latency to peak from stimulus onset of 252 ± 49 ms , 6 Emx1-Ai95mice; 414 ± 60 ms , 4 Emx1-Ai96 mice; 219 ± 60 ms , 4 Emx1-Ai93 mice; 1 . 37 ± 0 . 3 s , four wild-type mice; mean 10–90% rise and decay times 71 and 536 ms in Emx1-Ai95 mice , 225 and 1151 ms in Emx1-Ai96 mice , 69 and 752 ms in Emx1-Ai93 mice ) . In summary , all three Emx1-GCaMP6 mouse lines displayed brighter fluorescence and faster fluorescence changes than wild-type mice with the result that the changes in fluorescence were dominated by GCaMP6 . To generate retinotopic maps , we employed a spherically-corrected checkerboard visual stimulus drifting across the visual field at 0 . 043–0 . 048 Hz ( Figure 2A ) ( Kalatsky and Stryker , 2003; Marshel et al . , 2011 ) and mapped retinotopy in Emx1-GCaMP6 mice that were head-restrained and free to run on a rotating disk . The mean fluorescence change was greatest in visual cortex ( Figure 2B ) , where the signal-to-noise ratio was sufficient to identify visually-evoked changes in fluorescence in individual trials ( Figure 2C ) . To generate retinotopic maps , we averaged 10–40 presentations of the stimulus in each of the four cardinal directions , thereby reducing the potential effects of ongoing activity ( Figure 2D; Video 1 ) . We generated azimuth and altitude position maps ( Kalatsky and Stryker , 2003 ) that included retinotopic locations across the visual field from approximately −20 to +30 degrees in altitude and −10 to +90 degrees in azimuth ( Figure 2E , F ) . As described previously ( Sereno et al . , 1994; 1995; Garrett et al . , 2014 ) , phase maps were consolidated into a visual field sign map ( Figure 2G ) . 10 . 7554/eLife . 18372 . 005Video 1 . Example fluorescence movies from retinotopic mapping experiment . Example of fluorescence changes during retinotopic mapping . Left: response to checkerboard travelling from the lower to the upper visual field . Right response to checkerboard travelling from the nasal to temporal visual field . Each movie is the mean change in fluorescence ( △F ) of 40 trials . DOI: http://dx . doi . org/10 . 7554/eLife . 18372 . 00510 . 7554/eLife . 18372 . 006Figure 2 . Example of a GCaMP6 fluorescence-based retinotopic map . Example of a GCaMP6 fluorescence-based retinotopic mapping data set from an awake mouse , generated with the drifting checkerboard stimulus on a grey background . ( A ) A single image from the visual stimulus movie used to map in the nasal to temporal ( azimuth ) direction in an Emx1-Ai96 mouse . The checkerboard pattern was swept from left to right ( arrow ) on a grey background . ( B ) Greyscale image illustrating the amplitude of the fluorescence change at 0 . 043 Hz during azimuth mapping , normalized to the maximum amplitude in the image . Dashed white circle approximates the border of the cranial window . ( C ) Spatial average fluorescence during nasal ( N ) to temporal ( T ) mapping from the region of interest outlined in green in panel B . The timing of stimulus presentation is indicated below , where the black segments indicate the center position of the checkerboard bar from −14 to 132 degrees ( azimuth ) and the grey segments indicate that no stimulus was on the monitor . ( D ) Fractional fluorescence changes from the three regions marked in panel B . Each trace is the average of 10 presentations of the stimulus . ( E and F ) Altitude and azimuth maps for the same cranial window . ( G ) Field sign map derived from the altitude and azimuth maps . ( H ) The result of automated border identification , drawn on a brightfield image of the brain surface over visual areas . Named visual areas were identified manually , based on published maps of visual areas . DOI: http://dx . doi . org/10 . 7554/eLife . 18372 . 00610 . 7554/eLife . 18372 . 007Figure 2—figure supplement 1 . Schematic summary of border identification routine . Summary of analysis steps for the generation of borders from altitude and azimuth maps . The left column summarizes the steps from altitude and azimuth maps to the sign map . The right column summarizes generation of borders from the sign map . Image names match those in the example Python notebook available at https://github . com/zhuangjun1981/retinotopic_mapping Blue text indicates the main variables employed at each step . Variables are described in the example notebook . DOI: http://dx . doi . org/10 . 7554/eLife . 18372 . 00710 . 7554/eLife . 18372 . 008Figure 2—figure supplement 2 . Effects of sign map threshold on patch size , shape and visual coverage . ( A ) Comparison of borders ( black lines ) , calculated using threshold values ( variable signMapThr in analysis code ) of 0 . 2 , 0 . 3 and 0 . 4 . For each panel , borders were calculated ( and are overlaid onto ) the mean sign map ( Figure 3C ) , with all variables within the analysis code held constant except the sign map threshold . Thresholds of 0 . 2 and 0 . 4 are the limits of the range of values employed in the analysis of our data sets and a threshold of 0 . 3 is close to the mean value ( mean = 0 . 32 ) . ( B ) Overlaid images of patches . Patches are shown in black , with the result that invariant borders are readily visible in white . Note that borders between two patches are generally stable unless the threshold is raised enough to eliminate the border . For example , all the borders of V1 are virtually invariant across the threshold range from 0 . 2 to 0 . 4; borders between RL and LM and between LM and P are unaffected by the change in threshold from 0 . 2 to 0 . 3 , but are absent at a threshold of 0 . 4 . ( C ) Borders overlaid in color to emphasize the sensitivity the exterior borders of the map to changes in threshold . Exterior borders typically retract as the threshold is raised . ( D ) Effects of threshold on the visual coverage of each patch . For most patches , the change in coverage with threshold is mild and is less than the difference in coverage between patches . As expected , the change in coverage is more pronounced for patches on the periphery of the map that display weaker changes in fluorescence to the visual stimulus , such as patches LLA and RLL . DOI: http://dx . doi . org/10 . 7554/eLife . 18372 . 00810 . 7554/eLife . 18372 . 009Figure 2—figure supplement 3 . Eye movements and pupil area during retinotopic mapping . ( A ) Example of pupil position and area in an anesthetized mouse during azimuth mapping on a black background on which the stimulus moved in the nasal-to-temporal direction . Stimulus is shown schematically ( top left ) . Images of the pupil at time points 1 and 2 are overlaid with circles fit by analysis software that locates the reflection of the infrared LED ( inner , black circle ) and the perimeter of the pupil ( outer , white circle ) . Pupil position in horizontal and vertical dimensions were measured relative to the mean position , with ten individual traces in grey and the mean in black . Stimulus position is indicated as in Figure 4 . ( B ) Example of pupil position and area in an awake mouse during azimuth mapping on a black background . ( C ) Example of pupil position and area in an awake mouse during azimuth and altitude mapping on a grey background of 50% of maximum luminance . ( D ) Mean pupil location and area during grey-background retinotopic mapping with all four stimulus directions . Mean ( black line ) ± SEM ( grey area ) results from 11 mice . Horizontal lines indicate mean horizontal or vertical pupil position . DOI: http://dx . doi . org/10 . 7554/eLife . 18372 . 00910 . 7554/eLife . 18372 . 010Figure 2—figure supplement 4 . Comparison of maps with and without eye movements . Maps from an Emx1-Ai96 mouse , generated after sorting trials into those with and those without eye movement of two degrees or greater . ( A ) Eye position for individual trails , sorted by stimulus direction . ( B ) Resulting sign maps and border locations . ( C ) Comparison of patches . Borders between visual areas ( e . g . between V1 and LM/RL ) were largely unaffected by eye movements . The exterior borders of the sign map display greater differences , either as a result of eye movements or of averaging across the relatively small number of trials without eye movements . DOI: http://dx . doi . org/10 . 7554/eLife . 18372 . 01010 . 7554/eLife . 18372 . 011Figure 2—figure supplement 5 . Comparison of fluorescence changes and V1 coverage with stimuli on black and grey backgrounds . ( A ) Fractional fluorescence changes from three regions in V1 ( see Figure 2B ) from mice mapped under three conditions: mouse anesthetized , stimulus on black background; mouse awake , stimulus on black background; and mouse awake , stimulus on grey background . Each trace is the average of 10 presentations of the stimulus . Dashed vertical line marks the initial fluorescence transient following appearance of the stimulus at the nasal edge of the monitor . The timing of stimulus presentation is indicated below . When presented on a black background , the initial appearance of the checkerboard stimulus evoked an increase in fluorescence across much of cortex ( and a decrease in pupil size; see Figure 2—figure supplement 3A , B ) . This global fluorescence transient could shift the calculated phases of the peak responses , resulting in miscalculation of retinotopic position . ( B ) Comparison of coverage for V1 in one mouse , mapped with black and grey backgrounds . The apparent coverage of V1 was reduced when the stimulus was presented on a black background . The coverage was 3481 degrees2 with black background and 4032 degrees2 with grey background . ( C ) Population average of V1 coverage for 11 mice ( mean ± SEM , black and grey bars indicate black and grey backgrounds , respectively , p<0 . 05 , two-tailed t-test ) . Mapping using a black background resulted in a 21% reduction in coverage of V1 relative to mapping with a grey background ( coverage 2902 ± 349 degrees2 with a black background and 3655 ± 313 degrees2 with a grey background , 11 mice , p<0 . 05 , paired t-test ) . DOI: http://dx . doi . org/10 . 7554/eLife . 18372 . 011 The borders between field sign patches ( Figure 2H ) were identified using a numerical routine ( Figure 2—figure supplement 1 ) . Briefly , the sign map was lightly filtered and thresholded to create an initial parcellation of cortex into patches . The threshold was tuned manually , with little change in the incidence , shape , border locations and area of most patches over the range of threshold values employed ( 0 . 2–0 . 4 , mean ± SEM of 0 . 32 ± 0 . 004 , 14 mice; Figure 2—figure supplement 2 ) . Patches were split and merged to ensure that neighboring patches with the same sign had redundant representation of visual space of ≤10% ( see Materials and methods ) . If sufficiently large and stereotyped , changes in pupil location or size might confound generation of accurate retinotopic maps in awake mice . During mapping , we observed abrupt eye movements of up to ~5–10 degrees vertically and up to ~15–20 degrees horizontally ( Figure 2—figure supplement 3C ) . Movements occurred during all phases of the stimulus and mean displacement was <~1 degree ( Figure 2—figure supplement 3D ) , leading us to conclude that movement-related effects were eliminated by averaging in our experiments ( Figure 2—figure supplement 4 ) . To minimize systematic changes in pupil size , we employed a stimulus with a constant mean luminance , delivering the moving checkerboard pattern on a grey background with 50% of the maximum luminance ( Figure 2—figure supplement 5 , Videos 2 , 3 ) . Hence , in our experiments maps were largely unaffected by eye movements or changes in pupil size . 10 . 7554/eLife . 18372 . 012Video 2 . Lower-to-upper visual field visual stimulus . Lower to upper visual field moving checkerboard stimulus used for retinotopic mapping . DOI: http://dx . doi . org/10 . 7554/eLife . 18372 . 01210 . 7554/eLife . 18372 . 013Video 3 . Nasal-to-temporal visual stimulus . Nasal-to-temporal moving checkerboard stimulus used for retinotopic mapping . DOI: http://dx . doi . org/10 . 7554/eLife . 18372 . 013 Twelve discrete areas have been identified in mouse visual cortex: primary visual cortex ( V1 ) , lateromedial area ( LM ) , rostrolateral area ( RL ) , anterior area ( A ) , anteromedial area ( AM ) , posteromedial area ( PM ) , medial area ( M ) , posterior area ( P ) , postrhinal area ( POR ) , laterointermediate area ( LI ) , anterolateral area ( AL ) , and laterolateral anterior area ( LLA ) ( Dräger , 1975; Wagor et al . , 1980; Olavarria et al . , 1982; Olavarria and Montero , 1989; Wang and Burkhalter , 2007; Garrett et al . , 2014 ) . Each area contains one retinotopic map and therefore appears in field sign maps as a single , distinct , positive or negative field sign patch ( Sereno et al . , 1994 , 1995 ) . Across mice these patches are arranged in a stereotyped configuration with consistent positioning relative to each other . To identify regions in our maps objectively , we used the automated image analysis approach described by Garrett et al . ( 2014 ) , which identifies field sign patches that each contain one and only one map of retinotopic space . Most field sign patches were visible after only 15 min of imaging ( Figure 3A; 10 sweeps of the stimulus in each direction ) and maps were stable across imaging sessions ( Figure 3—figure supplement 1 ) . The ability to quickly generate maps that are comparable to those resulting from 1–2 hr of red-wavelength reflectance-based imaging under anesthesia ( Garrett et al . , 2014 ) is a practical advantage of GCaMP6 fluorescence-based mapping . Typically , we further reduced the effects on ongoing activity by presenting 20–40 sweeps of the stimulus in each direction ( imaging for 30–60 min ) . Like previous authors , we found that maps displayed consistent structure across mice ( Figure 3A ) . The arrangement of areas was broadly consistent with published maps of the mouse visual system ( e . g . Wang and Burkhalter , 2007; Marshel et al . , 2011; Garrett et al . , 2014 ) , but differed in several respects . Key differences include additional visual field sign patches , particularly medial to AM and PM ( Figure 3A ) . The narrow strip of tissue between AM/PM and retrosplenial cortex is generally termed MM or V2MM and is considered part of visual cortex ( Wang and Burkhalter , 2007; Franklin and Paxinos , 2007 ) , but its retinotopic structure remains uncharacterized . GCaMP6 fluorescence-based maps revealed consistent structure in MM , with a positive field sign patch medial to AM ( MMA ) and a negative field sign patch medial to PM ( MMP ) . Furthermore , there were one or more positive field sign patches medial to MMA and MMP , likely in retrosplenial cortex . The most consistent was a positive field sign region which our automated segmentation routine generally failed to separate from PM . Finally , in lateral visual cortex , we observed a negative field sign patch anterior and lateral to RL , which we termed RLL . 10 . 7554/eLife . 18372 . 014Figure 3 . Retinotopic organization of mouse visual cortex . ( A ) Field sign maps from six mice , illustrating differences between mouse lines and individual mice . The mouse line and duration of imaging are indicated on each map . Scale bar 0 . 5 mm . ( B ) Mean field sign maps for 4 Emx1-Ai96 and 10 Emx1-Ai93 mice , from 30–75 min of imaging . ( C ) Mean of the Emx1-Ai96 and Emx1-Ai93 field sign maps in panel B , with borders and area labels . ( D ) Map of variance of the visual field sign . Variance was calculated from visual field sign maps from 14 mice , after alignment as described for calculation of the mean field sign map . Whiter areas denote higher variance . Area borders are overlaid in white . ( E ) The probability of mapping different visual areas with GCaMP6 fluorescence in awake mice . Blue and red bars denote areas with negative and positive field signs , respectively . Results were derived from 14 mice ( 4 Emx1-Ai96 , 10 Emx1-Ai93 ) . Mouse numbers: V1 14/14 , LM 14/14 , LI 10/10 , AL 14/14 , LLA 10/13 , RL 14/14 , RLL 6/14 , AM 13/14 , PM 14/14 , MMA 14/14 , MMP 14/14 , M 5/9 , P 14/14 , POR 1/3 , where , for each area , the denominator indicates the number of mice in which the area was visible within the cranial window , determined manually . ( F ) Mean ± SEM fluorescence change for each visual area , derived from the △F/F spectral power . For each map power was normalized to that in V1 . S1 region was drawn manually towards the anterior extent of the cranial window . Mouse numbers: LM 14 , LI 10 , AL 14 , LLA 10 , RL 14 , RLL 5 , AM 12 , PM 14 , MMA 12 , MMP 12 , M 5 , P 14 , S1 14; from 14 mice ( 4 Emx1-Ai96 , 10 Emx1-Ai93 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 18372 . 01410 . 7554/eLife . 18372 . 015Figure 3—figure supplement 1 . Map stability . ( A and B ) Field sign maps generated from two Emx1-Ai96 mice , one at postnatal days 116 and 203 and a second at postnatal days 97 and 181 . ( C and D ) Images of surface vasculature , with border positions overlaid in white . ( E and F ) Comparison of the area of each patch across imaging sessions . DOI: http://dx . doi . org/10 . 7554/eLife . 18372 . 015 MMA , MMP and RLL appear in the mean field sign map , which summarizes the locations of field sign patches that mapped consistently across 14 Emx1-Ai96 and Emx1-Ai93 mice ( Figure 3C ) . Across these 14 mice , the maximum numbers of patches in an individual map were nine for positive and seven for negative field sign patches . Maps from Emx1-Ai96 and Emx1-Ai93 mice were similar ( Figure 3B ) , with 11 . 75 field sign patches in Emx1-Ai96 ( range 11–14 , 4 Emx1-Ai96 mice ) and 12 field sign patches in Emx1-Ai93 mice ( range 9–14 , 10 Emx1-Ai93 mice ) . Our results indicate that there is retinotopic organization in regions of the mouse cortex in which retinotopy had not been reported . In our mouse lines , GCaMP6 is present in neuronal somata and dendrites ( Figure 1—figure supplement 1 ) and presumably in axons . Accordingly , some of these new field sign patches may result from retinotopically-organized projections originating in surrounding visual areas . Since our results do not indicate whether somata in these regions are retinotopically-organized , we refer to these regions of extended retinotopic organization as 'patches' rather than visual 'areas' . We performed further calculations to characterize and visualize mouse-to-mouse variability and test the consistency of each field sign patch . To visualize mouse-to-mouse variability , we calculated a variance map in which the intensity of each pixel represents the variance of the field sign ( Figure 3D ) . Variance was lower towards the centers of field sign patches and higher towards the borders , but there was little difference in the mean variance per unit area between areas ( not shown ) , except for V1 , which displayed low variance per unit area . To quantify mouse-to-mouse consistency in area identification , we calculated the incidence of each field sign patch ( Figure 3E ) . 10 field sign patches ( V1 , LM , LI , AL , RL , AM , PM , MMA , MMP , P ) occurred in >85% of Emx1-Ai93 and Emx1-Ai96 mice ( ≥12 of 14 mice ) . LI and LLA were frequently along the lateral edge of the cranial window , which may account for the lower incidence of LLA in our maps ( 10 of 13 mice ) . Similarly , M was near the postero-medial edge of the window and was observed in 5 of 9 mice ( in 4 of 7 Emx1-Ai93 and 1 of 4 Emx1-Ai96 mice ) . POR was absent from our maps , probably because it was invariably outside the cranial window . Of the areas that were within the cranial window , the anterior patches A and RLL occurred least consistently . Area A is a putative negative field sign region medial to RL ( Wang and Burkhalter , 2007 ) which maps inconsistently with red-wavelength reflectance and autofluorescence imaging in anesthetized mice ( Marshel et al . , 2011; Garrett et al . , 2014; Tohmi et al . , 2014 ) . We observed a negative field sign patch in this location in 1 of 14 mice and a positive field sign patch ( which segmented separately from RL ) in 3 of 14 mice ( 2 of 10 Emx1-Ai93 mice and 1 of 4 Emx1-Ai96 mice ) . The presence of area A in GCaMP6 fluorescence maps from only 1 of 14 mice probably reflects the difficulty of mapping this area , in which projections from V1 are diffuse and their topographic organization appears weaker than that of many other areas ( Wang and Burkhalter , 2007 ) . RLL was observed in 60% of mice ( 3 of 10 Emx1-Ai93 and 3 of 4 Emx1-Ai96 mice ) . In summary , of the three new field sign patches , MMA and MMP appeared consistently across mice , whereas RLL was less consistent . We further tested the consistency of location of each patch in the mean sign map using k-means cluster analysis , as described previously ( Garrett et al . , 2014 ) . Cluster metric ( Ck ) values were LM 0 . 23; LI 0 . 42; AL 0 . 36; LLA 1 . 17; RL 0 . 42; RLL 1 . 29; AM 0 . 77; PM 0 . 63; MMA 0 . 69; MMP 1 . 0; P 0 . 39 . ( Maps were aligned to the centroid of V1 . ) Of the three new patches , MMA and MMP were each associated with a tightly-packed cluster of patch centroids that differed from a shuffled distribution . In contrast , and consistent with its relatively low incidence , RLL was associated with a non-significant cluster ( Ck = 1 . 29 ) . To summarize our results for the three new field sign patches , MMA and MMP mapped consistently: they appear on the mean field sign map , were present in almost every mouse and their locations relative to other field sign patches were sufficiently consistent that they were associated with a significant k-means cluster . In contrast , RLL was consistent enough to appear on the mean field sign map , but was present in only 6 of 14 mice and was associated with a non-significant k-means cluster . Our ability to map additional regions of retinotopic organization likely results from the superior signal-to-noise ratio of GCaMP6 fluorescence-based maps relative to intrinsic signals . Consistent with this hypothesis , the amplitudes of fluorescence changes were smaller in MMA , MMP , RLL and LLA than for other visual field sign patches ( Figure 3F ) . Of these four patches , LLA was first mapped recently with red reflectance-based imaging and extensive averaging ( Garrett et al . , 2014 ) and the remaining three have now been revealed with GCaMP6 fluorescence mapping . RLL exhibits the smallest mean change in fluorescence of any of the visual regions within our maps ( Figure 3F ) and this weak activation , and resulting low signal-to-noise ratio , likely accounts for the inconsistency of RLL in our maps ( Figure 3E ) . The new field sign patches identified in our GCaMP6 fluorescence-based maps extend far beyond the borders of V1 , in some cases by >2 mm . It is likely that some of these newly-mapped patches extend into architectonically-defined cortical areas surrounding visual cortex . For example , AM and PM are separated from retrosplenial cortex by V2MM . This narrow strip of tissue would appear large enough to accommodate MMA and MMP , but not the long medial extension of PM , which is likely , therefore , to be within retrosplenial cortex . Similarly , the architectonically-defined borders of V1 and barrel cortex are separated by a thin band of tissue , raising the possibility that RLL , and perhaps RL , might extend into barrel cortex . To compare retinotopic and chemoarchitectonic borders , after retinotopic mapping we processed tissue for cytochrome C oxidase staining . We labeled surface vasculature with a fluorescent dye by transcardial perfusion immediately before fixation , used the vasculature to align images acquired in vivo ( Figure 4A; Figure 4—figure supplement 1 ) , after fixation ( whole mount , Figure 4B ) , after flattening ( Figure 4C ) and after sectioning parallel to the cortical surface and staining ( Figure 4D ) and compared retinotopic and chemoarchitectonic borders from 4 Emx1-Ai96 mice ( Figure 4G–J ) . The results confirm that the retinotopic map extends into primary somatosensory cortex and retrosplenial cortex ( Figure 4G–I ) , with RLL mapping to the posterior whiskers ( posterior barrels ) of barrel columns B and C . 10 . 7554/eLife . 18372 . 016Figure 4 . Registration of functional retinotopic maps to chemoarchitectonic borders . ( A–D ) Images from key stages in the processing of tissue from an Emx1-Ai96 mouse , each aligned to the cytochrome C oxidase ( CO ) image . ( A ) Brightfield image of surface vasculature with overlaid field sign map . ( B ) Fluorescence image of whole-mount brain , after perfusion , in which a subset of the surface vasculature is labeled with DyLight 649-lectin conjugate . ( C ) fluorescence image of the flattened cortex . D: brightfield image of a section through layer four after CO staining . ( E ) Overlaid fluorescence images of surface vasculature in whole-mount ( red , panel B ) and after flattening ( green , panel C ) . ( F ) Overlaid images of the surface vasculature and CO staining in posterior barrel cortex and anterior V1 . The contrast of the vasculature image is inverted for clarity . Arrowheads indicate small , circular regions that do not stain for CO and likely result from transverse cuts through ascending/descending vessels . Note the alignment of these putative vessels with likely locations of ascending/descending vessels in the fluorescence image of surface vasculature . ( G ) Field sign map ( panel A ) aligned to chemoarchitectonic borders from the CO image ( panel D ) . Borders of primary visual cortex , auditory cortex , and of barrels in primary somatosensory cortex ) were drawn manually . Barrels in putative columns B and C are shaded grey . ( H–J ) Alignment of functional retinotopic maps and chemoarchitectonic borders for three additional Emx1-Ai96 mice . DOI: http://dx . doi . org/10 . 7554/eLife . 18372 . 01610 . 7554/eLife . 18372 . 017Figure 4—figure supplement 1 . Vessels common to images from live and fixed tissue . Example of vessel tracing through the series of images used for registration of sign maps to fixed tissue . Major vessels common to multiple images are traced on the images of Figure 4 . Green lines mark vessels that are present in all four images . Blue lines mark vessels that extend beyond the borders of the cytochrome oxidase image , but are visible in the other images . Note that not all vessels are labeled after fixation , with the result that some large vessels are visible only in the in vivo image . DOI: http://dx . doi . org/10 . 7554/eLife . 18372 . 017 Another prominent feature of the comparison between cytochrome oxidase-based chemoarchitectonic and GCaMP6-based functional maps is a mismatch in the location of the lateral border of V1 . The lateral border of chemoarchitectonically-defined V1 runs through functionally-defined areas RL and LM . The distance between the chemoarchitectonically-defined lateral border of V1 and that defined by functional retinotopy , measured along the LM/RL border , was 312 ± 88 µm ( 4 Emx1-Ai96 mice ) . One possible explanation for the mismatch is that the cytochrome oxidase-rich region of visual cortex extends beyond the reversal in retinotopy that defines the functional border of V1 . Alternatively , the apparent mismatch in border locations might be an experimental artifact arising from misalignment of the maps during fixation and subsequent histological processing , but misalignment on this scale is unlikely given our fluorescent marker-based registration process and direct alignment of images from live and fixed tissue . To further clarify the alignment of functional and architectonic boundaries , we identified cytoarchitectonic borders in Rorb-Ai93 mice . Rorb drives expression preferentially in layer four and upper layer five pyramidal neurons , leading to stronger labeling of primary sensory areas than of surrounding cortex ( http://connectivity . brain-map . org/transgenic/search ? page_num=0&page_size=29&no_paging=false&search_type=line-name&search_term=Rorb-IRES2-Cre ) . Resting GCaMP6 fluorescence was greater in primary sensory areas than in surrounding regions ( Figure 5A ) , enabling cytoarchitectonic borders to be identified by thresholding GCaMP6 fluorescence images ( Figure 5B ) . Consequently , GCaMP6 fluorescence-based retinotopic maps were inherently aligned to cytoarchitectonic borders in Rorb-Ai93 mice , eliminating the potential for misalignment resulting from tissue processing . Comparison of cytoarchitectonic and functional retinotopic borders again indicated a mismatch , with the lateral border of cytoarchitectonically-defined V1 being lateral of the functionally-defined border and running through functionally-defined areas RL and LM ( Figure 5C , D ) . The distance between the cytoarchitectonically-defined lateral border of V1 and that defined by functional retinotopy , measured along the LM/RL border , was 120 ± 38 µm ( 10 Rorb-Ai93 mice ) . 10 . 7554/eLife . 18372 . 018Figure 5 . Registration of functional retinotopic maps to cytoarchitectonic borders . ( A ) GCaMP6 fluorescence image from a Rorb-Ai93 mouse . Primary sensory areas are marked: S1v barrel cortex , V1 primary visual cortex . ( B ) Image in panel A after filtering and semi-automated identification of major cytoarchitectonic borders . ( C ) Mean retinotopic map/cytoarchitectonic border registration for 10 Rorb-Ai93 mice . Retinotopic maps and cytoarchitectonic borders were pooled across mice as described in the Materials and methods . Cytoarchitectonic borders are shown in black . ( D ) Mean sign map with patch notation , from Figure 3C . ( E ) Mean fluorescence image from 10 Emx1-Ai93 mice , after filtering , alignment and semi-automated identification of the borders of primary sensory areas and retrosplenial cortex ( RS ) . ( F ) Mean map/border registration for 10 Emx1-Ai93 mice . DOI: http://dx . doi . org/10 . 7554/eLife . 18372 . 018 A similar change in resting GCaMP6 fluorescence was observed at the borders of primary sensory areas in Emx1-Ai93 mice ( Figure 1—figure supplement 1F , G ) . The change in fluorescence was less pronounced than in Rorb-Ai93 mice , but borders were visible after averaging images of the resting GCaMP6 fluorescence from 10 Emx1-Ai93 mice ( Figure 5E ) . Here again , the lateral border of cytoarchitectonically-defined V1 was lateral of the functionally-defined border of V1 ( Figure 5F ) , by 236 µm ( measured along the LM/RL border ) . Three methods have been particularly influential in the generation of established maps of the mouse visual cortex: architectonics , using various stains , including cytochrome oxidase; functional retinotopy , primarily measured with electrodes; and projection-based retinotopy , in which axonal projections between visual areas were used to identify locations with matching retinotopy . Having measured the mismatch between borders based on architectonics and on functional retinotopy , we next sought to determine whether borders based on projection-based retinotopy more closely match functional retinotopic borders or architectonic borders , and particularly to determine whether there is a mismatch between the lateral border of V1 from projection-based retinotopy and that based on architectonics . To investigate the alignment of architectonic and projection-based retinotopic borders , we used data from the Allen Mouse Brain Connectivity Atlas ( Oh et al . , 2014; http://connectivity . brain-map . org/ ) . We selected projection data from 99 mice , each with a single injection of anterograde fluorescent tracer into V1 . These injections , and the resulting projection maps , were registered to a three-dimensional reference atlas of the mouse brain built from serial section image data sets from 1675 mice . The reference atlas includes tissue autofluorescence images and in the top projection , the major cytoarchitectonic areas ( including primary visual cortex , whisker and digit barrels in primary sensory cortex , primary auditory cortex and retrosplenial cortex ) are readily visible ( Figure 6A ) due to the enhanced autofluorescence of regions with increased myelination . 10 . 7554/eLife . 18372 . 019Figure 6 . Retinotopy of projections from V1 . ( A ) Locations of injections into V1 in 99 mice , selected from the Allen Brain Connectivity Atlas . Each point indicates an injection . Injection locations were registered to a 3D model of the mouse brain generated from 1675 brains and injections are illustrated on a top projection of the mean autofluorescence from the 3D model . Variations in autofluorescence clearly delineate major architectonic boundaries , including barrels in primary somatosensory cortex , primary auditory cortex , primary visual cortex and retrosplenial cortex . ( B and C ) Projection-based maps of connectivity with ipsilateral V1 . Colors indicate the distance from the geometric center of V1 ( black circle ) from which the strongest projection arises , along the anterior ( a ) – posterior ( p ) and medial ( m ) – lateral ( l ) axes of V1 . These maps are projection-based homologues of the azimuth and altitude maps generated from functional mapping ( Figure 4E , F ) . ( D ) Projection sign map generated from the maps in panels B and C . ( E ) Automated borders generated from the projection sign maps of panel D , overlaid onto the autofluorescence top projection from panel A . ( F ) Subregion corresponding to the box in panel E . A lower threshold was used to generate the visual area borders from the sign map , eliminating the gap between areas V1 , AL , RL and LM . DOI: http://dx . doi . org/10 . 7554/eLife . 18372 . 01910 . 7554/eLife . 18372 . 020Figure 6—figure supplement 1 . Generation of projection-based retinotopy maps . Schematic indicating the steps in generating the projection-based retinotopy map . References are provided where methods are published ( including the collection of the source data set ) . Information on data formats are provided and key steps are illustrated graphically . DOI: http://dx . doi . org/10 . 7554/eLife . 18372 . 020 For each mouse with an injection into V1 , we calculated the density of projections from V1 and combined the results across mice to derive the location within V1 with which each voxel within the brain was most strongly connected ( Figure 6—figure supplement 1 ) . After projecting the maximum connectivity in superficial cortex to the brain surface , we plot two pseudo-colored top-views in which color indicates the location in V1 to which each pixel in the image was most strongly connected ( Figure 6B , C ) . Due to the retinotopic organization of V1 , these two plots are analogous to maps of altitude and azimuth retinotopy ( Figure 3E , F ) . From these plots we generated the projection-based equivalent of a field sign map , a 'projection sign map' ( Figure 6D ) , used our numerical routine to derive area borders ( Figure 6E ) and overlaid these borders onto the autofluorescence top projection from the reference atlas ( Figure 6E ) . Due to the registration of all images in the Connectivity Atlas to the 3D reference atlas , the projection sign map and area borders are inherently aligned . Again , the results indicate that retinotopy extends into primary somatosensory and retrosplenial cortices , with RLL mapping to the posterior whiskers in posterior barrel cortex . LLA also appears to extend into auditory cortex . Regarding the relative positions of the borders of V1 , the myeloarchitectonically-defined lateral border of V1 ( here visible as a transition between relatively bright autofluorescence in V1 and dimmer autofluorescence more laterally ) was lateral to the lateral border of V1 defined by projection-based retinotopy ( Figure 6E , F ) . The distance between the myeloarchitectonically-defined lateral border of V1 and that defined by projection-based retinotopy , measured along the LM/RL border , was 180 µm . The above comparisons of retinotopic maps with architectonic borders using three different methods lead us to two main conclusions . Firstly , retinotopic organization extends into retrosplenial and primary somatosensory cortices in the mouse . RLL is entirely within barrel cortex . The retinotopic map observed in retrosplenial cortex was often continuous with the map in PM , and was not segmented into two distinct regions by our algorithm . The lateral architectonic boundary of retrosplenial cortex extends from the posterior extent of MMP and separates the patch identified as PM into two pieces: a lateral portion that runs parallel to the medial border of V1 and is similar to PM as described previously , and a medial extension within retrosplenial cortex . In subsequent figures we separate these two portions of PM along the approximate lateral boundary of retrosplenial cortex . Secondly , there is a mismatch between the borders of V1 as defined by architectonic markers and by retinotopy , with the architectonically-defined border of V1 lateral to the retinotopic border of V1 by up to ~300 µm ( four measurements: 312 , 120 , 236 and 180 µm; mean 212 µm ) . With what precision can we locate borders with widefield imaging , which presumably reports the mean retinotopy of many neurons at each location ? To answer this question , we compared widefield retinotopic maps with the receptive fields of layer 2/3 pyramidal neurons , measured using 2-photon microscopy in Emx1-Ai93 mice . After mapping cortex , we placed the mouse under a 2-photon microscope , directing the field of view to the V1-LM border . Before performing 2-photon measurements , we generated a local widefield retinotopic map through the microscope objective using an LED and camera ( Figure 7A–D ) , which confirmed that we had located the border region . By mapping through the microscope objective , we ensured that widefield and 2-photon measurements were aligned . 10 . 7554/eLife . 18372 . 021Figure 7 . Widefield borders match single-cell retinotopy at the V1-LM border . ( A , B and C ) Local altitude and azimuth and field sign maps generated under the 2-photon microscope by widefield imaging through the x16 objective . Images were acquired with the objective focused 200 µm below the pial surface of cortex . ( D and E ) Images of surface vasculature acquired with the objective focused on the pial surface . Borders of V1 and LM derived from the field sign map are marked in blue and red , respectively . ( F ) 2-photon fluorescence image acquired with the microscope objective focused 200 µm below the pial surface of cortex . Borders of V1 and LM derived from the field sign map are marked in blue and red , respectively . White outlines indicate 366 somatic ROIs . ( G ) Example fluorescence traces extracted from the three somatic regions ( before neuropil subtraction ) marked with arrowheads in panel F ( grey traces ) and the corresponding neuropil regions ( black traces ) . Fluorescence scale is in arbitrary units . A black horizontal line indicates zero fluorescence for each pair of traces . ( H ) On and Off receptive fields from an example cell , extending from −6 to 60 degrees in azimuth and −24 to 54 degrees in altitude . ( I ) Altitude and azimuth maps ( from summed receptive fields ) for the experiment illustrated in panel F . The color of each soma represents its receptive field center location . Of the 366 somata identified in this field of view , 336 displayed significant receptive fields ( maximum z-score ≥ 2 ) and are illustrated in panel I . Black lines mark the borders of V1 and LM . ( J ) Plots illustrating the distribution of single-cell altitude and azimuth as a function of minimum distance to the V1-LM border . Each point represents a single soma from the field of view illustrated in panel F . ( K ) Single-cell altitude and azimuth as a function of distance to the V1-LM border . Results from three experiments were pooled , yielding 964 somata with receptive fields . Each bar represents the mean and standard deviation of cells binned by distance from the border , in 20 µm bins . DOI: http://dx . doi . org/10 . 7554/eLife . 18372 . 02110 . 7554/eLife . 18372 . 022Figure 7—figure supplement 1 . On and Off receptive fields for an example cell . ( A ) ΔF/F traces displaying On ( red ) and Off ( blue ) responses for each pixel in the nasal receptive field . Each pixel of the stimulus subtended six degrees in altitude and six in azimuth . Each trace represents the mean ( line ) and standard error ( shaded area ) of 60 trials . Horizontal lines indicate zero ΔF/F and vertical line the onset of the stimulus . ( B ) Receptive field maps derived from the example in panel A . Lines represent 40% , 50% , 60% , 70% , 80% and 90% of the maximum z-score . DOI: http://dx . doi . org/10 . 7554/eLife . 18372 . 02210 . 7554/eLife . 18372 . 023Figure 7—figure supplement 2 . Neuropil tuning and subtraction . ( A ) Somatic ROIs ( grey ) and corresponding neuropil ROIs ( black ) for the three cells highlighted in Figure 7F and G . ( B ) Neuropil On and Off receptive fields for the cell illustrated in Figure 7H . ( C ) Plots illustrating the distribution of neuropil altitude and azimuth as a function of distance to the V1-LM border . Each point represents a single neuropil ROI from the field of view illustrated in Figure 7F and can be directly compared to the equivalent plot for somatic tuning in Figure 7J . ( D ) Distribution of the difference in somatic and neuropil tuning for V1 ( blue bars ) and LM ( red bars ) . Black line is the cumulative distribution ( right axis ) . These results indicate that most somata display similar tuning to the local neuropil , with a difference of <5 degrees for >90% of neurons . ( E ) Results of a numerical simulation to illustrate the effectiveness of neuropil subtraction . We examined the effects of adding an additional contaminating signal to each somatic trace . The additional signal was a neuropil trace tuned to 24 . 15 and 24 . 28 degrees in altitude and azimuth . To increase the effect of this additional contamination , we first multiplied the amplitude of the contaminating trace by 10 . From the modified traces , we recalculated the somatic altitude and azimuth maps with and without neuropil subtraction . As expected , in the absence of neuropil subtraction , the added signal shifted all somatic tuning towards the coordinates of the added signal , whereas with neuropil subtraction enabled , the somatic altitude and azimuth maps were almost identical to the unmodified maps ) . Left column: somatic altitude and azimuth maps . Center column: somatic altitude and azimuth maps after addition of enlarged neuropil signal , with no neuropil subtraction . Right column: somatic altitude and azimuth maps after addition of enlarged neuropil trace , with neuropil subtraction . The results of this test indicate that our neuropil subtraction routine corrects for neuropil contamination several times larger than that observed in our results . DOI: http://dx . doi . org/10 . 7554/eLife . 18372 . 023 We then measured the receptive fields of layer 2/3 pyramidal neurons in the same field of view using 2-photon excitation and a sparse noise visual stimulus . In the example illustrated in Figure 7 , we identified 366 neuronal somata in layer 2/3 ( Figure 7F ) and summing three experiments , identified 1276 somata . Neuropil contamination , the presence of fluorescence from surrounding GCaMP-labeled processes in the somatic pixels , is a characteristic of densely labeled tissue . We extracted fluorescence from the neuropil surrounding each neuron and , as expected , found that these local regions of neuropil displayed receptive fields ( Figure 7—figure supplement 2 ) . Many of the neuropil regions displayed similar retinotopy to their parent somata . To correct for neuropil contamination , for each soma we measured and subtracted neuropil fluorescence from the surrounding pixels . Numerical simulations indicated that this approach was effective in removing neuropil contamination from somatic fluorescence measurements and confirmed that the map of somatic retinotopy was not an artifact of neuropil contamination ( Figure 7—figure supplement 2 ) . Most somata displayed On and Off receptive subfields after neuropil subtraction ( e . g . Figure 7—figure supplement 1 ) . Further analysis was performed on cells with a receptive field with a maximum z-score ≥ 2 , which included 92% of neurons ( 336 of 366 ) in the example in Figure 7 and 76% ( 964 of 1276 ) of neurons across three experiments . For each neuron , we summed On and Off subfields and determined the center of the summed receptive field , then created somatic altitude and azimuth maps ( Figure 7I ) . Receptive field centers formed an orderly map of retinotopy with a progression of altitudes along the V1-LM border and azimuth changing parallel to the V1-LM border . Importantly , the gradient in somatic azimuth reversed at the border measured by widefield mapping . For most neurons , somatic altitude and azimuth were similar to the altitude and azimuth in the local widefield maps ( Figure 7A , B ) . For a more quantitative comparison , for each neuron we plot somatic altitude and azimuth as a function of distance from the V1-LM border ( defined by widefield fluorescence; Figure 7J ) . Somata with zero azimuth were close to the border ( distance = 0 ) and azimuth increased approximately linearly with distance from the border , into V1 and into LM . As expected , the gradient of the relationship between somatic azimuth and distance to the border was steeper in LM than in V1 , indicating greater cortical magnification in V1 than in LM near the border . Finally , we pooled results from three experiments , calculating the mean somatic altitude and azimuth as a function of distance from the V1-LM border in 20 µm bins . The relationship displayed an orderly progression of somatic azimuth with distance from the border ( Figure 7K ) . The lowest-azimuth bin was centered at −30 µm , indicating that the single-cell border was 30 µm lateral of the widefield retinotopic border . This 30 µm difference in retinotopic border locations is insufficient to account for the 100–300 µm mismatch between retinotopic and architectonic border locations . In addition to the new field sign patches and medial displacement of the lateral border of V1 , GCaMP6 fluorescence-based retinotopic maps differ from previous maps in two respects . Firstly , area P extends across the posterior border of V1 , from LM to PM . Secondly , area M is displaced medially relative to the location reported in Garrett et al . ( 2014 ) . One result of these differences is an almost continuous ring of field sign positive areas surrounding V1 , broken only by the field sign negative area AM . To assess the representation of visual space in V1 and of the surrounding extrastriate regions in the mouse , we plot the coverage of these regions in retinotopic coordinates ( Figure 8 ) . V1 included representation of the right visual hemifield between ~0 ( the vertical meridian ) and 90° azimuth and ~25–35° above and below the horizontal meridian ( Figure 8A ) . The 4 positive field sign patches around V1 ( LM , RL , PM , P; Figure 8A ) each represented a portion of the visual field and each was biased towards a different quadrant: LM was biased towards the upper nasal visual field , RL the lower nasal field , P upper temporal field and PM lower temporal field ( Figure 8C ) . The summed coverage of these four patches approximated that of V1 with modest overlap in coverage in two narrow strips of visual space ~0–15° above the horizon and ~40–50° from the vertical meridian . The intersection of these zones of overlap included the center of coverage of V1 ( altitude 7 . 4 ± 2 . 1 degrees , azimuth 37 . 8 ± 1 . 4 degrees , 14 mice ) , which was represented in areas LM , RL and PM . This overlap was also observed in maps of eccentricity , where a representation of the center of visual coverage was present for each area ( Figure 9B ) . 10 . 7554/eLife . 18372 . 024Figure 8 . Visual coverage across areas . ( A ) Coverage map of visual space by V1 and surrounding positive field sign patches . Top left , overview map of V1 ( grey ) and the four surrounding positive field sign patches ( red; LM , RL , PM , P ) . Top right , V1 coverage map . Locations represented in V1 are indicated in grey . Circle indicates the center of coverage of V1 at 7 . 4° altitude , 37 . 8° azimuth . Dashed line indicates the horizontal meridian . Center panels , coverage maps of positive field sign patches that border V1 ( LM , RL , PM , P ) . Each coverage map illustrates the one positive field sign patch ( in red ) , overlaid on the coverage of V1 ( black outline ) . Note that coverage of PM excludes retrosplenial cortex . Lowest panel , overlapping coverage of 5 areas ( V1 in grey; LM , RL , PM , P in red ) . ( B ) Coverage of negative field sign patches LI , AL and AM . Coverage of each patch ( blue ) is overlaid on the coverage of V1 ( black outline ) . ( C ) Coverage of remaining positive and negative field sign patches . DOI: http://dx . doi . org/10 . 7554/eLife . 18372 . 02410 . 7554/eLife . 18372 . 025Figure 8—figure supplement 1 . Expanded coverage . Plots illustrating the expansion of visual field coverage , based on published receptive field sizes . ( A ) Upper panels: visual field coverage for areas LM , P , RL and PM ( from Figure 8A ) and expanded coverage ( red shading ) . For each area , coverage expanded by the mean radius of single-cell receptive fields , taken from Wang and Burkhalter ( 2007 ) . The receptive field radius employed is in the lower right corner of each plot . Lower panel: overlapping coverage of LM , P , RL and PM . Circle indicates the center of coverage of V1 at 7 . 4° altitude , 37 . 8° azimuth . ( B ) Coverage and expanded coverage of V1 ( grey ) and of AL , AM and LI . ( C ) Coverage and expanded coverage of MMA and MMP . For MMA and MMP , the receptive field radius employed ( 40° ) was that of MM in Wang and Burkhalter ( 2007 ) . No expanded coverage estimates are provided for LLA , RS and RLL as their receptive field sizes were not provided by Wang and Burkhalter ( 2007 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 18372 . 02510 . 7554/eLife . 18372 . 026Figure 9 . Magnification and representation of visual space across visual cortex . ( A ) Color map of visual space in eccentricity coordinates . Color ( bar ) indicates the distance ( in degrees ) from the center of coverage of V1 ( 7 . 4 ± 2 . 1 degrees altitude , 37 . 8 ± 1 . 4 degrees azimuth ) . ( B ) Eccentricity map of mouse visual areas , using the color scheme indicated in panel A . ( C , D ) Altitude and azimuth contour plots of mouse visual cortex , overlaid on mean field sign borders . Dashed line represents horizontal meridian . Contours are at 5° intervals from −25 to 30° in altitude and 0 to 90° in azimuth . ( E ) Colored sector map of visual space , with the division between upper and lower visual fields at 7 . 4° altitude , 37 . 8° azimuth , which corresponds to the center of coverage of V1 . ( F ) Colored sector map of mouse visual cortex , with colors corresponding to those in panel B and denoting representation of the four quadrants of the visual field . DOI: http://dx . doi . org/10 . 7554/eLife . 18372 . 026 Like the ring of four positive field sign patches , negative field sign patches LI , AL and AM exhibited coverage that overlapped at the center of coverage of V1 ( Figure 8B ) . LI represented the upper visual field at ~30–40° azimuth; AL represented the nasal visual field at ~0–15° altitude; and AM represented the lower visual field at 10–30° azimuth ( Figure 8B ) . Of the field sign patches newly identified by GCaMP6 fluorescence mapping , the positive field sign patches MMA and in retrosplenial cortex exhibited the largest coverage , extending in a ~10° strip ~30–40° along and ~5–10° above the horizontal meridian , respectively , and the negative field sign patches RLL and MMP each represented ≤10° x 20° of the visual field ( Figure 8C ) . Like other patches towards the periphery of the mouse visual cortex ( areas POR and LLA; Garrett et al . , 2014 ) , the newly-identified field sign patches each represented smaller regions of the visual field around the center of coverage of V1 . The limited coverage of most field sign patches is also visible in eccentricity maps , in which representation of the peripheral visual field is limited to V1 and its immediate neighbors ( Figures 8A and 9B ) . Furthermore , neighboring field sign patches generally share a similar bias , with the lower visual field represented primarily in anterior field sign patches and upper visual field in posterior patches , the nasal visual field primarily in lateral patches and temporal visual field in medial patches ( Figure 9C–F ) . In short , almost all higher visual areas in the mouse display a strong bias in representation which is shared with their immediate neighbors , but all share a representation of the center of coverage which corresponds approximately to the center of gaze .
Our results indicate that retinotopic organization is more extensive in mouse cortex than previously appreciated . We observed five instances of extended retinotopy . Firstly , area P extends across the posterior extent of V1 . Tracer injections into V1 have revealed projections into this region in rat and mouse ( Olavarria and Montero , 1981 , 1984; 1989; Coogan and Burkhalter , 1993; Wang et al . , 2012 ) and recent maps have suggested the presence of one or more positive field sign patches posterior to V1 ( Garrett et al . , 2014 ) . The signal-to-noise ratio of our measurements may be particularly limited in this posterior region , along the edge of our cranial window . Furthermore , at its posterior extent cortex is folded and abuts the transverse sinus , and there may be additional retinotopic structure under the sinus or in the folded region that we were unable to access . This additional region may include representation of the center of coverage of V1 , which is absent from our map of coverage of P . We identified two field sign patches in the anterior and posterior aspects of MM , the narrow strip of visual cortex immediately lateral to retrosplenial cortex ( Wang and Burkhalter , 2007 ) . These two patches were labeled MMA and MMP . Variations in the density of neurons immunoreactive for nonphosphorylated neurofilament protein across medial visual areas have led to the suggestion that MM contains multiple distinct regions ( Van der Gucht et al . , 2007 ) . Furthermore , MM receives direct projections from V1 that terminate in distinct anterior and posterior regions ( Wang and Burkhalter , 2007 ) and likely correspond to MMA and MMP . Retinotopy extends into retrosplenial and barrel cortices . Retinotopic organization presumably facilitates the processing of multimodal information in these areas ( Olcese et al . , 2013 ) . For example , like RL ( Garrett et al . , 2014 ) , RLL is biased towards the nasal , lower field and may convey to barrel cortex visual information on whisker location and nearby objects . Similarly , the presence of a visual map in retrosplenial cortex is consistent with its visual responsiveness ( Murakami et al . , 2015 ) and its role in spatial memory and navigation ( Vann et al . , 2009 ) . We used a flickering checkerboard to drive cortical activity during retinotopic mapping , but some areas might be activated more effectively , and mapped more readily , with other visual stimuli . In primates , for example , neurons in higher visual areas are more strongly activated by complex stimuli ( Nassi and Callaway , 2009 ) . In the mouse , some higher areas are preferentially activated by high or low spatial and temporal frequencies ( Marshel et al . , 2011; Andermann et al . , 2011; Roth et al . , 2012; Tohmi et al . , 2014 ) , leading to the suggestion that simple stimuli with different spatial and temporal frequency characteristics might preferentially activate different regions of cortex . However , the checkerboard pattern includes sharp borders between black and white regions of the checkerboard that naturally include a broad band of spatial and temporal frequencies . Furthermore , retinotopically-organized regions occupy most of the territory between primary sensory areas in posterior cortex , suggesting that there are few additional visual areas to be discovered in the mouse . As a result , we would not expect substantial differences in maps of higher visual areas generated with checkerboards with distinct spatial or temporal frequency band characteristics . Our maps are derived from population imaging , which naturally imposes limits on the resolution with which we can map areas and borders . Furthermore , imaging of widefield fluorescence signals at the brain surface is strongly contaminated by vasculature artifacts , which we address by focusing below the surface of cortex . Defocusing has a blurring effect and can limit effective resolution . Widefield fluorescence signals in Emx1 mice are presumably the average of fluorescence from many neurons , with dendrites and axons providing fluorescence signals far from the soma . Processes probably cross regional borders , making it likely that dendritic and axonal signals from neurons on one side of the border contribute to fluorescence on the other side of the border . It is unclear the extent to which border-crossing processes blur borders since blurring will depend on many factors , including whether stimulus-locked changes in the fluorescence of processes more closely match stimulus-locked changes of the parent soma or of the local network into which the processes extend . Given the potential for blurring of borders , it is perhaps surprising that widefield and 2-photon images give border locations which are separated by only a few tens of micrometers . The cellular and laminar origins of visually-evoked changes in widefield GCaMP6 fluorescence remain unknown . GCaMP6 is likely expressed throughout excitatory neurons , including their axons , and it is possible that some of the observed retinotopic organization , such as that in retrosplenial and barrel cortices , results from retinotopically-organized axons . Indeed , the lack of retinotopy in retrosplenial and barrel cortices of Rorb-Ai93 mice suggests a lack of somatic retinotopy in layer four and upper layer five in these regions . We have used the term 'patch' to refer to regions of cortex identified by our analysis routine . We define patches as regions of cortex with retinotopic organization that are distinct from neighboring patches , either because of a reversal of chirality in the visual map or because the neighboring patches contain redundant representations of visual space . We use the term 'patch' to avoid implying any specific mechanistic basis for the retinotopic organization , such as somatic retinotopy . We have used 'region' simply to refer to part of cortex , with no intended implications regarding structure or properties . 'Visual area' is a more established term that implies the consistent identification of borders with multiple approaches , including anatomical and functional measurements ( Orban et al . , 2004; Wandell et al . , 2007 ) . Furthermore , we consider somatic retinotopy necessary in a visual area . In the absence of evidence of somatic retinotopy and supporting evidence from other techniques , we have , where possible , avoided using 'area' to refer to some patches . In maps of mouse visual areas drawn with the assistance of architectonic boundaries , the borders of V1 , AL , LM and RL all intersect at a common point ( e . g . Wang and Burkhalter , 2007 ) . Our results indicate that in retinotopic maps obtained with functional imaging and with projection-based mapping , the lateral border of V1 is more medial than the architectonic boundary . As a result , V1 and AL lack a common border . The territory between V1 and AL is occupied by RL and LM , which share a border extending ~100–300 µm . The architectonic border of V1 runs through RL and LM and typically intersects the medial tip of AL . Hence the medial tip of AL intersects the lateral border of architectonically-defined V1 , but not retinotopically-defined V1 . The border mismatch helps explain why the relative positions of visual areas , and their points of intersection , can differ on maps based on results from different techniques . The mismatch between the architectonic and retinotopic borders of V1 is consistent in our results across three methods , indicating that the architectonic border of V1 may correlate with functions other than retinotopic reversal . One possibility is that the cytochrome oxidase-rich region includes the binocular region around the lateral border of V1 , which would help explain the mismatch since areas RL and LM include representations of the binocular zone ( Wagor et al . , 1980 ) . Another possibility is that there is a thin , largely monocular zone along the medial edge of LM and RL which is cytochrome oxidase-rich ( Laing et al . , 2015 ) . It is unclear whether there is a mismatch in the architectonic and retinotopic borders of other visual areas , such as between LM and LI ( Wang et al . , 2011 ) and further studies will be required to establish the similarity of architectonic and retinotopic border locations for other visual areas . A prominent feature of GCaMP6-based retinotopic maps is the four positive field sign areas ( RL , LM , P and PM ) neighboring V1 that form an almost continuous ring , broken only by the negative field sign area AM . This arrangement resembles early visual areas in primates , in which V1 is a negative field sign region surrounded by V2 , and V2 is a continuous strip of positive field sign tissue broken only at a single point ( anterior tip of the calcarine sulcus; Gattass et al . , 2005; Sereno et al . , 1995; Warnking et al . , 2002; Silver and Kastner , 2009; Wilms et al . , 2010; Laumann et al . , 2015 ) . Early studies of mouse visual cortex revealed a retinotopic map in V1 and a second map anterior and lateral to V1 , leading this second region to be named V2 ( Wagor et al . , 1980 ) . Further studies revealed several overlapping maps within mouse and rat V2 , which was therefore split into several named visual areas , including LM and RL ( Olavarria and Montero , 1981 , 1984 , 1989; Malach , 1989; Olavarria et al . , 1982; Thomas and Espinoza , 1987; Coogan and Burkhalter , 1993; Wang and Burkhalter , 2007 ) . Subsequent studies have documented additional distinctions between the four positive field sign areas neighboring V1 . For example , a study of the laminar termination patterns of axonal projections between cortical areas suggested that the areas around V1 occupy different positions in the hierarchy of visual areas in rats ( Coogan and Burkhalter , 1993 ) . Consistent with this suggestion , medial and lateral regions of visual cortex , likely corresponding to LM , RL and PM , display different response latencies in mice ( Polack and Contreras , 2012 ) , and neurons in these areas differ in their mean receptive field sizes ( Wang and Burkhalter , 2007 ) and in their tuning to the spatial and temporal characteristics of visual stimuli ( Andermann et al . , 2011; Marshel et al . , 2011; Roth et al . , 2012; Glickfeld et al . , 2013; Tohmi et al . , 2014 ) . These studies all support the identification of multiple visual areas in the tissue bordering V1 , in contrast with primates where the tissue neighboring V1 is considered a single visual area ( V2 ) . Our results indicate that the visual field representations of the four areas around V1 ( LM , P , PM , RL ) are complementary , with the outline of the coverage of these four areas matching the outline of V1 coverage . Our results likely underestimate the range of visual coverage in each area and the overlap between areas . V1 included a visual coverage range of ~60° in altitude and ~90° in azimuth , comparable to measurements with reflectance-based imaging ( e . g . Garrett et al . , 2014 ) but smaller than the range measured with single-cell electrical recordings ( ~100° in altitude and ~150° in azimuth; Wagor et al . ( 1980 ) . Single-cell recordings are likely to provide a larger range for two main reasons . Firstly , widefield mapping identifies retinotopic position based on the peak of activity , thereby emphasizing the receptive field center . The outer limits at which visual stimuli can evoke activity in V1 will be expanded , by approximately the radius of the largest receptive fields of neurons at the edges of V1 . A more pronounced expansion in effective coverage is expected in higher visual areas , where receptive field sizes are larger ( Wang and Burkhalter , 2007 ) . Expanded coverage , calculated using published receptive field sizes ( Wang and Burkhalter , 2007 ) is illustrated in Figure 8—figure supplement 1 . For V1 , expanded coverage includes ~70° in altitude and ~100° in azimuth . Mean local receptive field size may change across a visual area , which possibly accounts for the mismatch of coverage at the borders of neighboring areas ( e . g . the V1-LM border , where the outer limits of the estimated V1 and LM coverage differ ) . Secondly , widefield fluorescence measures the average retinotopy of the local population of neurons . Single-cell receptive field centers vary around the mean retinotopy . For example , in V1 most pyramidal neuron receptive fields are within ~7 degrees of the local population average ( Bonin et al . , 2011 ) . How variability relative to the mean local receptive field center changes towards the borders of V1 or in other visual areas is unknown , but an expansion of coverage of 7 degrees at each border ( in addition to the expansion due to receptive field size ) would result in total coverage of V1 of ~85° in altitude and ~115° in azimuth , closer to the numbers from single-cell recordings . These differences between population and single-cell coverages may explain the small apparent regions of coverage of many higher visual areas , in which receptive field sizes and perhaps also the scatter in receptive field centers are greater than in V1 . Even allowing for some expansion of coverage for each visual area , our results indicate that no higher visual areas ( even those immediately surrounding V1 ) contain a complete description of the visual hemifield . Hence information on features that subtend more than approximately a quarter of the visual field will be routed to different early visual areas , with the result that information on features in different locations within a single visual scene will be processed in regions with different functional properties . Further studies will be needed to understand the advantages and limitations of processing information from different locations in visual space in functionally distinct regions of visual cortex .
In this study we employed six mouse lines: Ai93 , Ai95 , and Ai96 are floxed GCaMP6 reporter lines ( Madisen et al . , 2015 ) , which differ in the promoter/enhancer used to drive GCaMP6 expression and in the isoforms of GCaMP6 expressed . Ai95 and Ai96 lines employ a ROSA-CAG promoter to drive Cre-dependent expression of GCaMP6f and GCaMP6s , respectively ( Madisen et al . , 2015 ) . Stronger expression of GCaMP6f is achieved in Ai93 , using the TIGRE promoter . For most experiments reported here , these reporter lines were crossed with Emx1-IRES-Cre mice to drive expression in pyramidal neurons throughout neocortex . Ai95 and Ai96 express GCaMP6 in the presence of Cre and all experiments were performed on double transgenic mice hemizygous for Cre and for GCaMP6 . As Ai93 requires the presence of both Cre and tTA to drive expression , Ai93 mice were crossed with Emx1-IRES-Cre and CaMK2a-tTA to yield triple transgenic mice that were hemizygous for all three genes . We refer to these crosses as Emx1-Ai93 , Emx1-Ai95 and Emx1-Ai96 mice . For a small sub-set of experiments ( Figure 5J–L ) , Ai93 was crossed with Rorb-IRES2-Cre and CaMK2a-tTA to yield Rorb-Ai93 mice . Both sexes of mice were used and all mice were maintained on a B6/C57 background . The pattern of GCaMP6 expression was examined in two mice of each genotype at postnatal day 67–114 ( Emx1-Ai95 P75 , Emx1-Ai96 P67 , Emx1-Ai93 P67 , wild-type P114 ) . Brains were fixed by transcardial perfusion with 4% ( w/v ) paraformaldehyde . 100 µm-thick coronal sections were cut using a vibratome and mounted in Vectastain . Endogenous GCaMP6 was imaged by widefield or 2-photon fluorescence microscopy . Image analysis was performed in ImageJ . Retinotopic imaging was performed through a 5 mm diameter circular cranial window positioned over visual areas of the left hemisphere . The preparation was similar to that described previously ( Andermann et al . , 2010; 2011 ) . Briefly , under isoflurane anesthesia , a head restraint bar was attached to the skull using C & B Metabond ( Parkell ) and a 5 mm craniotomy opened at center coordinates 2 . 7 mm lateral , 1 . 3 mm anterior to lambda . The craniotomy was sealed with a stack of three #1 coverslips , attached to each other using optical adhesive ( Norland ) and to the skull with Metabond . The mouse was permitted to recover for at least seven days and conditioned to the head restraint and running wheel for several days before mapping . All experiments and procedures were approved by the Allen Institute Animal Care and Use Committee . Widefield fluorescence images were acquired with a 1:1 optical relay using two x1 PlanAPO dissecting microscope lenses ( Leica , 10450028 ) . Illumination was from a blue LED ( M470 , Thorlabs ) , via a bandpass filter ( 469/35 , Semrock ) and fluorescence was detected by a CCD camera ( Orca R2 , Hamamatsu ) via a 497 nm dichroic and 525/39 bandpass filter ( Semrock ) . Parts were mounted on a macroscope ( THT scope , Scimedia ) with its optical axis tilted 22 degrees in the coronal plane such that the optical axis was perpendicular to the cranial window . The focal plane of the microscope was positioned deep in cortex , thereby defocusing the surface vasculature during retinotopic mapping . Illumination and image acquisition were controlled with software written by JW using the Hamamatsu Video Capture Library for Labview , v . 2 . 0 . 2 . During imaging , the head was restrained via the implanted bar and the eyes were on a horizontal plane . Visual stimuli were displayed on a 40" LED TV ( Samsung 6300 ) , placed 13 . 5 cm from the right eye . The mouse was oriented with its midline at ~30° to the plane of the monitor . Visual coordinates were calculated with respect to the midline ( azimuth coordinates ) and the horizontal plane through the eyes ( altitude coordinates ) . The monitor covered approximately −10 to 130 degrees in azimuth and −50 to 60 degrees in altitude . The luminance of the stimulus monitor ranged from 0 . 05 ( black ) to 177 ( white ) cd/m2 . Awake mice were free to run on a 16 . 5 cm diameter disk . For imaging under anesthesia , mice were anesthetized with <1% isoflurane ( inhaled ) and chlorprothixene ( 2 . 5 mg/kg , intramuscular ) and silicon oil ( 10 , 000 molecular weight ) was applied to both eyes to prevent dehydration . Retinotopic maps were generated by sweeping a bar across the monitor ( Kalatsky and Stryker , 2003 ) . The bar contained a flickering black-and-white checkerboard pattern , with spherical correction of the stimulus to stimulate in spherical visual coordinates using a planar monitor ( Marshel et al . , 2011; Garrett et al . , 2014; Videos 2 and 3 ) . The pattern subtended 20 degrees in the direction of propagation and filled the monitor in the perpendicular dimension . The checkerboard square size was 25 degrees . Each square alternated between black and white at 6 Hz . To generate a map , the bar was swept across the screen ten times in each of the four cardinal directions , moving at nine degrees per second . To ensure that stimulus-evoked activity had subsided between sweeps , a gap of ≥5 s was inserted between sweeps , resulting in repetition of the stimulus at 0 . 048 Hz for vertically-moving stimuli and 0 . 043 Hz for horizontally-moving stimuli . During mapping , fluorescence images were acquired at 10 Hz with 2×2 binning , resulting in an effective pixel size of 12 . 9 µm at the sample . Across different mouse lines , fluorescence varied ~100 fold for each mouse and illumination intensity was adjusted to almost fill the camera well depth during periods of activity . Mean ± SEM illumination intensity for GCaMP mice was 89 ± 21 µW/mm2 ( range 19–210 µW/mm2 ) . A slight decline in fluorescence was generally observed during mapping , presumably due to photobleaching . Photobleaching was approximately linear with time and intensity , with fluorescence declining at 235% J−1mm2 ( 11 mice ) . At the mean illumination intensity during mapping ( 89 µW/mm2 ) , resting fluorescence declined at a mean rate of 0 . 02% per second or 1 . 25% per minute . Brief stimuli ( Figure 1 ) consisted of a white circle 20° in diameter , 50 ms in duration , centered at 60° azimuth and 0° altitude , on a black background . The white circle was displayed 20 times at 0 . 2 Hz . Images were acquired under 1 . 8 µW/mm2 illumination at ~64 Hz with 8×8 on-chip binning , resulting in an effective pixel size of 51 . 6 µm at the sample . All image analyses were performed in the Python programming environment , with OpenCV/SimpleCV libraries , after subtraction of a camera bias of 100 digitizer units . For retinotopic mapping experiments , our analysis followed the methods described by Garrett et al . ( 2014 ) with minor modifications . We first created △F movies: for each presentation of the checkerboard stimulus , from all frames of the movie we subtracted an image corresponding to the mean of the 2 s before the start of stimulation . We then created a stimulus-triggered mean △F movie for each of the four stimulus directions , averaging 10–40 trials in each direction . To generate azimuth and altitude position maps , from fluorescence versus time data for each pixel we extracted retinotopic positions from the phase of the first harmonic component of the Fourier series , with peak frequencies of 0 . 043 Hz ( 0 . 022–0 . 065 Hz band ) for azimuth and 0 . 048 Hz for altitude ( 0 . 024–0 . 072 Hz band ) maps , corresponding to the periodicity of the stimulus ( moving at 9° per second ) . To cancel the delay from stimulus to response , we calculated a pixel-by-pixel average of the visual response positions derived from movies for stimuli traveling in opposite directions ( for altitude , we averaged bottom-to-top and top-to-bottom movies; for azimuth , nasal-to-temporal and temporal-to-nasal movies ) . Azimuth and altitude maps were combined to generate a visual field sign map ( Sereno et al . , 1994; 1995; Garrett et al . , 2014 ) , where the visual field sign at each pixel is the sine of the angle between the local gradients ( derived with the numpy . gradient function ) in azimuth and altitude . The visual field sign map was converted into borders as described by Garrett et al . ( 2014 ) , Juavinett et al . ( 2016 ) , as outlined in Figure 2—figure supplement 1 . The visual field sign map was spatially filtered with a Gaussian kernel ( standard deviation range 6–10 µm , mean ± SEM of 8 . 14 ± 0 . 08 µm , 14 mice ) then thresholded to create a binary mask . For each map the threshold was tuned manually over a narrow range ( field sign values of 0 . 2–0 . 4 , mean ± SEM of 0 . 32 ± 0 . 004 , 14 mice ) . Each suprathreshold patch was dilated to yield a border width between patches of one pixel . Isolated pixels were eliminated with open/close operations . The binary mask was converted into an initial 'raw' patch map in which each pixel value was −1 , 0 or 1 . Patches were further processed with a split/merge routine in which patches with >10% redundancy in visual coverage were split using a watershed routine at the local minimum of the visual eccentricity map and , subsequently , adjacent patches with the same sign and <10% redundancy in visual coverage were merged . Patches smaller than 0 . 00166 mm2 ( 100 pixels ) were discarded . Python code which accepts altitude and azimuth maps and identifies patch borders is available at https://github . com/zhuangjun1981/retinotopic_mapping and Supplementary file 1 . The code includes an example data set and associated Python notebook illustrating each analysis step . Within the analysis routine , there are 13 variables that are set manually . Further explanation of these variables and the values employed in our analyses are stated in the code . Results from multiple mice were pooled to create mean azimuth , altitude and field sign maps . Maps for each mouse were centered on the centroid of V1 and rotated to align the major axis of the azimuth gradient . Differences in relative positions of mouse and monitor were corrected by defining the V1 , LM , RL border as 0° in altitude and azimuth , necessitating a correction of 0 . 3 ± 1 . 8 degrees in altitude and 14 . 9 ± 2 . 0 degrees in azimuth ( 14 mice ) . Mean azimuth and altitude maps were then calculated by vector summation ( Garrett et al . , 2014 ) . To create patch borders for pooled results from the average sign map ( Figure 3B and C ) , we thresholded the sign map at 0 . 3 and further processed the patches as described above ( split , merge , discard small patches ) . For brief , circular stimuli ( Figure 1F–J ) twenty trials were averaged and Frest was calculated as the mean of the 2 s pre-stimulus period . △F/F movies were generated by subtraction of Frest from all pixel intensities , before division by Frest . Fluorescence was extracted from the region of the image with the greatest peak △F/F ( which was generally V1 ) : after spatial filtering with a Gaussian kernel ( σ = 258 µm , corresponding to five pixels ) to reduce noise , a 516 µm ( 10 pixel ) square region of interest was centered on the brightest pixel in the maximum intensity projection of the movie ( dimmest pixel in wild-type mice , in which fluorescence declined with stimulation ) . Results in Figure 1 report the fluorescence intensity within this region . An image of the right eye was recorded by infrared ( IR ) imaging . The peri-ocular region was illuminated with an 850 nm LED ( Ostar SFH4750 ) and an image of the right eye was reflected by a low-pass dichroic mirror ( Semrock , FF750-SDi02-25×36 ) placed ~3 cm from the eye and between the eye and the stimulus monitor . Distance from the eye to the camera was ~30 cm . Pupil images were acquired via a Microsoft webcam chip through a Tamron CCTV lens ( 23FM50SP , focal length: 50 mm ) and long-pass filter ( Thorlabs FEL0800 ) . Acquisition was at 30 Hz with 320×240 pixels , effective pixel size 18 µm . To calculate pupil position and area , we first located and tracked the corneal reflection of the infrared LED ( the brightest object in the image ) . The LED reflection was masked to prevent it from interfering with pupil detection ( e . g . in the situation that the LED reflection was inside the image of the pupil ) . The image was then blurred and edges were detected and exaggerated using the OpenCV 'Canny edge detection' function ( Canny , 1986 ) . The pupil appeared as a dark , approximately round object . A region containing the pupil was manually selected and all spatially separate dark objects in the selected region were outlined . Each outline was subjected to an 'open' operation and all outlines with an area of less than 0 . 03 mm2 were discarded . For each outline , centroid location , area , average intensity and roundness were calculated and rank order similarity to the pupil in the previous frame was calculated for each parameter . The outline with the smallest summed rank across all parameters was identified as the pupil . Mice blinked periodically . The reflection of the LED was absent from frames in which the eyelid was closed , permitting automated identification of blink events . In the absence of a previous frame containing the pupil ( the first frame of the movie and first frame after a blink ) the pupil was identified as the largest , most circular dark outline , again determined with a minimum rank sum criterion . After pupil identification , pupil area was calculated by counting the number of pixels within the borders of the extracted outline and multiplying by the area of a single pixel ( 3 . 24×10−4 mm2 ) . Pupil motion in retinotopic coordinates ( changes in gaze angle ) was calculated under the assumption that the mouse eye is spherical , with a radius of 1 . 7 mm ( Remtulla and Hallett , 1985 ) . First , for each frame we determined the location of the centroid of the pupil and of the reflection of the IR-LED . From these values we calculated the position of the pupil ( in pixels ) relative to the LED reflection in horizontal and vertical planes ( Xi and Yi , where i represents frame number ) . The cardinal axes of the camera chip , which was mounted parallel to the optical table , were used to define the horizontal and vertical planes of pupil movements . Pupil positions in pixels were converted to changes in azimuth and altitude gaze angles by trigonometry: Δθazi = arcsin ( ΔXi/r ) ; Δθalt = arcsin ( ΔYi/r ) , where ΔXi and ΔYi represent the deviation of X and Y position in the ith frame from the mean location during the movie , and r was set to be 1 . 7 mm ( the average radius of mouse eye ball ) . Detailed protocol for registration can be found at Bio-protocol ( Zhuang et al . , 2018 ) . Retinotopic maps were compared to chemoarchitectonic borders via cytochrome C oxidase ( CO ) -stained tangential sections of flattened cortex from 4 Emx1-Ai96 mice . For each mouse , the retinotopic map was aligned to the bright field image of the surface vasculature within the cranial window . The mouse was transcardially perfused , sequentially , with saline ( 10 ml/min for 10 min ) ; 5 µg/ml DyLight 649-lectin conjugate ( Vector Laboratory , 5 ml/min for 5 min , to label vascular endothelium ) ; 5 min pause; 1% ( w/v ) paraformaldehyde ( PFA ) in PBS ( 5 ml/min for 20 min ) . After perfusion , a fluorescence image of labeled vasculature within the cranial window was acquired ( 640/690 nm excitation emission ) . The left cortex was isolated , flattened between glass slides , and post-fixed overnight in 4% PFA ( Wang and Burkhalter , 2007 ) . The fixed cortical sheet was cut into an asymmetrical shape to aid future alignment of fluorescence and brightfield images of the flattened tissue . A fluorescence image of the surface vasculature was acquired and the tissue was cut tangentially into 50 µm sections . Sections were stained for CO as described previously ( Tootell et al . , 1988; Wang et al . , 2012 ) . Images acquired at different stages of tissue processing were sequentially aligned to the CO image by manual warping using the TrakEm2 plug-in in ImageJ . Key registration steps included the alignment of vasculature images across live , whole-mount and flattened preps ( Figure 4E ) . Alignment of images of surface vasculature and CO-stained tissue , both in the flattened tissue , occurred in two steps: coarse , global alignment was via the edges of the asymmetrically-shaped tissue and fine , local alignment was by matching cross-sections of vertical blood vessels in the CO image to putative entry points of ascending/descending vessels in the fluorescence image ( Figure 4F ) . Alignment was optimized in anterior V1 and barrel cortex , with the likely result that alignment was most accurate near the visual/somatosensory border . Chemoarchitectonic borders were identified manually . Cytoarchitectonic borders were examined in 10 Rorb-Ai93 and 11 Emx1-Ai93 mice . Cytoarchitectonic borders were identified from a fluorescence image of the cortical surface via a semi-automated process . Illumination gradients were first removed by filtering the image with a Gaussian kernel ( σ = 1290 µm ) and subtracting the result . A median filter was used to remove small structures such as blood vessels and edges were detected using Canny edge detection implemented in Python ( OpenCV package ) . From the set of edges , a subset that best matched the borders of primary sensory areas were selected manually . Distances between borders were measured manually , along the axis of the LM/RL border . The projection-based map was derived from analysis of data from the Allen Mouse Brain Connectivity Atlas ( Oh et al . , 2014 ) ; http://connectivity . brain-map . org/ ) . The Allen Mouse Brain Connectivity Atlas is a large data set derived from many mice , each with a single injection of adenoassociated virus that drives expression of GFP , an anterograde tracer . After fixation , each mouse brain in the atlas was imaged via an automated imaging and sectioning microscope , and registered to a 3-dimensional template derived from the mean autofluorescence of 1675 mouse brains ( common coordinate framework v3 , http://help . brain-map . org//display/mouseconnectivity/API ) . Hence the atlas includes the location of each injection site and the distribution of fluorescently-labeled projections from the injection site , with each voxel registered to a three-dimensional template of the mouse brain ( Kuan et al . , 2015 ) . We selected data from 99 mice ( 35 wild-type ( Bl6 ) mice and 64 Cre mice ) , each with an injection into V1 . The data set from each mouse consists of a 3D map of projection density . We processed the data sets through several distinct steps ( Figure 6—figure supplement 1 ) , en route to creating two maps of connectivity anterior-posterior and medial-lateral connectivity which are comparable to maps of altitude and azimuth , respectively . In the first step , we pooled data from 99 mice by creating a weighted 3D map of connectivity , with the entry in each voxel being the center of mass of all source locations weighted by projection strength:L⇀proj= ∑i=1nWi . L⇀i∑i=1nWi where L⇀proj is a vector specifying the estimated source location ( in 3D ) for this target voxel . L⇀ is a 3D vector specifying the center of an injection site . W is a measure of projection strength: W=Ft/Fs Ft is the projection density ( derived from fluorescence intensity ) at the target pixel . Fs is the injection density ( derived from fluorescence intensity ) at the source pixels ( the injection site ) . n = 99 mice The output of this procedure was a 3D map ( a 3D array ) in which each voxel contained a vector ( L⇀proj ) with three entries: the x , y and z positions of the location in V1 from which the voxel received the strongest projection . From this 3D array , we projected L⇀proj to the pial surface of cortex . We first calculated equipotential surfaces , using a Laplace transform , with each point on a surface being equidistant on a normalized scale from pia ( distance = 0 ) to white matter ( distance = 1 ) ( http://help . brain-map . org/download/attachments/2818171/MouseCCF . pdf ? version=1&modificationDate=1432939552497 ) . At each surface location , the voxel with the greatest summed projection strength along a line orthogonal to the equipotential surfaces was identified and its L⇀proj was projected to the surface , where summed projection strength = ∑i=1nWi . Only L⇀proj values from 0 . 1 to 0 . 5 of the normalized cortical depth were projected to the pial surface . The result was a 2D map of L⇀proj entries . This 2D pial surface map was projected to the horizontal plane , yielding a 2D 'top view' of L⇀proj entries . From the L⇀proj top view , we generated two maps: one displaying the anterior-posterior location in V1 from which each pixel received the strongest projection , and the other displaying the medial-lateral-posterior location in V1 from which each pixel received the strongest projection . As vertical and horizontal retinotopy are represented in orthogonal axes in V1 , these two projection maps are comparable to maps of altitude and azimuth . After filtering ( σ = 43 μm ) , we used these two maps to generate the projection-based equivalent of a field sign map , which we term the 'projection sign map' ( also filtered , σ = 130 μm ) . We processed the projection sign map to locate borders and , thereby , draw a projection-based retinotopic map of visual cortex using the same numerical routine employed to derive borders from field sign maps . The first step in deriving borders is to threshold the sign map . For projection-based signs maps we employed a threshold of ±0 . 2 . Since a threshold of ±0 . 2 could leave gaps between visual areas , such as at the intersection of V1 , AL , RL and LM ( e . g . Figure 6E ) , we calculated a second projection-based field sign map using a threshold of ±0 . 1 . The locations of injection sites and the projection-based retinotopic map were displayed on a surface projection of autofluorescence derived from 1675 mouse brains , thereby revealing the registration of injection sites and the projection-based retinotopic map to the borders of major architectonically-defined cortical areas . 2-photon experiments were performed on a Sutter MoM . Before 2-photon imaging , a retinotopic map was generated using widefield GCaMP6 fluorescence , as described above . The mouse was then moved to the 2-photon microscope where a 'local' widefield retinotopic map was generated through the microscope objective using LED illumination , a sCMOS camera and the drifting checkerboard stimulus . The V1-LM border location ( which was later compared with single-cell retinotopy derived from 2-photon imaging ) was derived from the local widefield map . To ensure accurate registration of local widefield and 2-photon images , the local widefield map was generated with the microscope objective focused 200 µm below the pial surface of cortex . The 2-photon data set was collected immediately afterwards without axial or transverse translation of the field of view of the microscope , relative to the preparation . The field of view of the widefield image was greater than that of the 2-photon image . For accurate alignment of the two images , images of the surface vasculature were acquired under widefield and 2-photon illumination and used to guide a rigid transform of the widefield image , which was then cropped to the dimensions of the 2-photon image . 2-photon imaging was performed with 920 nm illumination from a Ti:sapphire laser ( Coherent Chameleon II ) , which was focused onto the prep with a x16/0 . 8 NA objective ( Nikon N16XLWD-PF ) , providing a 720×720 µm field of view . 512×512 pixel images ( 1 . 4 µm per pixel ) were acquired at 30 Hz . Emitted light was collected in the epifluorescence configuration through a 735 nm dichroic reflector ( FF735-DiO1 , Semrock ) and a 490–560 bandpass emission filter ( ET525/70 m-2P , Chroma Technology ) . Image acquisition was controlled using ScanImage software . Single-cell receptive field mapping was performed using a sparse noise stimulus consisting of black and white squares on a 50% grey background in pseudorandom order . Each square ( 6×6 visual degrees , 100 ms duration ) was displayed 60 times per polarity on an LCD monitor ( ASUS PA248Q , mean luminance: 50 cd/m2 ) . Fluorescence was extracted from the 2-photon time series by defining weighted somatic regions of interest ( ROIs ) using a PCA-ICA routine ( Mukamel et al . , 2009 , µ = 0 . 2 ) . A size filter was employed to eliminate ROIs smaller than 59 µm2 or larger than 395 . 5 µm2 . For each somatic ROI , a neuropil ROI was created by dilating the outer border of the somatic ROI by 5 and 15 pixels to define inner and outer limits of the neuropil ROI . The union of somatic ROIs was excluded from all neuropil ROIs . For each neuron we extracted two fluorescence values per time point using the two ROIs: Fmeasured from the somatic ROI and Fneuropil from the neuropil ROI . We then calculated the true somatic fluorescence ( Fc , without neuropil contamination ) assuming linear summation: Fc=Fmeasured+r∗Fneuropil . We estimated r , the contamination ratio , separately for each soma by gradient descendent regression with a smoothness regularization ( http://help . brain-map . org/download/attachments/10616846/VisualCoding_Overview . pdf ? version=1&modificationDate=1465258498093 ) . The mean ± standard deviation value of r for all neurons in our data set was 0 . 244 ± 0 . 189 . After neuropil subtraction , we calculated the stimulus-triggered average fluorescence for each stimulus pixel location and polarity . An example is illustrated in Figure 7—figure supplement 2A ) . For each stimulus location , ΔF/F integrals during the 300 ms following stimulus onset were calculated for white and black squares , yielding On and Off spatial receptive fields respectively . Baseline was defined as mean fluorescence within the 0 . 5 s window before stimulus onset . From On and Off receptive fields we calculated On and Off z-score maps by subtracting the mean of the pixel values in the map and dividing by the standard deviation of the pixel values in the map . z-score maps were smoothed with a Gaussian filter ( σ = 6 degrees ) and up-sampled by a factor of 10 with cubic interpolation . If the maximum of either the On or Off receptive field was greater than two , the neuron was considered responsive to the stimulus and was included in subsequent analyses . To calculate the receptive field center of each cell , for each receptive field map pixels with a z-score below a threshold were set to zero . The threshold employed was 40% of the greater of the maximum On and maximum Off z-scores . Pixels with z-scores above the threshold retained their original z-score values . On and Off thresholded receptive field maps were summed and the altitude and azimuth of the cell’s receptive field location were defined as weighted average coordinates of the combined receptive field . The soma-border distance for each soma was the shortest Cartesian distance between the border and the weighted average coordinates of the soma pixels . | Our eyes send information about the world around us to a region on the surface of the brain called the visual cortex , which is made up of a series of interconnected areas . Researchers have studied the anatomy and activity of these areas to generate maps that show how these areas are arranged . For example , architectonic borders are drawn where there are abrupt changes in the density of cells , or the degree to which they are stained by certain chemicals . Maps based on architectonics and those based on brain activity are thought to exhibit matching borders between visual areas . Zhuang et al . used animaging approach to produce detailed maps of the visual cortex of mice . The approach uses a fluorescent protein called GCaMP6 to indicate levels of activity in the brain while the mice were exposed to visual cues . Furthermore , Zhuang et al . added a second step to this approach to reveal the architectonic borders of the areas in the visual cortex . This made it possible to compare the locations of activity-based and anatomical borders in a single mouse . Zhuang et al . found that maps of the visual cortex based on architectonics do not completely match those based on activity . These findings help reconcile the differences between maps of mouse visual cortex produced by other studies . It is not clear whether a similar mismatch in architectonic and activity-based border locations exists in other animals , such as primates . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"neuroscience"
] | 2017 | An extended retinotopic map of mouse cortex |
Here , we describe that lysine-specific demethylase 1 ( Lsd1/KDM1a ) , which demethylates histone H3 on Lys4 or Lys9 ( H3K4/K9 ) , is an indispensible epigenetic governor of hematopoietic differentiation . Integrative genomic analysis , combining global occupancy of Lsd1 , genome-wide analysis of its substrates H3K4 monomethylation and dimethylation , and gene expression profiling , reveals that Lsd1 represses hematopoietic stem and progenitor cell ( HSPC ) gene expression programs during hematopoietic differentiation . We found that Lsd1 acts at transcription start sites , as well as enhancer regions . Loss of Lsd1 was associated with increased H3K4me1 and H3K4me2 methylation on HSPC genes and gene derepression . Failure to fully silence HSPC genes compromised differentiation of hematopoietic stem cells as well as mature blood cell lineages . Collectively , our data indicate that Lsd1-mediated concurrent repression of enhancer and promoter activity of stem and progenitor cell genes is a pivotal epigenetic mechanism required for proper hematopoietic maturation .
Epigenetic modifications , such as histone lysine methylation , promote or repress gene expression , depending on the specific lysine residue modified , the number of methyl moieties present , and the genomic positioning of the lysine modification ( Jenuwein , 2001; Kouzarides , 2007 ) . While active promoters are typically marked by dimethylation and trimethylation at Lys4 of histone H3 ( H3K4 ) around transcriptional start sites ( TSS ) , enhancer elements are characterized by high levels of H3K4 monomethylation and low levels of H3K4 trimethylation ( Heintzman et al . , 2007; Koch et al . , 2007 ) . The regulation of lysine methyl modifications is a dynamic process , tightly controlled by the opposing forces of lysine methyltransferases ( KMTs ) and lysine demethylases ( KDMs ) . Histone monomethylation , dimethylation , and trimethylation of H3K4 are mediated by a group of SET domain-containing lysine methyltransferases , for example , MLL1-5 and ASH1 ( Ruthenburg et al . , 2007 ) . Among KDMs , KDM2B is restricted to removal of trimethylated H3K4 , whereas the KDM5 family ( KDM5 A–D ) and NO66 demethylate H3K4me2/3 ( Cloos et al . , 2008; Lan et al . , 2008; Kooistra and Helin , 2012 ) . Lysine-specific demethylase 1 ( Lsd1/KDM1A ) and its homolog KDM1B , however , demethylate monomethylated and dimethylated H3K4 , but not H3K4me3 ( Shi et al . , 2004; Ciccone et al . , 2009 ) . Hence , Lsd1/KDM1A and KDM1B are the only KDMs known with substrate specificity for H3K4me1 , a crucial enhancer mark . Lsd1 mediates its repressive functions as part of the CoREST ( corepressor for element-1-silencing transcription factor; Lee et al . , 2005 ) or NuRD ( nucleosome remodeling and histone deacetylation; Wang et al . , 2009b ) repressor complexes , but has also been implicated in gene activation , however , only when in complex with androgen or estrogen receptors through demethylation of H3K9me1/me2 ( Metzger et al . , 2005; Ruthenburg et al . , 2007; Wissmann et al . , 2007 ) . Although the biochemical functions of Lsd1 have been studied in detail ( reviewed in Cloos et al . , 2008; Lan et al . , 2008; Kooistra and Helin , 2012 ) , mechanistic understanding of Lsd1 in complex biological systems is limited . Targeted deletion of Lsd1 in mice is lethal . In Lsd1−/− embryos , the egg cylinder fails to elongate and gastrulate , resulting in developmental arrest around embryonic day ( E ) 5 . 5 and loss of Lsd1−/− embryos by E7 . 5 ( Wang et al . , 2007 , 2009a ) . Human and murine Lsd1−/− embryonic stem cells ( ESCs ) have proliferation and differentiation defects ( Wang et al . , 2009a; Adamo et al . , 2011; Whyte et al . , 2012 ) . In addition , recent evidence suggests that Lsd1 may be a point of vulnerability for acute myeloid leukemia cells ( Harris et al . , 2012; Schenk et al . , 2012 ) . However , the significance of Lsd1 in adult differentiation processes remains largely unexplored . Here , we have examined the in vivo roles of Lsd1 in hematopoiesis through conditional inactivation in the mouse . We identified Lsd1 as an indispensible epigenetic governor of hematopoietic differentiation . Consequences of Lsd1 loss are profound , including defects in long-term repopulating hematopoietic stem cell ( LT-HSC ) self-renewal and stark impairment of LT-HSC as well as mature lineage hematopoietic differentiation . We found that Lsd1 represses genes that are normally expressed in hematopoietic stem and progenitor cells ( HSPCs ) and that failure to silence HSPC gene signatures during differentiation is incompatible with terminal maturation of multiple blood lineages resulting in severe pancytopenia .
We generated a conditional Lsd1 allele in which Cre-recombinase–mediated excision of exons 5 and 6 generates a frame shift and premature stop in the mRNA . Exons 5 and 6 encode a flavin adenine dinucleotide binding site and the N-terminal portion of the amine oxidase domain , both essential for Lsd1 enzymatic activity ( Figure 1—figure supplement 1 and ‘Material and methods’ ) . Consistent with findings of other laboratories ( Wang et al . , 2007 , 2009a ) , germline deletion of Lsd1 resulted in early embryonic lethality ∼E7 . 5 ( data not shown ) , thus precluding analysis of early blood cell differentiation . To bypass early embryonic lethality , we used VavCre to delete Lsd1 . VavCre typically enables gene deletion across all hematopoietic cells as early as embryonic day 9 . 5 ( Stadtfeld , 2004 ) . Lsd1fl/fl VavCre animals were born at Mendelian ratios , but died neonatally of severe anemia ( Figure 1A , B ) . Deletion of Lsd1 during embryonic development was incomplete , as evidenced by the presence of Lsd1 protein in knockout and control fetal liver ( FL ) lysates ( Figure 1C ) , precluding analysis of embryonic hematopoiesis . We could , however , study fetal hematopoiesis since Lsd1 protein was completely absent in newborn Lsd1fl/fl VavCre bone marrow ( BM ) samples ( Figure 1C ) . 10 . 7554/eLife . 00633 . 003Figure 1 . Deletion of Lsd1 results in pancytopenia . ( A ) Lsd1fl/fl VavCre cohort statistics 0 and 10 days after birth ( n = 146 pups ) . ( B ) Lsd1fl/fl and Lsd1fl/fl VavCre newborn pups . ( C ) Western blot analysis of E13 . 5 fetal liver and newborn bone marrow cells with indicated genotypes . ( D ) Differential PB counts of 5-day-old control and Lsd1fl/fl VavCre pups . Data are expressed as mean ± SEM; n ≥ 5 mice per group . ( E ) May Gruenwald Giemsa–stained blood smears of Lsd1fl/fl VavCre and control pups . ( F ) Cell counts of femurae and tibiae and E14 . 5 fetal livers of control and Lsd1fl/fl VavCre neonates and embryos , respectively . Data are expressed as mean ± SEM; n ≥ 8 per group for fetal livers and n ≥ 4 per group for bone marrow . ( G ) Immunophenotypic analysis of control and Lsd1fl/fl VavCre bone marrow hematopoietic stem and progenitor cells . ( H ) Frequency and absolute numbers of lin− Sca-1− c-Kit+ cells ( LS−K+; containing CMP , GMP , and CMP populations ) and lin− Sca-1+ c-Kit+ cells ( LS+K+; containing HSC and MPP populations ) . Data are expressed as mean ± SEM; n ≥ 4 per group . DOI: http://dx . doi . org/10 . 7554/eLife . 00633 . 00310 . 7554/eLife . 00633 . 004Figure 1—figure supplement 1 . Lsd1 targeting strategy . ( A ) Illustration of Lsd1 protein domain structure . ( B ) Mouse Lsd1 locus ( Chr4 . D3 ) drawn to scale; Untargeted wild-type allele ( exons 3–8 ) ; targeted allele +neo ( fl_neo/+ ) : wild-type exons 5 and 6 have been replaced with a loxP sites flanked cassette carrying a frt site flanked neomycin resistance gene and exons 5 and 6; targeted allele −neo ( fl/+ ) : targeted allele after FlpE-mediated neomycin cassette deletion; deleted allele: Knockout allele after cre-mediated deletion of exons 5 and 6 . ( C ) PCR genotyping of +/+; fl/+ , and fl/fl allele . ( D ) Western blot analysis of Lsd1fl/fl and Lsd1−/− embryonic stem cell lysates . DOI: http://dx . doi . org/10 . 7554/eLife . 00633 . 004 Blood counts of Lsd1fl/fl VavCre newborns were significantly reduced ( Figure 1D , E ) , as was total BM but not FL cellularity ( Figure 1F ) . These findings are indicative of defects in the most immature bone marrow cells . Indeed , the hematopoietic stem and progenitor cell compartment of Lsd1fl/fl VavCre animals was vastly distorted ( Figure 1G ) . Frequency and absolute numbers of lineage-negative Sca-1+ c-Kit+ cells ( i . e . , LS+K+; encompassing hematopoietic stem cells and multipotent progenitors [MPP] ) as well as lineage-negative Sca-1− c-Kit+ cells ( LS−K+ cells; encompassing myeloid progenitor cells ) were reduced over 30-fold in Lsd1fl/lf VavCre mice ( Figure 1H ) . The only cell population present in the lineage-negative compartment consisted of Sca-1+ c-Kit− cells ( LS+K− cells , a poorly defined population , which may contain early lymphoid cells; Harman et al . , 2008; Kumar et al . , 2008; Brickshawana et al . , 2011 ) . Together , these data suggested early hematopoietic differentiation defects in the absence of Lsd1 . To test whether deletion of Lsd1 during adult hematopoiesis also results in hematopoietic stem cell defects , Lsd1fl/fl mice were interbred with the Mx1Cre mouse line , which allows inducible deletion of Lsd1 throughout all hematopoietic lineages and hematopoietic stem cells ( Kühn et al . , 1995 ) . Injection of Lsd1fl/fl and Lsd1fl/fl Mx1Cre mice with the dsRNA poly ( I:C ) ( Figure 2—figure supplement 1A ) resulted in efficient Lsd1 inactivation and a massive reduction in peripheral blood counts ( Figure 2—figure supplement 1B–D ) . In keeping with this , the majority of Lsd1fl/fl Mx1Cre mice died with severe anemia 7–10 days after the final dose ( Figure 2—figure supplement 1E ) . As in Lsd1fl/fl VavCre animals , Lsd1fl/fl Mx1Cre animals were largely deficient in LS+K+ as well as LS−K+ cells , again at the expense of LS+K− cells ( Figure 2A ) . We suspected that loss of Lsd1 might alter the immunophenotype of stem and progenitor cells and therefore adopted an additional FACS-gating strategy to allow the potential identification of LT-HSCs . We inverted the commonly used stem cell gating strategy ( i . e . , LS+K+ CD150+ CD48− ) , by first gating on lineage-negative CD150+ CD48− cells , succeeded by Sca-1 and c-Kit analysis . With this approach , we identified a population of cells in Lsd1 knockout animals , resembling immunophenotypic LT-HSCs , albeit with slightly decreased c-Kit expression levels ( Figure 2B ) . Interestingly , the frequency of Lsd1 knockout lin− CD150+ CD48− Sca-1+ c-Kit+ cells was increased about fourfold ( p≤0 . 001 ) . In keeping with this , we found that twice as many Lsd1 knockout lin− CD150+ CD48− Sca-1+ c-Kit+ cells were proliferating , and their apoptosis levels were not significantly altered compared to Lsd1fl/fl cells ( Figure 2—figure supplement 2A , B ) . This suggests compensatory LT-HSC expansion due to downstream differentiation defects . We isolated Lsd1fl/fl and Lsd1fl/fl Mx1Cre LT-HSCs , assessed Lsd1 mRNA expression levels by qPCR , and validated that Lsd1fl/fl Mx1Cre LT-HSCs had Lsd1 deleted and did not represent Lsd1fl/fl Mx1Cre cells that had escaped excision ( Figure 2—figure supplement 2C ) . 10 . 7554/eLife . 00633 . 005Figure 2 . Lsd1 knockout HSCs fail to give rise to terminally differentiated cells . ( A ) Immunophenotypic analysis of control and Lsd1fl/fl Mx1Cre hematopoietic stem and progenitor cells 1 week after final poly ( I:C ) dose . ( B ) Alternative gating logic for analysis of Lsd1fl/fl Mx1Cre and control hematopoietic stem cells: lin− CD150+ CD48− cells were gated first , followed by analysis of Sca-1+ c-Kit+ cells . Frequency of lin− CD150+ CD48− Sca-1+ c-Kit+ cells is presented as percentage of lineage-depleted bone marrow cells . Data are expressed as mean ± SEM; n ≥ 4 mice per group . ( C ) Experimental outline of LT-HSC colony-forming cell assays . Control and Lsd1fl/fl Mx1Cre CD150+CD48− LS+K+ cells were isolated by FACS prior to poly ( I:C ) treatment . Forty LT-HSCs were plated in methylcellulose containing 1000 U/ml IFNα , to induce Lsd1 deletion in vitro . ( D ) Colony morphology of LT-HSC colony-forming cell assays ( top panel , scale bar: 500 µM ) and May Gruenwald Giemsa–stained cytospin preparations from LT-HSC colony-forming cell assays ( lower panel , scale bar: 40 µM ) . ( E ) Flow cytometric analysis with indicated antibodies of cells isolated from colony assays after 12 days of differentiation . ( F ) Number of colonies obtained after 12 days in colony-forming cell assay . Data are expressed as mean ± SEM; n ≥ 6 mice per group . ( G ) and ( H ) Absolute numbers of lineage-positive and lineage-negative cells per colony assay . Data are expressed as mean ± SEM; n ≥ 6 mice per group . ( G and H ) Absolute numbers of LS+K+ and Gr1+Mac1+ cells per colony assay . Data are expressed as mean ± SEM; n ≥ 6 mice per group . DOI: http://dx . doi . org/10 . 7554/eLife . 00633 . 00510 . 7554/eLife . 00633 . 006Figure 2—source data 1 . Source data for Figure 2—figure supplement 2 . Background corrected CT values of FACS-sorted single cells analyzed by BIOMARK Fluidigm microfluidics qPCR . DOI: http://dx . doi . org/10 . 7554/eLife . 00633 . 00610 . 7554/eLife . 00633 . 007Figure 2—figure supplement 1 . Mx1Cre-mediated deletion of Lsd1 results in pancytopenia . ( A ) Experimental design . ( B ) Western Blot analysis of three control ( Lsd1fl/fl ) and three experimental ( Lsd1fl/fl Mx1Cre ) spleen lysates 1 week after p ( I:C ) injection . ( C ) Genomic PCR from control and Lsd1fl/fl Mx1Cre for Lsd1 , Cre , and Hprt . ( D ) Differential PB counts of control and Lsd1fl/fl Mx1Cre mice . Data are expressed as the mean ± SEM; n ≥ 10 mice per group . ( E ) Kaplan–Meier plot of Lsd1fl/fl ( n = 27 ) and Lsd1fl/fl Mx1Cre ( n = 44 ) mice after final dose of p ( I:C ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00633 . 00710 . 7554/eLife . 00633 . 008Figure 2—figure supplement 2 . Hierarchical clustering of immunoprofiles derived from single sorted Lsd1 knockout CD150+ CD48− LS+K+ and LS+K− cells . ( A ) Frequency of Lsd1fl/fl and Lsd1fl/fl Mx1Cre CD150+ CD48− LS+K+ LT-HSCs in G1/G0 vs S/G2/M cell cycle phases . Unfractionated bone marrow cells were incubated with Hoechst 33342 , stained with indicated antibodies and analyzed by flow cytometry . Data are expressed as the mean ± SD; n = 3 mice per group . ( B ) Frequency of early ( Annexin V+ Dapi− ) and late ( Annexin V+ Dapi+ ) apoptotic Lsd1fl/fl and Lsd1fl/fl Mx1Cre CD150+ CD48− LS+K+ LT-HSCs determined by flow cytometry . Data are expressed as the mean ± SD; n = 3 mice per group . ( C ) Lsd1 mRNA expression levels in Lsd1fl/fl and Lsd1fl/fl Mx1Cre CD150+ CD48− LS+K+ LT-HSCs determined by qPCR . Data are expressed as the mean ± SEM; n = 3 mice per group ( D ) FACS gating logic for isolation of LS−K+ and GMP cells . ( E ) Gating logic for CD150+ CD48− LS+K+ LT-HSCs . ( F ) Heatmap of single cell high throughput qPCR immunoprofiling . Cells were sorted into 96 wells , 1 week after the final p ( I:C ) injection . Extracted RNA was subjected to BIOMARK Fluidigm microfluidics qPCR . Biomarker expression values were clustered ( average linkage method; Pearson correlation ) using MeV ( Multi Experiment Viewer 4 . 8; Saeed et al . , 2006 ) . Data are presented as normalized CT values . A total of 96 single cells were profiled ( i . e . , 12× Lsd1 knockout LT-HSCs; 28× Lsd1 knockout LS+K− cells; 12× wild-type LT-HSCs; 11× wild-type LS+K− cells; 12× plasmacytoid dendritic cells; 21× granulocyte/macrophage progenitor cells [GMPs] ) . *p<0 . 05; n . s . : not significant . DOI: http://dx . doi . org/10 . 7554/eLife . 00633 . 008 To validate the knockout cells as bona fide LT-HSCs , we performed qPCR-based immunophenotyping of single cells . We profiled single Lsd1 knockout LT-HSCs for >30 cell surface markers by highly multiplexed microfluidics qPCR ( for complete gene list and data set , see Figure 2—figure supplement 2; Figure 2—source data 1 ) . We further applied this method to determine whether immunophenotypically altered myeloid progenitor cells , such as granulocyte macrophage progenitors ( GMPs ) , were concealed in the Lsd1fl/fl Mx1Cre LS+K− cell population , and if not , to categorize these cells . We performed hierarchical clustering of the resulting gene expression matrices of Lsd1 knockout LT-HSCs and LS+K− cells , with data sets consisting of ∼30 defined wild-type hematopoietic cell types also characterized by multiplexed microfluidics qPCR ( Guo et al . , complete data set will be published elsewhere ) . Lsd1fl/fl Mx1Cre LT-HSCs clustered closest with wild-type LT-HSCs , further substantiating that knockout lin− CD150+ CD48− Sca-1+ c-Kit+ cells represent authentic LT-HSCs ( Figure 2—figure supplement 2E , F; LT-HSC cluster ) . Lsd1 knockout lin− Sca-1+ c-Kit− cells clustered with wild-type LS+K− ( Figure 2—figure supplement 2D , F; LS+K− cluster ) . However , about 50% of knockout LS+K− cells displayed features of plasmacytoid dendritic cells , characterized by high expression of toll-like receptor 9 and leukemia inhibitory factor receptor ( Guiducci , 2006; Heng et al . , 2008; http://www . immgen . org ) . Nevertheless , all knockout LS+K− populations clustered separately from wild-type granulocyte macrophage progenitors ( GMPs ) , suggesting that Lsd1 knockout animals indeed lack LS−K+ myeloid progenitor cells ( Figure 2—figure supplement 2D , F; GMP cluster ) . To test whether the lack of myeloid progenitor cells might be due directly to a LT-HSC differentiation defect , we cultured Lsd1-deficient lin− CD150+ CD48− Sca-1+ c-Kit+ cells in methylcellulose . LT-HSCs from control and Lsd1fl/fl Mx1Cre animals were purified prior to Lsd1 deletion . This strategy ensured that control and Lsd1fl/fl Mx1Cre cells would be comparable ( i . e . , immunophenotypically identical LT-HSCs ) . To delete Lsd1 in vitro , we added interferon alpha ( IFNα ) to control and Lsd1fl/fl Mx1Cre cultures ( Figure 2C ) , and although colony numbers were comparable between Lsd1 knockout and control , stark differences in colony and cell morphology were observed ( Figure 2D , F ) . Control LT-HSCs generated large colonies of multipotential granulocyte , erythroid , macrophage , and megakaryocyte progenitor identity ( CFU-GEMM ) , whereas Lsd1-deficient colonies were smaller and consisted almost exclusively of immature blasts . Lsd1 knockout colonies primarily consisted of lineage-negative c-Kit+ Sca-1+ hematopoietic progenitor cells , whereas wild-type colonies were largely composed of mature Gr1+ Mac1+ myeloid cells ( Figure 2E , G–J ) . Together , the absence of LS−K+ myeloid progenitors determined by flow cytometry , the single cell multiplex qPCR immunoprofiling data , and colony assays indicate that Lsd1 function is vital for efficient differentiation of LT-HSCs into more mature myeloid progenitor cells . Self-renewal capacity of Lsd1fl/fl Mx1Cre HSCs was assessed by competitive BM transplantation . Unfractionated BM cells from Lsd1fl/fl Mx1Cre or Lsd1fl/fl mice ( CD45 . 1+CD45 . 2+ ) were cotransplanted at a 1:1 ratio with wild-type ( WT ) unfractionated BM competitor cells ( CD45 . 2 ) into lethally irradiated recipient mice , and all mice were treated with poly ( I:C ) 5 weeks after establishment of chimerism ( Figure 3A ) . After poly ( I:C ) treatment , we evaluated the kinetics of peripheral blood ( PB ) multilineage donor contribution . The initial contribution of Lsd1fl/fl Mx1Cre or Lsd1fl/fl cells to mature myeloid and lymphoid lineages was equivalent ( Figure 3B ) . By 4 weeks after poly ( I:C ) , the contribution of Lsd1fl/fl Mx1Cre cells to the myeloid lineage was barely detectable and remained low for 12 weeks ( Figure 3B ) . Correspondingly , the contribution of Lsd1fl/fl Mx1Cre cells to the B and T lymphoid lineages decreased >50% at 4 weeks after poly ( I:C ) and progressively declined over 12 weeks ( Figure 3B ) . Likewise , at 12 weeks after poly ( I:C ) injection , Lsd1fl/fl Mx1Cre cells generated only 2–14% of mature cells in the spleen and 1–9% of mature cells in the bone marrow ( Figure 3C , D ) . These results suggested a defect in HSC self-renewal in the absence of Lsd1 . To address this directly , we evaluated the contribution of Lsd1fl/fl Mx1Cre donor cells among immature hematopoietic stem and progenitor cells . 12 weeks after poly ( I:C ) treatment , the level of Lsd1fl/fl Mx1Cre cells that either contributed to lineage-negative CD150+ Sca-1+ c-Kit+ hematopoietic stem cells , CD150− LS+K+ multipotent progenitor cells , or various lineage-negative myeloid progenitor cells ( i . e . , GMP , PreGM , PreMegE , MkP , PreCFU-E , CFU-E ) was virtually undetectable ( Figure 3E , F ) . In combination , these data implicate Lsd1 as crucial epigenetic guardian of HSC self-renewal and homeostasis . 10 . 7554/eLife . 00633 . 009Figure 3 . Lsd1 is essential for HSC self-renewal . ( A ) Experimental outline for competitive transplantation experiment . Equal numbers of unfractionated CD45 . 2 competitor bone marrow cells were mixed with equal numbers of either CD45 . 1/ . 2 Lsd1fl/fl Mx1Cre or CD45 . 1/ . 2 Lsd1fl/fl donor unfractionated bone marrow cells . Bone marrow chimerism was established for 5 weeks prior to poly ( I:C ) injection . ( B ) Peripheral blood donor chimerism of myeloid , B , and T lymphocyte lineages 4 , 8 , and 12 weeks after poly ( I:C ) treatment . n = 5 for CD45 . 1/ . 2 Lsd1fl/fl; n = 12 for CD45 . 1/ . 2 Lsd1fl/fl Mx1Cre . p values were determined using a two-way ANOVA test . ( C ) Representative FACS plots for spleen donor chimerism of myeloid , B , and T lymphocyte lineages 12 weeks after poly ( I:C ) treatment . ( D ) Average BM and Spleen donor chimerism of control or Lsd1fl/fl Mx1Cre cells at 12 weeks after transplant . Data are expressed as mean ± SD; n = 4 recipients per group; **p<0 . 01; ***p<0 . 001 . ( E ) Representative FACS plots displaying gating logic used to determine donor chimerism of CD150+ LS+K+ HSCs , CD150− LS+K+ multipotent progenitor cells and myeloid progenitor subpopulations ( Pronk et al . , 2007 ) . ( F ) Average HSC and progenitor cell donor chimerism of control or Lsd1fl/fl Mx1Cre cells 12 weeks posttransplant . Data are expressed as mean ± SD; n = 4 recipients per group; **p<0 . 01; ***p<0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 00633 . 009 Next , we evaluated whether the deletion of Lsd1 also affects differentiation of mature lineage cells . Indeed , mature Gr1high Mac1+ granulocytes were almost absent from the bone marrow of Lsd1fl/fl Mx1Cre mice ( Figure 4A , B ) . The concomitant accumulation of immature Gr1dim Mac1+ cells , a population containing immature granulocytes , bipotential granulocytic/monocytic precursor cells , and monocytes ( Hestdal et al . , 1991; Walkley et al . , 2002 ) , indicated a defect in granulocytic maturation . We tested whether increased apoptosis or reduced proliferation of Lsd1 knockout Gr1dim Mac1+ cells could lead to the loss of Gr1high Mac1+ cells , but could not observe differences compared to control ( Figure 4—figure supplement 1A , B ) . Further analysis with antibodies specific for neutrophils ( Neutr . 7/4 ) and macrophages/monocytes ( F4/80 ) confirmed loss of mature neutrophils ( i . e . , Neutr . 7/4+ F4/80−; Figure 4A , B ) in Lsd1fl/fl Mx1Cre animals with a concomitant increase in bipotential granulocytic/monocytic precursor cells ( i . e . , Neutr . 7/4+ F4/80+ ) . The number of Neutr . 7/4− F4/80+ monocytes was unaffected . Cytospin preparations of Gr1+ Mac1+ cells confirmed the lack of mature neutrophils and the accumulation of immature myeloid cells ( Figure 4C ) . Thus , loss of Lsd1 culminates in a differentiation block at the transition of immature Gr1dim Mac1+ to mature Gr1high Mac1+ granulocytes . Next , we used EpoRCre mice to perform erythroid-specific gene inactivation of Lsd1 ( Heinrich et al . , 2004 ) . As EpoRCre-mediated deletion of Lsd1 was lethal in the prenatal period , we analyzed Lsd1fl/fl EpoRCre embryos . Combinatorial FACS staining of the cell surface markers CD71 and Ter119 , or combination of CD71 with c-Kit , were used to stage erythroid maturation in vivo ( Zhang et al . 2003 ) . Lsd1 knockout embryos displayed a 300% increase of CD71high Ter119low proerythroblasts ( R2 ) , but a 20-fold reduction of reticulocytes and erythrocytes ( R5; Figure 4D , E ) . Similar results were obtained using c-Kit and CD71 ( Figure 4D , E ) . Thus , erythroid cell maturation in Lsd1-deficient embryos is impaired at the transition from proerythroblasts/R2 to basophilic erythroblasts/R3 . In keeping with this , enucleated mature definitive erythroid cells were markedly reduced in Lsd1-deficient fetal livers ( Figure 4—figure supplement 1C ) . Overall , Lsd1-deficient embryos were small and exhibited pale fetal livers at E13 . 5 , similarly consistent with compromised erythropoiesis ( Figure 4F ) . By E14 . 5 , embryos displayed grave developmental defects , most likely due to insufficient oxygenation ( Figure 4—figure supplement 1D ) . Cross sections of E13 . 5 fetal livers showed a disturbed architecture with multiple pyknotic cells displaying karyorrhexis , indicative of increased cell death ( Figure 4—figure supplement 1E ) , also confirmed by flow cytometry ( Figure 4—figure supplement 1A ) . Taken together , these data demonstrate that Lsd1 is essential not only for differentiation and self-renewal of hematopoietic stem cells but also for terminal granulocytic and erythroid differentiation . 10 . 7554/eLife . 00633 . 010Figure 4 . Lsd1 is required for terminal granulocytic and erythroid differentiation . ( A ) Flow cytometric analysis for indicated myeloid cell surface markers 1 week after poly ( I:C ) . ( B ) Quantification and statistical analysis of Gr1/Mac1 and Neutr . 7/4/F4/80 flow cytometric analysis . Data are expressed as mean ± SEM with n ≥ 5 mice per group . **p<0 . 01; ***p<0 . 001 . ( C ) Cytospin of Gr1+ Mac1+ cells from control and Lsd1 knockout bone marrow stained with May Gruenwald Giemsa . Scale bar: 20 µM . ( D ) Erythroid maturation staging of Lsd1fl/fl EpoRCre and control fetal liver cells: Gates R1 and R2 contain primitive erythroid progenitor cells ( CD71low Ter119low and CD71high Ter119low ) , R3 early and late basophilic erythroblasts ( CD71high Ter119high ) , R4 chromatophilic and orthochromatophilic erythroblasts ( CD71med Ter119high ) , and R5 late orthochromatophilic erythroblasts and reticulocytes ( CD71low Ter119high ) . CD71+ c-Kit+ cells are proerythroblasts; and CD71+ c-Kit− cells contain early basophilic erythroblasts . ( E ) Quantification and statistical analysis of CD71/Ter119 and CD71/c-Kit flow cytometric analysis . Data are expressed as mean ± SEM; n ≥ 10 embryos per group . **p<0 . 01; ***p<0 . 001 . ( F ) Embryonic day 13 . 5 Lsd1fl/fl EpoRCre and control embryos . DOI: http://dx . doi . org/10 . 7554/eLife . 00633 . 01010 . 7554/eLife . 00633 . 011Figure 4—figure supplement 1 . Lsd1fl/fl EpoRCre mice die in utero due to erythroid differentiation defects . ( A ) Frequency of early ( Annexin V+ 7-AAD− ) and late ( Annexin V+ 7-AAD+ ) apoptotic Lsd1fl/fl and Lsd1fl/fl Mx1Cre Gr1dim Mac1+ bone marrow cells ( left ) , as well as , early ( Annexin V+ 7-AAD− ) and late ( Annexin V+ 7-AAD+ ) apoptotic Lsd1fl/fl and Lsd1fl/fl EpoRCre CD71+c-Kit+ fetal liver cells determined by flow cytometry . Data are expressed as the mean ± SD; n ≥ 5 mice per group . n . s . : not significant , ***p<0 . 001 ( B ) Frequency of Lsd1fl/fl and Lsd1fl/fl Mx1Cre bone marrow cells , as wells as Lsd1fl/fl and Lsd1fl/fl EpoRCre CD71+c-Kit+ fetal liver cells in G1/G0 , S , and G2/M cell cycle phases . Unfractionated bone marrow/fetal liver cells were incubated with Hoechst 33342 , stained with indicated antibodies and analyzed by flow cytometry . Data are expressed as the mean ± SD; n ≥ 3 mice per group . ( C ) Peripheral blood cells of E13 . 5 Lsd1fl/fl EpoRCre and control embryos . Solid black arrow: primitive erythrocyte; black/white arrow: definitive erythrocyte . ( D ) E14 . 5 Lsd1fl/fl EpoRCre and control embryos . ( E ) Fetal liver sections of E13 . 5 Lsd1fl/fl EpoRCre and control embryos . DOI: http://dx . doi . org/10 . 7554/eLife . 00633 . 011 To investigate an underlying molecular basis for the profound multilineage hematopoietic differentiation defects , we performed global gene expression profiling of Gr1dim Mac1+ granulocytic precursor cells , CD71+ c-Kit+ proerythroblasts , and CD150+ CD48− LS+K+ hematopoietic stem cells . Several genes that are typically highly expressed in normal hematopoietic stem and progenitor cells were substantially upregulated in Lsd1 mutant cells ( e . g . , CD34 , 41-fold; HoxA9 , 35-fold; Sca-1 , 16-fold; Figure 5—figure supplement 1A ) . To pursue the potential significance of this observation , we performed gene set enrichment analysis ( GSEA; Subramanian et al . , 2005 ) using a gene signature derived from HSPCs ( LSK signature; Figure 5—source data 1; Krivtsov et al . , 2006 ) . Strikingly , the stem/progenitor cell gene signature was highly enriched across all of the assayed Lsd1-deficient cells ( Figure 5A–C; Gr1dim Mac1+ cells: NES: −2 . 1; FDR ≤ 10−4; CD71+c-Kit+ proerythroblasts: NES: −2 . 2; FDR ≤ 10−4; CD150+ CD48− LS+K+ HSCs: NES: −2 . 0; FDR ≤ 10−4 ) . Similar results were obtained using independently derived stem/progenitor cell signatures ( Figure 5—source data 2 ) . Quantitative PCR analysis of Lsd1 knockout cells for selected hematopoietic stem and progenitor cell genes confirmed the derepression of HSPC genes ( Figure 5D ) . Next , we assessed whether derepression of stem and progenitor cell genes in mature cells might affect the expression of the key transcriptional regulators that instruct and orchestrate granulocytic or erythroid differentiation ( such as C/EBPα or Gata1 ) . However , mRNA expression levels of these master transcription factors were not downregulated in Gr1dim Mac1+ cells or in CD71+c-Kit+ proerythroblasts , indicating that the block in terminal differentiation could not be ascribed simply to the reduced expression of these regulators ( Figure 5—figure supplement 1B ) . In order to identify signaling pathways that potentially drive or maintain the upregulation of a stem and progenitor cell gene signature , we again performed gene set enrichment analysis and found that Hox genes , which are typically associated with early hematopoiesis , were significantly upregulated in Lsd1 knockout cells ( NES: −1 . 46; FDR ≤ 0 . 02; Figure 5E ) . In keeping with this , we discovered significant enrichment of HoxA7-driven ( Z-score: 2 . 746; p=1 . 63 × 10−8 ) and HoxA9-driven ( Z-score: 2 . 314; p=1 . 13 × 10−3 ) gene expression networks ( Figure 5—figure supplement 1D ) by performing pathway enrichment analysis with the ‘Ingenuity Systems’ pathway analysis tool . We found that the expression of downstream HoxA9 target genes , including Bcl2 , Msi2 , and Sox4 , were also significantly upregulated ( NES: −1 . 78; FDR ≤ 0 . 001; Figure 5F ) . Based on these findings , we hypothesized that failure to repress earlier lineage gene expression programs impairs early , as well as later , hematopoietic maturation programs . 10 . 7554/eLife . 00633 . 012Figure 5 . Loss of Lsd1 results in derepression of stem and progenitor cell genes . ( A ) – ( C ) Gene set enrichment analyses showing upregulation of LSK signature genes in Lsd1fl/fl Mx1Cre Gr1dim Mac1+ myeloid cells ( A ) , Lsd1fl/fl EpoRCre CD71+ c-Kit+ erythroid cells ( B ) , and Lsd1fl/fl Mx1Cre CD150+ CD48− LSK cells ( C , see Figure 5—source data 1 for gene set ) . ( D ) Relative mRNA expression levels of stem/progenitor cell marker genes measured by real-time PCR . Mean ± SEM values are from four biological replicates of Lsd1fl/fl Mx1Cre Gr1dim Mac1+ cells normalized to expression in control cells ( dashed line ) . ( E ) and ( F ) GSEA analyses showing derepression of hematopoiesis related Hox genes and HoxA9 target genes in Lsd1fl/fl Mx1Cre Gr1dim Mac1+ myeloid cells ( see Figure 5—source data 1 for gene sets ) . ( G ) Genome-wide distribution of Lsd1 binding sites in granulocytic progenitor cells ( 32D ) . ( H ) Venn diagram of overlap between Lsd1 target genes and genes upregulated in Lsd1fl/fl Mx1Cre Gr1dim Mac1+ cells ( ≥1 . 5-fold , p≤0 . 05 ) . ( I ) IPA functional category analysis of direct Lsd1 target genes as defined in ( H ) . *p<0 . 05; **p<0 . 01; ***p<0 . 001; n . s . : not significant . DOI: http://dx . doi . org/10 . 7554/eLife . 00633 . 01210 . 7554/eLife . 00633 . 013Figure 5—source data 1 . Gene sets used for GSEA analyses . DOI: http://dx . doi . org/10 . 7554/eLife . 00633 . 01310 . 7554/eLife . 00633 . 014Figure 5—source data 2 . Normalized enrichment scores ( NES ) and false discovery rates ( FDR ) of GSEA analyses with stem/progenitor gene sets . DOI: http://dx . doi . org/10 . 7554/eLife . 00633 . 01410 . 7554/eLife . 00633 . 015Figure 5—source data 3 . Lsd1 occupancy peak position in 32D cells . DOI: http://dx . doi . org/10 . 7554/eLife . 00633 . 01510 . 7554/eLife . 00633 . 016Figure 5—figure supplement 1 . Deletion of Lsd1 results in upregulation of stem and progenitor cell genes . ( A ) Heatmap of normalized log2 expression values of control and Lsd1fl/fl Mx1Cre Gr1dim Mac1+ cells ( twofold or less changes; p≤0 . 05 ) . ( B ) Relative mRNA expression levels of myeloid and erythroid transcription factors measured by real-time PCR . Mean ± SEM values are from four biological replicates of Lsd1fl/fl Mx1Cre Gr1dim Mac1+ granulocytic progenitor cells or Lsd1fl/fl EpoRCre CD71+ c-Kit+ proerythroblasts normalized to expression in respective control cells ( dashed line ) . ( C ) GSEA plot showing the downregulation of mature granulocytic signature genes Lsd1fl/fl Mx1Cre Gr1dim Mac1+ cells vs controls . ( For gene sets see Figure 5—source data 1 . ) ( D ) Cartoon of HoxA7 and HoxA9 network . Genes colored in red are at least twofold upregulated ( p≤0 . 01 ) and genes colored in blue at least twofold downregulated ( p≤0 . 01 ) in Lsd1fl/fl Mx1Cre Gr1dim Mac1+ granulocytic progenitor cells . *p<0 . 05; **p<0 . 01; ***p<0 . 001; n . s . : not significant ( related to Figures 4 and 5 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00633 . 016 To further our mechanistic understanding of Lsd1-mediated regulation of gene expression in hematopoietic cells , we performed chromatin immunoprecipitation followed by next-generation sequencing ( ChIP-Seq ) . We focused our search for direct Lsd1 target genes within the granulocytic lineage . Sufficient cellular material for Lsd1 ChIP-Seq could not be prepared from primary Gr1dim Mac1+ cells to generate high-quality occupancy data . Therefore , we used an immature murine granulocytic cell line ( i . e . , 32D ) that can be differentiated into mature neutrophils upon addition of G-CSF ( Rovera et al . , 1987 ) . In total , we identified a total of 25 , 173 Lsd1 binding sites ( peaks ) from the ChIP-Seq data set . We chose 18 peaks for validation by ChIP-qPCR in primary Gr1dim Mac1+ cells . We found that 16 ( >85% ) of these binding sites were also occupied by Lsd1 in Gr1dim Mac1+ cells ( Figure 6—figure supplement 1A ) . Genome-wide analysis of Lsd1 peak distribution revealed only limited binding to proximal promoters ( 15% ) . Lsd1 binding peaks were predominantly positioned at intergenic ( 40% ) and intronic ( 38% ) regions ( Figure 5G; Figure 5—source data 3 ) . Next , we compared the Lsd1 ChIP-Seq binding sites with gene expression changes extracted from Gr1dim Mac1+ microarray data and found that Lsd1 occupied 55% of genes ( i . e . , 1361 out of 2473 ) that were at least 1 . 5-fold ( p≥0 . 05 ) upregulated in Lsd1 knockout cells ( Figure 5H ) . Using the Ingenuity Systems functional categorization tools on these direct Lsd1 target genes , we found that the majority of these genes are indeed involved in regulation of blood cell function ( Figure 5I ) . To assess the potential consequences of Lsd1 loss on global histone methylation levels , we analyzed Lsd1 substrates H3K4me1/me2 and H3K9me1/me2 by Western blot . No differences in global methylation levels were detected ( Figure 6—figure supplement 1B ) . Based on the distribution of regions occupied by , we sought to determine whether Lsd1 participates in the control of promoter regions , as well as distal enhancer elements . H3K4me2 and H3K4me3 typically mark active promoters , whereas distal enhancer elements are characterized by high levels of H3K4me1 and low levels of H3K4me3 ( Heintzman et al . , 2007; Koch et al . , 2007 ) . Active enhancers are distinguished from poised enhancers by the presence of H3K27 acetylation ( H3K27ac; Creyghton et al . , 2010; Rada-Iglesias et al . , 2011 ) . To determine the relative contribution of Lsd1 toward regulation of promoter or enhancer activities , we performed ChIP-Seq for promoter-associated H3K4me2 and H3K4me3 and the enhancer-associated H3K4me1 and H3K27ac marks in wild-type Gr1dim Mac1+ cells , and compared Lsd1 binding with the allocation of the respective histone marks . Peak overlap analysis of Lsd1 with the respective histone methylation marks confirmed that Lsd1 occupied about three times more putative enhancers ( Figure 6B ) than active promoters ( Figure 6A ) , implying that Lsd1 in fact regulates both promoter and enhancer activities . 10 . 7554/eLife . 00633 . 017Figure 6 . Lsd1 occupies enhancers and promoters of stem and progenitor cell genes and loss of Lsd1 is associated with increased levels of H3K4me1 and H3K4me2 on enhancers and promoters of HSPC genes . ( A ) Venn diagram of the overlap between Lsd1 and promoter-associated H3K4me2 and H3K4me3 peaks in wild-type Gr1dim Mac1+ cells . ( B ) Venn diagram of overlap between Lsd1 with enhancer-associated H3K4me1 and H3K27ac peaks in wild-type Gr1dim Mac1+ cells . ( C ) and ( D ) Box plots displaying log2 expression ratios of Lsd1 knockout over wild-type Gr1dim Mac1+ cells at KO-specific ( red bars ) and wild type–specific ( blue bars ) H3K4me1/me2/me3 and H3K27ac target genes . ( E ) and ( F ) Enrichment scores for KO-specific ( red bars ) , common ( black bars ) , and wt-specific ( blue bars ) H3K4me1/me2/me3 and H3K27ac peak associated target genes within LSK and granulocyte gene signatures , respectively . Enrichment scores were calculated as the ratio of overlap between H3K4me1/me2/me3 and H3K27ac target genes and LSK or granulocyte signature genes , compared to expected overlap at random . p values were determined using Fisher’s exact test . ( G ) Analyses of Lsd1 occupied genes within LSK and granulocyte signatures , respectively . Lsd1 target genes were classified into gene lists according to differential Lsd1 occupancy ( i . e . , Lsd1 binding at distal and promoter regions [red bars]; Lsd1 binding at distal regions only [black bars]; Lsd1 binding promoter regions only [blue bars] ) . Enrichment scores were calculated as the ratio of overlap between Lsd1 target genes with LSK or granulocyte signature genes , compared to expected overlap at random . p values were determined using Fisher’s exact test . ( H ) Representative ChIP-Seq tracks for LSK signature genes ( i . e . , Meis1 and Ctnnb1 [β-catenin] ) , which have Lsd1 bound at distal regulatory and promoter regions . Scale bar ticks represent 10 kb . DOI: http://dx . doi . org/10 . 7554/eLife . 00633 . 01710 . 7554/eLife . 00633 . 018Figure 6—source data 1 . Target genes of Lsd1 KO-specific and wt-specific histone modification peaks in Gr1dim Mac1+ cells . DOI: http://dx . doi . org/10 . 7554/eLife . 00633 . 01810 . 7554/eLife . 00633 . 019Figure 6—source data 2 . Target genes of Lsd1 in 32D cells ( distal; promoter; distal , and promoter ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00633 . 01910 . 7554/eLife . 00633 . 020Figure 6—figure supplement 1 . Validation of Lsd1 ChIP-Seq and normalization of histone ChIP-Seq data using ‘MAnorm’ . ( A ) In vivo chromatin occupancy was examined by ChIP-qPCR in primary murine Gr1dim Mac1+ myeloid cells . ChIP-qPCR validation for Lsd1 binding on LSK genes compared to rabbit IgG and negative control regions ( i . e . , Gnai1 TSS and Chr . 5 Gene Desert region ) . Data are plotted as means ± SD of two independent ChIP experiments . Primer sequences are available upon request . ( B ) Histone acid extracts from control and Lsd1fl/fl Mx1Cre whole bone marrow cells from two experimental animals per group were subjected to Western blot analysis with indicated antibodies . ( C ) and ( D ) Representative MA plots for H3K4me1 and H3K27ac from Lsd1fl/fl ( wild type ) and Lsd1fl/fl Mx1Cre ( Lsd1 KO ) FACS-sorted Gr1dim Mac1+ cells before and after ‘MAnorm’ normalization . The same normalization was also carried out for H3K4me2 and H3K4me3 ( not shown ) . The log2 ratio of peak read density ( M ) in Lsd1 KO ChIP-Seq samples vs it in the corresponding wild-type samples was plotted against the average log2 read density ( A ) for all peaks , and robust linear regression was applied to fit the global dependence between the M-A values of common peaks . Finally , the derived linear model was used as a reference for normalization and extrapolated to all peaks . The normalized M value was then used as a quantitative measure of differential binding in each peak region between Lsd1 KO and wild type , with peak regions associated with larger absolute M values exhibiting greater differences in binding . DOI: http://dx . doi . org/10 . 7554/eLife . 00633 . 02010 . 7554/eLife . 00633 . 021Figure 6—figure supplement 2 . Definition of Lsd1 knockout unique histone methylation peaks . ( A ) – ( D ) Histone modification peaks ( i . e . , H3K4me1 , H3K4me2 , H3K4me3 , and H3K27ac ) derived from Lsd1fl/fl ( wild type ) and Lsd1fl/fl Mx1Cre ( Lsd1 KO ) Gr1dim Mac1+ cell ChIP-Seq experiments were normalized using ‘MAnorm’ . The resulting normalized log2 peak read densities of Lsd1 KO and wild-type peaks were plotted against each other . Lsd1 KO peaks that exhibited ≥2-fold higher normalized log2 peak read density over wild-type normalized log2 peak read density ( the corresponding M value >1 ) were defined as Lsd1 KO-specific ( red circles ) and wild-type peaks that displayed a ≥2-fold higher normalized log2 peak read density over Lsd1 KO normalized log2 peak read density ( the corresponding M-value <−1 ) were defined as wild type–specific ( blue circles ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00633 . 021 To evaluate the effect of Lsd1 loss at enhancers and promoters in hematopoietic cells , we also performed ChIP-Seq for H3K4me1/me2/me3 and H3K27ac in Lsd1 knockout Gr1dim Mac1+ cells . All histone ChIP-Seq data were normalized using the ‘MAnorm’ program ( Shao et al . , 2012 ) , which allows for quantitative comparison between wild-type and knockout-specific ChIP-Seq peaks ( Figure 6—figure supplement 1C , D ) . The resulting normalized log2 peak read densities of Lsd1 knockout and wild-type data sets were plotted against each other to identify and categorize Lsd1 knockout-specific , wt-specific , and common peaks ( Figure 6—figure supplement 2A–D ) . We mapped cell type–specific peaks to the nearest Refseq annotated genes ( Figure 6—source data 1 ) and observed that both KO-specific promoter and enhancer peak target genes correlated highly with genes upregulated in Lsd1 knockout Gr1dim Mac1+ cells ( Figure 6C , D ) . These findings indicate that Lsd1 represses gene expression in association with changes in H3K4me1 and H3K4me2 levels at enhancers and promoters , respectively . To assess the contribution of KO-specific H3K4me1 and H3K4me2 methylation to the derepression of stem and progenitor cell genes , we determined the overlap of KO-specific and wt-specific peak target genes with the stem and progenitor cell gene signature ( LSK signature ) , compared to the expected genome-wide overlap at random . We found that both KO-specific enhancer-associated H3K4me1 ( ES: 2 . 31; p=4 . 5 × 10−14 ) and the promoter-associated H3K4me2 ( ES: 2 . 60; p=9 . 7 × 10−19 ) peak target genes were highly enriched among the LSK signature genes ( Figure 6E ) . Similar results were obtained for KO-specific H3K4me3 ( ES: 3 . 83; p=4 . 9 × 10−20 ) and H3K27ac ( ES: 2 . 98; p=1 . 3 × 10−34 ) peak target genes ( Figure 6E ) . We reasoned that genes downregulated in Lsd1 knockout Gr1dim Mac1+ cells should not show considerable enrichment of KO-specific methylation marks . Consequently , we determined the enrichment of histone methylation peaks within a mature granulocyte signature , which was downregulated in Lsd1 knockout Gr1dim Mac1+ cells ( NES: 2 . 8; FDR ≤ 10−4; Figure 5—figure supplement 1C ) . As expected , KO-specific peak target genes were not meaningfully enriched in the granulocyte gene signature ( Figure 6F ) . To assess direct involvement of Lsd1 in the regulation of HSPC genes , we examined the occupancy of Lsd1 on LSK and granulocyte signature genes . We divided Lsd1 target genes into three groups: genes that had Lsd1 either bound at distal sites and at the promoter ( 1 ) , only at distal regions ( 2 ) , or only at the proximal promoter ( 3 ) ( Figure 6G and Figure 6—source data 2 ) . We found that genes of group ( 1 ) were highly enriched ( ES: 2 . 02; p=2 . 53 × 10−11 ) and that genes of group ( 2 ) were slightly but significantly enriched ( ES: 1 . 15; p=0 . 02 ) in the LSK signature . As expected , none of the three groups were significantly enriched in the granulocyte signature ( Figure 6G ) , suggesting that Lsd1 is directly involved in the repression of hematopoietic stem and progenitor cell genes but not in the activation of mature granulocyte genes . Among the genes occupied by Lsd1 at distal sites and at the promoter were Meis1 , Gfi1b , Myb , β-catenin , and Runx1 ( Figure 6—figure supplement 1A; Figure 6—source data 2 ) , a set of genes with established roles in hematopoietic stem cells . These findings provide evidence of the relevance of this class of Lsd1 target genes . Representative tracks of genes bound by Lsd1 at distal regulatory and promoter regions ( Meis1 and β-catenin ) are depicted in Figure 6H . Collectively , we established that Lsd1 occupies distal regulatory regions as well as proximal promoters of stem and progenitor cell genes and that deletion of Lsd1 is associated with increased levels of the Lsd1 substrates H3K4me1 and H3K4me2 on enhancers and promoters of HSPC genes .
The biochemical functions of Lsd1 have been studied in considerable detail; yet , comprehensive characterization of its roles in tissue-specific differentiation , and particularly hematopoiesis and HSCs , remains largely unexplored . As Lsd1 has been proposed recently as a target for therapy in acute myeloid leukemia ( Harris et al . , 2012; Schenk et al . , 2012 ) , elucidation of its in vivo requirements is highly relevant to both normal and malignant hematopoiesis . Through study of blood lineage–specific conditional knockout mice , we established that Lsd1 is an indispensable factor for hematopoietic differentiation . Conditional deletion of Lsd1 resulted in severe pancytopenia , the consequence of combined defects in early hematopoietic stem cell differentiation and terminal blood cell maturation . We used Cre recombinase mouse strains to permit deletion of Lsd1 at early or later developmental stages . In this manner , we dissected the complex phenotypes ensuing from Lsd1 loss . We established that Lsd1 is crucial not only for immature hematopoietic stem cell differentiation , as might have been expected from the previous ES cell studies ( Wang et al . , 2009a; Adamo et al . , 2011; Whyte et al . , 2012 ) , but also critical for differentiation of mature hematopoietic cells . Taken together , we found that Lsd1-deficient HSCs were severely impaired in their capacity to mature into immature progenitors . Deletion of Lsd1 not only compromised early hematopoietic differentiation but also strongly interfered with terminal granulocytic and erythroid differentiation . Our data are most consistent with a requirement for Lsd1 at promoters and enhancers of stem and progenitor cell genes in hematopoietic cells . As a consequence , differentiating Lsd1 knockout cells aberrantly express HSPC genes normally expressed only in HSCs and progenitors . We propose that failure to silence these LSK-associated genes in Lsd1 mutants interferes with proper differentiation , culminating in detrimental consequences for hematopoiesis ( see Figure 7 for model ) . 10 . 7554/eLife . 00633 . 022Figure 7 . Cellular and molecular effects of Lsd1 loss during hematopoietic differentiation . ( A ) Characterization of the stem/progenitor cell compartment of Lsd1 knockout mice demonstrated the presence LT-HSCs ( CD150+ CD48− LSK ) , while having lost all LS−K+ myeloid progenitor cells ( i . e . , CMP , GMP , and MEP ) in the bone marrow . In line with this , Lsd1-deficient bone marrow cells could not contribute to multilineage hematopoiesis in competitive transplantation assays . Accordingly , Lsd1-deficient HSCs maintained an immature cellular morphology , an LS+K+ immunophenotype , and were incapable of giving rise to mature cells , in colony formation cell assays , which we used to directly test LT-HSC differentiation capability . Gene expression profiling of Lsd1-deficient CD150+ CD48− LSK cells demonstrated upregulation of hematopoietic stem and progenitor cell signature genes . Collectively , our data suggested that the inability to silence HSPC genes keeps Lsd1 knockout LT-HSCs from differentiating into myeloid progenitor cells and lineage commitment . ( B ) Lsd1 deficiency in granulocytic and erythroid cells resulted in erroneous derepression of stem and progenitor cell gene sets , which resulted in stark maturation defects in the mature erythroid and granulocytic cell lineages . These findings suggested that Lsd1 is required to maintain appropriate expression levels of HSPC genes during early , as well as late hematopoietic differentiation and that the failure to so , is incompatible with terminal differentiation . DOI: http://dx . doi . org/10 . 7554/eLife . 00633 . 022 Conditional deletion of Lsd1 in fetal ( VavCre ) as well as adult ( Mx1Cre ) HSCs resulted in pancytopenia , consistent with a prominent role for Lsd1 in HSC homeostasis . Indeed , we found that Lsd1-deficient hematopoietic stem cells were impaired in their in vitro differentiation capacity . Lsd1 knockout LT-HSCs mostly gave rise to cells with immature blast morphology and an LSK immunophenotype while being largely unable to contribute mature cell progeny in methylcellulose colony-forming cell assays . In a competitive bone marrow transplantation setting , Lsd1-deficient bone marrow cells were not only unable to significantly contribute to peripheral myeloid and B and T lymphoid lineages but more importantly CD150+ LS+K+ hematopoietic stem cells , CD150− LS+K+ multipotent progenitor cells , and various lineage-negative myeloid progenitor cells ( i . e . , GMP , PreGM , PreMegE , MkP , PreCFU-E , CFU-E ) were almost undetectable . While the reduced contribution of mature cells could be explained , among other things , by differentiation defects , the loss of Lsd1-deficient HSCs in the competitive transplantation setting indicates that Lsd1 is also vital for LT-HSC self-renewal . We also observed that Lsd1 is essential not only for early hematopoietic differentiation but also for the final steps of terminal blood cell maturation in multiple lineages . We found that mature Gr1high Mac1+ granulocytes were virtually absent from the peripheral blood and bone marrow of Lsd1fl/fl Mx1Cre mice . The concomitant increase of immature Gr1dim Mac1+ cells , a population containing immature granulocytes ( Hestdal et al . , 1991; Walkley et al . , 2002 ) , indicated a defect in granulocytic maturation . Similarly , we found that erythroid-specific Lsd1 deletion resulted in lethal embryonic anemia . Lsd1-deficient embryos were small and exhibited pale fetal livers at E13 . 5 , which could be ascribed to a 20-fold reduction of reticulocytes and erythrocytes . The combination of rapid lethality of Lsd1 knockout mice and the relatively long half-life of lymphoid lineage cells did not allow us to study the long-term effects of Lsd1 loss in lymphoid cells . Although we did not observe a decrease of peripheral B or T cells in Lsd1 Mx1Cre mice ( data not shown ) , future studies on the role of Lsd1 in lymphoid development are warranted . Some of our phenotypic discoveries ( i . e . , granulocytic and erythroid maturation defects ) are independently confirmed in a recent study that used shRNA-mediated knockdown rather than gene excision to affect Lsd1 expression ( Sprüssel et al . , 2012 ) . As might be anticipated , in vivo knockdown of Lsd1 revealed attenuated phenotypes when compared to the consequences of gene knockout . For example , we found that Lsd1 deletion with VavCre or Mx1Cre resulted in severe HSC differentiation and self-renewal defects resulting in complete loss of lineage-negative c-Kit+ myeloid progenitor cells , whereas Sprüssel et al . ( 2012 ) did not observe a defect in LT-HSCs . It seems likely that incomplete inactivation of Lsd1 by knockdown may account for these phenotypic differences . Two recent reports propose Lsd1 as a potential target for treatment of acute myeloid leukemia ( AML; Harris et al . , 2012; Schenk et al . , 2012 ) . These studies described that chemical inhibition of the enzymatic activity , as well as knockdown of Lsd1 , resulted in increased apoptosis and impaired leukemogenicity of MLL-AF9- and PML-RARα-transformed cells ( Harris et al . , 2012; Schenk et al . , 2012 ) . Two independent studies have described that RNAi-mediated or chemical inhibition of Lsd1 enhances γ-globin expression in human erythroid cells and quite modestly in human β-locus transgenic mice ( Shi et al . , 2013; Xu et al . , 2013 ) , which would alleviate symptoms of β-hemoglobinopathies such as sickle cell disease and β-thalassemia . Although these findings illustrate favorable features of Lsd1 as a potential drug target , Sprüssel et al . ( 2012 ) , Xu et al . , and this study demonstrate that Lsd1 serves critical roles in hematopoietic differentiation and that appreciable or complete inhibition of Lsd1 could have undesired side effects . Therefore , prospects for the utility of small-molecule inhibitors against Lsd1 for therapy of hematologic diseases will rest on the availability of an adequate therapeutic window . Importantly , in addition to analyzing phenotypic effects of Lsd1 loss , we also examined how Lsd1 deficiency leads to molecular and gene expression changes that culminate in hematopoietic deficiencies . Previously , no comprehensive analysis of Lsd1 function has been performed in adult cell types , such as hematopoietic cells . Previous reports studying effects of Lsd1 loss in hematopoietic cells were largely performed in cancer cell lines and did not include Lsd1 ChIP-Seq or analysis of global H3K4me1 methylation changes upon Lsd1 deletion ( Harris et al . , 2012; Schenk et al . , 2012 ) . Our studies are the first to integrate genome-wide Lsd1 occupancy data , H3K4me2 , and H3K4me1 mark status in primary wild-type and Lsd1 knockout cells with microarray data in an effort to reveal a more comprehensive picture of the biomolecular function of Lsd1 in hematopoietic cells . Through comprehensive , integrative epigenomic analyses , we established that Lsd1 function in hematopoietic cells is associated with reduced methylation of H3K4me2 at transcription start sites and reduced methylation of H3K4me1 at enhancers . Failure of proper H3K4me1 and H3K4me2 regulation is associated with derepression and upregulation of hematopoietic stem and progenitor cells signatures , which is ultimately incompatible with early HSC , as well as terminal blood cell , differentiation . The finding that Lsd1 acts not only at transcriptions start sites but also at enhancers is of particular interest , given the genome-wide enhancer mapping studies that highlight enhancers as major determinants of cell type–specific gene expression and cell fate decisions ( Lupien et al . , 2008; Heintzman et al . , 2009; Jin et al . , 2011; Xu et al . , 2012 ) . Enhancers function by recruiting sequence-specific transcription factors into chromatin-associated multiprotein complexes . However , the epigenetic mechanisms underlying the regulation and fine-tuning of enhancer activity are still incompletely understood . Our findings strongly suggest that Lsd1 serves a critical role in regulating enhancer activity in hematopoietic cells . Lsd1-occupancy data generated from murine and human embryonic stem cells suggested that Lsd1 occupies and regulates virtually all enhancers of embryonic stem cell genes ( Ram et al . , 2011; Whyte et al . , 2012 ) , whereas our data interestingly reveal that Lsd1 occupies only about 30% of active enhancers in mature myeloid cells . Although this is still a considerable number of enhancers , it clearly points toward a somewhat unexpected complexity of enhancer regulation in more mature cells compared to ES cells . In this regard , it is relevant to recall that ES cell chromatin exists in an unique and unusual configuration with widely dispersed open chromatin ( Efroni et al . , 2008 ) , and that following differentiation , there is extensive epigenetic reorganization ( Meshorer and Misteli , 2006; Efroni et al . , 2008; Hiratani et al . , 2010 ) . Recent genome-wide comparisons between fetal and adult hematopoietic cells have illustrated how the epigenetic profile of fetal blood cells differs from adult blood cells ( Xu et al . , 2012 ) . In light of the global differences in the architecture of embryonic , fetal , and adult epigenetic landscapes , it is noteworthy that our data suggest a somewhat differential requirement for Lsd1 in ES cells vs mature cells , and hence open new avenues to study how other histone demethylases , or other chromatin remodeling factors , participate in the regulation of enhancers during adult tissue differentiation . It is critical to discuss that Lsd1 deletion may impact gene expression in additional ways other than misregulation of histone demethylation . For example , loss of Lsd1 protein has been suggested to destabilize the CoREST complex ( Lee et al . , 2005; Foster et al . , 2010 ) , which has been demonstrated to result in increased global H3K56 acetylation ( Foster et al . , 2010 ) . Future studies will be necessary to dissect whether increased H3K56 acetylation also contributes to the hematopoietic phenotypes observed in this study . Lsd1 is also essential for the stability of DNA methyltransferase 1 ( Dnmt1; Wang et al . , 2009a ) . We have considered the possibility that a reduction in Dnmt1 levels might contribute to the observed phenotypes in the Lsd1 knockout animals . However , given the subtle hematopoietic phenotypes observed in conditional Dnmt1 knockout animals during steady-state hematopoiesis ( Bröske et al . , 2009; Trowbridge et al . , 2009 ) , we believe this to be unlikely . Collectively , we identified Lsd1 as indispensible for hematopoietic differentiation . We propose that Lsd1 is an epigenetic governor required broadly for repression of stem cell gene expression programs during embryonic stem cell as well as somatic cell differentiation . Given its crucial role in hematopoietic differentiation , the possibility of Lsd1 as a potential drug target in hematopoietic malignancies will have to be carefully evaluated in the context of potentially detrimental side effects .
Mx1Cre ( Kühn et al . , 1995 ) , VavCre ( Stadtfeld , 2004 ) , and EpoRCre ( Heinrich et al . , 2004 ) mice were described previously and maintained in HEPA filtered cages on a mixed C57BL/6J background . For Mx1Cre-mediated deletion , high–molecular weight poly ( I:C ) ( InvivoGen , San Diego , CA ) was administered via intraperitoneal injections at 12 . 5 µg/g body weight . Boston Children’s Hospital Animal Ethics Committee approved all the experiments . Exons 5 and 6 of Lsd1 , located on chromosome 4 in Mus musculus , were flanked with loxP sites . The deletion of these exons results in a frameshift , causing a premature stop in the mRNA , leading to nonsense-mediated decay of the Lsd1 transcript . Additionally , exons 5 and 6 encode both a flavin adenine dinucleotide ( an essential cofactor for Lsd1 enzymatic activity ) binding site and the N-terminal portion of the amine oxidase domain ( Figure 1—figure supplement 1A ) . Total genomic DNA isolated from CJ7 ES cells was used as a template to generate short ( 1 . 5-kb ) and long ( 6 . 2-kb ) arms of homology as well as a DNA fragment carrying exons 5 and 6 . The homology arms and exons 5 and 6 were cloned using standard techniques into a vector carrying a neomycin resistance cassette to be used for positive selection and expressing thymidine kinase for negative selection ( i . e . , pBS XF PGKNeo FX TK ( r ) , kind gift from Huafeng Xie , Dana-Farber Cancer Center , Boston , MA ) . The sequence-verified targeting vector was linearized and electroporated into feeder-dependent CJ9 ES cells ( 129sv background ) , followed by selection with 300 μg/ml G418 and ganciclovir . Homologous recombination was identified via Southern blot analysis . For the short arm , EcoRI-digested genomic DNA was screened with a 505-bp external probe . This probe detects a 7 . 5-kb fragment from the wild-type allele and a 5 . 1-kb fragment from the targeted allele ( Figure 1—figure supplement 1B ) . For the long arm , XbaI-digested DNA was screened with a 295-bp external probe . This probe detects a 12-kb fragment from the wild-type allele and a 9 . 5-kb fragment from the targeted allele ( Figure 1—figure supplement 1B ) . To ensure that the positive clones only carried a single integration of the targeting vector , positive clones were also analyzed using a probe specific for the neomycin resistance cassette . Positively targeted ES cell clones were further genotyped for the presence of loxP sites . To delete the neomycin resistance gene cassette , 10 million Lsd1fl_neo/+ ES cells were electroporated with a vector encoding the FlpE recombinase and seeded onto puromycin-resistant DR4 feeder cells . 24 hr after electroporation , puromycin selection ( 1 µg/ml ) was applied for 2 days . Puromycin-resistant clones were picked after 10 days and assayed for the deletion of the neomycin resistance cassette using a PCR-based strategy . ES cell clones with a normal karyotype and containing one allele with exons 5 and 6 of Lsd1 flanked by loxP sites ( Lsd1fl/+ ) were injected into c57Bl/6J blastocysts to generate chimeric mice . High-degree chimeric mice ( 80–95% ) were interbred with c57Bl/6J mice to obtain Lsd1fl/+ germline offspring . Genotyping primers are available upon request . Single-cell suspensions of BM were prepared from pooled femurs , tibiae , and iliac crest bones . PB was collected from the submandibular vein , and CBC’s were determined using a Hemavet 950 FS ( Drew Scientific Group , Waterbury , CT ) . RBCs were lysed with ammonium chloride buffer prior to staining . Prior to staining , cells were incubated with purified anti-mouse CD16/CD32 antibody ( FC Block ) for a minimum of 15 min . LT-HSCs and myeloid progenitors were isolated and analyzed as in Ema et al . ( 2006 ) and Pronk et al . ( 2007 ) with minor modifications . In brief , bone marrow cells were isolated by crushing iliac crest bones , femurae , and tibiae in PBS containing 0 . 2% BSA and 2 mM EDTA . Isolated bone marrow cells were then subjected to Ficoll ( Ficoll-Paque Premium; GE Healthcare ) density gradient centrifugation . Cells from buffy coat were washed twice with PBS/0 . 2% BSA/2 mM EDTA and then labeled at subsaturating levels with the biotin mouse lineage depletion kit ( eBioscience , San Diego , CA ) . Lineage-positive stained cells were magnetically depleted with Biotin Binder Dynabeads ( Invitrogen , Grand Island , NY ) . The remaining cells were then stained with lineage cocktail CD150 , CD48 , c-Kit , and Sca-1 antibodies . Dead cells were excluded from analysis using 7-Aminoactinomycin D ( 7-AAD; Becton Dickinson , BD ) or 4′ , 6-diamidino-2-phenylindole ( DAPI; Invitrogen ) . Antibody clones: Gr1 ( RB6-8C5 ) , CD11b ( M1/70 ) , B220 ( RA3-6B2 ) , CD71 ( R17217 ) , Ter119 , CD5 ( 53-7 . 3 ) , CD48 ( BCM1 ) , CD16/32 ( 93 ) , CD105 ( MJ7/18 ) , Sca-1 ( D7 ) , c-Kit ( 2B8 ) CD41 ( MWReg30 ) , CD45 . 1 ( A20 ) , CD45 . 2 ( 104 ) , all from eBioscience; CD150 ( TC15-12F12 . 2 ) from Biolegend ( San Diego , CA ) ; F4/80 ( CI:A3-1 ) from Serotec ( Raleigh , NC ) ; Neutrophil 7/4 ( clone 7/4 ) from Abcam ( Cambridge , MA ) . Lineage cocktail is defined as Gr1 , Mac1 , B220 , CD3 , Ter119 , and CD5 . Flow cytometric analysis was performed on a FACSCalibur or LSRFortessa and sorting on a FACSAria I ( both Becton Dickinson , BD , Franklin Lakes , NJ ) and data were analyzed with FlowJo ( Tree Star , Inc . , Ashland , OR ) . For bone marrow transplantation experiments , congenic female C57Bl/6J ( CD45 . 2+ ) mice ( JAX Mice , Bar Harbor , ME ) were irradiated with a split dose of 11 Gy . Donor BM cells were obtained from a cross of Lsd1fl/fl or Lsd1fl/fl Mx1Cre C57BL/6 and B6 . SJL mice . Lsd1fl/fl or Lsd1fl/fl Mx1Cre ( CD45 . 1+/CD45 . 2+ ) BM cells were coinjected retro-orbitally with competitor BM ( CD45 . 2+ ) cells at a ratio of 1:1 ( 2 × 106 cells each ) . Mice were maintained on antibiotic treated water . Chimerism in the peripheral blood was determined 5 weeks after reconstitution and prior to Lsd1 deletion . Mice with established chimerism were injected three times with poly ( I:C ) every other day . Peripheral blood chimerism was monitored every 4 weeks for 12–16 weeks . Donor contribution in peripheral blood , mature bone marrow lineage , bone marrow stem cell and progenitor cell compartment , and spleen were measured by cell surface staining for the CD45 . 1 and CD45 . 2 markers as well as genomic DNA analysis of sorted populations . For cell cycle analysis , cells were resuspended in prewarmed DMEM +2% FCS at a concentration 106/ml followed by a 1-hr incubation at 37°C with 4 µg/ml Hoechst 33342 ( Sigma , St Louis , MO ) . Cells were washed twice and incubated with antibodies as indicated . Doublets and dead cells were excluded prior to cell cycle analysis . For detection of early and late apoptotic cells , Annexin V staining was performed according to the manufacturer’s protocol . For LT-HSC methylcellulose colony assays , 40 double-sorted LSK CD150+CD48− cells from Lsd1fl/fl or Lsd1fl/fl Mx1Cre ( pre-poly ( I:C ) ) animals were plated per 1 . 1 ml of Methocult M3434 ( Stem Cell Technologies , Vancouver , Canada ) containing 1000 U/ml mouse interferon alpha ( R&D Systems , Minneapolis , MN ) to activate Mx1Cre-mediated deletion in vitro . Single colonies were picked and PCR genotyped to verify Lsd1 deletion . RNA was isolated using the RNeasy Plus Mini Kit ( Qiagen , Valencia , CA ) according to the manufacturer’s protocol . cDNA was synthesized with the iScript cDNA synthesis kit ( Bio-Rad ) . Real-time quantitative RT-PCR was performed using the iQ SYBR Green Supermix ( Bio-Rad ) and analyzed by real-time PCR on a MyiQ real-time PCR instrument ( BioRad , Hercules , CA ) . Relative expression was quantified using the ΔΔCt method as described previously ( Livak and Schmittgen , 2001 ) . Real-time PCR primers are available upon request . Individual primer sets ( total of 40 ) were pooled to a final concentration of 0 . 1 µM for each primer . Individual cells were sorted directly into 96-well PCR plates loaded with 10 µl RT-PCR master mix ( 5 . 0 µl CellsDirect reaction mix; Invitrogen; 1 . 0 µl primer pool; 0 . 2 µl RT/Taq enzyme; Invitrogen; 3 . 8 µl nuclease-free water ) in each well . Sorted plates were immediately frozen on dry ice . Cell lysis and sequence-specific reverse transcription were performed at 50°C for 30 min . The reverse transcriptase was inactivated by heating to 95°C for 2 min . Subsequently , in the same tube , cDNA went through sequence-specific amplification by denaturing at 95°C for 15 s , and annealing and amplification at 60°C for 5 min for 20 cycles . These preamplified products were diluted fivefold prior to analysis with Universal PCR Master Mix ( Applied Biosystems , Grand Island , NY ) , EvaGreen Binding Dye ( Biotium , Hayward , CA ) , and individual qPCR primers in 96x96 Dynamic Arrays on a BioMark System ( Fluidigm , San Francisco , CA ) . Ct values were calculated using the system’s software ( BioMark Real-Time PCR Analysis; Fluidigm ) . Bone marrow Gr1dim Mac1+ cells , bone marrow Lin−CD150+CD48−c-Kit+Sca-1+ cells , and fetal liver CD71+c-Kit+ cells from knockout and control mice were isolated using a FACSAria I ( BD , Franklin Lakes , NJ ) and sorted twice to achieve >95% purity . Total RNA was extracted with the RNeasy Micro Kit ( Qiagen , Valencia , CA ) , treated with DNaseI , and reverse transcribed with iScript ( BioRad , Hercules , CA ) . RNA extracted from Lin−CD150+CD48−c-Kit+Sca-1+ cells was amplified with the Ovation Pico WTA RNA Amplification System 2 ( NuGEN Technologies , San Carlos , CA ) . Single-stranded cDNA amplification products were purified using QIAquick PCR Purification Kit ( Qiagen , Hercules , CA ) and labeled with the FL-Ovation cDNA Biotin Module V2 ( NuGEN Technologies , San Carlos , CA ) . Hybridization to Affymetrix GeneChip Mouse Genome 430 2 . 0 arrays ( GeneChip Mouse Genome 430A for the erythroid samples ) , washing , and scanning was performed by the Dana-Farber Cancer Institute microarray core facility . CEL files were imported into the Gene Pattern Software Suite ( Reich et al . , 2006 ) for GC-RMA normalization and extraction of signal intensities . Independent biological repeats were combined by averaging the signal intensities of each probe represented on the microarray . Data were further analyzed by gene set enrichment analysis ( GSEA; Subramanian et al . , 2005 ) and IPA ( Ingenuity Systems , Redwood City , CA; www . ingenuity . com ) : The IPA functional analysis identified the biological functions and/or diseases that were most significant to the data set . Molecules from the data sets that met the ≥1 . 5-fold cutoff at p≤0 . 05 were associated with biological functions and/or diseases in the Ingenuity Knowledge Base were considered for the analysis . Right-tailed Fisher’s exact test was used to calculate a p-value determining the probability that each biological function and/or disease assigned to that data set is due to chance alone . GSEA provides a general statistical method to test for the enrichment of sets of genes in expression data , and has been particularly useful in identifying molecular pathways at play in complex gene expression signatures , as recently reported ( Subramanian et al . , 2005 ) . GSEA considers a priori defined gene sets , for example , genes in a signature such as the self-renewal associated signature or members of a pathway . It then provides a method to determine whether the members of these sets are overrepresented at the top ( or bottom ) of a gene list of markers that have been ordered by their correlation with a specific phenotype or class distinction , and produces a gene set–gene list specific Enrichment Score ( ES ) . The running enrichment score ( red line ) is graphed vs the gene number in a gene list ordered based on the correlation of interest . Simply , the higher the ES score and the earlier in the ordered gene list the max ES score is obtained , the greater the enrichment of the gene set . GSEA setting applied in this study: dataset was collapsed to gene symbols and 1000 permutations were run ( Permutation type: Gene Set ) . Fifty to 100 million 32D cells or 5 × 106 Gr1dim Mac1+ cells were cross-linked with 1% formaldehyde for 10 min at room temperature . Cross-linking was stopped with 125 mM glycine for 5 min , and cells were rinsed twice with 1× PBS . Lsd1-ChIP samples were first cross-linked with 1 . 5 mM EGS in PBS ( ethylene glycol-bis [succinimidyl succinate]; Sigma ) for 20 min at room temperature before continuing to formaldehyde fixation . Cells were resuspended , lysed in lysis buffer , and sonicated to solubilize and shear cross-linked DNA . The samples were sonicated using a Heat Systems ( Newtown , CT ) Ultrasonic Processor with a microtip . Samples were kept on an ethanol ice bath at all times . The resulting whole-cell extract incubated overnight at 4°C with approximately 5 µg antibody . Protein A Dynabeads ( Invitrogen ) were used for collection of chromatin . Beads were washed 2× with 10 mM Tris–HCl pH 7 . 4 , 1 mM EDTA , 0 . 1% SDS , 1% Triton X-100 , 0 . 1% NaDOC; 2× with 10 mM Tris–HCl pH 7 . 4 , 300 mM NaCl , 1 mM EDTA , 0 . 1% SDS , 1% Triton X-100 , 0 . 1% NaDOC; 2× with 10 mM Tris–HCl pH 8 . 0 , 250 mM LiCl , 2 mM EDTA , 0 . 5% NP40 , 0 . 5% NaDOC; and 2× with TE . Bound complexes were eluted from the beads ( 50 mM Tris–Hcl , pH 8 . 0 , 10 mM EDTA , and 1% SDS ) by heating at 65°C for 1 hr with occasional vortexing . Cross-linking was reversed by overnight incubation at 65°C . Ten percent of input DNA was also treated for cross-link reversal . For H3K4me1-occupied genomic regions , we performed ChIP-Seq experiments using Abcam ab8895 rabbit polyclonal antibody . The antibody was raised with a synthetic peptide conjugated to KLH derived from within residues 1–100 of human H3K4me1 . Antibody specificity was previously determined in Meissner et al . ( 2008 ) . For H3K4me2-occupied genomic regions , we performed ChIP-Seq experiments using Milipore ( Billerica , MA ) ( 07-030 ) rabbit polyclonal antibody . KLH conjugated synthetic peptide ( ARTMe2KQTAR-GC ) corresponding to amino acids 1–8 of human Histone H3 . Antibody specificity was previously determined in Boggs et al . ( 2001 ) . For H3K4me3-occupied genomic regions , we performed ChIP-Seq experiments using Milipore ( 04-745 ) rabbit monoclonal antibody . The antibody was raised with a synthetic Peptide containing the sequence [RTtrimKQ] in which lysine 4 is trimethylated on human histone H3 . Antibody specificity was previously determined in Flowers et al . ( 2009 ) . For H3K27Ac-occupied genomic regions , we performed ChIP-Seq experiments using Abcam ab4729 rabbit polyclonal antibody . The antibody was raised with a synthetic peptide conjugated to KLH derived from within residues 1–100 of human histone H3 , acetylated at K27 . Antibody specificity was previously determined in Creyghton et al . ( 2010 ) . For Lsd1-occupied genomic regions , we performed ChIP-Seq experiments using Abcam ab17721 rabbit polyclonal antibody . The antibody was raised with a synthetic peptide conjugated to KLH derived from within residues 800 to the C-terminus of human LSD1 . Antibody specificity was previously determined in Whyte et al . ( 2012 ) . Statistical analyses were performed using GraphPad Prism 5 ( GraphPad Software , La Jolla , CA ) . Unless specified differently , p values were calculated using the Student’s t test ( two-tailed ) . The GEO accession numbers for all unpublished gene expression and ChIP-Seq data reported in this paper are GSE40440 and GSE40605 . | Our blood contains many different types of cells . Red blood cells carry oxygen around the body , whereas white blood cells are a key part of our immune system . All these different types of blood cells are derived from special cells in our bone marrow called hematopoietic stem cells . The type of blood cell that the stem cell becomes depends on the genes that are expressed as proteins in that stem cell . Gene expression can be controlled in a number of ways , including epigenetic process that influence the expression of genes without altering the underlying sequence of bases in the DNA . For example , DNA is wrapped around histone proteins and the addition of a methyl group to these proteins , a process known as histone methylation , can increase the expression of a gene , whereas the removal of a methyl group ( demethylation ) can repress gene expression . Lysine-specific demethylase 1 ( Lsd1 ) is an enzyme that is known to mediate the demethylation of lysine amino acids on histone proteins . The role of Lsd1 in embryonic stem cells has been widely studied , and deletion of the gene that codes for Lsd1 is known to result in the death of mice embryos . However , very little is known about its roles in the later stages of mammalian development . Here , Kerenyi et al . use new genetic tools to knock out the gene for Lsd1 at different stages of development in order to examine its impact on the formation of new blood cells . They find that Lsd1 is required for the successful differentiation of hematopoietic stem cells into different types of blood cells , and that knocking out Lsd1 results in a severe loss of white and red blood cells . Moreover , they show that the lack of Lsd1 causes problems during both the early and later stages of development . Kerenyi et al . go on to demonstrate that Lsd1 regulates the activity of promoters and enhancers of various genes associated with hematopoietic stem cells . They also show that knocking out the Lsd1 gene results in impaired silencing of these genes , and that the incomplete expression of these genes is not compatible with the maturation of blood cells . Lsd1 has recently been proposed as the potential target for the treatment of leukemia and other blood disorders . However , the fact that a loss of Lsd1 function has adverse effects during both the early and later stages of blood cell development suggests that research into drugs that target Lsd1 should not begin until a suitable time window for the administration of such drugs can be identified . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"stem",
"cells",
"and",
"regenerative",
"medicine",
"chromosomes",
"and",
"gene",
"expression"
] | 2013 | Histone demethylase Lsd1 represses hematopoietic stem and progenitor cell signatures during blood cell maturation |
Sleep is a behavior that is found in all animals that have a nervous system and that have been studied carefully . In Caenorhabditis elegans larvae , sleep is associated with a turning behavior , called flipping , in which animals rotate 180° about their longitudinal axis . However , the molecular and neural substrates of this enigmatic behavior are not known . Here , we identified the conserved NK-2 homeobox gene ceh-24 to be crucially required for flipping . ceh-24 is required for the formation of processes and for cholinergic function of sublateral motor neurons , which separately innervate the four body muscle quadrants . Knockdown of cholinergic function in a subset of these sublateral neurons , the SIAs , abolishes flipping . The SIAs depolarize during flipping and their optogenetic activation induces flipping in a fraction of events . Thus , we identified the sublateral SIA neurons to control the three-dimensional movements of flipping . These neurons may also control other types of motion .
Sleep is a widespread phenomenon found in all animals that have a nervous system and that have been studied carefully . It is defined by behavioral criteria and can be identified in animals by the lack of voluntary movements , assumption of a specific posture , an increased arousal threshold , and homeostatic regulation ( Campbell and Tobler , 1984; Cirelli and Tononi , 2008; Allada and Siegel , 2008 ) . While sleep appears to be a relatively inactive state , at least if seen from a behavioral point of view , it is essential and likely serves several functions ranging from optimizing behavior , basic control of energy metabolism , macromolecule synthesis and clearance , to memory consolidation and gaining of deeper insight into logical problems ( Cirelli and Tononi , 2008; Diekelmann and Born , 2010; Stickgold and Walker , 2004; Stickgold , 2012; Mackiewicz et al . , 2007; Siegel , 2009; Xie et al . , 2013 ) . Because of its importance , sleep is under homeostatic control that ensures that enough sleep takes place ( Borbély , 1982; Porkka-Heiskanen , 2013 ) . The profound physiological and behavioral changes during sleep are controlled by the nervous system . Central to the control of sleep are sleep-active sleep-promoting neurons that release inhibitory neurotransmitters , such as GABA and neuropeptides , at sleep onset ( Saper et al . , 2005 , 2010 ) . Genetic analysis of sleep in different model organisms has uncovered sleep regulatory mechanisms and has shown that many pathways have conserved functions across species , supporting the view that sleep is evolutionarily ancient and conserved ( Sehgal and Mignot , 2011; Singh et al . , 2014; Zimmerman et al . , 2008 ) . C . elegans is an established model system to study the molecular basis of behavior . The hermaphrodite has a small nervous system containing 302 neurons with a known and invariant connectivity . The short generation time and molecular accessibility make it an attractive model to study ( Brenner , 1974; White et al . , 1986 ) . At the end of each larval cycle , C . elegans molt . Before shedding the old cuticle , larvae go through a developmental stage and behavioral state called lethargus during which they don’t feed . Locomotion behavior during lethargus can be described as quiescence bouts that are intermitted by motion bouts ( Iwanir et al . , 2013; Nagy et al . , 2014a , 2014b ) . Developmentally controlled quiescence behavior fulfills the behavioral criteria that define sleep in other organisms , such as decreased voluntary movement , an increased arousal threshold , reversibility , and homeostatic regulation , and is thus called sleep ( Iwanir et al . , 2013; Nagy et al . , 2014a; Raizen et al . , 2008; Trojanowski et al . , 2015; Trojanowski and Raizen , 2016; Driver et al . , 2013; Schwarz et al . , 2011 , 2012; Turek et al . , 2016; Cassada and Russell , 1975 ) . C . elegans larvae display behavioral quiescence also during dauer diapause ( Cassada and Russell , 1975; Gaglia and Kenyon , 2009 ) . In the adult , stress , satiety , and reduced insulin signaling can induce behavioral quiescence ( Gaglia and Kenyon , 2009; Van Buskirk and Sternberg , 2007; Hill et al . , 2014; You et al . , 2008 ) . How these different types of quiescence are related and whether all of them fulfill the definitions for sleep currently is under investigation . It appears that different types of quiescence are controlled by both distinct and overlapping mechanisms ( Trojanowski et al . , 2015; Trojanowski and Raizen , 2016; Kayser and Biron , 2016 ) . Here we study developmentally controlled sleep behavior in the larva . The timing of lethargus is controlled by LIN-42 , a homolog of the circadian regulator PER , that oscillates with the sleep-wake and developmental rhythm ( Jeon et al . , 1999; Monsalve et al . , 2011 ) . Crucial to the induction of larval sleep behavior during lethargus is a single neuron called RIS , a GABAergic and peptidergic neuron , which depolarizes at the onset of sleep and actively induces sleep by releasing inhibitory neurotransmitters including FLP-11 peptides ( Turek et al . , 2016 , 2013 ) . Peptide release to control behavioral states appears to be a common theme in sleep regulation in various species ( Turek et al . , 2016; Richter et al . , 2014; Nelson et al . , 2013 , 2014 ) . Thus , RIS and mammalian sleep-active neurons are functionally similar , because both are active at sleep onset and actively induce sleep through inhibitory transmitters ( Saper et al . , 2005 ) . C . elegans is typically cultured on the flat surface of agarose plates where it lies on its side and propels itself by ‘crawling’ , a locomotion behavior consisting of undulating body movements created by sinusoidal dorsal and ventral muscle contraction . Most motor neurons that control undulating body movement are found in the ventral cord . These neurons innervate both dorsal and ventral muscle . The body wall muscles run along the long axis of the body and are organized into four quadrants , two dorsal and two ventral quadrants , which run along the side of the dorsal cord or the ventral cord , respectively . Separate groups of cholinergic motor neurons exist that drive forward or backward movement through excitatory neuromuscular junctions onto dorsal or ventral muscle . In addition , GABAergic motor neurons play a modulatory , facilitating role in movement through contralateral inhibition . Thus , the motor circuits for crawling on plane surfaces are relatively well defined ( Zhen and Samuel , 2015 ) . In contrast , natural environments are rugged three-dimensional landscapes through which animals have to navigate . Simple dorsal and ventral contractions can hardly explain three-dimensional movements , which would require additional left-right body movements and thus a more complex regulation of musculature though a separate innervation of each muscle quadrant . C . elegans shows a number of additional types of movements that appear to be more complex than locomotion on a plane surface such as burrowing into substrates , nictation , which is a dispersal behavior during which worms assume an upright posture , and turning from their left to their right side ( Lee et al . , 2012; Beron et al . , 2015; Vidal-Gadea et al . , 2015 ) . Left-right turning , or flipping , occurs mostly during lethargus . It is a movement in which the worm turns 180° around its longitudinal axis . It was suggested that this behavior might help loosen the old cuticle , but causality was never tested . Flipping does not normally occur outside of lethargus except after stressful handling of the worms ( Singh and Sulston , 1978 ) . Most flips were shown to occur during quiescence bouts but they can also be found during motion bouts . They are usually initiated in worms that contain a relaxed body posture that contains only one body bend , which is the typical posture during lethargus . By contrast , worms outside of lethargus typically contain more than one body bend . The emergence of flips from a single-bend posture suggests facilitation by this posture , consistent with the idea that biomechanical constraints exist for this behavior . Thus , hypothetically , flipping occurs mostly during lethargus because this stage is characterized by reduced body bends . Flipping is associated with sleep behavior in the sense that most flips emerge from quiescence bouts during which there is a sleep-specific , relaxed , posture ( Tramm et al . , 2014 ) . Even though flipping behavior is long known , the molecular and neural substrates and molecular mechanisms through which worms flip are unknown . Here , we study left-right turning during larval sleep in C . elegans as an example for a sleep-associated behavior . We identified mutants in the NK2 homeobox gene ceh-24 that do not flip during sleep . Analysis of ceh-24 combined with calcium imaging and optogenetics suggests that the sublateral SIA neurons are required for flipping but also for other types of movements . The SIAs appear to act as motor neurons that innervate all of the four muscle quadrants separately , and they are thus ideally suited to control complex movements such as flipping .
Flips have been reported to occur mostly during lethargus quiescence bouts ( Tramm et al . , 2014 ) . To corroborate these findings , we looked at wild type worms and analyzed their locomotion speed before flipping . We cultured L1 larvae in microfluidic chambers that were filled with bacterial food and filmed their development and behavior from the mid L1 stage until the early L2 larva using time-lapse video microscopy ( Bringmann , 2011; Turek et al . , 2015 ) . We identified lethargus in these movies by the absence of feeding as seen by the cessation of pharyngeal pumping and manually scored the videos for flipping by monitoring the side on which the worms were lying by identifying the position of the developing gonad , which is located ventrally ( Figure 1A ) . We measured the movement speed in the five seconds before each flip and plotted a frequency distribution of speeds versus the frequency distribution of flipping . The speed distribution showed that most flips occurred during low movement speeds , consistent with the idea that flips emerge from quiescence bouts ( Figure 1B ) ( Tramm et al . , 2014 ) . To test whether sleep behavior only coincides with or actually is required for flipping we analyzed a mutant that has greatly impaired sleep . We used a deletion in aptf-1 , a conserved regulator of sleep , which completely eliminates locomotion quiescence during lethargus , and counted flips in this mutant ( Turek et al . , 2013; Kucherenko et al . , 2016 ) . Mutant worms went through the molting cycle , but flipping during lethargus was strongly reduced in aptf-1 ( - ) , consistent with the idea that sleep facilitates flipping ( Figure 1C ) . Thus , our results support the view that most flips occur during periods of low mobility that are characteristic of sleep . Also , sleeping behavior appears to facilitate flipping . 10 . 7554/eLife . 24846 . 003Figure 1 . left-right turning is associated with sleep and appears to be facilitated by this behavior . ( A ) Flipping during sleep: Images showing flipping in an example time lapse movie of an L1 larva cultured in a hydrogel microcompartment . Shown are DIC images . At time point 0s the developing gonad is located on the left side ( arrowhead ) , at time point 0 . 5 s the larva turns and at time points 1s and 1 . 5s the developing gonad can be seen on the right side . The right plot shows the behavior over time during the L1 to L2 stage for one animal . Lethargus is defined here as a lack of feeding as seen as a lack of pharyngeal pumping and it is the phase in which sleep occurs , i . e . movement is strongly reduced and quiescence bouts occur . Nose speed measurements show the reduction of mobility during sleep . Flips are displayed in green . ( B ) Most flips occur during phases of low mobility . Shown is a frequency distribution of flips as a function of movement five seconds before the flip . Compare also images in A ) . ( C ) Flipping is strongly reduced in aptf-1 ( - ) , which is lacking sleep behavior , consistent with the view that sleep behavior facilitates flipping . DOI: http://dx . doi . org/10 . 7554/eLife . 24846 . 003 No mutants have been reported that have normal locomotion quiescence but lack flipping behavior . In order to dissect sleep-specific flipping behavior , we decided to identify such mutants . We looked through existing movies that we had taken previously in a screen to identify sleep mutants . For the initial screen , mutant strains were cultured inside agarose hydrogel microchambers and were filmed and scored for flipping as described above . We found one strain that appeared not to flip ( methods ) . This strain carried a deletion in ceh-24 . To characterize the ceh-24 mutant phenotype , we cultured worms again inside microfluidic chambers made from agarose hydrogel and quantified both locomotion quiescence and flipping behavior in two ceh-24 mutants . The ceh-24 ( cc539 ) allele identified in the screen deletes almost the entire gene and likely is a molecular null . ceh-24 ( tm1103 ) is an independently isolated allele that also deletes large parts of the gene . Lethargus was clearly identifiable in the wild type and in the mutants by a phase of feeding cessation as seen by the absence of pharyngeal pumping , which was followed by cuticle shedding . Both mutants showed strongly reduced flipping behavior . Whereas wild type worms flipped 0 . 4/h during wake and 6 . 4/h during sleep , both ceh-24 mutants had strongly reduced flipping , 0/h during wake and 0 . 1/h during sleep , with most individuals showing no flipping at all ( Figure 2A ) . The presence of a flipping defect in two independently isolated strains carrying ceh-24 deletions strongly suggested that ceh-24 is required for flipping . To confirm that ceh-24 is required for flipping we performed a rescue analysis by introducing a transgene carrying the ceh-24 gene driven by its own promoter into a ceh-24 deletion background . The ceh-24 transgene rescued flipping confirming that the flipping defect seen in ceh-24 mutants was indeed caused by deletion of this gene ( Figure 2C ) . Locomotion quiescence during sleep appeared normal in ceh-24 ( - ) worms as judged from an analysis of locomotion speeds and extraction of quiescence bouts . Speeds were normal and quiescence bout duration and frequency were indistinguishable in wild type and ceh-24 ( - ) larvae ( Figure 2B , Figure 2—figure supplement 1 ) . Thus , we conclude that ceh-24 is required for flipping . 10 . 7554/eLife . 24846 . 004Figure 2 . ceh-24 is required for left-right turning and escape from an indentation . ( A ) Flipping is strongly reduced in ceh-24 ( - ) : The left plot shows the reduction of total number of flips scored in a movie that lasted from two hours before sleep onset until two hours after sleep end ( Kolmogorov Smirnov test , wild type [N = 17] , ceh-24 ( tm1103 ) [N = 12] , ceh-24 ( cc539 ) [N = 7] ) . The right plot shows the flip rate per hour during and outside of lethargus ( paired Wilcoxon Signed Ranks test for comparison of sleep versus wake , and Kolmogorov Smirnov test for comparisons between genotypes , wild type [N = 17] , ceh-24 ( tm1103 ) [N = 12] , ceh-24 ( cc539 ) [N = 7] ) . ( B ) Sleep behavior appears normal in ceh-24 mutants: Shown is the reduction of nose speed during lethargus compared with the speed outside of lethargus in percent ( Kolmogorov Smirnov test , wild type [N = 5] , ceh-24 ( tm1103 ) [N = 6] , ceh-24 ( cc539 ) [N = 6] ) . Also shown are quiescence bouts displayed as the fraction of worms that was quiescent and the mean quiescence bout duration . ( C ) A ceh-24 transgene array rescues the flipping defect in ceh-24 ( - ) : a rescue transgene containing of the ceh-24 promoter and the ceh-24 coding region was transformed into ceh-24 ( - ) mutant worms and the total number of flips was scored ( Kolmogorov Smirnov test , animals were selected that expressed the array and had normal-looking sublateral neurons with proper sublateral processes , ceh-24 ( tm1103 ) [N = 24] , rescue [N = 21] ) . ( D ) Cuticle shedding appears normal in ceh-24 mutant worms as judged by the time from resumption of pumping until completion of cuticle shedding ( Kolmogorov Smirnov test , wild type [N = 17] , ceh-24 ( tm1103 ) [N = 12] , ceh-24 ( cc539 ) [N = 7] ) . ( E ) Escape from an indentation is impaired in ceh-24 ( - ) : Young adult worms were placed into shallow indentations that were printed into agarose and the time the worms needed to crawl out was scored ( two-sample t-test , wild type [N = 29] , ceh-24 ( tm1103 ) [N = 39] , ceh-24 ( cc539 ) [N = 19] ) . Chamber dimension was 700 μm × 700 μm , 65 μm deep . *** denotes p<0 . 001 , n . s . denotes p>0 . 05 . DOI: http://dx . doi . org/10 . 7554/eLife . 24846 . 00410 . 7554/eLife . 24846 . 005Figure 2—figure supplement 1 . Sleep behavior appears normal in ceh-24 ( - ) . Frequency distribution of nose speeds during lethargus and outside of lethargus in wild type , ceh-24 ( tm1103 ) , and ceh-24 ( cc539 ) . Relates to Figure 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 24846 . 00510 . 7554/eLife . 24846 . 006Figure 2—figure supplement 2 . Locomotion and response to mechanical stimulation of adults on an agar surface appears normal in ceh-24 ( - ) . ( A ) Crawling speed of wild type and ceh-24 ( tm1103 ) . ( B ) The number of spontaneous reversals of wild type and ceh-24 ( tm1103 ) . ( C ) The fraction of worms responding to gentle mechanical stimulation in wild type and ceh-24 ( tm1103 ) . Relates to Figure 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 24846 . 006 Flipping has been suggested to be required for cuticle loosening and shedding . We hence measured the length of the nonpumping period and the time it took to fully shed the cuticle . Pharyngeal pumping and resumption could be scored from DIC movies of worms grown in microfluidic chambers . Typically , after pumping restarted , the worm crawled out of the old cuticle , which could also be scored manually from the same DIC movies . The time point at which the worm had crawled out of the cuticle completely , i . e . also the tail had completely left the cuticle , was scored as the completion of the molt . Despite the absence of flipping , the time the animals took from resuming pumping until complete shedding of the old cuticle was not significantly altered in ceh-24 mutant animals and we could not detect any obvious defects in the new cuticle ( Figure 2D ) . Thus , we could not find any contribution of flipping to the molting process , suggesting that flipping rather is related to other processes . Is ceh-24 also required for other behaviors or is it specifically required for flipping ? To test for other defects in movement we looked at larvae in microchambers and at adults on a planar and on a structured surface . Compared with the wild type , ceh-24 ( - ) L1 larvae grown in microchambers showed slightly slowed down nose movement ( Figure 2—figure supplement 1 ) . Adult mutant worms crawling on a planar NGM plate moved with normal speed and showed a normal fraction of forward and reverse movement ( Figure 2—figure supplement 2A–B ) . In order to challenge the worms we assayed the ability of adults to escape from an indentation in the agar surface . We cast agar surfaces containing small box-shaped indentations by using PDMS molds and placed individual worms into these indentations . The worms then quickly crawled out of the indentation . We measured the time each individual worm needed to escape . Wild type worms pushed their heads over the rim of the indentation and crawled out without any signs of effort in about two minutes . ceh-24 mutant worms could also eventually escape from the indentation , but needed four times longer for this ( Figure 2E ) . Thus , ceh-24 ( - ) had problems crawling out of an indentation . This is likely not caused by a defect in sensory systems , because ceh-24 ( - ) have normal chemotaxis to both volatile and non-volatile odorants ( Harfe and Fire , 1998 ) . Also , ceh-24 ( - ) responded normally to mechanical stimulation ( Figure 2—figure supplement 2C ) . These results suggest that ceh-24 mutant worms have defects not only in flipping but also in other types of movements . Previous work showed that ceh-24 is expressed in both muscle ( m8 muscle of the pharynx and vulva ) and in three types of sublateral motor neurons including the two SMDV neurons , the four SIB neurons , and the four SIA neurons ( Harfe and Fire , 1998; Kennerdell et al . , 2009 ) . ceh-24-expressing sublateral neurons are cholinergic and they are still present in ceh-24 mutant animals ( Harfe and Fire , 1998; Duerr et al . , 2008; Pereira et al . , 2015 ) . The molecular functions of ceh-24 , however , are unknown . We confirmed the expression pattern of ceh-24 with a transgenic insertion that drove mKate2 expression from the ceh-24 promoter ( Figure 3A ) . To find out how ceh-24 acts we looked at the morphology of the sublateral neurons and at their neurotransmitter function in ceh-24 mutant worms . We crossed a ceh-24 deletion into the fluorescence reporter driven by the ceh-24 promoter and imaged the sublateral processes using spinning disc microscopy . In the wild type , straight sublateral processes were found to run along the muscle quadrants . In ceh-24 mutant worms , expression from the ceh-24 promoter was increased . Processes did not run straight along the muscle but diverted from it or branched . Processes also ended prematurely and did not extend as far posteriorly as in the wild type ( Figure 3B ) . Cholinergic function is determined by the expression of two genes , unc-17 , a synaptic vesicle acetylcholine transporter and cha-1 , a choline acetyltransferase gene that is required for the biosynthesis of acetylcholine . unc-17 and cha-1 are co-transcribed in an operon . Expression of this operon can be assayed by using an unc-17 promoter reporter , which detected cholinergic function in 159 out of 302 neurons including the SMD , SIB , and SIA neurons ( Duerr et al . , 2008; Pereira et al . , 2015; Rand and Russell , 1984; Alfonso et al . , 1994 , 1993; Rand , 2007 ) . To assay for cholinergic function , we looked at the expression of a GFP reporter driven from the unc-17 promoter . To clearly identify sublateral neurons among the many unc-17-expressing neurons , we also crossed a red fluorescent mKate2 marker driven by the ceh-24 promoter into this strain . In the wild type , expression from the cha-1 promoter was seen in SIA , SIB , and SMD neurons . However , in ceh-24 mutant worms , expression was either undetectable or greatly reduced in all the sublateral neurons tested ( Figure 3C ) . Thus , ceh-24 is required for normal process morphology and cholinergic transmitter expression in SIA , SIB , and SMD . 10 . 7554/eLife . 24846 . 007Figure 3 . ceh-24 is required for sublateral process formation and cholinergic function . ( A ) ceh-24 is expressed in the sublateral SMB , SIB , and SIA neurons as well as in the m8 muscle cells: An adult animal that is expressing mKate2 from the ceh-24 promoter is shown in DIC and in fluorescence imaging . Cartoons show worm and cell body outlines . ( B ) Sublateral processes are not formed properly in ceh-24 ( - ) : processes ended prematurely and branched in ceh-24 ( - ) in all animals tested ( Kolmogorov Smirnov test , p<0 . 001 ) ( N = 10 , 10/10 wt animals had no branching and 10/10 mutant animals had branching defects ) . ( C ) Expression of the synaptic vesicle acetylcholine transporter gene unc-17 is abolished or strongly reduced in the SIA neurons in ceh-24 ( - ) . The plot shows the quantification of unc-17-expressing neurons ( Kolmogorov Smirnov test , wild type [N = 7] , ceh-24 ( tm1103 ) [N = 9] ) . ( D ) ceh-17 mutants , which have a branching defect in the SIAs , have normal unc-17 expression and have a flip defect ( Kolmogorov Smirnov test , wild type [N = 7] , ceh-17 ( np1 ) [N = 14] ) . *** denotes p<0 . 001 , ** denotes p<0 . 01 , * denotes p<0 . 05 , n . s . denotes p>0 . 05 . DOI: http://dx . doi . org/10 . 7554/eLife . 24846 . 007 Which of the cellular defects prevented flipping ? Defects in sublateral process formation or lack of unc-17 expression ? Mutation of ceh-17 , a paired-like homeobox gene , also compromises process outgrowth in SIA neurons in a way that resembles the defects we observed in ceh-24 mutant worms ( Pujol et al . , 2000 ) . We hence tested if ceh-17 mutant worms were also flipping defective and whether they were also defective in unc-17 expression of the SIA neurons . In ceh-17 mutant worms , flipping was reduced by about half and in contrast to ceh-24 mutant worms they had no reduction in unc-17 expression in the SIA neurons ( Figure 3D ) . This experiment suggests that the proper process morphology of the SIA neurons is at least partially required for flipping and we tested the role of cholinergic function next . The sublateral processes lie inside a thin layer of hypodermis close to and running along the body wall muscle . Each of the sublateral neurons extends one sublateral process into one muscle quadrant . The sublateral processes contain periodic swellings containing synaptic vesicles and presynaptic densities suggesting that the sublateral neurons are the motor neurons that innervate adjacent muscle quadrants . The innervation pattern suggests that an individual control of each muscle quadrant by the sublateral neurons would be possible ( White et al . , 1986; Altun and Hall , 2011 ) . Our results above suggest that ceh-24 is required for flipping not only because it controls process outgrowth but also because it specifies a cholinergic transmitter expression . To test the hypothesis that cholinergic function in sublateral neurons is required for flipping , we used cell-specific RNAi to knock down cha-1 . We expressed double-stranded RNA corresponding to cha-1 from the ceh-24 promoter to knock down cha-1 . Indeed , cha-1 RNAi in ceh-24-expressing cells led to a flipping defect ( Figure 4A ) . Because the only cholinergic cells known to express ceh-24 are SMD , SIB , and SIA , cha-1 RNAi in these cells likely caused the phenotype . To identify which of the three types of sublateral neurons require cha-1 for flipping , we performed a rescue mosaic analysis . We used an extrachromosomal array expressing the RNAi construct plus a fluorescent mKate2 marker that co-expressed with the RNAi construct . The array was transmitted to and expressed in only a subset of ceh-24-expressing cells , generating a mosaic expression of the cha-1 RNAi transgene in a subset of ceh-24-expressing cells . We cultured and filmed 370 individuals inside microfluidic compartments and scored flipping behavior during L1 sleep . Afterwards , we increased the imaging resolution to check the transgene expression and manually selected those worms that had a clear expression of the array in only a subset of the sublateral neurons . For these , we determined their flip rate . We found one single individual that expressed the RNAi construct only in the SIA neurons and this individual was flipping defective . Also flipping defective were three individuals that expressed the RNAi construct in both SIA and SMD neurons , whereas individuals that carried the transgene only in the SMDs , were not flipping defective . Also , expression in the SIBs was not associated with a flipping defect ( Figure 4A ) . The mosaic analysis suggested that cha-1 expression in the SIA neurons but not in the other neurons is crucial for flipping , yet the numbers of individuals that had specific expression patterns that could be obtained from the mosaic analysis were small despite the large number of animals tested . To confirm the requirement of cha-1 in the SIA neurons we used another transgene that drove cha-1 RNAi only in SIA using the ceh-17 promoter . ceh-17 also expresses in the ALA neuron but this neuron is not cholinergic , which should result in a SIA-specific knockdown of cha-1 ( Duerr et al . , 2008; Pereira et al . , 2015; Pujol et al . , 2000; Suo and Ishiura , 2013 ) . Again , we measured flipping as before . Flipping was greatly reduced by SIA-specific cha-1 RNAi ( Figure 4B ) . To further confirm a role of the SIAs in flipping , we laser-ablated one or two of the four SIAs . We did not ablate all four SIAs , because this appeared to cause pleiotropic effects that likely were unspecific and were probably caused by large doses of laser irradiation . As expected , ablation of some of the SIAs led to a reduction of flipping ( Figure 4C ) . To test whether the SIAs are also required for other movements , we assayed for the capacity of SIA::cha-1RNAi and ceh-17 ( - ) worms to escape from an indentation . Like ceh-24 ( - ) worms , both types of SIA-defective worms had problems crawling out of the indentation ( Figure 4D ) . We also tried to rescue the flipping defect in ceh-24 ( - ) by expression of ceh-24 from the ceh-17 promoter , which expresses in the SIAs but not in other sublateral neurons . This transgene rescued neither the process outgrowth nor the flipping defect ( Figure 4—figure supplement 1 ) . Perhaps , the ceh-17 promoter did not recapitulate the proper spatiotemporal expression pattern or expression level required for rescue ( Levin et al . , 2012 ) . Thus , the evidence for a role of ceh-24 in the SIAs is indirect . The loss-of-function results imply that cha-1 expression and thus cholinergic function in the sublateral SIA neurons is required for flipping and other types of motion . Our Data are consistent with a model in which the SIAs act as cholinergic motor neurons to execute specific movements . 10 . 7554/eLife . 24846 . 008Figure 4 . Cholinergic function in the SIA neurons is required for flipping . ( A ) Mosaic analysis of cha-1 RNAi in the sublateral neurons suggests a crucial role for cholinergic transmission from the SIAs: A total of 370 individuals carrying an array expressing double-stranded RNA corresponding to cha-1 were filmed to score flipping . After filming , individuals with restricted expression in a subset of neurons were selected . ( B ) cha-1 RNAi in SIA only ( driven by the ceh-17 promoter ) confirms a crucial role for cholinergic transmission from these neurons ( wild type [N = 10] , ceh-14 ( tm1103 ) [N = 10] , pceh-17::cha-1 RNAi [N = 13] ) . ( C ) Ablation of some of the SIAs leads to a flipping defect ( mock ablated [N = 13] , one SIA neuron ablated [N = 19] , two SIA neurons ablated [N = 3] ) . ( D ) cha-1 RNAi in SIA leads to a defect in escaping from an indentation ( wild type [N = 29] , ceh-17 ( np1 ) [N = 25] , pceh-17::cha-1RNAi [N = 33] ) . The Kolmogorov Smirnov test was used for all experiments , *** denotes p<0 . 001 , ** denotes p<0 . 01 , * denotes p<0 . 05 . DOI: http://dx . doi . org/10 . 7554/eLife . 24846 . 00810 . 7554/eLife . 24846 . 009Figure 4—figure supplement 1 . Lack of rescue of ceh-24 ( - ) using ceh-24 ( + ) driven by the ceh-17 promoter . ( A ) Quantification of flipping in ceh-24 ( tm1103 ) and ceh-24 ( tm1103 ) ; Ex[pceh-17::ceh-24::SL2mKate2] . ( B ) Fluorescence images of outgrowth defects in ceh-24 ( tm1103 ) ; Ex[pceh-17::ceh-24::SL2mKate2] . DOI: http://dx . doi . org/10 . 7554/eLife . 24846 . 009 How does the activity of the SIAs generate flipping ? We used calcium imaging of the sublateral neurons to investigate their activities during flipping and analyzed the activity of the SIA neurons . We expressed the calcium indicator GCaMP6s , which increases green fluorescence upon calcium binding , from the ceh-24 promoter and cultured worms in microfluidic compartments to film neural activity using time-lapse microscopy and behavior across the sleep-wake cycle ( Chen et al . , 2013 ) . First , we selected and analyzed short sequences of frames covering 80 s that included a flip in the middle of the movie sequence . Individual sublateral neurons were identified manually in each frame and their fluorescence intensity was quantified . The GCaMP signal in the SMD neurons during flipping was low and we could not extract any signals . SIB neurons showed calcium sensor activation , but the SIB neurons were close to each other so that signal extraction for individual cells was not possible . Because the SIA neurons are crucial for flipping and because their signal could be extracted with single cell resolution , we focused on the analysis of these neurons . SIA neurons strongly activated during flipping as seen by a rise in calcium indicator fluorescence . While all of the four individual SIA neurons showed activation over baseline , the activation maximum of each of these four neurons differed , with strong activation of one or two neurons and with less activation of the other SIAs . The individual SIA that activated strongest varied between flips , i . e . different unequal activation patterns of the four SIAs were associated with flipping ( Figure 5A , individual traces before normalization are shown in Figure 5—figure supplements 1–2 ) . During activation , calcium transient maxima typically occurred at the same time for all four SIAs . I . e . , rather than activating sequentially , the sublateral neurons appeared to activate simultaneously . Thus , SIA neurons activated all at the same time but the transient maximum of the individual neurons differed . Flipping is not associated with just one particular activation pattern but with several activation patterns . We could not detect differences in timing of activation of the individual SIAs , i . e . the maximum of the calcium transient in each of the four SIAs occurred at the same time . This may be due to slow response times of the indicator or slow acquisition by the EMCCD camera . While subtle differences in timing may contribute to flip generation , our measurements rather suggest that timing differences of SIA activations are less important for flipping . Together , these data suggest a model in which longitudinal rotation is generated through activation of SIA neurons activating muscle quadrants on one side of the worm , which should result in muscle contraction and bending of either the left or the right side ( see model Figure 7 ) . Because the worms are confined to a plane , this bending could result in a longitudinal rotation in agreement with a previously proposed theoretical model of flipping that suggested that confinement to a plane is essential for flipping ( see discussion and model in Figure 7 ) ( Tramm et al . , 2014 ) . 10 . 7554/eLife . 24846 . 010Figure 5 . Sublateral SIA neurons are active during flipping . ( A ) The SIA neurons activate during flipping: calcium measurements were aligned to the flip . SIA activation and flipping coincided . SIA activation was seen both for flips from the right side to the left side and for flips from the left side to the right side . Histograms show statistical tests for calcium transient maxima ( Wilcoxon signed ranks test , right to left flip [N = 14] , left to right flip [N = 12] ) . ( B ) SIA neurons show calcium transients both outside and during lethargus: SIA calcium peaks were compared in three conditions , outside lethargus , during lethargus without coincidental flipping , and during flipping ( Mann-Whiney U test , nine individual worms were analyzed , numbers of activation transients analyzed were: 17 during flip , 63 during lethargus without flipping , 148 outside lethargus without flipping ) . ( C ) The probability that an activation of the SIAs coincides with a flip is increased during lethargus . ( Mann-Whiney U test , N = 9 ) . *** denotes p<0 . 001 , ** denotes p<0 . 01 , * denotes p<0 . 05 , n . s . denotes p>0 . 05 . DOI: http://dx . doi . org/10 . 7554/eLife . 24846 . 01010 . 7554/eLife . 24846 . 011Figure 5—source data 1 . SIA activation transients are highly variable outside of lethargus . The SIAs activate with different transient strengths relative to each other and variably during and outside of lethargus: Individual SIA activation patterns during wake were unequal regarding their transient strengths and highly variable between activation transients . Each row ( A–H ) contains three example traces from the same individual worm , one example is shown for a flip , one for a trace containing transients during lethargus without flipping and one for a trace outside of lethargus . DOI: http://dx . doi . org/10 . 7554/eLife . 24846 . 01110 . 7554/eLife . 24846 . 012Figure 5—figure supplement 1 . Sample calcium transients during left-to-right flipping . Individual calcium traces during left-to-right flipping before normalization . DOI: http://dx . doi . org/10 . 7554/eLife . 24846 . 01210 . 7554/eLife . 24846 . 013Figure 5—figure supplement 2 . Sample calcium transients during right-to-left flipping . Individual calcium traces during right-to-left flipping before normalization . DOI: http://dx . doi . org/10 . 7554/eLife . 24846 . 01310 . 7554/eLife . 24846 . 014Figure 5—figure supplement 3 . SIA baseline calcium sensor signals are reduced during lethargus and heatmap of calcium signal maxima . ( A ) Baseline calcium signal levels are significantly reduced during lethargus . ( B ) Heatmap of the calcium transient during flipping and during outside of flipping for the SIA neurons for 17 individual animals . DOI: http://dx . doi . org/10 . 7554/eLife . 24846 . 014 The activation of the sublateral neurons during flipping is consistent with their role in this behavior . But why do flips mostly occur around the time of lethargus ? The simplest model would be that the SIA neurons would activate only during lethargus and would thus confine flipping to this developmental stage . However , because the sublateral neurons appear to be also involved in other types of movements these neurons could also be active outside of flipping . We hence looked at sublateral neuron activation also outside of flipping by scoring all activation transients during and outside of lethargus . The neural calcium activity of the SIAs can be described as a baseline activity plus transients above this baseline . We analyzed time-lapse movies that covered the time before , during , and after lethargus and quantified baseline activity and calcium transients by counting their frequencies and maxima and compared calcium transients during three conditions: flipping during lethargus , no flipping during lethargus , and no flipping outside lethargus ( Figure 5B , individual sample traces before normalization can be found in Figure 5—source data 1 ) . Outside of lethargus , many calcium transients were visible . During lethargus , baseline calcium activity was slightly but significantly reduced ( Figure 5—figure supplement 3A ) . Calcium transients , however , were still visible ( Figure 5C , Figure 5—source data 1 and Figure 5—figure supplement 3B ) . During lethargus , the SIA sublateral neurons activated 0 . 6 times per minute , and 19% of activation events were associated with a flip . Outside of lethargus , sublateral neurons activated 0 . 5 times per minute , but only 1% of these events coincided with flipping ( Figure 5C ) . Thus , like in many other neurons , baseline calcium activity appeared to be reduced during lethargus in the SIAs . However , calcium transients still occurred during lethargus , consistent with the idea that the SIAs are required for flipping . The majority of SIA calcium transients were not associated with flipping both in and outside of lethargus , but the probability that a transient coincided with a flip was increased in lethargus . Could this be explained by different activation patterns of the SIAs , with only a subset of activity patterns leading to a flip ? There was variability in the activation pattern of the SIAs both for transient patterns that were associated with flipping and for transient patterns that were not associated with flipping ( Figure 5—source data 1 and Figure 5—figure supplement 3B ) . Thus , from our calcium data , we could not explain why a certain activation transient pattern would cause a flip and why others would not . Perhaps , other factors determine whether SIA activation results in a flip . Calcium imaging showed that flipping correlated with activation transients in the SIA neurons . But is SIA activation also causative for flipping ? To test this idea we activated the SIA sublateral neurons optogenetically by expressing the Channelrhodopsin variant ReaChr , which can be activated by green-to-red light , in the SIA neurons using the ceh-17 promoter and by providing green light stimuli during and outside of lethargus ( Lin et al . , 2013; Urmersbach et al . , 2016 ) . We cultured worms expressing ReaChr in the SIA neurons in the presence of retinal in microfluidic chambers and activated ReaChr repeatedly every 30 min using a 5 s flash of green/orange light . We filmed the behavior before , during , and after the stimulation using DIC imaging . From these movies , we manually determined the stage of the animal ( lethargus or not ) and scored whether the worms flipped during ReaChr activation . ReaChr activation resulted in flipping in a fraction of stimulations and the efficiency depended on the behavioral state: Whereas stimulation outside of lethargus led to flipping in 8% of stimulations , during lethargus it led to a flip rate in 17% of stimulations ( Figure 6A ) . Thus , optogenetics showed that the SIAs can trigger flipping in a fraction of stimulations , consistent with the observation that not every activation of the SIAs coincides with flipping . The probability that sublateral neuron stimulation leads to flipping was increased during lethargus , supporting the hypothesis that lethargus generates a facilitating environment for flipping . Calcium imaging suggested that the strength of SIA activation transients differed in the four individual SIA neurons . However , we have no functional evidence that only unequal SIA activation patterns can lead to flipping . Thus , one might ask whether optogenetic induction of flipping through SIA activation is consistent with the idea that unequal activation of the four SIAs is required for flipping . We hence quantified the expression of ReaChr from the extrachromosomal array in the SIAs and found that the expression levels differed in the individual neurons ( Figure 6—figure supplement 1 ) . Thus , perhaps , optogenetic activation recapitulated an unequal activation pattern , but this is speculative . Alternatively , the unequal activation pattern of the SIAs may not be crucial for flipping and flipping may also be triggered by equal activation . Targeted illumination approaches could be used in the future to control the activation of the SIAs individually and to test the role of the unequal activation patterns ( Guo et al . , 2009; Leifer et al . , 2011; Stirman et al . , 2011 ) . 10 . 7554/eLife . 24846 . 015Figure 6 . Optogenetic activation of SIA neurons can induce flipping with a higher chance during lethargus . ( A ) ReaChR-induced depolarization of the SIAs triggers flipping with a higher chance during lethargus . The SIAs were filmed and were activated every 30 min with yellow light and we scored whether worms flipped during these short movies . SIA activation could induce flipping in and outside of lethargus , but the likelihood of triggering a flip was higher during lethargus . Consistent with the calcium imaging data , the majority of SIA activations did not coincide with flipping . Flipping was impaired in hyperactive egl-30 mutant worms and increased in hypoactive worms . As controls we used worms of the same strain that did not visibly express the SIA::ReaChr array . Wild type ( no transgene N = 25 , transgene N = 38 ) , egl-30 ( tg26 ) ( no transgene N = 20 , transgene N = 40 ) , egl-30 ( n686 ) no transgene N = 15 , transgene N = 14 ) . ( B ) Endogenous flipping is impaired in hyperactive worms and increased in hypoactive worms . Wild type ( N = 17 ) , egl-30 ( tg26 ) ( N = 13 ) , egl-30 ( n686 ) ( N = 11 ) . Kolmogorov-Smirnov test was used for comparisons between genotypes , paired Wilcoxon Signed Ranks test for sleep-wake comparisons , *** denotes p<0 . 001 , ** denotes p<0 . 01 , * denotes p<0 . 05 , n . s . denotes p>0 . 05 . DOI: http://dx . doi . org/10 . 7554/eLife . 24846 . 01510 . 7554/eLife . 24846 . 016Figure 6—figure supplement 1 . Heatmap showing variability in extrachromosomal array expression of ReaChr in the SIA neurons . ( A ) Heatmap of mKate2 expression ( fused to ReaChr ) in individual SIAs for multiple individual animals . Compare with Figure 6 . DOI: http://dx . doi . org/10 . 7554/eLife . 24846 . 016 Why does flipping occur mostly during lethargus and why does it only occur during a small fraction of SIA activations ? One hypothetical model consistent with our calcium imaging data is that lethargus generates a facilitating environment during which the likelihood that sublateral activation transients result in flipping is increased . With other words , conditions outside of sublateral neuron activation could either inhibit or promote flipping . Lethargus is characterized by systemic changes in neuronal activity , leading to reduced muscle activity and a relaxed posture . Recently , a biomechanical model was proposed in which a relaxed body posture with fewer body bends facilitates flipping of worms confined to a plane . Consistent with this result , ectopic flipping could be observed in a V-ATPase mutant that showed a relaxed body posture outside of lethargus ( Iwanir et al . , 2013; Nagy et al . , 2014b; Schwarz et al . , 2011 , 2012; Tramm et al . , 2014 ) . We hence set out to directly test the idea that altered activity levels influence the probability that SIA activation leads to flipping . We investigated mutations that change the activity of neurons and muscles globally . We chose two mutants in egl-30 , a G alpha q protein required for the activity of excitable cells , because loss- and gain-of-functions exist for this gene and because the role of G alpha q in facilitating synaptic transmission is understood quite well . While null mutants of egl-30 are lethal , hypomorphic reduction-of-function mutants such as egl-30 ( n686 ) are slow and have a more relaxed body posture . Hypermorphic gain-of-function mutations in egl-30 such as egl-30 ( tg26 ) have increased activity of excitable cells and are behaviorally hyperactive ( Brundage et al . , 1996; Trent et al . , 1983; Park and Horvitz , 1986; Schwarz and Bringmann , 2013 ) . We imaged egl-30 ( n686 ) and egl-30 ( tg26 ) across the developmental cycle in microfluidic compartments and quantified flipping . Reduction-of-function mutation of egl-30 led to an increase in flipping rate during lethargus and these mutant worms also showed prominent flipping outside of lethargus . Gain-of-function mutation of egl-30 led to a decrease of flipping in and outside of lethargus ( Figure 6B ) . Next , we optogenetically activated SIA neurons again using ReaChr as described above and measured the efficiency by which neural activation caused flipping . Flipping probability per ReaChr activation was increased in the egl-30 reduction-of-function mutant , and was decreased in the gain-of-function mutant ( Figure 6A ) . Thus , changes in the global activity of excitable cells appear to affect the probability that sublateral neuron activation leads to flipping . While increased behavioral activity is associated with a decreased likelihood that sublateral neuron activation causes flipping , a decreased behavioral activity is associated with an increased probability of flipping . These results are consistent with a model in which flipping is the result of two processes: First , depolarization of SIA neurons , which provides a neuronal basis for triggering this active behavior . And second , a facilitating environment during lethargus that may consist of a relaxed body posture that increases the probability that SIA activation leads to flipping ( Tramm et al . , 2014 ) .
During lethargus , both motion bouts and quiescence bouts occur . Quiescence bouts are characterized by strongly reduced locomotion , reduced sensory responsiveness , and a more relaxed body posture . Quiescence bouts are intermitted by motion bouts during which larvae show low levels of locomotion but do not feed . Animals during lethargus motion bouts are more responsive to stimulation compared with animals during a quiescence bout yet have dampened sensory neuron activity and reduced body curvature compared with animals outside of lethargus ( Iwanir et al . , 2013; Raizen et al . , 2008; Schwarz et al . , 2011; Cho and Sternberg , 2014 ) . Quiescence bouts fulfill the behavioral criteria that define sleep but the exact nature of mobility bouts is not clear yet . At present a parsimonious yet hypothetical interpretation may be that motion bouts present a state that is in between wake and quiescence bout . Most flips emerge from periods in which there is little locomotion and the worm displays a relaxed body posture typically containing only one body bend ( Iwanir et al . , 2013; Tramm et al . , 2014 ) . Flipping also appears to depend on sleeping behavior , as it is strongly reduced in aptf-1 mutants that do not show any quiescence bouts during lethargus . The occurrence of flips during periods of low mobility may seem counterintuitive at first , as flipping constitutes a motion and a flip could thus be , by definition , regarded as a motion bout . However , x-y movements of the nose and body typically define a motion bout ( Iwanir et al . , 2013; Nagy et al . , 2014b ) . Flipping consists only of a rotation along the longitudinal axis . During this movement , there is almost no x-y movement of the nose or body ( see Figure 1A for an illustration and compare with [Tramm et al . , 2014] ) . A motion threshold cut-off is used in imaging studies to define quiescence bouts . Small movements typically occur also during quiescence bouts . The small motions associated with flipping often fall below the threshold of a motion bout and thus flipping occurs , by this definition , mostly during quiescence bouts ( Nagy et al . , 2014b ) . Irrespective of the classification into motion bouts and quiescence bouts , flipping occurs preferentially during phases of low mobility . CEH-24 is a conserved NK2 class homeodomain protein that is homologous to mammalian NKX2-1 , whose mutation leads to cancer cell survival and progression in humans ( Yamaguchi et al . , 2013 ) . Work on the mammalian homolog showed that this protein is required for organogenesis of the thyroid , lung , pituitary , and ventral forebrain ( Stanfel et al . , 2005; Kimura et al . , 1996 ) . Here we found that CEH-24 is required for left-right flipping and other types of movements . CEH-24 is expressed and plays a role in the sublateral neurons , where it is required for proper formation of sublateral processes and for the expression of acetylcholine . Thus , the regulation of neural development appears to be a conserved function of CEH-24 . These results strongly suggest that ceh-24 acts in flipping through controlling process morphology and acetylcholine expression in the sublateral neurons . Cell-specific RNAi against the acetylcholine-synthesizing gene cha-1 showed that acetylcholine expression in SIA neurons is required for flipping and suggests that the SIAs are required for flipping . While the most likely model is that ceh-24 acts in the SIA neurons to control cholinergic function required for flipping , the evidence for a role of ceh-24 in the SIAs is indirect , because a rescue of ceh-24 ( - ) specifically in the SIAs was not possible . Also , it is possible that other sublateral neurons contribute to flipping in addition to the SIAs . Through loss-of-function experiments we have identified cholinergic function in the SIAs to be crucial for flipping and thus found a crucial role of the SIAs in flipping . These neurons were initially thought to have relatively little output onto other neurons ( White et al . , 1986 ) . Recent reinvestigation of the connectivity of the C . elegans nervous system using electron microscopy suggests that additional postsynaptic partners exist ( wormwiring . org ) . Thus , it is formally possible that sublateral neurons act through other neurons to control muscle activity required for flipping . Electron microscopic reanalysis also found synaptic vesicles and presynaptic densities along the sublateral chord , suggesting that the sublateral neurons act as motor neurons and that the sublateral processes directly innervate the four muscle quadrants ( Altun and Hall , 2011 ) . Separate control of each muscle quadrant has previously been proposed to underlie flipping , but the neural substrates were not known ( Tramm et al . , 2014 ) . The separate innervation of left and right muscle quadrants may not only be an anatomical prerequisite for flipping , but could potentially also underlie other types of movements . While the sublateral neurons have long been known , their functions have remained unidentified . Here we provide the first functional evidence for a role of the sublateral SIA neurons , which is the control of flipping , a three-dimensional body movement . In addition , the SIAs also have roles in other types of movements , as demonstrated by assaying the ability to escape from an indentation . Consistent with this view , SIA activation transients occur outside of lethargus and most transients do not coincide with a flip . During lethargus , the baseline calcium activity is reduced , but depolarization transients still occur . Dampening of calcium activity during lethargus has also been reported for other neurons ( Iwanir et al . , 2013; Schwarz et al . , 2011; Turek et al . , 2013; Schwarz and Bringmann , 2013; Cho and Sternberg , 2014; Choi et al . , 2013 ) . Despite the reduced baseline activity of the SIAs during lethargus , the SIAs still show calcium transients during this stage consistent with the view that these neurons trigger flipping . Flipping likely requires a separate activation of the left and right body wall muscle quadrants . Consistent with this view , SIA neurons activated with different calcium maxima with regard to the individual neuron with one or two of the four neurons activating strongest , and the other SIA neurons activating less . Most of the SIA activation events do not lead to a flip suggesting that flipping depends not only on activation of the SIAs , but also on other factors . Outside of lethargus , neuronal activity is high and promotes an undulating body posture that forces the larva to lie on its side ( Zhen and Samuel , 2015 ) . The stronger the dorso-ventral body muscles contract , the stronger the curvature and the more force needs to get exerted to turn the worm about its longitudinal axis . During lethargus , worms engage in sleeping behavior and neuronal and muscle activity is lower and the body muscles are more relaxed ( Iwanir et al . , 2013; Schwarz et al . , 2012; Tramm et al . , 2014 ) . If confined to a plane , less force should be necessary to turn the worm about its longitudinal axis if it is only slightly bent compared with an undulating worm that contains more than one bend , as multiple body bends could fix the worm on its side . The same activation strength of the SIAs that may not be sufficient to turn the worm outside of lethargus may become sufficient to turn it during lethargus . As a consequence , flipping would occur more often during lethargus . Thus , the physiological changes that trigger sleep behavior during lethargus would also facilitate flipping . This would explain why flipping is associated with sleeping behavior during lethargus . For example , imagine a sleeping larva that is concavely bent on the ventral side . It could activate SIA neurons leading to a contraction of the right dorsal muscle quadrant . Such a posture would be energetically unfavorable because now the strongest muscle contraction would occur on the right dorsal side , but the dorsal side is convexly bent . A more energetically favorable configuration would be that the dorsal side would be concavely bent . To transition to this energetically more favorable configuration the muscle contraction could either move tail and nose to the opposite side or the worm could rotate about its longitudinal axis to achieve a more bend dorsal side . The first option is impaired due to the confinement of the worm to a plane . Thus , the transition to the posture in which the more contracted side also is the concavely bend side rather occurs through longitudinal rotation ( Figure 7 , [Tramm et al . , 2014] ) . This behavior does not occur during waking behavior outside of lethargus , because the worm is fixed by its undulating body posture , preventing longitudinal rotation . In summary , hypothetically , flipping could be possible only during lethargus for two reasons: first , because the sublateral neurons still depolarize , and second because the posture is relaxed . This model would not need a complicated choreography in which different sublateral neurons activate sequentially to achieve flipping . While a timed choreography would result in more efficient flipping , we did not find evidence for such a model in our calcium imaging experiments ( Tramm et al . , 2014 ) . The biomechanical facilitation and the lack of a sequential choreography could together explain why SIA neuron activation results in flipping in only a small faction of events . Together , these data support the hypothesis that both , activity of the SIAs and a relaxed body posture , are required for flipping . 10 . 7554/eLife . 24846 . 017Figure 7 . Model for SIA neuron-controlled flipping . ( A ) Anatomy of the innervation of body wall muscles by sublateral neurons . Cross-section through C . elegans showing the ventral and dorsal cords that control undulating 2D movement . Body wall muscle is organized into four quadrants that run along the long axis of the worm . Along the four body wall muscle quadrants run the sublateral processes , which each emerge from one sublateral SIA neuron ( SIAVL , SIAVR , SIADL , and SIADR ) . Thus , the separate innervation of the four body wall muscle quadrants by the four sublateral SIA processes provides an ideal anatomical basis for controlling three-dimensional movements . ( B–C ) Model for SIA-induced flipping: B ) In a worm that is outside of lethargus , the body wall muscles are under tension and the SIA neurons may act to control specific movements . Flipping is prohibited because the worm is confined to the plane by the undulating body posture . ( C ) During lethargus , sleep behavior occurs and the overall activity of neurons and muscles is reduced and the worm assumes a less-bent posture . Activation of the SIAs may trigger flipping in some cases in which the musculature on the convex side contracts . This favors a bent posture with the side of the contraction occurring on the concave side . To transition to this posture , the worm rotates 180 degrees around its longitudinal axis . Sublateral SIA neurons likely act as motor neurons that control this three-dimensional motion through separate innervation of the four body wall muscle quadrants . DOI: http://dx . doi . org/10 . 7554/eLife . 24846 . 017 What are the functions of left-right turning during lethargus ? We could not find evidence for a role of flipping in molting . The shedding of the cuticle occurs several minutes after the worm has resumed pumping and locomotion activity . Vigorous movements that only slow down once the worm has shed its old cuticle coincide with cuticle shedding . This rather suggests that cuticle shedding is an active process that is triggered not by flipping but by vigorous movements after the period of behavioral quiescence . By contrast , flipping rather appears to be a gentle movement . What then may the function of flipping be ? Flipping leads to small changes in the posture , but does not constitute locomotion . Brief muscle activations or changes in the body posture may stimulate metabolic processes in otherwise quiescent muscles , but this idea is purely speculative . We initially set out to understand the molecular and cellular basis of flipping . But the low frequency of SIA neuron activation that result in flipping and the strong activity of these neurons outside of flipping suggest that the major role of the SIA neurons is not in flipping . We found that the ability of ceh-24 mutant worms to crawl was almost normal . Crawling out of an indentation , however , was strongly impaired , suggesting that sublateral neurons are required for locomotion , maybe especially important for specific types of movements . The culture of C . elegans on plane agar surfaces has led to strong progress in the understanding of undulating locomotion . Natural environments , such as the rotting biological material on which C . elegans is typically found in the wild require three-dimensional navigation and locomotion ( Kiontke and Sudhaus , 2006 ) . Recent work is starting to establish assays for the study of such types of locomotion . One example is burrowing , with a downward tendency , using magnetic cues for orientation ( Beron et al . , 2015; Vidal-Gadea et al . , 2015 ) . Another is nictation , a dispersal behavior , in which dauer larvae stand on their tails displaying undulating body movements trying to attach to a larger animal in order to get dispersed and potentially reach new food sources ( Lee et al . , 2012 ) . It would be interesting to test the role of the sublateral neurons , and in particular the role of the SIAs , in these and other types of movements . Because ceh-24 ( - ) does not appear to have defects in sensory function and because cha-1 RNAi in the SIAs cause a motion defect it seems most likely that ceh-24 and the SIAs play a role in motor function rather than in sensory function . How could sensing of a three-dimensional environment be coupled to navigating through it ? The CEP neurons sense the mechanical properties of the surrounding on which C . elegans crawls ( Kindt et al . , 2007; Kang et al . , 2010; Sawin et al . , 2000 ) . There are four CEP neurons in the head , each with one sensory cilium projecting into the nose . The sensory cilia are positioned so that two of them are located on the ventral side and two of them are located on the dorsal side . In both the ventral and in the dorsal neuron pair one neuron is located on the left and one neuron is located on the right side . Thus , mechanical stimuli could be sensed in all quadrants of the nose . Intriguingly , the CEP neurons have direct connections to the SIA neurons with each CEP connecting to only one SIA . The connectivity directly matches the quadrant in which the CEP cilium is located to the quadrant the sublateral SIA process is located ( White et al . , 1986 ) . Thus , these anatomical connections could allow a direct control of three-dimensional sensing with three-dimensional movement . In summary , we identify the molecular and neuronal bases for flipping in C . elegans , and this may be the starting point for also studying other types of motion .
C . elegans was grown on Nematode Growth Medium ( NGM ) agarose plates seeded with E . coli OP50 and were kept at 25°C ( Brenner , 1974 ) . The following strains and alleles were used: N2: wild type PD4588: ceh-24 ( cc539 ) V . HBR1000: ceh-24 ( tm1103 ) V . IB16: ceh-17 ( np1 ) I . HBR352: egl-30 ( tg26 ) I . ( Created from CG21 by backcrossing 2x into wild type to remove the him background ) MT1434: egl-30 ( n686 ) I . LX929: vsIs48[unc-17::GFP] . VN411: vnEx128[ceh-17::cha-1sense , ceh-17::cha-1antisense , lin-44::gfp , pBluescript] . VN412: vnEx129[ceh-17::cha-1sense , ceh-17::cha-1antisense , lin-44::gfp , pBluescript] . HBR1077: goeIs247[pceh-24::GCaMP6s::mKate2-unc-54–3`utr , unc-119 ( + ) ] . HBR1094: ceh-24 ( tm1103 ) V; goeIs247[pceh-24::GCaMP6s::mKate2-unc-54–3`utr , unc-119 ( + ) ] . HBR1130: vsIs48[unc-17::GFP]; goeIs253[ceh-24::ReaChr::mKate2-unc-54–3'utr , unc-119 ( + ) ] . HBR1131: ceh-24 ( tm1103 ) V; vsIs48[unc-17::GFP]; goeIs253[ceh-24::ReaChr::mKate2-unc-54–3'utr , unc-119 ( + ) ] . HBR1231: goeEx470[pceh-24::mGFP::unc-54–3'utr , unc-119 ( + ) , pceh-24::cha-1RNAi::unc-54–3'utr , unc-119 ( + ) , pceh-24::cha-1RNAirev::unc-54–3'utr , unc-119 ( + ) ] . HBR1265: unc-119 ( ed3 ) III; goeEx501[pceh-24::ceh-24gene::SL2-mKate2-unc-54–3'utr , unc-119 ( + ) , [pceh-24::mGFP::unc-54–3'utr , unc-119 ( + ) ] . HBR1567: ceh-17 ( np1 ) I; vsIs48[unc-17::GFP]; goeIs253[ceh-24::ReaChr::mKate2-unc-54–3'utr , unc-119 ( + ) ] . HBR1568: egl-30 ( tg26 ) ) ; goeEx464[pceh-17::ReaChr::mKate2-unc-54–3'utr , unc-119 ( + ) ] . HBR1569: egl-30 ( n686 ) I; goeEx464[pceh-17::ReaChr::mKate2-unc-54–3'utr , unc-119 ( + ) ] . HBR1800: unc-119 ( ed3 ) III;goeEx683[pceh-17::ceh-24gene::SL2-mKate2-unc-54–3´utr , unc-119 ( + ) ]; ceh-24 ( tm1103 ) V . The deletion allele ceh-24 ( tm1103 ) was backcrossed 10 times against N2 to generate HBR1000 . All transgenic insertions were backcrossed at least two times against N2 to remove the unc-119 ( - ) background . The following primers were used to detect the deletion in ceh-24 ( tm1103 ) using a three primer PCR: Primer 1 – CCA GGT AGA CTT CCA GGC AA Primer 2 – ACG ACG GAA AAG AAG AGT CCT Primer 3 – GGG GTG AGC TTC CAT CTT CA All constructs were cloned using the Multisite Gateway system ( Invitrogen , Carlsbad , CA ) into pCG150 ( Merritt and Seydoux , 2010 ) . All constructs obtained from LR reactions were sequenced for verification . Sequences used for cha-1 RNAi were described previously ( Suo and Ishiura , 2013 ) . The following Gateway entry clones were constructed for this study: We generated transgenic strains by microparticle bombardment or by microinjection using unc-119 ( ed3 ) rescue as a selection marker ( Wilm et al . , 1999; Praitis et al . , 2001 ) . Injection concentration was 60 ng/µl for all constructs . For genetic screening we used time-lapse movies from ~40 strains that we had from a previous screen for sleep mutants . These mutants were initially selected because we suspected that they had defective sleep behavior , but they turned out to have normal sleep as judged by the time-lapse movies ( Turek et al . , 2013 ) . For the sleep screen , we had filmed between two and four individuals in microfluidic compartments made from agarose hydrogel ( 190 × 190 μm , 10 μm deep ) using DIC microscopy and a time lapse protocol with one frame each five seconds . Movies were then scrolled through quickly for the presence of flips as identified by the side of the developing gonad . The examiner was blinded for the underlying mutations during scoring of flipping . For subsequent experiments , the examiner was not blinded any longer . For PD4588 , no flips were detected in four out of four animals filmed . While the number of strains screened was quite low , we did not continue screening but rather focused on characterizing ceh-24 . All long-term imaging experiments were carried out using agarose microchamber imaging ( Bringmann , 2011; Turek et al . , 2015 ) . Nose tracking was performed manually . Calcium imaging was performed similarly as described before using GCaMP3 . 35 and co-expression of mKate2 as an expression control ( Schwarz et al . , 2011 , 2012; Turek et al . , 2013; Schwarz and Bringmann , 2013 ) . For calcium imaging , we used an Andor iXon ( 512 × 512 pixels ) EMCCD camera and LED illumination ( CoolLed ) using standard GFP and Texas Red filter sets ( Chroma ) . Exposure times were in the range of 5–20 ms and allowed imaging of moving worms without blurring . The EMCCD camera triggered the LED through a TTL ‘fire’ signal to illuminate only during exposure . LED intensity was in the range of 15–30% . EM gain was between 0 and 100 . All calcium-imaging experiments were done in agarose microchambers . Typically , 15–50 individuals were cultured in individual microchambers that were in close vicinity . Animals were filmed by taking a z-stack in a continuous mode ( Figure 4A ) or in a ‘burst mode’ , which means that a short movie of z stacks was taken every 15 min ( Figure 4B , C , D ) . Each burst consisted of 50 z-stacks ( with each stack consisting of 19 planes with 1 μm resolution ) with a frame rate of 20 pictures / second . Individual compartments were repeatedly visited by using an automatic stage ( Prior Proscan2/3 ) set to low acceleration speeds . Before each fluorescent measurement , we took a brief DIC movie to assess the developmental stage and behavioral state . Larvae that showed pharyngeal pumping were scored as being in the wake state . We manually selected a z plane for each neuron we wanted to quantify ( Figure 4A ) and projected fluorescence pictures to extract left and right neurons using Andor IQ Software ( Figure 4B , C , D ) . Fluorescence signals were cut out using semi-automated homemade Matlab routines , in which neurons were manually identified and were then cut out automatically . We scored frequency and intensity of the peaks by counting them and measuring their maximum size ( Figure 4B , C , D ) . The baseline was determined separately for each activation transient by averaging five frames before each transient . We noticed that the baseline was slightly reduced in the SIA neurons by approximately 16% during sleep compared with wake . We computed the activation difference over baseline . To allow for a better comparison of transients during sleep and wake we normalized transient strength in wake by using the baseline during sleep for each individual animal . Long-term imaging was similarly described previously ( Turek et al . , 2016 ) . All imaging experiments were carried out at 20°C . Quiescence bout analysis was carried out similar to described ( Iwanir et al . , 2013; Nagy et al . , 2014b ) . Movement below a threshold of 1 . 65 µm/s was used to define a quiescence bout and movement above this threshold was used to define a motion bout . The length and frequency of the quiescence bouts were extracted using a Matlab routine . ReaChr experiments were performed inside agarose microchambers as described ( Turek et al . , 2016; Lin et al . , 2013 ) . We grew hermaphrodite mother worms on a medium that was supplemented with 0 . 2 mM all trans Retinal ( Sigma ) . We then placed the eggs from these mothers together with food into microchambers without any further Retinal supplementation . We stimulated ReaChr with an LED of 585 nm with about 0 . 1 mW/mm ( Cirelli and Tononi , 2008 ) as measured with a light voltmeter . We took 10 frames with a rate of 2/s before stimulation , i . e . filmed for 5 s , then , we stimulated ReaChr for 10 s , during which we continued with image acquisition , and then finally we filmed the worms again after stimulation for 5 s . This protocol was repeated for the same individual worm every 30 min . Using an automated stage , we imaged 10–50 individuals in one overnight round of experiments . The experiment was controlled by Andor iQ software . For mosaic analyses , after ReaChr stimulation and imaging , we added 25 mM levamisole onto the chambers to immobilize the worms . We then took high-resolution ( 1000x ) stacks of the mKate2 expression signal to identify neurons expressing mKate2 . We used spinning disc imaging with an Andor Revolution spinning disc system using a 488 nm laser and a 565 nm laser , a Yokogawa X1 spinning disc head , a 100x oil objective and an iXon EMCCD camera . Z stacks were taken and a maximum intensity projection calculated using iQ software as described in ( Schwarz et al . , 2011 ) . Laser ablation were as described previously . Briefly , we used a 355 nm laser focused to a near-diffraction limited spot ( Rapp Opto , DPSL-355/14 , direct coupling ) . All ablations were performed at 1000x magnification in a strain expressing mKate2 in the sublateral neurons . Neurons were identified and ablated in late embryos using mKate2 fluorescence without using anesthetics . We verified the successful ablation of neurons using mKate2 fluorescence , and worms were discarded when they showed unspecific laser damage or when the neurons of interest were not successfully ablated . We used mock-ablated worms that were treated like ablated worms except that they were not irradiated with the laser ( Turek et al . , 2013 ) . Movement measurements on a plane surface: To characterize movement on a plane surface , young adult worms were placed onto an NGM ( Nematode Growth Plate ) seeded with OP50 bacteria . Animals were recorded using a camera mounted onto a Leica MZ15 Stereomicroscope using dia illumination for 7 min with 1 frame per second . Backwards and forwards movement was scored manually and also the position of the nose was tracked manually to determine movement speeds . Response to mechanical stimulation: Responsiveness to gentle touch was tested as previously described ( Chalfie et al . , 2014 ) . Worms were touched briefly at their head behind the pharynx using an eyelash glued to a holding stick . An individual was scored as responding to mechanical stimulation if it stopped movement or initiated a backward movement . Each individual was tested ten times and a total of 30 individuals were tested per condition . Escape from an indentation: Indentations were box shaped with the following dimensions: 700 μm x 700 μm ( xy ) and 65 μm deep . Indentations were cast from 3% agarose in S-Basal by using a PDMS stamp as described ( Bringmann , 2011 ) . Fresh indentations were cast for each individual experiment . Unlike in the long-term culture experiments , chambers were not sealed with a glass coverslip but were left open . No food was added to the chambers . Individual young adult worms were picked into the indentation . The sample was then placed onto a time-lapse microscope and the behavior of the animal was recorded immediately with two frames per second at 100x magnification using bright field microscopy . The time point at which a worm had left the indentation was scored manually and was defined as the first point in time when the entire body of the worm was outside the indentation . Statistical tests used were Mann Whitney U test , Kolmogorov-Smirnov test , Wilcoxon Signed Ranks test , or Student’s t-test using Origin software . The specific test used is described in the figure legends . Error bars are SEM throughout . | Although sleeping individuals do not move voluntarily , they are not completely immobile . Both people and animals regularly change position in their sleep , but it is not known why these movements occur or what regulates them . One of the simplest animals known to require sleep is the nematode worm Caenorhabditis elegans , which is often used by researchers to study the molecular basis of behavior . In common with more complex animals , worms go to sleep lying on either their left or right side and then switch periodically between the two . This “flipping” behavior is typically not seen outside of sleep . By screening worms with mutations in different genes , Schwarz and Bringmann identified one mutant that does not flip during sleep . The mutant lacked a gene called ceh-24 , which is normally active in a set of four neurons known as SIAs . These are a type of motor neuron; that is , neurons that control the contraction of muscles . The body wall muscles of C . elegans run along the length of its body and are organized into “quadrants” that each cover a quarter of the worm . Schwarz and Bringmann show that unlike other C . elegans motor neurons , SIA neurons control each quadrant separately . By activating specific SIA neurons the worms can contract the muscles on each side of the body independently , and thereby flip from one side to the other . Further investigation revealed that the SIA motor neurons can also control other types of complex movement . Additional experiments are now needed to determine how the neurons support these behaviors . Another challenge will be to work out the purpose of posture changes during sleep for C . elegans and other animals . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"neuroscience"
] | 2017 | Analysis of the NK2 homeobox gene ceh-24 reveals sublateral motor neuron control of left-right turning during sleep |
Recognition of Nod factors by LysM receptors is crucial for nitrogen-fixing symbiosis in most legumes . The large families of LysM receptors in legumes suggest concerted functions , yet only NFR1 and NFR5 and their closest homologs are known to be required . Here we show that an epidermal LysM receptor ( NFRe ) , ensures robust signalling in L . japonicus . Mutants of Nfre react to Nod factors with increased calcium spiking interval , reduced transcriptional response and fewer nodules in the presence of rhizobia . NFRe has an active kinase capable of phosphorylating NFR5 , which in turn , controls NFRe downstream signalling . Our findings provide evidence for a more complex Nod factor signalling mechanism than previously anticipated . The spatio-temporal interplay between Nfre and Nfr1 , and their divergent signalling through distinct kinases suggests the presence of an NFRe-mediated idling state keeping the epidermal cells of the expanding root system attuned to rhizobia .
Perception of Nod factors by LysM receptor kinases , NFR1 and NFR5 in Lotus japonicus ( Broghammer et al . , 2012 ) , triggers tightly coordinated events leading to root nodule symbiosis ( Madsen et al . , 2003; Radutoiu et al . , 2003 ) . Minutes after the activation of receptors , a signalling cascade ( Stracke et al . , 2002; Antolín-Llovera et al . , 2014 ) leading to regular calcium oscillations in the root hair cells located in the susceptible zone is initiated ( Miwa et al . , 2006 ) . These oscillations are interpreted by the Calcium Calmodulin Kinase ( CCaMK ) ( Miller et al . , 2013 ) , which activates a set of regulators that launch transcription of symbiosis specific genes in the outer root cell layers ( Yano et al . , 2008; Hirsch et al . , 2009 ) . Progression of the symbiotic signalling events from epidermis into the cortex is necessary for nodule organogenesis and infection thread formation . NIN , a transcriptional regulator , and cytokinin signalling have been implicated in this epidermal to cortex signalling ( Murray et al . , 2007; Tirichine et al . , 2007; Vernié et al . , 2015 ) . Mutations in Nfr5 and its homologs in pea and M . truncatula eliminate all Nod factor-induced physiological , molecular and cellular responses ( Madsen et al . , 2003; Arrighi et al . , 2006 ) . However , some or several of these responses are retained in the Ljnfr1 , Mtlyk3 and Pssym37 mutants ( Radutoiu et al . , 2003; Smit et al . , 2007; Zhukov et al . , 2008 ) raising the possibility that modular receptor complex formation regulated in a spatio-temporal manner might contribute to Nod factor signalling . The LysM receptor kinase family has greatly expanded in legumes through whole genome or tandem duplications ( Zhang et al . , 2009; Lohmann et al . , 2010; Kelly et al . , 2017 ) . In L . japonicus , four genes , Lys1 , Lys2 , Lys6 and Lys7 , are closely related to Nfr1 . Lys1 and Lys2 are located in tandem and at ~10 kb distance from Nfr1 ( Lohmann et al . , 2010 ) . Interestingly , a similar chromosomal organisation of NFR1-type receptors was reported in all studied legumes , as well as in genomes outside of Leguminosae clade raising the possibility that gene duplication leading to tandem NFR1-type receptors preceded the evolution of the legume family ( De Mita et al . , 2014 ) . The precise role of these NFR1 paralogs and their homologs is unknown apart from the chitin receptors L . japonicus LYS6 ( now CERK6 ) and Medicago truncatula LYK9 ( Bozsoki et al . , 2017 ) . Nonetheless , key details about the signalling competencies of LYS2 , LYS6 and LYS7 were obtained from functional complementation analyses in the nfr1 mutant using the LjUbiquitin promoter . Only the intracellular kinase regions of LYS6 and LYS7 , but not that of LYS2 , could restore nodulation and/or infection when coupled to the NFR1 Nod factor-binding domain ( Nakagawa et al . , 2011 ) . Here , we show that LYS1 is an epidermal LysM receptor contributing to the NFR1-NFR5 mediated signalling in a spatio-temporal manner . This gene is primarily expressed in epidermal cells of the susceptible zone where roots are competent for initiation of symbiosis , and has a restricted signalling capacity leading to Nin activation in the outer root cell layers . Our findings provide evidence for a complex Nod factor signalling where LYS1 activity in the outer root cell layers aids in maintaining a normal calcium spiking interval in the root hairs , integral transcript responses in the susceptible root zone , and initiation of nodule primordia on the expanding root system . The Lys1 gene is therefore renamed Nfre , in accordance with the identified role of this gene during Nod factor signalling in the epidermal layer .
NFRe is predicted to encode a LysM receptor protein with a typical intracellular kinase domain ( Figure 1A , Figure 1—figure supplement 1 ) . Based on the close similarity to NFR1 we investigated its biochemical and molecular properties . For this , we analysed the binding capacity of NFRe towards purified pentameric M . loti R7A Nod factor ligand ( Bek et al . , 2010 ) . The NFRe ectodomain was expressed in insect cells using a recombinant baculovirus induced-expression system ( Kawaharada et al . , 2015 ) . Pure protein was obtained after four steps of purification and the homogeneity was confirmed by size exclusion chromatography ( Figure 1—figure supplement 2 ) . Biolayer interferometry ( BLI ) ( Kawaharada et al . , 2015 ) was used for receptor-ligand affinity measurements since this technique is well suited for handling sparingly soluble hydrophobic compounds like Nod factor . To enable ligand immobilization on streptavidin biosensors M . loti R7A Nod factor and chitin pentamer ( ( GlcNAc ) 5 , CO5 ) were conjugated to a biotinylated linker using N-glycosyl oxyamine chemistry ( Bohorov et al . , 2006; Villadsen et al . , 2017 ) ( Figure 1—figure supplement 3 ) . Affinity measurements showed that the ectodomain of NFRe has the capacity to bind M . loti Nod factor with an equilibrium dissociation constant ( KD ) of 29 . 1 ± 7 . 1 μM ( Figure 1B , E ) . Next , we tested whether NFRe has the capacity to bind chitin but no signal was observed for CO5 ligands in this system ( Figure 1B ) . To test our immobilised ligands , we performed the same binding experiment with purified NFR1 ectodomain ( Figure 1—figure supplement 2 ) , which gave a KD of 34 ± 6 . 3 μM to M . loti R7A Nod factor and no binding to CO5 ( Figure 1C , E ) . As a positive control for our chitin ligand we additionally expressed and purified the Arabidopsis CERK1 ectodomain ( Figure 1—figure supplement 2 ) and measured an affinity of 59 μM to the immobilized CO5 ( Figure 1D , E ) , which is very similar to the previously reported KD of 66 μM measured by isothermal titration calorimetry ( Liu et al . , 2012 ) . In short , our binding studies show that NFRe has the capacity to perceive Nod factor with comparable affinity as seen for the NFR1 and both receptor ectodomains distinguish Nod factor from pentameric chitin ligands in a BLI binding assay ( Figure 1E ) . NFRe is a challenging and low expressed protein and further biochemical ligand competition studies are required to fully define the specificity and receptor capacity of NFRe . Next , we assessed the activity of the intracellular kinase domain of NFRe ( Figure 1—figure supplement 1 ) . E . coli-produced NFRe kinase transphosphorylated the myelin basic protein ( MBP ) substrate and autophosphorylated ( Figure 1F , lanes 1–3 ) , showing that NFRe encodes a protein kinase with in vitro activity similar to NFR1 ( Madsen et al . , 2011 ) . Alanine substitutions of three critical amino acids from the catalytic loop ( D418 ) , Mg2 +binding loop ( D436 ) , or P+1 loop ( T459 ) abolished the phosphorylation activity of NFRe ( Figure 1F , lanes 4–12 ) showing that conserved residues from NFRe kinase are critical for its biochemical activity . Together , our results from biochemical studies demonstrate that Nfre encodes a LysM receptor kinase that can perceive Nod factor and has an active kinase . Since we now know that NFRe is an active LysM receptor with properties comparable to NFR1 in these in vitro assays , we next investigated the signalling capacities of NFRe compared to NFR1 in Lotus roots . We tested this by expressing NFRe in the nfr1-1 mutant line containing the symbiotic Nin:GUS reporter ( Radutoiu et al . , 2003 ) . Activation of the Nin promoter in Nfre transformed roots of nfr1-1-Nin:GUS plants , or the development of nodule and/or infection threads would indicate activation of symbiotic signalling . While nfr1-1 roots transformed with the Nfr1 gene developed bona fide root nodules and induced Nin promoter , those transformed with the empty vector , and thus expressing the native Nfre gene did not show any responses to inoculation with rhizobia ( Table 1 and Figure 2—figure supplement 1 ) . These results indicate that Nfre , in its native status cannot replace the functions of NFR1 . On the other hand , p35S-Nfre led to strong activation of the Nin promoter after inoculation with M . loti ( Figure 2A and Figure 2—figure supplement 1 ) . This symbiotic induction was however , only detected in the outer root layers ( Figure 2A ) , and it was not followed by formation of nodules or infection threads even after 5 weeks of exposure to M . loti ( Figure 2—figure supplement 1 ) . This differed from the p35S-Nfr1-mediated signalling that induced Nin expression in both epidermal and cortical cells ( Figure 2B ) and led to formation of infected nodules ( Figure 2—figure supplement 1 ) . To understand whether this particular and cell layer specific activation of Nin by NFRe is a result of the expression of any LysM receptor of the NFR1-type , or specific to NFRe , Nin activation was assayed in nfr1-1-Nin:GUS plants transformed with the Lys paralogs of NFR1 ( Lohmann et al . , 2010 ) . Under similar conditions , Lys2 , Cerk6 or Lys7 driven by 35S promoter could not activate the Nin promoter , or induce nodule or infection thread formation in the nfr1-1 mutant . ( Table 1 and Figure 2—figure supplement 1 ) . These results demonstrate that in the presence of M . loti , NFRe , like NFR1 , can initiate a symbiotic signalling cascade leading to Nin induction in Lotus roots , and that the cellular effects of this signalling are receptor- , and expression-dependent . Previous studies based on transcript measurement showed that Nfre is expressed in Lotus roots ( Lohmann et al . , 2010 ) . However , our results from the nfr1-1 complementation studies revealed that expression of Nfre from p35S promoter is needed to induce observable Nin activation after rhizobia inoculation ( Figure 2—figure supplement 1 ) . To further understand the cause of this differential signalling we characterised the spatio-temporal regulation of Nfre in detail using GUS and tYFPnls ( triple YFP-nuclear localised ) reporter fusions , and measured the levels of Nfre transcript by quantitative RT-PCR . In uninoculated roots the Nfre promoter ( 2 , 6 kb ) was preferentially active in root hair epidermal cells , in the susceptible zone of the root , and in the root tip ( Figure 2C , E , and Figure 2—figure supplement 2 ) . This differed from Nfr1 that is expressed in the whole uninoculated root ( Radutoiu et al . , 2003; Kawaharada et al . , 2017 ) ( Figure 2D ) . The expression pattern of Nfre did not change after inoculation with M . loti ( Figure 2F and Figure 2—figure supplement 2 ) , indicating that , unlike Nfr1 ( Radutoiu et al . , 2003; Kawaharada et al . , 2017 ) and Figure 2—figure supplement 2 ) , the expression of Nfre is not symbiotically regulated . Analyses of Nfre transcript levels in wild type roots either treated with Nod factor or inoculated with M . loti compared to control roots , further confirmed the unaltered expression observed from Nfre promoter studies ( Figure 2—figure supplement 2 ) . Direct comparison of Nfr1 and Nfre transcript levels in uninoculated wild type roots showed a 3-fold higher level for Nfr1 . Interestingly this difference was reduced significantly after Nod factor treatment ( 8 hr post treatment ) or M . loti inoculation ( 2 and 3 dpi post inoculation ) , where Nfr1 expression was down regulated , while Nfre maintained a low , but constant level ( Figure 2—figure supplement 2 ) . In summary , Nfre and Nfr1 differ in their expression level and pattern in uninoculated roots , and follow a differential regulation during root nodule symbiosis . These differences could therefore , at least in part , account for the differential signalling capacities of the two LysM receptors . NFRe has the capacity to bind Nod factors in vitro ( Figure 1B ) and to induce a symbiotic signalling in planta when expressed in the nfr1-1 mutant from the 35S promoter ( Figure 2A ) . This prompted us to ask whether NFRe plays a role in root nodule symbiosis . Homozygous mutant plants from three independent alleles with exonic insertion of LORE1 retroelement ( Mun et al . , 2016 ) ( Figure 1A and Supplementary file 1 ) formed significantly fewer nodules compared to wild type when grown in a binary association with M . loti ( Figure 3A ) . The contribution of NFRe to root nodule organogenesis became even more evident when wild type and nfre mutants were grown in soil and were exposed to the native bacterial community . After 9 weeks , nfre mutants developed only half the number of wild type nodules ( Figure 3—figure supplement 1 ) . The shoot biomass and the general plant fitness were significantly reduced ( Figure 3—figure supplement 1 ) . Wild type plants had well-developed green pods , while nfre mutants had only few open flowers ( Figure 3—figure supplement 1 ) . Analyses of the dynamics of nodule primordia formation on plate-grown plants , revealed that nfre mutants , besides a noticeable reduced nodulation at the early time point ( two wpi ) , had a significantly lower ability to reinitiate nodule formation on the expanding root system ( five wpi ) ( Figure 3B ) . Unlike nodule organogenesis , the formation of infection threads ( IT ) appeared not to be affected by mutations in the Nfre . A similar number of ITs were present in wild type and nfre root hairs at 9 or 14 dpi ( Figure 3C ) . The mature nodules formed on nfre appeared normally infected ( Figure 3—figure supplement 1 ) , and the proportion of pink/total nodules formed by soil-grown wild type and nfre plants was similar ( Figure 3—figure supplement 1 ) , indicating a normal infection process in the nfre mutants . To further investigate the role of NFRe in Nod factor signalling we analysed its requirement for induction and maintenance of nuclear-associated calcium oscillations ( spiking ) after Nod factor treatment . Root hairs of wild type ( n = 50 ) and nfre-1 ( n = 46 ) stable transgenics expressing the nuclear localised YC3 . 6 ( Yellow Cameleon ) showed clear signs of calcium oscillations after M . loti Nod factor treatment ( Figure 3D ) ( app . 80% of the analysed cells responded ) . Closer inspection of the spiking frequency revealed that the average inter-spike interval was significantly longer in the nfre cells ( 106 s ) compared to wild type ( 86 s ) , indicating that NFRe contributes to a constant interval length of calcium oscillations ( Figure 3—figure supplement 2 ) . Next , we used RNA sequencing to investigate the requirement for Nfre in the transcriptional changes induced by M . loti Nod factor in the susceptible zone of the root at 24 hr after treatment . Genes that were differentially expressed ( DEGs adjusted p<0 . 05 ) ( Materials and methods ) in Nod factor treated roots compared to water control were identified in wild type , nfre-1 and nfre-2 mutants . A large proportion of these ( 44 out of 90 ) , which includes Nin , expansins , nodulins , receptors , transporters and transcription factors , were regulated by Nod factor in wild type but not in the nfre roots , indicating that their appropriate regulation in the susceptible zone , at 24 hr after Nod factor treatment is dependent on an active NFRe ( Figure 3—figure supplement 3 and Supplementary file 2 ) . Other symbiosis-related genes like NFY-A , subtilase , and two genes encoding the cytokinin-induced message were found among the 13 DEGs in wild type and nfre mutants . Only one gene ( an expansin ) was found regulated by the Nod factor in both nfre mutants but not in wild type . Our biochemical in vitro data based on BLI measurements shows that the NFRe ectodomain does not bind chitopentaose , suggesting that NFRe might not be involved in chitin signalling . To test this hypothesis in planta we measured the induction of reactive oxygen species ( ROS ) in response to CO8 or CO4 in the nfre mutants and wild type . We found that wild type , nfre-1 and nfre-2 mutants produced comparable levels of ROS , indicating that NFRe is unlikely to be involved in chitin signalling ( Figure 3—figure supplement 4 ) . Together , these results show that NFRe represents an influential component of the epidermal Nod factor signalling in L . japonicus , promoting intracellular signalling that leads to optimal calcium spiking , activation of gene transcription and efficient nodule organogenesis on the expanding root system . The clear difference observed between NFR1 and NFRe in their ability to induce Nod factor signalling and spatial activation of the Nin promoter in M . loti inoculated nfr1-1-Nin:GUS roots prompted us to identify the molecular determinants for this differential regulation . A chimeric receptor ( NeK ) containing the NFR1 extracellular domain followed by the transmembrane and intracellular kinase regions of NFRe ( NFR1 ectodomain-NFRe kinase- NeK ) ( Figure 1—figure supplement 1 ) was constructed . This receptor was expressed in nfr1-1-Nin:GUS to test its capacity to induce activation of Nin promoter after M . loti inoculation . We observed that the signalling capacity of NeK receptor was similar to that of the NFRe , namely only epidermal induction of the Nin promoter ( Table 2 and Figure 2—figure supplement 1 ) . This provides evidence for the presence of a molecular determinant for specific Nin induction in the intracellular regions of NFR1 and NFRe receptors . Alignment of the two kinases identified several divergent regions ( Figure 1—figure supplement 1 ) , but clear differences were found in the activation segment ( Figure 1—figure supplement 1 ) . Based on these differences , and previous knowledge ( Nakagawa et al . , 2011 ) that this region is crucial for kinase signalling and substrate recognition , we hypothesised that a specific NFR1/NFRe activation segment determines the specificity of the downstream signalling . We tested this hypothesis by swapping the NFRe activation segment with the corresponding region of NFR1 in the NeK receptor ( NFR1 ectodomain-NFRe kinase with the Activation segment of NFR1 -NeKA1 ) ( Figure 1—figure supplement 1 ) . In contrast to the NeK receptor that induced Nin in the outer root cell layers of the nfr1-1 mutant , NeKA1 led to cortical activation of Nin and nodule formation ( Table 2 and Figure 2—figure supplement 1 ) . These results show that the activation segment in NFR1 and NFRe determines the downstream signalling output in Lotus roots after M . loti inoculation . Genetic and molecular studies established that a concerted NFR1-NFR5 signalling induces the nitrogen-fixing symbiosis ( Radutoiu et al . , 2003; 2007 ) . Here , we present evidence that in L . japonicus NFRe assists the development of root nodule symbiosis . Therefore , we hypothesised that NFRe-dependent signalling also involves NFR5 . For this , we investigated the biochemical capacity of the NFRe kinase to transphosphorylate the intracellular NFR5 pseudokinase . The in vitro kinase assays showed that NFR5 is a substrate for the NFRe kinase . ( Figure 4A , lanes 1–3 ) and that this transphosphorylation was dependent on the activation segment of the NFRe kinase . Mutation of T459 to A abolished the kinase activity of NFRe , while exchanging the native segment with the corresponding region of NFR1 maintained transphosphorylation ( Figure 4A , lanes 4–6 , 7–9 ) . These results corroborate the observed nodule formation in the nfr1-1 expressing the NeKA1 receptor ( Table 2 and Figure 2—figure supplement 1 ) . Next , we analysed the localisation and molecular properties of full-length NFRe and NFR5 using heterologous expression in Nicotiana benthamiana . The YFP tagged NFRe protein was found to localize to the plasma membrane and to co-localise with the plasma membrane marker , AtPIP2 , like previously observed for NFR1 and NFR5 ( Madsen et al . , 2011 ) ( Figure 4—figure supplement 1 ) . Bimolecular fluorescence complementation ( BiFC ) analyses based on split YFP revealed that NFRe formed homomeric complexes alone and heteromeric complexes when co-expressed with either NFR1 or NFR5 ( Figure 4B and Figure 4—figure supplement 1 ) . The formation of heteromeric complex with NFR5 was not affected by kinase inactivation ( Figure 4B ) . Like in the case of NFR1-NFR5 co-expression ( Madsen et al . , 2011 ) , a signalling cascade leading to leaf cell death , dependent on an active NFRe kinase , was identified in N . benthamiana leaves co-expressing NFRe and NFR5 ( Figure 4—figure supplement 1 ) . Finally , we analysed whether the NFRe-dependent activation of Nin in L . japonicus roots was dependent on NFR5 . Nfre driven by 35S promoter failed to induce Nin:GUS symbiotic reporter in the nfr5-2 mutant background ( Figure 4—figure supplement 1 and Supplementary file 3 ) . These results demonstrate that NFRe can interact with , and trans-phosphorylates NFR5 kinase , and induce a signalling cascade dependent on the NFR5 receptor . Collectively , our results from biochemical studies of the extracellular and intracellular domains of NFRe , together with those obtained from mutant and functional analyses provide evidences for the involvement of NFRe ensuring a robust signalling for symbiosis with nitrogen-fixing rhizobia .
Nod factor binding by NFR1-NFR5 LysM receptors is required to induce nodule organogenesis and infection thread formation in L . japonicus ( Radutoiu et al . , 2003; 2007 ) . Here , we show that the NFR1-NFR5 signalling cascade operates on the framework provided by the epidermal LysM receptor NFRe . NFRe and NFR1 share biochemical and molecular properties that is similar Nod factor-binding affinity , and chitopentaose differentiating capacity when assessed by biolayer interferometry ( Broghammer et al . , 2012 ) , functional kinases dependent on fully operative domains ( Madsen et al . , 2011 ) , capacity to phosphorylate , and induce a signalling cascade dependent of NFR5 ( Madsen et al . , 2011 ) . In spite of these similarities , NFR1 and NFRe have evolved distinct biological properties defined by specific spatio-temporal expression , and downstream signalling cascades controlled by diverged kinases . The epidermis of the expanding root system is continuously exposed to Nod factors produced by rhizobia present in the rhizosphere . Nevertheless , the number and the location of primordia guiding the epidermal infection threads are precisely determined . Complex regulatory networks involving transcriptional regulators , hormones , shoot- and root-derived signals ( Ferguson et al . , 2010; Miyata et al . , 2013; Sasaki et al . , 2014; Miri et al . , 2016; Roy et al . , 2017 ) , as well as tightly controlled receptor signalling ( Mbengue et al . , 2010; Kawaharada et al . , 2017 ) , collaborate to coordinate how many nodules the plant develops . With this framework in mind , a working model is emerging when considering our findings ( Figure 4C ) . This model incorporates the interplay of the NFR1 and NFRe in the epidermis , ensuring efficient and robust signalling in the susceptible zone of the expanding root system . In the absence of the symbiont , Nfre has a low and constant expression in the susceptible zone , while Nfr1 outnumbers Nfre in terms of expression level and spatial distribution ( Figure 4C-1 ) . Once the symbiotic process is initiated , the expression of NFR1 is rapidly downscaled in the susceptible zone ( Figure 4C-2 ) . A sustained expression of NFRe in the epidermal cells of the susceptible zone could ensure an idling signalling , keeping the expanding root system tuned in to rhizobia ( Figure 4C-3 ) . NFR1 acts as a master switch triggering recurrent symbiotic events in a fast and efficient manner from NFRe-attuned epidermal cells ( Figure 4C–4 ) . In general , protein-carbohydrate interactions are usually weak and low-affine ( micromolar-millimolar range ) ( Holgersson et al . , 2005 ) and signalling therefore , emerges as being controlled by ligand multivalency and/or by receptor multiplicity ( Kiessling and Pohl , 1996; Rabinovich , 2002; Vasta et al . , 2012 ) . In line with this notion studies of receptors present at the plant and mammalian plasma membrane revealed a conserved strategy to ensure specific , instantaneous , switchable and evolvable downstream signalling; namely , increased responsiveness and specificity via combinatorial systems ( Ostrom et al . , 2001; Piñeyro , 2009; Bodmann et al . , 2015; Bücherl et al . , 2017 ) . The signalling properties of NFRe remain to be determined , but our findings based on the properties of this LysM receptor kinase , together with the symbiotic phenotypes of nfre mutants unveil a more complex signalling operating in the epidermal cells of L . japonicus than anticipated from studies of the basic and essential receptor-components . It is possible that multiple LysM receptors assemble into functional signalling complexes where signalling specificity is the result of the nature of the complex , rather than isolated LysM receptors alone . The mechanistic details of NFR1-NFRe signalling remain to be discovered , but we envision that differences might exist among legumes , since nfr1 in Lotus and lyk3 or sym37 in Medicago and pea have different symbiotic phenotypes ( Radutoiu et al . , 2003; Smit et al . , 2007; Zhukov et al . , 2008 ) , indicating distinct evolutionary trajectories after separation of the IRLC ( Inverted Repeat-lacking clade ) legumes ( Sprent , 2008 ) . Tandem NFR1-type receptors are found in all legumes and in non-legume species as well ( Liang et al . , 2013; De Mita et al . , 2014 ) . Ample comparative phylogenomics and trans-complementation studies targeting tandem duplicated LysM receptors will greatly help determining their evolutionary impact and their role in different plant species .
Clustal Omega was used to prepare multiple sequence alignment for phylogenetic analysis . The region between 55 and 251 in this alignment was realignment to adjust the positions of CXC motif . This alignment was used for the phylogenetic analysis with Neighbor Joining . The distance was measured with Jukes-Cantor and the bootstrap was 1000 replicates . These alignment and phylogenetic analyses were performed in the CLC Main Workbench v7 . 9 . 1 . The amino acid sequence of OsCERK1 ( Os08g0538300-01 ) is available in The Rice Annotation Project Data Base ( rap-db ) . The other sequences below are available in NCBI: AtCERK1 ( NP_566689 ) , NFR1 ( CAE02590 ) , NFRe ( AB503681 ) , LYS2 ( AB503682 ) , EPR3 ( AB503683 ) , LYS4 ( AB503685 ) , LYS5 ( AB503686 ) , LYS6 ( AB503687 ) , LYS7 ( AB503688 ) . NFRe and NFR1 ectodomain boundaries were defined by secondary structure prediction performed with PSIPRED ( Buchan et al . , 2013 ) . Their signal peptides were predicted using the SignalP 4 . 1 server ( Petersen et al . , 2011 ) . The AtCERK1 ectodomain boundaries were designed based on the reported crystal structure ( Liu et al . , 2012 ) . The predicted ectodomain sequences were codon-optimized for insect cell expression and synthesized with an N-terminal gp67 secretion signal peptide and a c-terminal hexa-histidine tag ( GenScript , Piscataway , USA ) and inserted into the pOET4 transfer vector ( Oxford Expression Technologies ) . Recombinant AcMNPV baculoviruses were produced in Sf9 cells cultured with SFX ( Hyclone ) or TNM-FH medium ( Sigma-Aldrich ) supplemented with 10% ( v/v ) FBS ( Gibco ) , 1% ( v/v ) chemically defined lipid concentrate ( Gibco ) and 1% ( v/v ) Pen/Strep ( 10 , 000 U/ml , Life Technologies ) . The FlashBac Gold kit ( Oxford Expression Technologies ) was utilized according to the manufacturer’s instructions . Viruses were amplified until a third passage virus culture of 500 mL was obtained . For large scale protein expression Sf9 cells were infected with 5% ( v/v ) of the passage three virus solution and cultured in suspension with serum-free SFX insect cell medium ( Hyclone ) or BD BaculoGold MAX-XP medium ( BD Biosciences , discontinued ) supplemented with chemically defined lipid concentrate and Pen/Strep as described above . The culture was maintained in a shaking incubator at 26°C for five days , after which the medium was harvested by centrifugation in a Sorvall RC5plus centrifuge ( SLA-1500 rotor ) at 6000 rpm at room temperature for 25 min . Subsequently , the cleared medium was dialyzed against 10 volumes of buffer A ( 50 mM Tris-HCl pH 8 , 200 mM NaCl ) for one day at 4°C with the buffer being exchanged at least four times . The proteins were loaded on a HisTrap excel column ( GE Healthcare ) equilibrated with buffer A and recirculated over 3 days at 4°C using a peristaltic pump . After a washing step with buffer W ( 50 mM Tris-HCl pH 8 , 500 mM NaCl and 20 mM imidazole ) proteins were eluted with buffer B ( 50 mM Tris-HCl pH 8 , 200 mM NaCl , 500 mM imidazole ) . Imidazole was removed by dialyzing against buffer A and the purity was improved by a second IMAC purification step using a HisTrap HP column ( GE Healthcare ) . The NFRe ectodomain was dialyzed against a low salt buffer ( 50 mM Tris-HCl and 50 mM NaCl ) before purification on a MonoQ column ( GE Healthcare ) . The resulting flow-through containing NFRe was collected and concentrated in a Vivaspin column ( 10 kDa cut-off , Sartorius Stedim biotech ) . NFRe and NFR1 were finally purified by size exclusion chromatography using a Superdex 200 10/300 GL column ( GE Healthcare ) and AtCERK1 using a Superdex 75 10/300 GL size exclusion column ( GE Healthcare ) in SEC buffer ( Phosphate buffered saline , pH 7 . 2 , 500 mM NaCl ) . At each purification step , yield and purity were assayed by SDS-PAGE . Biotin conjugates were synthesized using a two-step procedure according to Figure 1—figure supplement 3 . O- ( 2-Aminoethyl ) -N-methyl hydroxylamine trifluoroacetic acid salt was prepared as described previously ( Bohorov et al . , 2006 ) , and all other chemicals were purchased from Sigma-Aldrich and used without further purification . Nod factors from Mesorhizobium loti , strain R7A , NodMl-V ( C18:1Δ11Z , Cb , Me , AcFuc ) , containing three main species ( 3-O-acetylated , 4-O-acetylated , or non-acetylated fucosyl unit ) were purified as described previously ( Bek et al . , 2010 ) . Purified R7A Nod factor ( 3 . 6 mg , 2 . 29 μmol , 5 mM ) was dissolved in 0 . 62 M NaOAc buffer , pH 4 . 5 , containing 50% acetonitrile , and O- ( 2-aminoethyl ) -N-methyl hydroxylamine trifluoroacetic acid salt ( 150 mM , 30 equiv . ) was added . The resulting mixture was allowed to react at room temperature for 16 hr , after which it was concentrated under a nitrogen flow . The intermediate product was purified by semipreparative HPLC on an UltiMate 3000 instrument fitted with a Waters 996 photodiode detector , using a Phenomenex Luna 5 μm , C18 ( 2 ) , 100 Å , 250 × 100 mm semi-preparative column . An isocratic elution at 40% acetonitrile in water , 5 mL/min for 30 min was used . The intermediate eluted at 9 . 5–11 . 5 min . Conjugate formation was confirmed by HR-MS ( ES+ ) : calcd for [M + H , 1Ac]+ = 1573 . 8081 , found 1573 . 8185 . The purified intermediate was dissolved at a concentration of 10 mM in 50 mM sodium tetraborate buffer , pH 8 . 5 , containing 50% acetonitrile . NHS-dPEG4-biotin ( 15 mM , 1 . 5 equiv . ) was added . The resulting mixture was allowed to react at room temperature for 16 hr . The biotin conjugate product was purified by semipreparative HPLC as described above . A gradient of 5–100% acetonitrile in water over 40 min , running at 5 mL/min , was used . The conjugate eluted after 21 min ( 68% acetonitrile ) . The chromatogram displayed a broad product peak due to the presence of the three species differing in substitution on the fucosyl residue . The biotin-R7A Nod factor conjugate ( 18% yield ) was quantified using the HABA/avidin biotin quantification kit ( Pierce ) . HR-MS ( ES+ ) : calcd for [M + 2 hr , 1Ac]2 += 1024 . 0175 , found 1024 . 0184 , and calcd for [M + 2 hr , 0Ac]2 += 1003 . 0122 , found 1003 . 0129 ( Figure 1—figure supplement 3 ) . High-resolution mass spectra ( HR-MS ) were obtained using a Dionex Ultimate 3000 UHPLC instrument ( Thermo ) coupled to a Bruker Impact HDII QTOF mass spectrometer . The synthesis of a biotin-chitopentaose ( GlcNAc ) five conjugate was performed essentially as for the biotin-R7A Nod factor conjugate . The product was purified by HPLC using a Phenomenex Luna 5 μm , C18 ( 2 ) , 100 Å , 250 × 100 mm semi-preparative column , using a gradient of 5–100% acetonitrile in water , 5 mL/min for 40 min . The product eluted after 12 . 7 min ( 35% acetonitrile ) . The final yield of the biotin- ( GlcNAc ) five conjugate was determined to be 9% . HR-MS ( ES+ ) : calcd for [M + 2 hr]2 += 790 . 3552 , found 790 . 3557 ( Figure 1—figure supplement 3 ) . Binding of NFRe , NFR1 and AtCERK1 ectodomains to biotin-R7A Nod factor and biotin- ( GlcNAc ) 5 ( CO5 ) was measured on an Octet RED biolayer interferometer ( Pall ForteBio ) . Biotinylated R7A Nod factor and ( GlcNAc ) 5 , were immobilized on streptavidin biosensors ( for kinetics , Pall ForteBio ) at a concentration of 250 nM for 5 min . Immobilization levels of biotin-R7A and biotin- ( GlcNAc ) five were followed during immobilization and amounted to approximately 2 . 4 nm and 0 . 4 nm of saturation , respectively . Interaction with NFRe , NFR1 and AtCERK1 ectodomains was measured in dilution series at protein concentrations ranging from 0 . 5 to 64 μM ( NFRe ) , 0 . 78–100 μM ( NFR1 ) or 0 . 93–160 µM ( AtCERK1 ) for 10 min . Subsequently , dissociation was recorded for 5 min . All steps were conducted in phosphate-buffered saline , pH 7 . 4 , 500 mM NaCl , 0 . 01% Tween20 . Parallel background measurements using biosensors immobilized with free biotin were subtracted from R7A Nod factor and ( GlcNAc ) five curves to correct for unspecific binding . Sensorgrams were processed using ForteBio Data Analysis 7 . 0 ( Pall ForteBio ) . Equilibrium dissociation constants from steady-state analysis were calculated in GraphPad Prism 6 ( GraphPad Software ) by nonlinear regression using the response at equilibrium ( Req ) plotted against protein concentration . The NFRe , NFR1 and NFR5 kinase domains were predicted using TMHMM Server v . 2 . 0 and PSIPRED secondary structure prediction . NFRe kinase domain was cloned into pET-30 Ek/LIC vector ( Novagen ) , NFR1 and NFR5 kinase domains were cloned into pET-32 Ek/LIC vectors ( Novagen ) . Three NFRe kinase domain mutants were created using the Quikchange Lightning Site-Directed Mutagenesis Kit ( Agilent Technologies ) according to the manufacturers’ instructions . For the chimeric kinase of NeKA1 , cDNA fragment was assembled by PCR as described previously ( Heckman and Pease , 2007 ) . NFRe kinase was expressed into E . coli BL21-CodonPlus , NFR1 and NFR5 kinase into E . coli Rosetta 2 . Cultures were grown until OD600 = 0 . 8 and cold-shocked for 30 min in an ice bath . Protein expression was subsequently induced with 1 mM IPTG and left overnight to shake at 20°C . Cells were harvested by centrifugation at 3300 rpm in a Sigma swing-out rotor 13855 and afterwards resuspended in 100 ml lysis buffer ( 50 mM Tris-HCl pH8 , 400 mM NaCl , 1 mM Benzamidine , 20 mM Imidazole , 5 mMβ-mercaptoethanol and 10% ( v/v ) glycerol ) . Resuspended pellets were broken by sonication and cell debris removed by centrifugation at 12000 rpm ( F21S-8 × 50 y rotor , ThermoFisher ) . The resulting supernatant was loaded on a Ni-NTA IMAC affinity column ( ThermoFisher ) equilibrated with lysis buffer at 4°C using a peristaltic pump . After a wash step with buffer W-kinase ( 50 mM Tris-HCI pH 8 , 1 M NaCl , 1 mM Benzamidine , 50 mM Imidazole , 5 mM β-mercaptoethanol and 10% Glycerol ) to remove contaminants , kinases were eluted with buffer B-kinase ( 50 mM Tris-HCl pH 8 , 300 mM NaCl , 500 mM imidazole , 5 mM β-mercaptoethanol and 10% Glycerol ) . His-tagged TEV protease ( homemade ) was added to the eluted proteins at a 1:100 ( w:w ) ratio and dialysed against a dialysis buffer ( 50 mM Tris-HCl pH 8 , 300 mM NaCl , 5 mM β-mercaptoethanol and 10% Glycerol ) overnight at 4°C . The cleaved kinase domain proteins were subjected to a second round of IMAC affinity column purification and collected in the flow-through . The kinase domain proteins were concentrated in a Vivaspin column ( 10 kDa cut-off ) before being subjected to size exclusion chromatography using either a Superdex 75 or Superdex 200 10/300 GL columns on ÄKTA Purifier system ( both GE Healthcare ) . Purification was performed by isocratic elution in Buffer GF ( 50 mM Tris-HCl pH 8 , 300 mM NaCl and 5 mM β-mercaptoethanol ) . After each purification step , yield and purity were assayed using SDS-PAGE . 4 µg of purified kinase domain proteins were incubated in Kinase Activity Buffer ( 2 mM MnCl2 , 2 mM NaCl , 2 mM MgCl2 , 1 mM ZnCl2 , 50 mM HEPES pH 7 and 100 µM ATP ) and 2mCi ATP , [gamma-32P] ( PerkinElmer ) in a total reaction volume of 20 µL . 2 µg Myelin Basic Protein ( Sigma Aldrich ) and 4 µg NFR5 kinase domain were added to the appropriate reactions . Additionally , controls without Myelin Basic Protein , ATP [gamma-32P] were made . The reactions were left to incubate for 1 hr at room temperature before loading and running on a 15% SDS-PAGE Gel . After staining with Coomassie Brilliant Blue , the gel was transferred on a phosphor plate and exposed overnight before scanning on a Typhoon Scanner ( GE Healthcare ) . Lotus japonicus , ecotype B–129 Gifu ( Handberg and Stougaard , 1992 ) is the wild type used for all experiments . Homozygous nfre ( previously called lys1 ) mutants were identified in the LORE1 collection ( Fukai et al . , 2012; Urbański et al . , 2012 ) and the primers used for genotyping are listed in Supplementary file 1 . Seeds were sterilized and germinated and the 3 days old seedlings were transferred to the corresponding conditions below . Mesorhizobium loti , strain R7A labelled with GFP or dsRed , and NZP2235 were used for phenotypic analyses . An inoculum density of OD600 = 0 . 02 was used for all studies . Agrobacterium rhizogenes AR1193 ( Stougaard et al . , 1987 ) was used for hairy root transformation . A . tumefasiens AGL1 was used for the transient expression in leaves of N . benthamiana . The various constructs used for L . japonicus transformation were assembled using Golden Gate Cloning ( Engler et al . , 2014 ) , and constructs for N . benthamiana were assembled using Gateway system with 35S promoter driving the expression . The details of each construct and primers for cloning are presented in Supplementary file 1 . All constructs were confirmed by sequencing . The seedlings for hairy root transformation were moved to half-strength B5 media and transformed as described previously ( Stougaard , 1995 ) . The composite plants were transferred to Magenta boxes containing sterile clay granule substrate or to sterile agar plates supplemented with ¼ B and D media and inoculated . Transformed roots were incubated in GUS staining buffer [0 . 5 mg/mL X-Gluc , 50 mM phosphate buffer ( pH7 . 0 ) , 5% methanol , 1 mM K4 ( Fe ( CN ) 6 ) , 1 mM K3 ( Fe ( CN ) 6 ) , 0 . 05% Triton X-100] at 37°C for 18 hr in dark . The samples were washed with 50 mM phosphate buffer ( pH7 . 0 ) and stored in 70% ethanol at 4°C . GUS stained roots and nodules were observed using a Leica M165FC stereomicroscope and Leica DFC 310 FX camera system . Three to five representative samples were used for generating transversal sections , as described previously ( Gavrilovic et al . , 2016 ) . For promoter activity using tYFPnls , transformed roots were fixed with paraformaldehyde and cleared as described previously ( Warner et al . , 2014 ) . The samples were analysed on a ZEISS confocal microscope LSM780 . The whole root images were obtained using Z-stack and tail scan tools , the images of root surface were obtained using Z-stack tool . Final images were generated by Maximum Intensity Projection in ZEN software ( ZEISS ) or ImageJ . For transcript measurement by quantitative RT-PCR , 3 days seedlings were moved to agar plates supplemented with 1/4 B and D media and whole roots of 12 days-old ( Figure 2—figure supplement 2 ) or 14 days old ( Figure 3—figure supplement 3 ) plants were harvested after specific treatment as specified in each experiment . The mRNA was isolated from whole roots ( Figure 2—figure supplement 2 ) or the susceptible zone ( Figure 3—figure supplement 3 ) using Dynabeads mRNA DIRECT TM kit ( Invitrogen ) . RevertAid Reverse Transcriptase ( Fermentas ) was used for cDNA synthesis . The quantitative RT-PCR was performed with LightCycler 480 II and LightCycler 480 SYBR Green I Master mix ( Roche ) . ATP-synthase ( ATP ) , Ubiquitin-conjugating enzyme ( UBC ) and Protein phosphatase 2A ( PP2A ) were used as reference genes . The three biological ( each consisting of 40 plants in Figure 2—figure supplement 2 or 30 plants in Figure 3—figure supplement 3 ) and three technical repetitions were used to calculate the geometric mean of the relative transcript levels and the corresponding upper and lower 95% confidence . Sterile agar plates or clay granule substrate supplemented with ¼ B and D media was used for phenotypic analysis in laboratory and greenhouse conditions ( Figure 3A , B ) . Cologne soil ( Zgadzaj et al . , 2016 ) with no additional nutrients or inoculum was used for plant phenotyping in Figure 3—figure supplement 1 . The 3 day old seedlings were transferred to agar plates supplemented with ¼ B and D media . After 3 days growth , the plants were inoculated with M . loti R7A labelled with dsRed . The infection threads were counted at 9 and 14 days after inoculation . Seedlings were grown on agar plates supplemented with 1/4 B and D with 12 . 5 µg/mL AVG for 1–2 weeks . One seedling was transferred to a glass slide and Nod factor treatment was performed using 10–8 M M . loti R7A Nod factor solution . The samples were analysed on a confocal microscope LSM780 ( ZEISS ) and a water lens ( W plan-Apochromat 40x/1 . 0 DIC M27 , ZEISS ) . YC3 . 6 was excited at 458 nm , and emissions from ECFP and cpVenus were split into different detectors and collected at 463 to 509 and 519 to 621 nm . Calcium spiking was monitored for up to 3 hr after the Nod factor treatment on each root . Several regions of the same root were monitored for 10 to 30 min , and minimum five nuclei were monitored on each root . In total , 50 nuclei from wild type and 46 nuclei from nfre-1 root hairs were monitored . The fluorescence intensity data collected in the first 10 min for each nucleus was analysed by CaSA software ( Russo et al . , 2013 ) . For calculation to the mean time between Ca2+ spikes ( inter spike interval , ISI ) for each genotype , the mean of ISI for one cell was used . For RNA sequencing , 3 days seedlings were moved to agar plates supplemented with 1/4 B and D media and susceptible zone of 14 days-old plants was harvested after specific treatment as specified in Figure 3F . The total RNA was isolated from the susceptible zone ( 15 mm root pieces ) using Nucleo spin RNA plant ( Macherey-Nagel ) . Total RNA ( >0 . 8 µg ) from two biological replicas per sample was used by GATC Biotech ( Germany ) to prepare random primed cDNA library and for sequencing with Illumina HiSeq: read length 1 × 50 bp . For the analysis of the RNA sequencing data the read trimming and mapping were performed by CLC genomics workbench 9 . 5 . 3 using Lotus japonicus v3 . 0 at Lotus base ( https://lotus . au . dk/ ) ( Mun et al . , 2016 ) , as reference . Differentially expressed genes ( log2 fold change >0 or<0 , adjusted p value < 0 , 05 ) were determined using the DESeq2 R package , with the ‘fittype’ parameter set to ‘local’ and the ‘betaprior’ parameter to ‘true’ . The HTS filter R package was integrated in the DESeq2 pipeline before calling for differentially expressed genes , in order to remove from the analysis the genes with low read counts . Venn diagrams were generated with the VennDiagram R package ( Chen and Boutros , 2011 ) . Seedlings were germinated and grown on a stack of wet filter paper in upright position at 21°C under 16/8 hr light/dark conditions . Roots of 7 day old seedlings were cut to 0 . 5 cm pieces , collected to white 96 well flat bottom polystyrene plates ( Greiner Bio-One ) and kept overnight in sterile water in darkness at room temperature to recover from stress before the treatment . ROS measurements were conducted in a Varioskan Flash Multimode Reader ( Thermo Scientific ) in luminometric measurement mode . The reaction mixture consisted of the respective elicitor , 20 µM luminol ( Sigma ) and 5 µg/ml horseradish peroxidase ( Sigma ) . As elicitor 1 µM tetra-N-acetyl-chitotetraose , CO4 ( Megazyme ) or octa-N-acetyl-chitooctaose , CO8 ( IsoSep ) was used . In the negative control wells water was replacing the elicitor . In one measuring well six roots ( 10 mg root material ) was used . In one repetition three wells were measured for every treatment for every genotype . At least two repetitions were conducted with similar results . N . benthamiana , infiltrated leaves were analysed after 3 days using a Zeiss LSM510 MetaConfocal microscope . The leaves were infiltrated with 0 . 8 M mannitol to induce plasmolysis . The samples were mounted in 30% glycerol on the slide . For cell death in N . benthamiana , infiltrated leaves were observed after 4 days . RNA-seq reads were deposited at ArrayExpress ( accession: E-MTAB-5855 ) . | Microbes – whether beneficial or harmful – play an important role in all organisms , including plants . The ability to monitor the surrounding microbes is therefore crucial for the survival of a species . For example , the roots of a soil-growing plant are surrounded by a microbial-rich environment and have therefore evolved sophisticated surveillance mechanisms . Unlike most other plants , legumes , such as beans , peas or lentils , are capable of growing in nitrogen-poor soils with the help of microbes . In a mutually beneficial process called root nodule symbiosis , legumes form a new organ called the nodule , where specific soil bacteria called rhizobia are hosted . Inside the nodule , rhizobia convert atmospheric dinitrogen into ammonium and provide it to the plant , which in turn supplies the bacteria with carbon resources . The interaction between the legume plants and rhizobia is very selective . Previous research has shown that plants are able to identify specific signaling molecules produced by these bacteria . One signal in particular , called the Nod factor , is crucial for establishing the relationship between these two organisms . To do so , the legumes use specific receptor proteins that can recognize the Nod factor molecules produced by bacteria . Two well-known Nod factor receptors , NFR1 and NFR5 , belong to a large family of proteins , which suggests that other similar receptors may be involved in Nod factor signaling as well . Now , Murakami et al . identified the role of another receptor called NRFe by studying the legume species Lotus japonicus . The results showed that NFRe and NFR1 share distinct biochemical and molecular properties . NRFe is primarily active in the cells located in a specific area on the surface of the roots . Unlike NFR1 , however , NFRe has a restricted signaling capacity limited to the outer root cell layer . Murakami et al . found that when NRFe was mutated , the Nod factor signaling inside the root was less activated and fewer nodules formed , suggesting NRFe plays an important role in this symbiosis . NFR1-type receptors have also been found in plants outside legumes , which do not form a symbiotic relationship with rhizobia . Identifying more receptors important for Nod-factor signaling could provide a basis for new biotechnological targets in non-symbiotic crops , to improve their growth in nutrient-poor conditions . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"plant",
"biology"
] | 2018 | Epidermal LysM receptor ensures robust symbiotic signalling in Lotus japonicus |
A set of chemical reactions that require a metabolite to synthesize more of that metabolite is an autocatalytic cycle . Here , we show that most of the reactions in the core of central carbon metabolism are part of compact autocatalytic cycles . Such metabolic designs must meet specific conditions to support stable fluxes , hence avoiding depletion of intermediate metabolites . As such , they are subjected to constraints that may seem counter-intuitive: the enzymes of branch reactions out of the cycle must be overexpressed and the affinity of these enzymes to their substrates must be relatively weak . We use recent quantitative proteomics and fluxomics measurements to show that the above conditions hold for functioning cycles in central carbon metabolism of E . coli . This work demonstrates that the topology of a metabolic network can shape kinetic parameters of enzymes and lead to seemingly wasteful enzyme usage .
An essential trait of living systems is their ability to reproduce . This fundamental ability makes all living organisms autocatalytic by definition . Moreover , autocatalytic metabolism is considered to be one of the essential components of life ( Ganti et al . , 2003 ) . In this work , we focus on autocatalytic cycles in chemical reaction systems , in the context of metabolic networks . The components we consider are the metabolites of the system , with autocatalytic cycles being formed using the reactions of the metabolic network . An illustrative example for a metabolic autocatalytic cycle is glycolysis . In glycolysis , 2 ATP molecules are consumed in the priming phase , in order to produce 4 ATP molecules in the pay off phase . Therefore , in order to produce ATP in glycolysis , ATP must already be present in the cell . Subsequently , autocatalysis of ATP in glycolysis ( also referred to as ‘turbo design’ ) results in sensitivity to mutations in seemingly irrelevant enzymes ( Teusink et al . , 1998 ) . Autocatalytic cycles have also been shown to be optimal network topologies that minimize the number of reactions needed for the production of precursor molecules from different nutrient sources ( Riehl et al . , 2010 ) . Metabolic networks often require the availability of certain intermediate metabolites , in addition to the nutrients consumed , in order to function . Examples of obligatorily autocatalytic internal metabolites in different organisms , on top of ATP , are NADH , and coenzyme A ( Kun et al . , 2008 ) . We find that other central metabolites , such as phospho-sugars and organic acids , are autocatalytic under common growth conditions . The requirement for availability of certain metabolites in order to consume nutrients implies metabolic processes must be finely controlled to prevent such essential metabolites from running out; in such cases metabolism will come to a halt . Autocatalytic cycles present control challenges because the inherent feed-back nature of autocatalytic cycles makes them susceptible to instabilities such as divergence or drainage of their intermediate metabolites ( Teusink et al . , 1998; Fell et al . , 1999; Reznik and Segrè , 2010 ) . The stability criteria typically represent one constraint among the parameters of the cycle enzymes . For large cycles , such as the whole metabolic network , one such constraint adds little information . For compact autocatalytic cycles embedded within metabolism , one such constraint is much more informative . We thus focus our efforts on analyzing small autocatalytic cycles . Finding the unique constraints that metabolic autocatalytic cycles impose is essential for understanding the limitations of existing metabolic networks , as well as for modifying them for synthetic biology and metabolic engineering applications . A key example of an autocatalytic cycle in carbon metabolism is the Calvin-Benson-Bassham cycle ( CBB ) ( Benson et al . , 1950 ) . The carbon fixation CBB cycle , which fixes CO2 while transforming five-carbon compounds into two three-carbon compounds , serves as the main gateway for converting inorganic carbon to organic compounds in nature ( Raven et al . , 2012 ) . The autocatalytic nature of the CBB cycle stems from the fact that for every 5 five-carbon compounds the cycle consumes , 6 five-carbon compounds are produced ( by the fixation of 5 CO2 molecules ) . Beyond the CBB cycle , we show that most of the reactions and metabolites in the core of central carbon metabolism are part of compact ( i . e . consisting of around 10 reactions or fewer ) metabolic autocatalytic cycles . Some of the autocatalytic cycles we find are not usually considered as such . The span of autocatalytic cycles in central carbon metabolism suggests that the constraints underlying their stable operation have network-wide biological consequences . In this study , we present the specific requirements metabolic autocatalytic cycles must meet in order to achieve at least one , non-zero , steady state which is stable in respect to fluctuations of either metabolites or enzyme levels close to the steady state point . The mathematical tools we use are part of dynamical systems theory ( Strogatz , 2014 ) . We identify the kinetic parameters of enzymes at metabolic branch points out of an autocatalytic cycle as critical values that determine whether the cycle can operate stably . We show that the affinity of enzymes consuming intermediate metabolites of autocatalytic cycles must be limited to prevent depletion of these metabolites . Moreover , we show that the stable operation of such cycles requires low saturation , and thus excess expression , of these enzymes . Low saturation of enzymes has previously been suggested to stem from a number of reasons in different contexts: ( A ) to achieve a desired flux in reactions close to equilibrium , for example in glycolysis ( Staples and Suarez , 1997; Eanes et al . , 2006; Flamholz et al . , 2013 ) ; ( B ) to provide safety factors in the face of varying nutrient availability , for example in the brush-border of the mouse intestine ( Weiss et al . , 1998 ) ; ( C ) to accommodate rapid shifts in demand from the metabolic networks in muscles with low glycolytic flux ( Suarez et al . , 1997 ) ; ( D ) to allow fast response times , for example to pulses of oxidative load in erythrocytes , resulting from their adherence to phagocytes ( Salvador and Savageau , 2003 ) . Our findings add to these reasons the essential stabilizing effect of low saturation of branch reactions on the stability of fluxes through autocatalytic cycles . We use recent fluxomics and proteomics data to test the predictions we make . We find them to hold in all cases tested where autocatalytic cycles support flux . Our analysis demonstrates how the requirement for stable operation of autocatalytic cycles results in design principles that are followed by autocatalytic cycles in-vivo . The results and design principles presented here can be further used in synthetic metabolic engineering applications that require proper functioning of autocatalytic cycles .
Different definitions exist for autocatalytic sets in the context of chemical reaction networks ( Hordijk and Steel , 2004; Eigen and Schuster , 2012; Kun et al . , 2008 ) . Here we define an autocatalytic cycle as a set of reactions and metabolites that form a cycle , and that , when the reactions are applied to the substrates at the given stoichiometric ratios , increase the amount of the intermediate metabolites . A minimal example of a metabolic autocatalytic cycle is shown in Figure 1 , where an internal metabolite joins with an external assimilated metabolite to give rise to 1+δ copies of the internal metabolite , representing an increase by δ copies . For stable operation , δ copies have to branch out of the cycle , and this consumption must be robust to small fluctuations in enzyme levels and metabolite concentrations . For a formal , mathematical definition , see Materials and methods section "Formal definition of an autocatalytic metabolic cycle" . 10 . 7554/eLife . 20667 . 003Figure 1 . A basic autocatalytic cycle requires an internal metabolite to be present in order to assimilate the external metabolite into the cycle , increasing the amount of the internal metabolite by some amount , δ . DOI: http://dx . doi . org/10 . 7554/eLife . 20667 . 003 While rarely discussed as such , a systematic search in the central carbon metabolism core model of E . coli ( see Materials and methods section "Systematic identification of autocatalytic cycles in metabolic networks" ) shows the ubiquity of compact autocatalytic cycles . On top of the previously discussed CBB cycle ( Figure 2 , example I ) , we show two other prominent examples:10 . 7554/eLife . 20667 . 004Figure 2 . Three representative autocatalytic cycles in central carbon metabolism: ( I ) The Calvin-Benson-Bassham cycle ( yellow ) ; ( II ) The glyoxylate cycle ( magenta ) ; ( III ) A cycle using the phosphotransferase system ( PTS ) to assimilate glucose ( cyan ) . Assimilation reactions are indicated in green . Arrow width in panels represent the relative carbon flux . DOI: http://dx . doi . org/10 . 7554/eLife . 20667 . 00410 . 7554/eLife . 20667 . 005Figure 2—figure supplement 1 . An autocatalytic cycle assimilating ribose-5-phosphate using the pentose phosphate pathway . This cycle contains a direct input reaction ( rpi , dashed line ) allowing the cycle to operate with broader sets of kinetic parameters than cycles missing this feature . A knockout strain where rpi is eliminated , does not grow under ribose despite having the theoretical ability to do so . DOI: http://dx . doi . org/10 . 7554/eLife . 20667 . 00510 . 7554/eLife . 20667 . 006Figure 2—figure supplement 2 . An autocatalytic cycle assimilating dhap while consuming gap using the fba reaction in the gluconeogenic direction . This cycle contains a direct input reaction ( tpi , dashed line ) allowing the cycle to operate with broader sets of kinetic parameters than cycles missing this feature . According to fluxomics data this cycle does not operate in vivo as a more energy efficient alternative in growth under glycerol is to use the tpi reaction and proceed in the glycolitic direction in the lower part of glycolysis . A knockout strain where tpi reaction is eliminated , does not grow under glycerol despite having the theoretical ability to do so . DOI: http://dx . doi . org/10 . 7554/eLife . 20667 . 006 Two additional examples are presented in Figure 2—figure supplements 1 and 2 and discussed below . The ubiquity of compact autocatalytic cycles in the core of central carbon metabolism motivates the study of unique features of autocatalytic cycles , as derived below , which may constrain and shape the kinetic parameters of a broad set of enzymes at the heart of metabolism . To explore general principles governing the dynamic behavior of autocatalytic cycles , we consider the simple autocatalytic cycle depicted in Figure 3A . This cycle has a single intermediate metabolite , X . We denote the flux through the autocatalytic reaction of the cycle by fa , such that for any unit of X consumed , it produces two units of X . The autocatalytic reaction assimilates an external metabolite ( denoted A ) , which we assume to be at a constant concentration . We denote the flux through the reaction branching out of the cycle by fb . Biologically , fb represents the consumption of the intermediate metabolite X . In the cycles we find in central carbon metabolism , the branch reactions provide precursors that support growth through subsequent reactions . We thus also sometimes consider fb to represent biomass generation . 10 . 7554/eLife . 20667 . 007Figure 3 . Analysis of a simple autocatalytic cycle . ( A ) A simple autocatalytic cycle induces two fluxes , fa and fb as a function of the concentration of X . These fluxes follow simple Michaelis-Menten kinetics . A steady state occurs when fa=fb , implying that X˙=0 . The cycle always has a steady state ( i . e . X˙=0 ) at X=0 . The slope of each reaction at X=0 is Vmax/KM . A steady state is stable if at the steady state concentration dX˙dX<0 . ( B ) Each set of kinetic parameters , Vmax , a , Vmax , b , KM , a , KM , b determines two dynamical properties of the system: If Vmax , b>Vmax , a , then a stable steady state concentration must exist , as for high concentrations of X the branching reaction will reduce the concentration of X ( cyan domain , cases ( I ) and ( II ) ) . If Vmax , bKM , b<Vmax , aKM , a , implying that Vmax , bVmax , a<KM , bKM , a , then zero is a non-stable steady state concentration as if X is slightly higher than zero , the autocatalytic reaction will carry higher flux , further increasing the concentration of X ( magenta domain , cases ( I ) and ( IV ) ) . At the intersection of these two domains a non-zero , stable steady state concentration exists , case ( I ) . DOI: http://dx . doi . org/10 . 7554/eLife . 20667 . 007 For simplicity in the derivation , we assume irreversible Michaelis-Menten kinetics for the two reactions . Even though fa should follow bisubstrate velocity equation , assuming constant concentration of A reduces the bisubstrate equation to a simple Michaelis-Menten equation . The apparent kinetic constants of the equation depend on the constant value of A ( see Materials and methods section "Connecting bisubstrate reaction kinetic constants with simple Michaelis-Menten constants" ) . We extend our analysis to bisubstrate reaction equations in the next section . We therefore assume that:fa=Vmax , aXKM , a+Xfb=Vmax , bXKM , b+X where Vmax is the maximal flux each reaction can carry and KM is the substrate concentration at which half the maximal flux is attained . Physiologically , these kinetic parameters must be positive . Using these simple forms allows us to obtain an analytic solution . We discuss more general cases below . We characterize the metabolic state of this system by the concentration of the metabolite X . We note that knowing the concentration of X suffices in order to calculate the fluxes originating from it , fa and fb , thus fully defining the state of the system . A steady state of the system is defined as a concentration , X* , which induces fluxes that keep the concentration constant , such that the total in-flux to X is exactly equal to the total out-flux from it . In our example , the outgoing flux from X is fa+fb and the incoming flux to X is 2fa , so at steady state it holds that: ( 1 ) X˙=dXdt=2fa- ( fa+fb ) =0 Intuitively , at steady state , the branch reaction must consume all the excess intermediate metabolite that is produced by the autocatalytic reaction . Indeed , expanding the condition above gives:fa=fb⇒Vmax , aX∗KM , a+X∗=Vmax , bX∗KM , b+X∗ which is satisfied either if X*=0 or if: ( 2 ) X*=Vmax , bKM , a-Vmax , aKM , bVmax , a-Vmax , b implying that: ( 3 ) X*KM , a=Vmax , bVmax , a-KM , bKM , a1-Vmax , bVmax , a The concentration of X cannot be negative , and thus we get a constraint on the kinetic parameters for which a positive steady state exists . Either both the numerator and the denominator of Equation 3 are positive , such that:1 > Vmax , bVmax , a > KM , bKM , a , or both are negative , such that:1 < Vmax , bVmax , a < KM , bKM , a These constraints are graphically illustrated in Figure 3B , cases ( III ) and ( I ) . In order to gain intuition for this relationship we note that VmaxKm is the slope of the Michaelis Menten function at X=0 . The existence of a positive steady state can be used to get that:X∗ > 0⇒Vmax , bKM , a−Vmax , aKM , bVmax , a−Vmax , b > 0⇒Vmax , bKM , b−Vmax , aKM , aVmax , a−Vmax , b > 0 The last inequality above implies that in order for a positive steady state to exist , the reaction with higher maximal flux must have a shallower slope at X=0 . Mathematically , the constraint states that if Vmax , a > Vmax , b then Vmax , aKM , a < Vmax , bKM , b . Alternatively , if Vmax , a < Vmax , b then Vmax , aKM , a > Vmax , bKM , b . This condition can be intuitively understood , as the reaction with shallower slope at X=0 has smaller fluxes for small values of X compared with the other reaction , so unless it has higher fluxes than the other reaction for large values of X ( meaning that its maximal flux is higher ) , the two will never intersect ( see Figure 3B ) . While having a steady state at positive concentration is an essential condition to sustain flux , it is not sufficient in terms of biological function . The steady state at positive concentration must also be stable to small perturbations . Stability with respect to small perturbations is determined by the response of the system to small deviations from the steady state , X* ( at which , by definition X˙=0 ) . Assuming X=X*+ΔX , stability implies that if ΔX is positive then X˙ needs to be negative at X*+ΔX , reducing X back to X* , and if ΔX is negative , X˙ will need to be positive , increasing X back to X* . It then follows that in order for X* to be stable , dX˙dX < 0 at X=X* , implying that upon a small deviation from the steady state X* ( where X˙=0 ) , the net flux X˙ will oppose the direction of the deviation . For the simple kinetics we chose , the stability condition dictates that: ( 4 ) dX˙dX|X=X∗=Vmax , aKM , a ( KM , a+X∗ ) 2−Vmax , bKM , b ( KM , b+X∗ ) 2 < 0 The analysis is straightforward for the case of X*=0 , yielding that 0 is a stable steady state concentration if Vmax , bKM , b > Vmax , aKM , a , corresponding to the area above the diagonal in Figure 3B , where Vmax , bVmax , a > KM , bKM , a . These cases are denoted as cases ( II ) and ( III ) . If the relation is reversed ( i . e . Vmax , bKM , b < Vmax , aKM , a ) , then 0 is an unstable steady state . The criterion that is of interest , however , is the criterion for stability of the non-zero steady state , X*=Vmax , bKM , a-Vmax , aKM , bVmax , a-Vmax , b . In this case , substituting X* in Equation 4 gives the opposite condition to that of X*=0 . This steady state is thus stable if Vmax , bKM , b<Vmax , aKM , a , corresponding to the magenta domain in Figure 3B , containing cases ( I ) and ( IV ) , and unstable otherwise . The stability criterion can be generally stated in metabolic control terms ( Fell , 1997 ) using the notion of elasticity coefficients of reactions , defined as:εXf=∂f∂XXf In these terms , stability is obtained if and only if the elasticity of the branch reaction at the positive steady state concentration is greater than the elasticity of the autocatalytic reaction:dfbdX|X=X∗ > dfadX|X=X∗⇒εXfb > εXfa The complete analysis is summarized in Figure 3B . Domain ( I ) is the only domain where a positive , stable steady state exists . Domains ( I ) and ( III ) are the domains at which a positive steady state concentration exists , but in domain ( III ) that steady state is not stable . The domains below the diagonal ( cases ( I ) and ( IV ) ) are the domains where X*=0 is an unstable steady state concentration , so that if another steady state exists , it is stable , but in domain ( IV ) no positive steady state exists . The domains above the diagonal ( cases ( II ) and ( III ) ) are the domains where X*=0 is a stable steady state concentration , so that the other steady state , if it exists , is unstable . Aside from existence and stability , a quantitative relationship between the affinity of the biomass generating , branching reaction and the flux it carries can be obtained . This relationship is opposite to the standard one , meaning that unlike the common case where the flux f increases when the affinity becomes stronger , in this case , because the steady state concentration increases when KM , b becomes weaker ( Equation 7 in Materials and methods section "Steady state concentration dependence on kinetic parameters of autocatalytic and branch reactions" ) , fb also increases when KM , b becomes weaker . To conclude , for this simple cycle , we get that in order for a positive-concentration stable steady state to exist ( case ( I ) ) , two conditions must be satisfied: ( 5 ) {Vmax , b > Vmax , aVmax , bKM , b <Vmax , aKM , a The first requirement states that the maximal flux of the biomass generating , branching reaction should be higher than the maximal flux of the autocatalytic reaction . This requirement ensures a stable solution exists , as large concentrations of X will result in its reduction by the branch reaction . The second requirement states that for concentrations of X that are close enough to 0 , the autocatalytic reaction is higher than the branch reaction ( as can be inferred from the slopes ) . This requirement implies that the two fluxes will be equal for some positive concentration of X , ensuring a positive steady state exists . As this requirement further implies that below the positive steady state the branch reaction will carry less flux than the autocatalytic reaction , it follows that small deviations of the concentration of X below the steady state will result in an increase in its concentration by the autocatalytic reaction , driving it back to the steady state . Meeting the second constraint has another consequence . Interestingly , these conditions imply that if KM , b < KM , a then no positive stable steady state can be achieved . Specifically , changes to the expression levels of the enzymes catalyzing fa and fb only affect Vmax , a and Vmax , b , and therefore do not suffice to attain a stable positive steady state . This indicates that stability of autocatalytic cycles , that are represented by the model analyzed above , depends on inherent kinetic properties of the enzymes involved and cannot always be achieved by modulating expression levels . We suggest this property to be a design principle that can be critical in metabolic engineering . To evaluate the validity of our analysis of autocatalytic cycles we searched for growth conditions under which the autocatalytic cycles we identified in central carbon metabolism carry substantial flux in-vivo . We used recent in-vivo flux measurements in E . coli from Gerosa et al . ( 2015 ) . According to the data , two autocatalytic cycles carry substantial flux under at least one of the growth conditions measured: a cycle using the PTS carries significant fluxes in growth on glucose and on fructose; the glyoxylate cycle carries significant flux in growth on acetate and on galactose . As noted above , we predict a design principle for functioning autocatalytic cycles: that at least one branch reaction should have a steeper response than the corresponding autocatalytic reaction at steady state . This requirement is sufficient , but not necessary , for the autocatalytic cycle to be at a stable steady state point . Moreover , having more than one branch point at which the branch reaction has a steeper response than the autocatalytic reaction increases the robustness of the steady state flux in the cycle as shown in the Materials and methods section "Multiple unsaturated branch reactions increase convergence speed and dampen oscillations" . An outcome of the relationship between the steepness’s of responses is a reverse relationship between the saturation levels of the corresponding reactions ( Equation 35 ) . In order to evaluate the saturation level of a reaction under a given condition , two values must be obtained: To estimate the maximal capacity of a reaction we followed the procedure described in ( Davidi et al . , 2016-22 ) ( see Materials and methods section "Evaluating maximal flux capacity of reactions under a given condition" ) . We used the data from ( Gerosa et al . , 2015 ) to identify the major branch points in each functioning cycle and the actual flux in them under each of the relevant conditions . The results are presented in Figure 6 and are provided , with the relevant calculations , in Supplementary file 1 . 10 . 7554/eLife . 20667 . 010Figure 6 . Major branch points and relative enzyme saturation in operating autocatalytic cycles . Solid arrow width represents carbon flux per unit time . Shaded arrow width represents the maximal carbon flux capacity per unit time , given the expression level of the catalyzing enzyme . In all cases there is enough excess capacity in the branching reactions to prevent the cycle from overflowing . A 4% flux from pep to biomass was neglected in growth under glucose and fructose . Only in one out of the nine branch points observed ( the branch point at fbp in growth under fructose ) , the outgoing reaction is significantly more saturated than the autocatalytic reaction . ( * ) A branch point at which the branching reaction is more saturated than the autocatalytic reaction , which may result from neglecting fructose transport directly as f6p when deriving fluxes ( see text ) . DOI: http://dx . doi . org/10 . 7554/eLife . 20667 . 010 Our results show that for any of the four functioning autocatalytic cycle cases , in at least one branch point the biomass generating branch reaction has a larger maximal flux capacity , and is considerably less saturated than the respective autocatalytic reaction , in accordance with our predictions . Moreover , out of nine branch points analyzed , in six branch points the branching reactions were significantly less saturated than the autocatalytic reactions , in two branch points the saturation levels were similar , and only in one branch point the autocatalytic reaction was less saturated than the branching reaction . The branch point at which the autocatalytic reaction is less saturated than the branch reaction is the branch point from fructose-1 , 6-bisphosphate in growth on fructose as the carbon source . The high saturation of the branch reaction arises as a large flux is reported for the fbp reaction , whereas the corresponding enzyme is not highly expressed under this condition . The large reported flux through fbp arises due to assuming a single transport pathway for fructose , entering the cycle as fructose-1 , 6-bisphosphate . However , an alternative fructose transport pathway is known to occur for the concentration at which the measurements were made ( Kornberg , 1990 ) . The alternative transport pathway produces fructose-6-phosphate from external fructose . Therefore , any flux through the alternative transport pathway should be directly deducted from the flux through fbp . Assuming 20% of the consumed fructose uses this pathway suffices in order to balance the saturation levels at the fructose-1 , 6-bisphosphate branch point . We made two negative control analyses to examine whether other reasons do not underlie the trend we find . First , we compared the saturation levels at the same branch points in growth conditions at which the autocatalytic cycles do not function , but the reactions carry flux . We find that for these cases , only 4 out of 9 cases satisfy the low branch saturation condition ( Supplementary file 1 ) . Second , we searched for branch points out of non-autocatalytic cycles and tested whether in such points branch reactions are also consistently less saturated than their corresponding cycle reactions . We found two flux-carrying cycles: the TCA cycle , carrying flux in glucose , fructose , and glycerol growth , and a cycle consisting of the pentose-phosphate pathway combined with gluconeogenesis , carrying flux in acetate , glycerol , and succinate growth . Out of the total six conditions-branch points cases , in three the branch reaction was less saturated than the cycle reaction , and in three the cycle reaction was less saturated than the branch reaction ( Supplementary file 1 ) . We therefore conclude that , for cases that do not involve autocatalysis , the saturation of branch versus cycle reactions seems evenly distributed . The consistently lower saturation values of biomass generating branch reactions demonstrate that the expressed enzymes have enough capacity to prevent the autocatalytic cycle from increasing the concentration of intermediate metabolites infinitely . Moreover , the lower saturation values of the biomass generating reactions suggest that at the steady state point their derivatives are higher , ensuring stable operation of the cycle . Another demonstration of the autocatalytic mechanism being at play is in the CBB cycle , which is not a part of the metabolic network of wild type E . coli , and for which no flux measurements are available . This cycle has been recently introduced synthetically into E . coli and was shown to carry flux in it , given further metabolic engineering of central carbon metabolism ( Antonovsky et al . , 2016 ) . The experimentally observed key evolutionary event enabling the functioning of the CBB cycle , was a mutation affecting the kinetic properties of the main branching reaction out of the CBB pathway , prs , weakening its affinity to its substrate , ribose-5p . The observed weakening of affinity of prs is directly in line with our predictions on the relationship between the affinity of branch reactions and the affinity of the corresponding cycle reactions ( see Materials and methods section of Antonovsky et al . , 2016 ) . The other examples of autocatalytic cycles we found did not carry flux in any of the conditions for which data were available . The pentose-phosphate cycle variants do not carry flux in any of the measured conditions , which is expected given that growth on ribose was not measured . The gluconeogenic FBA with ED pathway cycle also did not carry flux in any of the measured conditions . Although glycerol could have been a potential carbon source to use this pathway , the metabolic network allows for a more energy efficient growth by using the tpi reaction , as was indeed observed . To conclude , existing data supports predictions made by our model , given the requirement for stable steady state operation of autocatalytic cycles . This agreement between predictions and measurements is especially encouraging given the highly limited information on kinetic properties , concentrations , and fluxes under various growth conditions . Allosteric regulation can modulate the kinetic properties of enzymes at branch points , and of the cycle in general . As such , the relevant condition for the existence of a stable positive steady state should hold for the updated kinetic properties as defined following the effect of allosteric regulation . We further analyze the ability of specific allosteric interactions to support fast convergence and stability of autocatalytic cycles in the Materials and methods section "Allosteric regulation can improve network performance" . We compare the expected beneficial allosteric interactions against the allosteric regulation network of the two functioning autocatalytic cycles we identified , the PTS-using autocatalytic cycle and the glyoxylate cycle ( Supplementary file 1 , regulation data were taken from Keseler et al . ( 2013 ) and Schomburg et al . ( 2004 ) ) . For the PTS cycle , we find that there are a total of 12 allosteric interactions , 7 inhibitions and five activations . Out of these 12 interactions , 11 interactions follow our expectations in terms of the type of the regulating metabolite ( assimilated metabolite , cycle intermediate , or branch product ) , the regulated reaction ( cycle reaction , branch reaction , or the reverse of a branch reaction ) , and the direction of the regulation ( activation or inhibition ) . One interaction , the activation of fba by pep , does not follow our expectation . For the glyoxylate cycle , we find that there are a total of 13 interactions , 12 inhibitions and one activation . Out of these 13 interactions , 8 interactions follow our expectations and 5 do not . The lack of significant agreement between the expected regulation direction and the actual regulation found for this cycle is consistent with the observation in Gerosa et al . ( 2015 ) that TCA cycle fluxes are regulated mainly by transcription and not by reactants levels . It is important to note that allosteric regulation serves many roles , and that the metabolic network faces many more challenges than just the support of stable autocatalysis . Therefore , the agreement we find between existing allosteric interactions and the expected regulation scheme supporting autocatalysis does not suggest that the autocatalytic nature of the PTS is the only , or even main underlying reason for these allosteric interactions .
Our study into the dynamics and stability of autocatalytic cycles suggests design principles applicable to both systems biology , that aims to understand the function of natural networks , and in the context of synthetic biology , in the effort to express novel heterologous cycles . While autocatalytic cycles are often overlooked in the study of metabolism , we find that such cycles are at the heart of central carbon metabolism . Our autocatalytic modeling framework gives concrete predictions on saturation levels of branch reactions for operating autocatalytic cycles . We find these predictions agree well with empirically measured fluxomics and proteomics data sets . Given that there are other suggestions ( Staples and Suarez , 1997; Weiss et al . , 1998; Suarez et al . , 1997 ) that may underlie the low saturation of branch reactions , we compare the saturation levels of branch reactions versus their corresponding cycle reactions both under conditions when the autocatalytic cycle does not function , and for branch points out of non-autocatalytic cycles . Both tests show no bias towards low saturation of branch reactions out of non-autocatalytic cycles , contrary to the clear trend we find for reactions branching out of autocatalytic cycles . Our findings thus support the addition of stability of intermediate metabolites of autocatalytic cycles as an explanation for the seemingly wasteful expression of enzymes ( Salvador and Savageau , 2003 , 2006 ) . The model we present can also highlight metabolic branch points at which the kinetic efficiency of enzymes is constrained due to stability requirements of a corresponding autocatalytic cycle . A common concept in synthetic biology is that the successful implementation of novel pathways requires the expression of functional enzymes in the correct amounts in the target organism . Here we show that in the context of autocatalytic cycles , such expression modulation may not suffice . Specifically , changes to the substrate affinity of enzymes at branch points of the cycle may be required in order for the novel pathway to function . Another aspect of our findings is that while it is common to assume that strong affinity and high catalytic rate are desirable traits for enzymes , such seeming improvements may actually lead to instability and thus to non functional metabolic cycles . Furthermore , for reactions branching out of autocatalytic cycles , weaker affinities increase the steady state concentration of intermediate metabolites , resulting in higher fluxes both through the cycle , and through the branch reaction , suggesting an unconventional strategy for optimizing fluxes through such reactions . We note that because allosteric regulators modify the affinity of the enzymes they target , such regulators can potentially be used to restrict the affinity of branch reactions only when the autocatalytic cycle functions . An experimental demonstration of these principles in-vivo is the recent implementation of a functional CBB cycle in E . coli by introducing the two genes missing for its function ( Antonovsky et al . , 2016 ) . The successful introduction of the genes did not suffice to make the cycle function , and further directed evolution was needed in order to achieve successful operation of the cycle . Strikingly , most evolutionary changes occurred in branch points from the cycle ( Antonovsky et al . , 2016 ) . The change which was biochemically characterized in the evolutionary process was the decrease of the value of kcatKM of phosphoribosylpyrophosphate synthetase ( prs ) , one of the enzymes responsible for flux out of the CBB cycle , corresponding to the branch reaction in our simple model . This is beautifully in line with the predictions of our analysis that suggest that decreasing VmaxKM of branch reactions can lead to the existence of a stable flux solution . Our observation regarding the stabilizing effect of input fluxes into an autocatalytic cycle can provide means to mitigate the stability issue in synthetic biology metabolic engineering setups . In such setups , conducting directed evolution under gradually decreasing input fluxes , such as those achieved in a chemostat , allows for a pathway to gradually evolve towards sustainable , substantial flux . Finally , while our work focuses on cycles increasing the amount of carbon in the system , we note that autocatalysis can be defined with respect to other quantities such as energy ( e . g . ATP investment and production in glycolysis [Teusink et al . , 1998] ) , non-carbon atoms , reducing power , or other moieties ( Reich and Selkov , 1981 ) . As autocatalysis is often studied with relation to the origin of life , our analysis may be useful in studying synthetic autocatalytic systems such as the one recently described in Semenov et al . ( 2016 ) . The analysis we present here can thus be of relevance for the analysis of metabolic networks in existing organisms and for the design of novel synthetic systems .
Given a metabolic network composed of a set of reactions and metabolites , the following criteria can be used to define a subset of the network that is an autocatalytic cycle: First we define a metabolic cycle . A set of irreversible reactions ( for reversible reactions only one direction can be included in the set ) and metabolites forms a cycle if every metabolite of the set can be converted , by sequential application of reactions in the set ( where two reactions can be chained if a metabolite in the set is a product of the first reaction and a substrate of the second reaction ) , to every other metabolite in the set . A cycle is autocatalytic if the reactions of the cycle can be applied , each reaction at an appropriate , positive number of times , such that the resulting change in the amount of each of the metabolites forming the cycle is non-negative , with at least one metabolite of the cycle having a strictly positive change . The same definition can be stated in terms of reaction vectors and a stoichiometric matrix . If a metabolic network has n metabolites , indicated by the numbers 1 to n , then every reaction , r , in the network can be described as a vector Vr in ℤn , such that the i’th coordinate of Vr specifies how much of metabolite i the reaction r produces ( if r consumes a metabolite , then the value at the coordinate representing the metabolite is negative ) . With this notation , a set of metabolites: M=m1⋯mj and a set of reactions , R=r1⋯rk form an autocatalytic cycle if: We implemented an algorithm to systematically search for autocatalytic cycles in metabolic networks . The algorithm is not comprehensive , in the sense that there may be autocatalytic cycles that will not be identified by it . Further work will enable a more advanced algorithm to identify additional autocatalytic cycles in full metabolic networks . We used the algorithm on the core carbon metabolism network of E . coli ( Orth et al . , 2010 ) . In our framework , a metabolic network is defined by a set of reactions , ( R¯ ) . Each reaction is defined by a set of substrates and a set of products , with corresponding stoichiometries Ri= ( S , P , NS , NP ) , such that Ri describes the reaction ∑jNjSSj→∑kNkPPk . The algorithm works as follows: The algorithm assumes reactions consume exactly one molecule of any of their substrates and produce exactly one molecule of any of their products , an assumption that holds for the core model of E . coli , but not in metabolic networks in general . The simple cycle steady state concentration , X* , is given in Equation 2 . Taking the derivative of this expression with respect to KM , a , KM , b , Vmax , a , and Vmax , b , under the assumption that the kinetic parameters satisfy the stability conditions in Equation 5 gives: ( 7 ) ∂X∗∂KM , a=Vmax , bVmax , a−Vmax , b < 0∂X∗∂KM , b=−Vmax , aVmax , a−Vmax , b > 0∂X∗∂Vmax , a=Vmax , b ( KM , b−KM , a ) ( Vmax , a−Vmax , b ) 2 > 0∂X∗∂Vmax , b=Vmax , a ( KM , a−KM , b ) ( Vmax , a−Vmax , b ) 2 < 0 So that X* increases when KM , a decreases or Vmax , a increases ( or both ) corresponding to activation of fa . On the other hand , X* decreases when KM , b decreases or Vmax , b increases ( or both ) corresponding to activation of fb . Three standard equations are commonly used to describe the flux through irreversible bisubstrate reactions ( Leskovac , 2003 ) . We show that , under the assumption that the assimilated metabolite maintains constant concentration , these equations reduce to simple Michaelis-Menten equations . We derive the expressions for the apparent Michaelis-Menten constants , KM and Vmax , as functions of the kinetic constants of the bisubstrate reaction and the concentration of the assimilated metabolite . While the substrates in these equations are generally denoted as A and B , here , to keep the notation consistent , we will denote by A the assimilated metabolite and by X the internal metabolite of the cycle . The simplest equation describing a bisubstrate reaction assumes substituted enzyme ( Ping Pong ) mechanism ( Imperial and Centelles , 2014 ) . As this equation is symmetric with respect to the two substrates , we can arbitrarily decide which of the two substrates is the assimilated metabolite , and which is the internal metabolite . We get that the flux through the reaction is:f=VmaxAXKXA+KAX+AX Rearranging to get the dependence of the flux on X in a Michaelis-Menten like form we get that: ( 8 ) f=VmaxAKA+AXKXAKA+A+X which gives apparent Michaelis-Menten kinetic constants of: ( 9 ) V~max=VmaxAKA+AK~M=KXAKA+A The second bisubstrate reaction mechanism we consider is the ternary enzyme complex with random binding order of the two substrates . As this equation is also symmetric with respect to the two substrates , we can again arbitrarily decide which of the two substrates is the assimilated metabolite , and which is the internal metabolite . We get that the flux through the reaction is:f=VmaxAXKi , AKX+KXA+KAX+AX Rearranging to get the dependence of the flux on X in a Michaelis-Menten like form we get that: ( 10 ) f=VmaxAKA+AXKi , A+AKA+AKX+X which gives apparent Michaelis-Menten kinetic constants of: ( 11 ) V~max=VmaxAKA+AK~M=Ki , A+AKA+AKX The other equation describing a ternary enzyme complex bisubstrate reaction assumes ordered binding of the substrates . Because in ordered binding the equation is asymmetric with respect to the two substrates , analyzing this reaction is further split according to which of the two substrates is assumed to be the assimilated metabolite with constant concentration . If the first binding metabolite is assumed to be the assimilated metabolite we get that: ( 12 ) f=VmaxAXKi , AKX+KXA+AX=VmaxXKi , A+AAKX+X which gives apparent Michaelis-Menten kinetic constants of: ( 13 ) V~max=VmaxK~M=Ki , A+AAKX If the first binding metabolite is assumed to be the internal metabolite we get that: ( 14 ) f=VmaxAXKi , XKA+KAX+AX=VmaxAKA+AXKi , XKAKA+A+X which gives apparent Michaelis-Menten kinetic constants of: ( 15 ) V~max=VmaxAKA+AK~M=Ki , XKAKA+A To summarize , the most common equations describing bisubstrate reactions reduce to equations of the same form as Michaelis-Menten equations , under the assumption that one of the metabolites maintains a constant concentration . The apparent kinetic constants of the Michaelis-Menten equivalent equations depend on the kinetic constants of the bisubstrate reactions , as well as on the concentration of the assimilated metabolite . In Equation 5 we obtain constraints on the kinetic parameters of Michaelis-Menten reactions that ensure the existence and stability of a positive steady state . We observe that these constraints imply that even if the maximal rates of the two reactions can be easily modified , if KM , b<KM , a then such changes cannot suffice in order to satisfy the existence and stability constraints . Here , we map the same constraints from Equation 5 onto bisubstrate autocatalytic reactions . This mapping results in constraints on the assimilated metabolite concentration , as well as on the kinetic parameters of the bisubstrate autocatalytic reactions . We show that in all ternary enzyme complex bisubstrate reaction schemes , there is a lower bound on the concentration of the assimilated metabolite , below which the system cannot attain a stable positive steady state . We further show that the nature of the bisubstrate reaction qualitatively affects the ability to satisfy the stability constraints by changing expression levels alone . In the cases of substituted enzyme mechanism , random binding order ternary complex , and ordered binding ternary complex , with the assimilated metabolite binding first , unless the kinetic parameters of the participating enzymes satisfy specific inequalities , changes to the maximal reaction rates alone cannot suffice in order to satisfy the existence and stability constraints . However , in the case of ordered binding ternary complex with the internal metabolite binding first , changes to the maximal reaction rates alone suffice in order to allow for stable steady state to occur , given high enough concentration of the assimilated metabolite . We analyze each of the four possible bisubstrate reaction schemes separately below . Equation 2 shows the dependency between the steady state concentration of the internal metabolite X , X* , and the kinetic parameters of the reactions in the system . Substituting the dependencies of the apparent kinetic parameters from Equations 11 , 13 , and 15 into Equation 2 gives the dependency of X* on the kinetic parameters of the bisubstrate reactions and the concentration of the assimilated metabolite , A . We get for these three cases respectively that: ( 24 ) X∗=VmaxAKA+AKM , b−Vmax , bKi , A+AKA+AKXVmax , b−VmaxAKA+A ( 25 ) X∗=VmaxKM , b−Vmax , bKi , A+AAKXVmax , b−Vmax ( 26 ) X∗=VmaxAKA+AKM , b−Vmax , bKi , XKAKA+AVmax , b−VmaxAKA+A Assuming the kinetic parameters satisfy the stable steady state conditions derived in Equations 17 , 20 , and 22 , we note that when A is equal to its lower bound , the numerator in all three cases is 0 , resulting in X*=0 . Furthermore , as A decreases towards its lower bound , X* decreases resulting in a decrease in both fb and fa ( for the two latter cases this is trivial to show , as the terms involving A increase and decrease monotonically in accordance with their effect on X* . In the first case , taking the derivative of the numerator w . r . t . A shows the derivative is always positive , resulting in the same conclusion ) . Interestingly , in the first and last cases , if Vmax > Vmax , b , then an upper bound on the concentration of A also exists . As the concentration of A approaches this upper bound , the denominator approaches 0 resulting in an increase in the concentration of X* towards infinity . The simple model assumed both the autocatalytic and the branch reactions are irreversible . Here we assume the branch reaction is reversible , and let Y denote its product . For simplicity , we further assume that Keq=1 , noting that this assumption can always be satisfied by measuring the concentration of Y in units of KeqX . We recall that the reversible Michaelis-Menten equation states that:fb=Vmax , b ( X-Y ) KX+X+KXKYY We assume that a third reaction , fc , irreversibly consumes Y . While assuming fc follows irreversible Michaelis-Menten kinetics is analytically tractable , the analysis is simpler , and as informative , under the assumption that fc=DY for some constant D . This simplification is equivalent to assuming fc follows Michaelis-Menten kinetics with Vmax , cKM , c≈D , and Vmax , c >> max ( Vmax , a , Vmax , b ) . We start by deriving the necessary conditions for steady state existence . Because at steady state fa=fc , it follows that: ( 27 ) Vmax , aX*KM , a+X*=DY*⇒Y*=Vmax , aDX*KM , a+X* Furthermore , as at the steady state fa=fb , we get that:Vmax , aX*KM , a+X*=Vmax , b ( X*-Y* ) KX+X*+KXKYY* Substituting Y* from Equation 27 gives:Vmax , aX*KM , a+X*=Vmax , b ( X*-Vmax , aDX*KM , a+X* ) KX+X*+KXKYVmax , aDX*KM , a+X* Which is satisfied when X*=0 ( implying that X*=Y*=0 is a steady state ) , or when X* satisfied the quadratic equation:0= ( X∗ ) 2+2KM , aVmax , b− ( KM , a+KX ) Vmax , a−KXVmax , a2KYD−Vmax , aVmax , bDVmax , b−Vmax , aX∗+KM , a ( Vmax , bKM , a−Vmax , aKX−Vmax , aVmax , bD ) Vmax , b−Vmax , a Albeit intimidating , this quadratic equation can be used to derive the conditions for the existence of a positive steady state . Only if both of the roots of this equation are negative , no positive steady state exists . We recall that the two roots of a quadratic equation of the form 0=aX2+bX+c are negative iff:{b=2KM , aVmax , b− ( KM , a+KX ) Vmax , a−KXVmax , a2KYD−Vmax , aVmax , bDVmax , b−Vmax , a > 0c=KM , a ( Vmax , bKM , a−Vmax , aKX−Vmax , aVmax , bD ) Vmax , b−Vmax , a > 0 As in the irreversible case , the sign of Vmax , b-Vmax , a determines the required condition on the numerators . We assume that Vmax , b > Vmax , a , noting that if Vmax , b < Vmax , a , a positive steady state cannot be globally stable because for X such that fa ( X ) > Vmax , b , the system will diverge regardless of the value of Y . Under the assumption that Vmax , b > Vmax , a , the denominator of both b and c is positive , meaning a positive steady state exists only if the nominators of b or c ( or both ) are negative . Thus , two options may arise . We now turn to analyze the stability of the steady state . For a steady state to be stable , the eigenvalues of the Jacobian matrix must have negative real values . In our system it holds thatX˙=fa-fbY˙=fb-fc We use the following notation:α=dfadX=Vmax , aKM , a ( KM , a+X ) 2βx=∂fb∂X=Vmax , b ( KX+Y ( 1+KXKY ) ) ( KX+X+KXYKY ) 2βy=∂fb∂Y=-Vmax , b ( KX+X ( 1+KXKY ) ) ( KX+X+KXYKY ) 2dfcdY=D We can use this notation to write the Jacobian matrix as:J= ( α-βx-βyβxβy-D ) which gives a characteristic polynomial of: ( α-βx-λ ) ( βy-D-λ ) +βyβx=0 In order for the real values of the roots of the characteristic polynomial to be negative it must hold that b>0 and c>0 , where b and c are now the coefficients of the quadratic equation aλ2+bλ+c=0 . We therefore get that:{b=βx−α−βy+D > 0c= ( α−βx ) ( βy−D ) +βyβx=βxD+αβy−αD > 0 We denote by f* the steady state flux in the system , such that f*=fa=fb=fc We note that for MM kinetics and positive concentrations it holds that:α > 0βx > 0−βy > βxβx+βy=−f∗1+KXKYKX+X+KXYKY First , we note that if α≥βx then the steady state cannot be stable as , looking at the value of c , we see that in such a case ( βx−α ) D < 0 and since αβy < 0 , c<0 violating the stability conditions . However , because we assume that Vmax , b>Vmax , a , then for Y=Y* , at X=0 , fb<fa , but for X→∞ , fb→Vmax , b and fa→Vmax , a , so that fb>fa . It then follows that , because the two fluxes can only intersect once for positive X and fixed Y , at the steady state point , where fa=fb , α<βx , so this condition is satisfied . We note that this condition is sufficient to ensure that b>0 . We also note that as α<βx , a large enough value of D exists at which the steady state is stable , concluding that if D is large enough , then a stable steady state exists if:{Vmax , b>Vmax , aVmax , bKX<Vmax , aKM , a If D is small , such that D<Vmax , a/KM , a , and Vmax , b>Vmax , a ( implying that α<βx ) , we need to check what other conditions are necessary in order to ensure that βxD+αβy−αD>0 . We look at the limit Vmax , b→∞ . At this limit , the quadratic equation for X* converges to:0= ( X∗ ) 2+ ( 2KM , a−Vmax , aD ) X∗+KM , a2−Vmax , aKM , aD For this equation , c<0 , implying that one of the roots is negative and one is positive . The positive root is:X*=Vmax , aD-KM , a As this X* is finite , we get that when Vmax , b→∞ , Y* also converges to Vmax , aD-KM , a . At this limit , βx increases infinitely and βy decreases infinitely , but βx+βy converges to:-f* ( 1+KXKY ) KX+ ( Vmax , aD-KM , a ) ( 1+KXKY ) that is constant . Therefore , rearranging c such that:c= ( βx+βy ) D−βy ( D−α ) −αD>0 we note that as Vmax , b increases , the dominant term becomes −βy ( D−α ) >0 ensuring that c>0 and therefore stability . On the other hand , when Vmax , b→Vmax , a , we note that because fc<Vmax , a , Y* is bounded by Y∗<Vmax , aD , but X*→∞ . Thus , both α and βx diminish like 1X*2 , and βy diminishes like 1X . The dominant term in c=βxD+βyα-αD therefore becomes ( βx−α ) D>0 so again stability is maintained . Therefore , for small values of D , as long as Vmax , b>Vmax , a , a positive stable steady state exists both in the lower limit of Vmax , b→Vmax , a , and in the upper limit of Vmax , b→∞ . Our conclusions are therefore as follows: As in the irreversible case , Vmax , b>Vmax , a is a necessary condition for the existence of a globally stable steady state . For large values of D , the reversible reaction is far from equilibrium , resulting in an additional condition , equivalent to the condition we obtained for the irreversible case , namely that Vmax , b/KX is upper bounded by a term that is larger than Vmax , a/KM , a , but approaches it as D increases . This condition is sufficient for the existence and stability of the steady state . For small values of D , a steady state always exists ( given that Vmax , b>Vmax , a ) . We can show that this steady state is stable both when Vmax , b→∞ , and when Vmax , b→Vmax , a . We therefore conclude that in this case , no further restrictions apply on KX , KY , or KM , a but rather that a steady state can always be achieved at most by changing Vmax , b . Qualitatively , the cases we analyze show that , on top of the required Vmax , b>Vmax , a condition , the second condition is that either the slope of fc=D is smaller than Vmax , a/KM , a , or that the maximal slope of fb , Vmax , b/KX , is smaller than Vmax , a/KM , a . We analyze the stability criteria for the autocatalytic cycles depicted in Figure 5A and B . We start by writing the relevant equations for the autocatalytic cycle depicted in Figure 5A . In this system , there are two intermediate metabolites , X1 and X2 , two reactions that form the cycle , fa1 and fa2 , and two branch reactions , fb1 and fb2 . We assume , without loss of generality , that the autocatalytic reaction ( the reaction that increases the amount of carbon in the cycle ) is fa2 and that the autocatalysis is in a 1:2 ratio . The equations describing the dynamics of the system are thus:X1˙=2fa2−fa1−fb1X2˙=fa1−fa2−fb2 We note that in steady state , where X . 1=X . 2=0 , because the autocatalysis is in a 1:2 ratio , it must hold that fb1+fb2=fa2 , meaning the total outgoing flux balances the total increase of intermediate metabolites due to autocatalysis . Given that a steady state of the system exists for some value ( X1* , X2* ) , we can evaluate the condition for stability . In multi-variable systems , stability dictates that the real part of the eigenvalues of the Jacobian matrix must all be negative . We define αi=∂fai∂Xi and βi=∂fbi∂Xi for i=1 , 2 . We note that as we assume Michaelis Menten kinetics , αi>0 and βi≥0 , where βi=0 is the case where there is no flux branching out at i . We then get that the Jacobian matrix is:J= ( - ( α1+β1 ) 2α2α1- ( α2+β2 ) ) Solving for the characteristic polynomial gives:0= ( λ+α1+β1 ) ( λ+α2+β2 ) −2α1α2=λ2+ ( α1+β1+α2+β2 ) λ+ ( α1+β1 ) ( α2+β2 ) −2α1α2 that has two negative roots when: ( α1+β1 ) ( α2+β2 ) −2α1α2>0⇒ ( 1+β1α1 ) ( 1+β2α2 ) >2 which is satisfied if β1>α1 or β2>α2 . Therefore , if either β1>α1 or β2>α2 at the steady state , then the steady state is stable . The two-metabolites cycle case can be easily extended to a larger number of intermediate metabolites and reactions , as is depicted in Figure 5B . For this extension , we again assume , without loss of generality , that the autocatalytic reaction is the last reaction , fan , and that the autocatalysis is in a 1:2 ratio . In this case , steady state implies that the concentration of each intermediate metabolite is conserved , meaning that for all i>1: ( 28 ) Xi˙=0⇒fai−1−fai−fbi=0⇒fai−1≥fai ( for i=1 , as fan is the autocatalytic reaction , we get that 2⋅fan≥fa1 ) . Also , because at steady state the total outgoing flux from the cycle must balance the total incoming flux into the system , which is the amount of autocatalysis carried out by fan , we get that:∑i=1nfbi=fan ( due to our assumption of a 1:2 autocatalytic ratio ) implying that for all i: ( 29 ) fbi≤fan We stress that Equation 29 is only valid if the autocatalysis is in up to a 1:2 ratio . Deriving a stability criterion for the multiple-reaction case , we get that in this case a steady state is stable if there exists i such that βi>αi ( see section 9 below ) . To conclude , for the straightforward extension of the simple model to multiple reactions with a single autocatalytic reaction , steady state implies that for all i: ( 30 ) fbi≤fan≤fai Where the left inequality is due to Equation 29 and the right inequality is due to Equation 28 . A sufficient condition for such a steady state point to be stable is that at the steady state point there exists at least one branching point i at which the derivative of the branch reaction is larger than the derivative of the equivalent autocatalytic reaction: ( 31 ) βi>αi Stability analysis of a model complex autocatalytic cycle with n reactions in the cycle results in the following Jacobian matrix: ( 32 ) J= ( - ( α1+β1 ) 0⋯02αnα1- ( α2+β2 ) ⋯00⋮⋮⋱⋮⋮00⋯- ( αn-1+βn-1 ) 000⋯αn-1- ( αn+βn ) ) The characteristic polynomial of this matrix is given by: ( 33 ) 0=∏i=1n ( λ+αi+βi ) -2∏i=1nαi To extract the conditions under which all the roots of the characteristic polynomial have negative real parts we use Rouche’s theorem . Our strategy will be as follows: We will define a contour that contains only numbers with negative real parts . We will show that all the roots of the polynomial 0=∏i=1n ( λ+αi+βi ) lie within the area this contour encloses . We will find the conditions for which |∏i=1n ( λ+αi+βi ) |>2∏i=1nαi on the contour , satisfying the premise of Rouche’s theorem . We will then claim that under these conditions all the roots of the polynomial in Equation 33 must also lie inside the contour , and therefore must have negative real parts . Given that all the roots of this polynomial have negative real parts , we will conclude that the eigenvalues of the Jacobian matrix at the steady state point all have negative real parts , making the steady state point stable . Using the Jacobian matrix from Equation 32 we can analyze the effect of multiple low saturation branch points on convergence to steady state . The analysis shows that the more i’s exist for which βi>0 , and the larger βi is ( resulting in lower saturation of fbi ) , the faster the convergence of the cycle to steady state will be . We denote by X*→ the steady state vector of the concentrations of the intermediate metabolites . We denote by X→=X*→+ΔXj a state where for all the intermediate metabolites that are not Xj , their concentration is the same as the steady state concentration , and Xj differs by a small amount , ΔXj , from its steady state concentration . We let F denote the fluxes function of the system such that F ( X ) =X˙|X→ . Evaluating the dynamics of the system at X→ by noting that F ( X→ ) ≈F ( X*→ ) +J⋅ΔXj=J⋅ΔXj ( where F ( X*→ ) =0 as X*→ is a steady state ) results in F ( X→ ) k=0 for all k≠j , j+1 . For Xj such that j≠n we get F ( X→ ) j≈- ( αj+βj ) ΔXj and for Xj+1 we get F ( X→ ) j+1≈αjΔXj . Therefore , the difference from the steady state decreases proportionally to βj ( and cycles to the next intermediate metabolite , Xj+1 ) . For j=n , we get that F ( X→ ) j≈- ( αj+βj ) ΔXj , as for j≠n , but F ( X→ ) 1≈2αjΔXj where the factor of 2 is due to the effect of the assimilating reaction , that causes an amplification of the deviation from steady state ( an amplification that is dampened by subsequent reactions along the cycle if the conditions for stable steady state are satisfied ) . It therefore follows that any increase in βj , for any j , increases the speed of convergence to steady state and reduces the propagation of deviations from steady state for Xj . Because of the linearity of matrix multiplication , an arbitrary deviation from X*→ can always be decomposed to individual deviations with respect to every intermediate metabolite , making the analysis above valid for such deviations as well . Thus , to keep deviations from steady state at check , it is beneficial to increase βj , for all j , which implies decreasing the saturation of fbj . It turns out that for the Michaelis-Menten kinetics equations , the following useful lemma can be used to connect theoretical observations on the relationships of derivatives to physiological observations on affinities and saturation levels . We define the saturation level of a reaction as the ratio between the flux it carries , and the maximal flux it can carry , given the expression level of the relevant enzyme , that is:S ( X ) =f ( X ) Vmax=XKM+X Given this definition we can show that if two Michaelis-Menten reactions consume the same metabolite , X , and at a given concentration , X* , it holds that fa ( X* ) ≥fb ( X* ) , then if: ( 34 ) dfbdX|X=X∗>dfadX|X=X∗ then it follows that: ( 35 ) {KM , b>KM , aSb ( X∗ ) <Sa ( X∗ ) Proof: expanding the condition that fa ( X* ) ≥fb ( X* ) , we get that: ( 36 ) Vmax , bX*KM , b+X*≤Vmax , aX*KM , a+X*⇒Vmax , bKM , b+X*≤Vmax , aKM , a+X* Expanding the premise of the lemma in Equation 34 gives us that:dfbdX|X=X∗>dfadX|X=X∗⇒Vmax , bKM , b ( KM , b+X∗ ) 2>Vmax , aKM , a ( KM , a+X∗ ) 2 Because Equation 36 holds , it follows that:KM , bKM , b+X∗>KM , aKM , a+X∗⇒11+X∗KM , b>11+X∗KM , a⇒KM , b>KM , a setting the affinity of the autocatalytic enzyme as a lower bound for the affinity of the branch enzyme . Finally , given this relation of affinities it follows that:KM , b>KM , a⇒X∗+KM , b>X∗+KM , a⇒X∗X∗+KM , b<X∗X∗+KM , a⇒Sb ( X∗ ) <Sa ( X∗ ) concluding the proof . We note that a multiple reaction autocatalytic cycle at a stable steady state point satisfies Equations 30 and 31 , so the lemma applies . To evaluate the maximal flux capacity of a reaction under a prescribed growth condition , given expression level and flux data for a set of conditions , we follow the procedure described in Davidi et al . ( 2016 ) . For each reaction , under every condition , we divide the flux the reaction carries ( obtained from Gerosa et al . , 2015 ) by the amount of the corresponding enzyme expressed under that condition ( obtained from Schmidt et al . , 2016 ) . We thus get a flux per enzyme estimate for the given reaction under each of the conditions . We define the enzyme maximal in-vivo catalytic rate as the maximum flux per unit enzyme it carries across all conditions analyzed ( noting that this is actually only a lower bound on this rate ) . Multiplying the enzyme maximal catalytic rate by the enzyme amount at each condition results in an estimate of the maximal possible flux through the given reaction under the relevant condition . In this section we touch upon the potential ( in somewhat simplified and naively non-rigorous terms ) of allosteric regulation to improve the properties of autocatalytic cycles . The constraint on the affinity of the branch reaction imposed by the stability requirement ( Equation 35 ) may be suboptimal under other flux modes . Furthermore , allosteric regulation can be used to accelerate the rate at which an autocatalytic cycle converges to its stable steady state mode . While many allosteric regulation schemes exist ( Leskovac , 2003 ) , all of these schemes affect the affinity of the regulated enzyme , and some of these schemes also affect the maximal rate . We qualitatively analyze the expected regulation benefits for autocatalytic cycles . From the perspective of the simple model , we recall that X˙=fa-fb . If the cycle is such that some steady state concentration , X* , is the desired value for biological function , then for levels of X below X* convergence will be faster if fa is increased and fb is decreased , compared with their values at X* . Conversely , for levels of X above X* , convergence will be faster if fa is decreased and fb is increased , compared with their values at X* . Convergence to X* can therefore be accelerated if , for example , X activates the branch reactions and inhibits the cycle reactions . The assimilated metabolite can also allosterically regulate the reactions of the cycle . We assume that the desired steady state , denoted X^ , does not depend on the concentration of the assimilated metabolite , A . Under this assumption , we further assume that X^ is attained for some constant concentration of the assimilated metabolite , A^ . It then follows that because the autocatalytic activity is higher when A>A^ , then in order to maintain X* close to its desired level , when A>A^ , fa should be inhibited , and fb should be activated , but when A<A^ , fa should be activated , and fb should be inhibited . Therefore , to increase the robustness of the steady state concentration to changes in the concentration of the assimilated metabolite , the assimilated metabolite should inhibit the cycle reactions and activate the branch reactions . Another possible class of regulators are the products of the branch reactions . Taking a somewhat simplistic view , if the level of Y , the product of a branch reaction is low , this can indicate that the cycle does not carry sufficient flux to supply the demand for Y . Regulation can then be used to increase X* . From Equation 7 , we get that the steady state concentration , X* , increases as KM , b increases and Vmax , b decreases , corresponding to inhibition of fb , and that X* decreases as KM , a increases and Vmax , a decreases , corresponding to inhibition of fa . So , to tune autocatalytic fluxes to match the demands of Y , regulation should increase X* when Y is low , by activating the recycling and autocatalytic reactions and inhibiting the branch reactions . On the other hand , regulation should decrease X* when Y is high , by inhibiting the autocatalytic reactions and activating the branch reactions . Therefore , to synchronize the demand of the cycle product with the cycle flux , the cycle branch products should inhibit the cycle reactions and activate the branch reactions . Finally , we note that in the autocatalytic cycles we identify in central carbon metabolism , there are also reactions that operate in the reverse direction to the branch reactions , such that they consume products of the cycle and produce intermediate metabolites of the cycle . As such reactions are mirror images of branch reactions , we expect them to be oppositely regulated to branch reactions . We find that these predictions hold for the cycle using the PTS , that is known to be allosterically controlled , but not for the glyoxylate cycle , which is known to be transcriptionally controlled ( Gerosa et al . , 2015 ) . | Many bacteria are able to produce all the molecules they need to survive from a limited supply of nutrients . This allows the bacteria to thrive even in harsh environments where other organisms struggle to live . The bacteria act as miniature chemical factories to convert nutrients into the desired molecules via a series of chemical reactions . Some molecules are made in sets of reactions termed autocatalytic cycles . These reaction sets require a molecule to be present in the cell in order to produce more of that molecule; like how a savings account needs to contain some money before it can generate more via interest . Bacteria have many different enzymes that each drive specific chemical reactions . In order for an autocatalytic cycle to work properly , the cell needs to maintain adequate supplies of the molecule it is trying to make and all of the “intermediate” molecules in the cycle . If less of an intermediate molecule is produced , for example , the cell needs to reduce the demand for that molecule by controlling later chemical reactions in the cycle . Bacteria control chemical reactions by regulating the activities of the enzymes involved , but it is not clear exactly how they regulate the enzymes that drive autocatalytic cycles . Barenholz et al . combined two approaches called proteomics and fluxomics to study autocatalytic cycles in a bacterium known as E . coli . The experiments suggest several core principles allow autocatalytic cycles to work smoothly in the bacteria . The next step is to apply these principles to different kinds of molecules produced in bacterial cells . A future challenge is to search for other structures that regulate chemical reactions in E . coli and other bacteria . Extending our understanding of autocatalytic cycles and other pathways of chemical reactions is essential for designing and engineering new reactions in bacteria . Such knowledge can be used to modify bacteria to produce valuable chemicals in environmentally friendly ways . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"computational",
"and",
"systems",
"biology"
] | 2017 | Design principles of autocatalytic cycles constrain enzyme kinetics and force low substrate saturation at flux branch points |
Animal behaviors are commonly organized into long-lasting states that coordinately impact the generation of diverse motor outputs such as feeding , locomotion , and grooming . However , the neural mechanisms that coordinate these distinct motor programs remain poorly understood . Here , we examine how the distinct motor programs of the nematode C . elegans are coupled together across behavioral states . We describe a new imaging platform that permits automated , simultaneous quantification of each of the main C . elegans motor programs over hours or days . Analysis of these whole-organism behavioral profiles shows that the motor programs coordinately change as animals switch behavioral states . Utilizing genetics , optogenetics , and calcium imaging , we identify a new role for dopamine in coupling locomotion and egg-laying together across states . These results provide new insights into how the diverse motor programs throughout an organism are coordinated and suggest that neuromodulators like dopamine can couple motor circuits together in a state-dependent manner .
As animals explore their environments , their nervous systems transition between a wide range of internal states that influence how sensory information is processed and how behaviors are generated ( Anderson and Adolphs , 2014; Artiushin and Sehgal , 2017; Liu and Dan , 2019 ) . These internal states of arousal , motivation , and mood typically alter many ongoing behaviors , impacting motor circuits that control diverse behavioral outputs such as feeding , grooming , and locomotion . A full understanding of how internal states are generated should explain how they coordinate the production of many complex motor outputs . Neuromodulators play a central role in generating internal states . For example , neuropeptides like orexin and pigment dispersing factor ( PDF ) promote brain-wide states of wakefulness in mammals and flies , respectively ( Saper et al . , 2010; Taghert and Nitabach , 2012 ) . In awake animals , norepinephrine controls arousal and attention ( Aston-Jones and Cohen , 2005; Carter et al . , 2010 ) , the neuropeptides Tac2 and CRH induce states of heightened anxiety ( Füzesi et al . , 2016; Kormos and Gaszner , 2013; Zelikowsky et al . , 2018 ) , and the neuropeptides AgRP and NPY promote behaviors associated with hunger ( Chen et al . , 2019; Krashes et al . , 2013 ) . The anatomical organization of neuromodulatory systems makes them well-suited to impact a wide range of motor outputs: the projections of neuromodulator-producing neurons are typically very diffuse , targeting many brain regions . However , our mechanistic understanding of how neuromodulatory regulation of CNS circuits is propagated to diverse motor outputs remains limited . In the simple nematode C . elegans , it should be feasible to determine how internal states influence every behavioral output of the animal . The C . elegans nervous system consists of 302 neurons , which act on 143 muscle cells ( White et al . , 1986 ) . C . elegans generates a well-defined repertoire of motor programs , each with a devoted motor circuit and muscle group: locomotion , egg-laying , feeding , defecation , and postural changes of the head and body ( de Bono and Maricq , 2005; Collins et al . , 2016; Pirri et al . , 2009; Schafer , 2005; Stephens et al . , 2008 ) . Quantitative studies of C . elegans locomotion have shown that animals switch between long-lasting behavioral states . In an extreme case , C . elegans animals cease all behaviors as they enter sleep-like states during development and after periods of stress ( Nichols et al . , 2017; Raizen et al . , 2008; Van Buskirk and Sternberg , 2007 ) . Awake animals in a food-rich environment switch between roaming and dwelling states , where they either rapidly explore the environment or restrict their movement to a small area ( Ben Arous et al . , 2009; Flavell et al . , 2013; Fujiwara et al . , 2002; Stern et al . , 2017 ) . After removal from food , C . elegans generates an area-restricted search state before switching to a dispersal state ( Gray et al . , 2005; Hills et al . , 2004; López-Cruz et al . , 2019; Wakabayashi et al . , 2004 ) . Like the internal states of mammals , these behavioral states in C . elegans last from minutes to hours and the transitions between states are abrupt . It remains unclear how the full repertoire of C . elegans behaviors is coordinated as awake C . elegans animals switch between behavioral states . Neuromodulation in C . elegans plays a pivotal role in behavioral state control ( Chase and Koelle , 2007; Li and Kim , 2008 ) . Serotonin initiates and maintains dwelling states , while the neuropeptide PDF initiates and maintains roaming states ( Choi et al . , 2013; Flavell et al . , 2013; Horvitz et al . , 1982; Rhoades et al . , 2019; Sawin et al . , 2000; Stern et al . , 2017 ) . In addition , dopamine , tyramine , and octopamine influence behaviors associated with the presence or absence of food ( Alkema et al . , 2005; Chase et al . , 2004; Horvitz et al . , 1982; Sawin et al . , 2000; Stern et al . , 2017 ) . The C . elegans sleep-like state is also strongly influenced by neuropeptides , including NLP-8 , FLP-13 , and FLP-24 , and ligands for the NPR-1 receptor ( Choi et al . , 2013; Iannacone et al . , 2017; Nath et al . , 2016; Nelson et al . , 2014 ) . Command-like neurons that release neuromodulators are capable of driving state changes: the serotonergic NSM neurons induce dwelling states ( Flavell et al . , 2013; Rhoades et al . , 2019 ) and the peptidergic ALA neuron evokes a sleep-like state ( Hill et al . , 2014; Nath et al . , 2016; Nelson et al . , 2014 ) . However , there is not a one-to-one mapping between neuromodulators and specific states . For example , although serotonin release from NSM drives dwelling , serotonin release from ADF neurons has no apparent effect on dwelling and instead impacts other behaviors ( Flavell et al . , 2013 ) . Likewise , ALA releases at least three neuropeptides that have unique yet overlapping functions to inhibit downstream behaviors during sleep ( Nath et al . , 2016 ) . However , each of these neuropeptides is also produced by other neurons that do not induce sleep . Thus , individual neuromodulators may exert different effects during different states and their combinatorial actions might give rise to the widespread behavioral changes that accompany each state . In this study , we examine how neuromodulation coordinates diverse behavioral outputs to give rise to state-dependent changes in behavior . First , we describe a new imaging platform that permits simultaneous , automated quantification of each of the main C . elegans motor programs . Then we analyze hours-long behavioral recordings to fully characterize the behavioral states of this animal . Finally , we identify a dopaminergic pathway that couples multiple motor programs together as animals switch behavioral states , revealing that dopamine promotes egg-laying in a locomotion state-dependent manner . Our study provides new insights into how the diverse motor programs throughout an organism are coordinated and suggests that neuromodulators like dopamine can couple motor circuits together in a state-dependent manner .
To examine how the distinct C . elegans motor programs are coordinated , we designed and constructed tracking microscopes and an accompanying software suite that permits simultaneous , automated measurement of numerous behaviors ( Figure 1A; Figure 1—figure supplement 1 ) . The microscope , which shares several design features with previously described microscopy platforms ( Yemini et al . , 2013; see also Faumont and Lockery , 2006; Nguyen et al . , 2016; Venkatachalam et al . , 2016 ) , collects brightfield images of individual animals ( Figure 1B ) at a frequency of 20 Hz and resolution of 1 . 4 um/pixel , which is sufficient to capture the most rapid and small-scale movements of C . elegans , such as pharyngeal motion ( Avery and You , 2012; Lockery et al . , 2012 ) . A closed-loop tracking system reliably keeps animals in view over hours or days , while live data compression makes storage of approximately 1 . 7 million images per animal per day feasible . Although the parameters for data collection that we use here are tailored to C . elegans recordings , this low-cost , open-source microscopy platform should be useful for recordings of many small animals ( see Materials and methods for links to parts list and build tutorial ) . We developed a software suite that allows us to automatically extract measurements of each of the main C . elegans motor programs from these video recordings . Because C . elegans is transparent , each animal movement – including movements inside the body , like those of the pharynx – is visible during brightfield imaging . Using this software suite , we are able to extract: body/head posture , locomotion , egg-laying , defecation , and pharyngeal pumping ( i . e . feeding ) . Extracting body posture and locomotion was straightforward , except during omega bends ( see Materials and methods ) ( Stephens et al . , 2008; Figure 1B; Figure 1—figure supplement 2A–D ) , However , the detection of egg-laying , defecation , and pumping required us to implement tailored machine vision algorithms ( Figure 1—figure supplement 2E–O; see Materials and methods ) . To ensure the reliability of our measurements , we compared the automated scoring of egg-laying , defecation , and pumping to manually scored data and found a high level of concurrence ( Figure 1—figure supplement 2G , J , O ) . These advances now allow us to extract near-comprehensive records of each animal’s behavior from hours-long recordings . The resulting datasets capture behavioral outputs over multiple timescales: from milliseconds-scale postural changes to hours-long behavioral states ( Figure 1C–E shows an example dataset at multiple time resolutions ) . To examine how the distinct motor programs of C . elegans are coordinated over time , we first analyzed the relationships between each behavioral variable . We examined data from 30 adult well-fed wild-type animals , each recorded for six hours on a homogenous E . coli food source . To determine how each behavioral output varies as a function of locomotion , we examined average behavioral outputs across different animal velocities ( Figure 1F ) . Egg-laying events were predominant during rapid forward locomotion , while defecation events were most commonly observed during slow or reverse movement . We also examined typical body postures as a function of velocity in two ways . First , we identified the most commonly observed body postures at different velocities ( top of Figure 1F shows most common postures out of 100 reference postures; see below ) . Second , we plotted 2D histograms of the eigenmodes of the angles along the body when animals traveled at different velocities ( Figure 1F , middle ) ( Stephens et al . , 2008 ) . Differences in the amplitudes of the eigenmodes correspond to differences in body postures . From these histograms and from the set of most common body postures , it is evident that the postures that animals express co-vary considerably with velocity , suggesting that animals sample different body postures at different velocities ( Figure 1F ) . We also examined average behaviors that surround egg-laying and defecation events by plotting event-triggered averages ( Figure 1—figure supplement 3 ) . These analyses revealed many of the same relationships between the behavioral outputs , but also showed that egg-laying and defecation events each occur during stereotyped posture and locomotion changes , as has been previously reported ( Collins et al . , 2016; Hardaker et al . , 2001; Nagy et al . , 2015 ) . We conclude from these analyses that there is extensive coordination between the distinct C . elegans motor programs . To understand how these behaviors are coupled together over longer time scales , we sought to identify the long-lasting behavioral states that wild-type animals generate , so that we could determine which motor outputs are observed in each state . Unsupervised learning can reveal the underlying states that generate observed behavioral variables ( Berman et al . , 2014; Marques et al . , 2018; Wiltschko et al . , 2015 ) . We and others have previously applied hidden Markov models ( HMMs ) to C . elegans locomotion parameters , which identifies roaming and dwelling states , as well as quiescence/sleep under certain conditions ( Flavell et al . , 2013; Gallagher et al . , 2013 ) . However , there may be a broader set of behavioral states that differ along other behavioral axes . To identify such states , we performed unsupervised discovery on the animal’s body posture ( Figure 2A ) , which is the most complex behavioral variable and is coordinated with the other motor programs . To describe postural changes in a compact manner , we used a previously described approach where we learned a compendium of reference postures that encompasses the broad range of body postures that animals display ( Figure 2A–B; Schwarz et al . , 2015 ) . To describe posture at any time point , we match the actual posture of an animal to its most similar match in the compendium . We then constructed a transition matrix that describes the probability of switching from each posture to the others ( Figure 2C ) . Clustering the rows of this matrix revealed a striking , symmetric block-like structure , indicating that there are groups of postures , which we term ‘posture groups , ’ that animals transition back and forth between ( Figure 2C–D; 11 . 9-fold higher transition rate to postures within the same group ) . The postures within individual groups included those emitted during both forward and reverse movement ( Figure 2—figure supplement 1 ) , indicating that the time spent in a given posture group can span multiple forward-backward transitions . This organization suggests some degree of long-term stereotypy in how animals emit their postures while exploring a bacterial food lawn . To more precisely characterize this long-term stereotypy , we fit an HMM where the emissions are essentially these posture groups ( see Materials and methods; Figure 2—figure supplement 2A–C ) . Based on Bayesian information criteria ( BIC ) estimates , a model with nine hidden states provided the best fit to the recorded data ( Figure 2—figure supplement 2D ) . Importantly , we reliably converged to the same model parameters , even when training was performed on different sets of animals and from different random starting conditions ( Figure 2—figure supplement 2E–F ) . These results suggest that the posture-HMM can provide a reliable description of postural changes over time . To understand the behavioral states that were captured by the HMM ( Figure 2—figure supplement 3A ) , we first examined locomotion parameters in each of the states . One of these behavioral states consisted of high forward velocity , low angular speed and typically lasted from tens of seconds to many minutes ( Figure 2E ) . Based on these parameters , this state is equivalent to the previously defined roaming state ( Figure 2E; labeled ‘Roam’ ) ( Ben Arous et al . , 2009; Flavell et al . , 2013; Fujiwara et al . , 2002 ) . Animals traveled at lower velocities during the other eight states , but , in contrast to quiescence/sleep states , they maintained high pumping and defecation rates in each ( Figure 2E ) . To maintain consistency with previous literature , we describe these as dwelling sub-modes ( Dwell1-8; see Video 1 for examples ) . Quiescence states were not recovered from this analysis because this state was almost never observed under our recording conditions ( only five ~20 s bouts identified in ~180 hr of data; Figure 2—figure supplement 3B ) . Because these dwelling sub-modes had not been previously described , we characterized them further . Whereas five of these states ( Dwell2 , 3 , 4 , 5 , 8 ) reflected almost completely paused movement , the other three states consisted of characteristic movements: animals displayed steady , slow forward locomotion in Dwell1 ( Figure 2E; see movement direction and average motion ) , a high incidence of active reversing in Dwell7 ( Figure 2E; see average motion and long reversal rates ) , and a wider range of head and neck movements in Dwell6 ( Figure 2—figure supplement 3A; see standard deviation of angles along body ) ; Moreover , animals displayed stereotyped body postures in each of the dwelling sub-modes that were significantly different from one another ( Figure 2—figure supplement 3A; Video 1 ) . For example , animals in Dwell4 displayed a reliably flat body posture with only very shallow bends ( Figure 2—figure supplement 3A; see most common postures ) . The average duration of each sub-mode was ~10 s ( Figure 2—figure supplement 4A–B ) and animals transitioned between them in a non-random fashion: the transition rates between many sub-modes were close to zero , whereas other transition rates were quite high ( Figure 2—figure supplement 4C–D ) . This suggests that animals transition between these distinct sub-modes of dwelling in an organized fashion . We considered whether segmenting dwelling into these discrete sub-modes provides a better description of postural changes during dwelling than simpler models where animals continuously switch between postures without any further temporal structure . To investigate this , we asked whether posture sequences generated by the posture-HMM provided a better match to posture sequences from real animals , compared to simpler alternatives . We compared the posture-HMM to two alternative models: ( 1 ) a model involving continuous transitions between postures , where the probability of each posture transition was determined by our measurements from real animals ( Figure 2C ) ; and ( 2 ) a model involving continuous transitions between postures , where the probability of each posture transition was determined by the similarity between postures . We generated synthetic posture sequences from each model and quantified how well they matched the sequences from real animals by measuring the average duration of time between each of the 100 compendium postures in the synthetic and real animal data ( Figure 2—figure supplement 5A ) . This metric captures the higher order patterns in which animals emit their postures , but does not assume any particular underlying structure . Based on this criterion , the posture sequences generated by the posture-HMM were a significantly better match to the sequences from actual animals , compared to the alternative models ( Figure 2—figure supplement 5B ) . These results , together with the BIC analysis described above ( Figure 2—figure supplement 2D ) , suggest that the posture-HMM provides a robust description of how animals transition between distinct postures during dwelling . Having examined how posture and locomotion differ across these distinct states , we next quantified the occurrence of other motor programs in roaming states and each dwelling sub-mode ( Figure 2E ) . Egg-laying rates were highest during roaming , whereas the dwelling sub-modes had smaller , but significant , differences in their egg-laying rates . This is consistent with previous observations that egg-laying is often accompanied by increased movement ( Hardaker et al . , 2001; McCloskey et al . , 2017 ) . Previous work showed that animals display an acute reduction in speed immediately after egg-laying ( Hardaker et al . , 2001 ) . We found that this rapid speed change is present after egg-laying events during roaming and dwelling , but is less pronounced when eggs are laid during roaming ( Figure 2—figure supplement 6 ) . Defecation rates were also significantly different between the nine states , but feeding rates were the same across the states ( Figure 2E ) . Taken together , these data suggest that animals transition between a high-velocity/high-egg-laying roaming state and different sub-modes of dwelling in which they display distinct body postures and motor programs . Animals also displayed stereotyped changes in multiple motor programs at the moments when they switched between behavioral states ( Figure 2—figure supplement 7 ) . We characterized these effects by examining event-triggered averages timed to each type of behavioral state transition . We found that the state transitions were accompanied by transient and reliable changes in velocity , pumping , and defecation rates ( Figure 2—figure supplement 7 ) . For example , transitions from Dwell2 to Dwell4 were preceded by a high incidence of defecation motor program ( DMP ) events ( immediately before the transition ) , followed by a transient increase in velocity and a reduction in feeding . These observations further confirm that the posture-HMM identified state transitions that correspond to reliable changepoints in behavior and suggest that animals display reliable , multimodal behavioral trajectories as they switch between these stereotyped postural states . We next sought to identify and characterize the neural circuits that allow animals to coordinate their motor programs across behavioral states . Here , we focused on the coupling between the roaming state and egg-laying , since this was the most robust form of motor program coupling that we detected and it could be easily measured using our new microscopy platform . We considered whether the increased frequency of egg-laying during roaming could be fully explained by animals laying eggs more frequently when they move faster . Thus , we examined egg-laying frequency as a function of velocity within roaming and within dwelling . While there was a positive correlation between velocity and egg-laying rate within each state , we observed a universally higher egg-laying rate during roaming compared to dwelling , even when matched for velocity ( Figure 2F ) . These results suggest that animals lay more eggs while traveling at high speeds , particularly during high-speed roaming states . To test whether neuromodulation is critical for the coupling between locomotion state and egg-laying , we examined mutants lacking biogenic amine and neuropeptide neuromodulators . For each mutant , we characterized the time spent in each behavioral state and the motor outputs within each state ( Figure 3A ) . States were defined consistently across genotypes using parameters learned from wild-type animals . Serotonin-deficient tph-1 mutants displayed reduced dwelling , higher speeds during roaming , reduced feeding , and fewer egg-laying events , as has been previously described ( Figure 3A–B; Avery and You , 2012; Flavell et al . , 2013; Hobson et al . , 2006; Horvitz et al . , 1982 ) . However , the increase in egg-laying rates during roaming states was largely intact , suggesting that serotonin is dispensable for this form of motor program coupling . pdfr-1 mutants , deficient in PDF neuropeptide signaling , displayed a broad defect in roaming states: they spent less time roaming , traveled at lower velocity while roaming , and did not bias their egg-laying to the state as robustly as wild-type animals . Thus , although pdfr-1 mutants are defective in locomotion and egg-laying during roaming , this may be part of a general deficit in their ability to display roaming states ( Flavell et al . , 2013 ) , rather than a specific deficit in motor program coupling . tdc-1 mutants that are defective in tyramine and octopamine displayed increased time in the roaming state . However , their egg-laying rates were still higher while roaming . tbh-1 mutants defective in octopamine synthesis showed a mild increase in their roaming velocity , but no deficit in egg-laying . Overall , these mutants point to important roles for neuromodulation in regulating behavioral states in C . elegans , but do not provide insights into the coupling of egg-laying with the roaming state . In contrast , dopamine-deficient cat-2 mutants displayed a striking loss of coupling between egg-laying and the roaming state . Although these animals displayed robust roaming states , their egg-laying rate was dramatically reduced during roaming ( Figure 3A ) . In contrast , their egg-laying rate during the dwelling state was unaltered . cat-2 animals travel at a higher velocity than wild-type animals while roaming ( Figure 3A ) , but still display the same postures ( Figure 3—figure supplement 1A ) , indicating that they are not broadly defective in the locomotor components of roaming ( Sawin et al . , 2000 ) . Because they have unaltered egg-laying rates during dwelling , which is the more prevalent state , there is only a modest reduction in egg-laying overall in these mutant animals ( Figure 3C ) . This suggests that they are not broadly defective in egg-laying , a finding that is consistent with the fact that they were not recovered as egg-laying defective ( egl ) in forward genetic screens . To ensure that this motor coupling phenotype was caused by the loss of the cat-2 gene , we examined a second , independent null allele of cat-2 and found that it caused a similar phenotype ( Figure 3C–D; Figure 3—figure supplement 1B ) . In addition , we rescued cat-2 expression in the mutant via a transgene and found that this restored higher egg-laying rates during roaming ( Figure 3—figure supplement 1C ) . These data suggest that dopamine is necessary for proper coupling of egg-laying to the roaming state and highlights the value of simultaneously quantifying multiple ongoing motor programs within the animal . We considered whether the reduced egg-laying rates of cat-2 mutants during roaming , but not dwelling , might be due to a floor effect , where the egg-laying rates during dwelling cannot be further reduced . To test for potential floor effects , we recorded mutant animals with deficits in the egg-laying motor circuit ( egl-1 ) . These animals showed a dramatic reduction in egg-laying during both roaming and dwelling states ( Figure 3—figure supplement 1D ) , suggesting that floor effects do not account for the cat-2 phenotype . To further characterize the dopaminergic pathway that couples egg-laying and locomotion during roaming , we examined mutants lacking each of the seven known dopamine receptors in C . elegans ( Figure 3E; Figure 3—figure supplement 1E; Figure 3—figure supplement 2 ) . None of these single mutants showed any egg-laying phenotypes . However , we found that mutant animals lacking the two D2-like dopamine receptors , dop-2 and dop-3 , displayed reduced egg-laying rates during roaming , but not dwelling , closely matching the cat-2 mutant phenotype . These results suggest that dop-2 and dop-3 act together to regulate egg-laying during the roaming state . In analyzing these mutant datasets , we considered whether the egg-laying phenotypes of the dopamine mutants could be due to an indirect effect , where increased forward velocity during roaming in dopamine pathway mutants reduces egg-laying . However , an analysis of velocity and egg-laying rates across the receptor mutants suggests that this is not the case: dop-2;dop-3 double mutants display normal velocities during roaming ( Figure 3—figure supplement 1E ) , but have dramatically reduced egg-laying rates during roaming ( Figure 3E ) . This interpretation that the egg-laying effects are not due to locomotion changes is further corroborated by our optogenetics findings below . Altogether , these mutant analyses suggest that dopamine acts through the D2-like receptors DOP-2 and DOP-3 to promote egg-laying during the roaming state . Animals might coordinate their locomotion and egg-laying in order to lay their eggs in locations that are advantageous for the survival of their progeny . For example , it might be adaptive for animals to broadly disperse their eggs across beneficial environments . We considered how the increased frequency of egg-laying during the roaming state might impact the spatial distribution in which eggs are laid as animals explore a food source . Specifically , we asked whether eggs laid during the roaming state are more dispersed throughout the environment than eggs laid during dwelling , and whether deficits in dopamine signaling reduce the dispersal of eggs . First , to examine the proximity of eggs to one another , we measured the average distance between each egg and its k nearest neighbors ( varying the number of k from 1 to 10 ) . As expected , we found that eggs laid during roaming were farther apart from other eggs compared to those laid during dwelling ( Figure 4A ) . To gain a more complete view of how animals distributed their eggs , we next measured the extent to which animals dispersed their eggs across the entire area that they explored . This was quantified by virtually superimposing a grid on the animal’s movement path and counting the number of the squares where at least one egg was laid ( normalized to the number of squares where the animal traveled; Figure 4B ) . This analysis also revealed a greater dispersal of eggs during roaming ( Figure 4C ) . Moreover , dopamine-defective mutants ( both cat-2 and dop-2;dop-3 ) displayed a significant reduction in egg dispersal ( Figure 4C ) . Consistent with these results , cat-2 and dop-2; dop-3 mutants both laid a higher fraction of their eggs during dwelling , as compared to roaming ( Figure 4—figure supplement 1 ) . These data suggest that the coupling of egg-laying to the roaming state increases the dispersal of eggs across a food source , which could have important implications for the survival of an animal’s progeny . Our genetic analyses are consistent with the possibility that dopamine signaling plays a direct role in elevating egg-laying rates during the roaming state . To examine whether dopamine acutely regulates egg-laying , we performed optogenetic studies to alter the activity of the neurons that release dopamine . To silence dopaminergic neurons , we used the dat-1 promoter ( Flames and Hobert , 2009; Jayanthi et al . , 1998 ) to drive expression of the GtACR2 anion channel ( Govorunova et al . , 2015 ) in all four dopaminergic cell types: CEPV , CEPD , ADE , and PDE ( no single-neuron promoters have been described for any of these four cell types ) . Exposing these animals to light caused them to display lower egg-laying rates that persisted until the light was terminated ( Figure 5A; additional controls in Figure 5—figure supplement 1A ) . We asked whether this effect reflected a reduction in egg-laying during roaming , dwelling , or both states by separately analyzing egg-laying rates during each of the states during the lights-on period ( Figure 5B ) . This analysis indicated that the silencing of dopaminergic neurons reduced egg-laying rates while animals were roaming , but only had a mild , non-significant effect during dwelling . We examined the other behaviors automatically quantified by the tracking microscopes but observed no significant effects of dopaminergic neuron silencing on feeding or defecation , and only a small , transient effect on locomotion ( Figure 5—figure supplement 1B–E ) . These data corroborate our genetic findings and suggest that dopaminergic neurons function to acutely promote egg-laying in adult roaming animals . To examine whether exogenously increasing dopaminergic activity levels is sufficient to drive egg-laying , we acutely activated the four dopaminergic cell types . For these experiments , we used the dat-1 promoter to drive expression of CoChR , a light-gated cation channel ( Klapoetke et al . , 2014 ) . Activation of dat-1::CoChR for three minute durations caused animals to display elevated egg-laying rates ( Figure 5C; minimal effects on velocity , Figure 5—figure supplement 1B and F–H; additional controls in Figure 5—figure supplement 1I ) . This increase in the rate of egg-laying persisted throughout the period of light exposure and was not necessarily time-locked to the moment of light onset . By contrast , acute activation of the egg-laying motor neuron HSN is known to trigger egg-laying events within seconds of light exposure ( Emtage et al . , 2012; Leifer et al . , 2011 ) . These data suggest that increased dopaminergic neuron activity increases the frequency of egg-laying events . The observation that dopamine is necessary for proper egg-laying rates during roaming , but not dwelling , might be explained by elevated dopamine release during roaming . Alternatively , dopaminergic activity might be similar across states , but downstream circuits might integrate their detection of dopamine with a locomotion state variable , such that dopamine only enhances egg-laying during roaming . To begin to distinguish between these possibilities , we asked whether the ectopic activation of dopaminergic neurons via optogenetics could increase egg-laying during roaming , dwelling , or both states . We found that light exposure to dat-1::CoChR animals elevated egg-laying during the dwelling state , but not during roaming ( Figure 5D ) . The finding that the exogenous activation of dopaminergic neurons during dwelling is sufficient to increase egg-laying argues against a model where downstream circuits only respond to dopamine during roaming , but not dwelling . The finding that dat-1::CoChR activation does not further enhance egg-laying during roaming may reflect an occlusion effect , perhaps due to the already high egg-laying rates during roaming . We again considered whether these effects of dopamine on egg-laying could be explained by an indirect effect on locomotion . We analyzed the velocity changes induced by optogenetic dopaminergic neuron silencing and activation . These two manipulations , which caused opposite effects on egg-laying , caused the same velocity change: a transient decrease in speed at light onset and a transient increase in speed at light offset ( Figure 5—figure supplement 1A ) . Therefore , the effects of these opposing optogenetic manipulations on egg-laying cannot be plausibly explained as an indirect consequence of altering locomotion . Animals have dramatically lower egg-laying rates when they are removed from their food source . Because dopamine signaling is thought to be elevated in the presence of food ( Sawin et al . , 2000; Tanimoto et al . , 2016 ) , we asked whether increasing dopaminergic neuron activity in the absence of food was sufficient to increase egg-laying rates . Strikingly , we found that activation of dat-1::CoChR could drive animals to lay eggs in the absence of food at significantly higher rates ( Figure 5E–F; velocity in Figure 5—figure supplement 1J ) . Altogether , these data suggest that native dopaminergic neuron activity in adult animals increases the probability of egg-laying during roaming states , and that exogenous dopaminergic neuron activity during dwelling or in the absence of food is sufficient to enhance egg-laying rates . The above data suggest that dopamine release can enhance egg-laying and that native dopaminergic signaling in wild-type animals primarily promotes egg-laying during the roaming state . One possible explanation for these observations is that dopamine release might predominate during high-speed roaming . Thus , we next examined the native activity patterns of dopaminergic neurons in freely-moving animals . We constructed a transgenic strain expressing GCaMP6m under the dat-1 promoter and recorded GCaMP signals in freely-moving animals using widefield imaging , as has been previously described ( Flavell et al . , 2013; Rhoades et al . , 2019 ) . The CEPD , CEPV , and ADE classes of neurons appeared to have static calcium levels while animals explored food and did not display dynamics that were dependent on locomotion speed ( Figure 6—figure supplement 1A ) . However , the dopaminergic PDE neurons displayed robust calcium dynamics as animals freely explored a food lawn ( Figure 6A–B ) . PDE neurons have short ciliated sensory dendrites that protrude through the cuticle along the dorsal side of the animal in the posterior section of the body . The PDE axon travels along the ventral nerve cord , from the posterior to the anterior end of the animal . Notably , PDE is the only dopaminergic neuron whose neurite passes in close proximity to the egg-laying circuit and vulval muscles . To quantify how PDE activity changes during locomotion , we first quantified the extent to which PDE dynamics depend on locomotion . We found that fluctuations in PDE calcium levels were significantly increased during high-speed movement ( Figure 6A ) , particularly during roaming ( Figure 6—figure supplement 1B ) , a relationship that resembles how animal movement is coupled to egg-laying ( Figure 2F ) . As animals moved , PDE activity was highly correlated with the animal’s body curvature . PDE was most active when the dorsal body wall muscles were contracted on the posterior end of the body ( Figure 6B shows an example; Figure 6C shows correlations ) . This correlation between posture and PDE activity was present during roaming and dwelling ( Figure 6C; Figure 6—figure supplement 1C ) . However , when animals are roaming , repeating waves of muscle contractions pass through the body from head to tail . Thus , one result of the correlation between PDE activity and body curvature was that PDE activity oscillated during roaming and was significantly elevated during a specific phase of each sinusoidal propagating wave ( Figure 6D shows an example; Figure 6E shows averages ) . Because PDE has an exposed sensory ending , we examined whether this activity pattern was impacted by the presence of bacterial food in the environment . Indeed , we found that PDE displayed significantly reduced dynamics during forward locomotion in the absence of food ( Figure 6—figure supplement 1D ) . These relationships between PDE fluorescence and body posture were not observed when imaging PDE::GFP ( instead of GCaMP ) , indicating that they were not due to motion artifacts ( Figure 6—figure supplement 1E–F ) . These data indicate that PDE calcium levels oscillate during forward movement on food , with a reliable activity peak during a specific phase of the forward propagating wave during roaming . Previous studies suggested that egg-laying events also depend on body curvature ( Collins et al . , 2016 ) , so we also quantified the frequency of egg-laying events during each sinusoidal propagating wave of muscle contractions during roaming . Indeed , egg-laying events were strongly biased to a specific phase of the sinusoidal wave ( Figure 6F ) . Notably , the phase of maximal PDE activity overlapped with and shortly preceded the phase of maximal egg-laying . These postural changes that preceded egg-laying during roaming were distinct from those during that preceded egg-laying events during dwelling , even though animals adopted similar postures in both states at the precise moment of egg-laying ( Figure 6—figure supplement 1G ) . These data indicate that the animal’s gait during roaming leads to a unique coupling of body curvature to egg-laying , in a manner that causes PDE activity to shortly precede the typical body posture for egg-laying . This relationship raised the possibility that dopamine release during roaming might not only promote egg-laying events overall , but might also bias them to occur during a specific phase of the sinusoidal wave . To test this possibility , we examined the distribution of egg-laying events in dopamine-deficient cat-2 mutants and found that there was a significant decrease in the phase-dependence of egg-laying ( Figure 6G ) . This difference was not due to a general change in body postures during roaming , since wild-type animals and cat-2 mutants displayed similar postures during forward propagating waves while roaming ( Figure 6—figure supplement 1H–I ) . Together , these data suggest that the activity of dopaminergic PDE neurons is phase-locked to egg-laying events during roaming and that dopamine signaling is required for proper coordination of body posture and egg-laying during roaming . We next sought to identify the downstream circuits through which dopaminergic neurons act to elevate egg-laying rates . To begin to identify these downstream components , we examined whether specific neurons or neurotransmitters were necessary for dopamine-dependent egg-laying . Egg-laying in C . elegans requires contraction of the vulval muscles , which receive many synapses from HSN and VC neurons and a smaller number of synapses from cholinergic VA/VB neurons and GABAergic VD neurons . We optogenetically activated dopaminergic neurons in mutant animals lacking each of these four inputs: ( 1 ) egl-1 mutants lacking HSNs , ( 2 ) lin-39 mutants lacking VCs , ( 3 ) acr-2 mutants with reduced cholinergic neuron activity , and ( 4 ) unc-25 mutants with abolished GABAergic transmission ( Figure 7A shows fold-change in egg-laying during lights-on , compared to lights-off ) . dat-1::CoChR activation still elevated egg-laying rates in acr-2 mutants , suggesting that cholinergic transmission in ventral cord neurons is not essential for these effects . However , dat-1::CoChR activation had a reduced effect in egl-1 and lin-39 mutants and failed to elevate egg-laying rates in unc-25 mutants . The finding that HSN and VCs are required for dopamine-induced egg-laying is not surprising , since these motor neurons are centrally involved in driving egg-laying , but the role of GABAergic signaling in egg-laying has not been well-studied , so we examined this interaction more closely . In C . elegans , GABA is produced by a small set of neurons in the head and tail , and by the VD/DD motor neurons in the ventral cord . To clarify which neurons mediate dopamine-dependent egg-laying , we activated dat-1::CoChR in unc-30 mutants , which lack GABA in the VD/DD neurons , but not in the other GABAergic cell types in the head and tail ( Figure 7A ) . We found that dat-1::CoChR failed to elevate egg-laying in unc-30 mutants , indicating that GABA production by VD/DD neurons is critical for this effect . A previous study showed that application of muscimol , a GABAA receptor agonist , strongly inhibits egg-laying ( Tanis et al . , 2009 ) , but the effects of reducing GABAergic signaling on egg-laying had not been previously examined . Thus , we performed experiments to clarify the role of native GABAergic signaling in egg-laying . Here , we compared wild-type animals to two independent unc-25 null mutant strains , which both have attenuated GABA synthesis . Compared to wild-type animals , unc-25 mutants had a slightly lower brood size ( ~220 eggs per animal vs . ~280 in wild-type ) and laid their eggs over their first 5–6 days of adulthood , versus 3–4 days in wild-type animals ( Figure 7B ) . As a result , unc-25 mutants displayed lower overall egg production during their first day of adulthood , but higher egg production as five- and six-day old adults ( Figure 7B ) . We examined state-dependent egg-laying in these animals and found that , after adjusting for their lower egg production , one-day old unc-25 mutants displayed a relative increase in egg-laying during dwelling , as compared to roaming ( Figure 7C ) . To test whether GABA also regulates egg-laying across environmental conditions , we examined egg-laying in the absence of food and found that unc-25 mutants displayed an elevated egg-laying rate off food , compared to wild-type animals ( Figure 7D ) . Altogether , these results indicate that native GABAergic signaling regulates egg-laying , specifically playing an inhibitory role during dwelling and in the absence of food . This inhibitory role is consistent with the previous finding that muscimol inhibits egg-laying ( Tanis et al . , 2009 ) . Given that GABAergic signaling is required for dopamine-dependent egg-laying , one possible explanation for these effects is that dopamine release may inhibit GABAergic neurons that function to inhibit egg-laying .
Behavioral states impact the generation of diverse motor outputs , but the neural mechanisms that allow them to exert these widespread effects are poorly understood . To examine this problem at whole-organism scale , we developed a new method that permits simultaneous , automated quantification of diverse C . elegans motor programs over long time scales . Analysis of these near-comprehensive records of animal behavior show that there is extensive coordination between the different motor programs of this animal . By combining this approach with genetics and optogenetics , we uncovered a new role for dopamine in promoting egg-laying in a locomotion state-dependent manner . Our results provide new insights into how the diverse motor programs throughout an organism are coordinated and suggest that neuromodulators like dopamine can couple motor circuits together in a state-dependent manner . To understand how internal states are represented in the brain , it is critical to obtain a quantitative description of the full repertoire of behaviors that are influenced by a given state . The transparency of C . elegans allows us to observe each of the motor actions of this animal , even the movements of internal muscle groups like the pharynx . Thus , by using tracking microscopes and machine vision software , we simultaneously quantified the diverse motor programs of this animal . These datasets revealed widespread coordination among distinct motor programs . For example , roaming and dwelling states that were previously described based on locomotion parameters also show robust differences in other motor programs , like egg-laying . Moreover , dwelling states can be segmented into different sub-modes based on reliable differences in body posture and other motor programs . We note that so far our analyses have been limited to well-fed adult animals exploring a homogeneous E . coli food source . The structure of C . elegans behavior could be quite different when animals are different ages ( Stern et al . , 2017 ) , exposed to different stimuli , or in different physiological states . For example , quiescence/sleep was not observed under our recording conditions , but this state occurs frequently in other conditions . Several factors likely contribute to the coupling between diverse motor programs , such as: changing internal state variables and recruitment of neuromodulators ( Donnelly et al . , 2013; Flavell et al . , 2013; Pirri et al . , 2009; Raizen et al . , 2008; Van Buskirk and Sternberg , 2007 ) , corollary discharges and related motor signals that allow motor circuits to interact ( Gordus et al . , 2015; Liu et al . , 2018 ) , and proprioceptive or environmental feedback ( Goodman and Sengupta , 2019; Hu et al . , 2011; Li et al . , 2006; Wen et al . , 2012; Yeon et al . , 2018 ) . Combining the whole-organism behavioral profiling approach described here with recently-developed whole-brain calcium imaging approaches ( Kato et al . , 2015; Nguyen et al . , 2016; Venkatachalam et al . , 2016; Yemini and Hobert , 2020 ) could yield a more complete understanding of which neural mechanisms are at work . To begin to clarify neural mechanisms that underlie the coordination between motor programs , we investigated a particularly robust form of motor program coupling: the increased frequency of egg-laying during high-speed roaming states . In wild-type animals exposed to a food source , egg-laying rates are approximately six-fold higher during roaming as compared to dwelling . Based on our analysis of the spatial distribution in which eggs are laid , this coupling allows animals to effectively disperse their eggs throughout a food resource . This might serve as an effective ‘bet-hedging’ strategy to increase the likelihood that at least some of the progeny develop in a beneficial environment . We found that dopamine signaling was required for motor program coupling: mutants with reduced dopamine levels had lower egg-laying rates during roaming but not dwelling . Previous work has shown that native dopamine signaling in C . elegans is involved in driving slow locomotion in response to a bacterial food lawn , an effect called the basal slowing response ( Sawin et al . , 2000 ) . Indeed , we observed increased forward velocity in cat-2 mutants ( Figure 3A ) . However , multiple lines of evidence suggest that the effects of dopamine on egg-laying are separable from those on locomotion: ( 1 ) dop-2; dop-3 dopamine receptor mutants have reduced egg-laying during roaming , but normal roaming velocities ( Figure 3E; Figure 3—figure supplement 1E ) , and ( 2 ) optogenetic silencing and activation of dopaminergic neurons have opposite effects on egg-laying , even though they have the same modest effect on locomotion . We have not yet mapped out where the dop-2 and dop-3 receptors function to control egg-laying , though it is intriguing that dop-3 is known to be expressed in VD/DD neurons ( Chase et al . , 2004 ) in light of our observation that optogenetic dopaminergic neuron activation fails to elevate egg-laying in GABA-deficient mutants ( Figure 7A ) . Further studies will be necessary to determine if dopamine acts directly on GABAergic neurons or signals to a different set of dop-2- and dop-3-expressing neurons to regulate egg-laying . How does dopamine influence egg-laying rates during the roaming state , but not during dwelling ? One possibility is that dopamine release dynamics differ in roaming compared to dwelling . Alternatively , downstream targets could integrate their detection of dopamine with a locomotion state variable , such that dopamine only enhances egg-laying during roaming . Our results argue against the latter , since elevating dopaminergic neuron activity during dwelling is sufficient to increase egg-laying . Moreover , in vivo calcium imaging revealed increased calcium fluctuations in dopaminergic PDE neurons during high-speed roaming . Specifically , we observed that PDE calcium levels oscillate during forward movement , peaking during a stereotyped phase of the forward propagating bend that overlaps with and shortly precedes the peak phase of egg-laying during roaming . We have not yet determined the underlying mechanism that drives these changes in PDE activity . However , since PDE is a ciliated sensory neuron that expresses mechanoreceptors like trp-4 ( Kang et al . , 2010; Li et al . , 2011; Sawin et al . , 2000 ) , it is possible that PDE might be activated by the animal’s own movement or by the increased flow of external food along the body during high-velocity roaming . In favor of the latter possibility , we found that PDE dynamics were reduced in the absence of food . Thus , it is possible that PDE receives environmental feedback that indicates the degree of movement along a food source . This type of environmental feedback is known to occur for other sensory modalities – for example , optic flow provides animals with visual feedback of how they are progressing through a visual scene . Although the dopaminergic ADE , CEPV , and CEPD neurons did not display calcium dynamics correlated to behavioral changes , it is possible that tonic levels of dopamine release from these neurons could also play a functional role in egg-laying . The egg-laying circuit consists of the HSN command neuron and VC , VB , and VD neurons that innervate the vulval muscles . We mapped out downstream effectors of dopamine by optogenetically activating dopaminergic neurons in mutant backgrounds lacking candidate neurons and neurotransmitters . These experiments showed that HSN and VC neurons are important for dopamine’s effects on egg-laying and that GABAergic signaling in VD/DD neurons is required for these effects . Recent studies have shown that GABA receptors are found on a multitude of C . elegans neurons that do not receive direct synaptic inputs from GABAergic neurons , including HSN and VCs ( Yemini and Hobert , 2020 ) . Given that VD/DD neurons have GABA release sites in close physical proximity to the egg-laying circuit , it is possible that GABA acts extrasynaptically to influence egg-laying . It is widely appreciated that neuromodulation contributes to behavioral state control , but the widespread behavioral changes that occur across states make it challenging to understand each neuromodulatory system’s specific contribution . The systematic approach described here reveals that in C . elegans dopamine elevates egg-laying rates during the roaming state . Previous work showed that PDF neuropeptides drive high-speed locomotion typical of the state and have a more modest effect on egg-laying . The combined actions of these neuromodulators and others likely give rise to the full set of behavioral parameters that define the roaming state , though the nature of their interactions will require additional studies . Recent studies of mammalian anxiety states have shown that individual neuromodulators like Tac2 neuropeptides can themselves also function in parallel circuits to control different behaviors ( Zelikowsky et al . , 2018 ) . Studies of fixed action patterns in invertebrates also revealed parallel functions of single neuropeptides in distinct circuits ( Kim et al . , 2006; Scheller et al . , 1982 ) . The whole-organism behavioral profiling approach that we describe here allows the behavioral changes that differ across states to be more fully characterized , as opposed to being analyzed one behavior at a time . Such a level of understanding will be essential to reveal the neural mechanisms that underlie the generation of these brain-wide states .
Nematode culture was conducted using standard methods ( Brenner , 1974 ) . Populations were maintained on NGM agar plates supplemented with E . coli OP50 bacteria . Wild-type was C . elegans Bristol strain N2 . For genetic crosses , all genotypes were confirmed using PCR . Transgenic animals were generated by injecting DNA clones plus fluorescent co-injection marker into gonads of young adult hermaphrodites . All assays were conducted at room temperature ( ~22°C ) . Animals were staged approximately 72 hr prior to recording . 10 adult animals were picked to an NGM plate seeded with E . coli OP50 and left to lay eggs for approximately one hour . Adult animals were removed , and eggs were allowed to develop into adults at room temperature until recorded . For experiments that did not involve optogenetics , animals were left undisturbed for the entire developmental period . For experiments that included optogenetics , animals were picked as L4s to an NGM plate seeded with a bacterial lawn with 50 µM all-trans-retinal ( ATR ) . The assay plates were standard 10 cm diameter petri dishes , with low-peptone ( 0 . 2 g/L ) NGM , seeded with 200 µL OP50 . 50 µM ATR was added for recordings involving optogenetic stimulation . Roughly circular lawn shapes were created with a spreader , and plates were left to dry overnight at room temperature . We used low-peptone plates because they limited the extent of growth of the bacterial lawn . Thick bacterial lawns could obscure the worm and/or contain deep tracks in them , which reduced the quality of worm tracking . Thus , thin bacterial lawns were more desirable for these recordings . For each on-food recording , a single 72 hr old adult animal was picked to a plate . Plates were taped onto the microscope stage open and face-down on top of 3 evenly placed spacers to prevent condensation on the stage . For animals recorded in the absence of food , a copper ring ( with ~7 cm inner diameter ) made of filter paper dipped in 0 . 02M copper chloride solution was positioned on the plate to prevent the animal from reaching the plate edge . For these recordings , animals were washed in M9 twice and placed on the no-food plate using a glass pipette . Kim wipes were used to remove any remaining liquid . Recordings began 10 min after animals were positioned on the assay plate and were approximately six hours long for on-food recordings and 1–2 hr long for off-food recordings . Optogenetic stimulation was performed with a 532 nm laser at an intensity of 150 µW/mm2 for three minutes every ten minutes . All data were included in analyses , except for a small number of datasets that had very poor quality . The criteria for exclusion , which were uniformly applied , were: ( 1 ) if the posture could not be determined in >10% of the frames , or ( 2 ) if the head vs tail could not be confidently assigned . While gross locomotor behavior could be quantified online , most other metrics were not , due to computational requirements . We developed an offline analysis pipeline in R ( RRID:SCR_001905 ) to 1 ) calculate additional locomotion-related metrics and also quantify 2 ) body posture 3 ) pumping 4 ) defecation and 5 ) egg-laying . In vivo calcium imaging of the four dopaminergic cell types was conducted on one-day-old wild-type animals . Animals were positioned on thin , flat slices of NGM agar with a PDMS outer barrier so that animals could not crawl off the agar . OP50 E . coli was seeded uniformly on the agar , except for off-food videos . Animals were permitted to equilibrate on the agar slides for 10 min prior to the beginning of each recording . Animals were then recorded with a 4x/0 . 2NA Nikon objective and an Andor Zyla 4 . 2 Plus sCMOS camera . Blue light output to animals was 9–20% output from an X-Cite 120LED system for 10 ms of each exposure . Brightfield images ( to collect postural and behavioral information ) and fluorescence images were collected in an alternating fashion , made possible by alternating which light source was active during each camera exposure . The total frame rate was 20fps , so that each channel ( GCaMP and brightfield ) had an effective frame rate of 10fps . GCaMP data and behavioral data were thus collected with a 50 ms lag , though for the purposes of our analyses we dis-regarded this small time lag . GCaMP fluorescence was tracked using a previously-described ImageJ ( RRID:SCR_003070 ) tracking macro ( Flavell et al . , 2013 ) and behavioral data were extracted using the software that was developed for the tracking microscope analysis ( described above ) . The details of all statistical tests carried out in this study can be found in Supplementary file 1 . | Animals generate many different motor programs ( such as moving , feeding and grooming ) that they can alter in response to internal needs and environmental cues . These motor programs are controlled by dedicated brain circuits that act on specific muscle groups . However , little is known about how organisms coordinate these different motor programs to ensure that their resulting behavior is coherent and appropriate to the situation . This is difficult to investigate in large organisms with complex nervous systems , but with 302 brain cells that control 143 muscle cells , the small worm Caenorhabditis elegans provides a good system to examine this question . Here , Cermak , Yu , Clark et al . devised imaging methods to record each type of motor program in C . elegans worms over long time periods , while also dissecting the underlying neural mechanisms that coordinate these motor programs . This constitutes one of the first efforts to capture and quantify all the behavioral outputs of an entire organism at once . The experiments also showed that dopamine – a messenger molecule in the brain – links the neural circuits that control two motor programs: movement and egg-laying . A specific type of high-speed movement activates brain cells that release dopamine , which then transmits this information to the egg-laying circuit . This means that worms lay most of their eggs whilst traveling at high speed through a food source , so that their progeny can be distributed across a nutritive environment . This work opens up the possibility to study how behaviors are coordinated at the level of the whole organism – a departure from the traditional way of focusing on how specific neural circuits generate specific behaviors . Ultimately , it will also be interesting to look at the role of dopamine in behavior coordination in a wide range of animals . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"neuroscience"
] | 2020 | Whole-organism behavioral profiling reveals a role for dopamine in state-dependent motor program coupling in C. elegans |
Sensory integration difficulties have been reported in autism , but their underlying brain-circuit mechanisms are underexplored . Using five autism-related mouse models , Shank3+/ΔC , Mecp2R308/Y , Cntnap2−/− , L7-Tsc1 ( L7/Pcp2Cre::Tsc1flox/+ ) , and patDp ( 15q11-13 ) /+ , we report specific perturbations in delay eyeblink conditioning , a form of associative sensory learning requiring cerebellar plasticity . By distinguishing perturbations in the probability and characteristics of learned responses , we found that probability was reduced in Cntnap2−/− , patDp ( 15q11-13 ) /+ , and L7/Pcp2Cre::Tsc1flox/+ , which are associated with Purkinje-cell/deep-nuclear gene expression , along with Shank3+/ΔC . Amplitudes were smaller in L7/Pcp2Cre::Tsc1flox/+ as well as Shank3+/ΔC and Mecp2R308/Y , which are associated with granule cell pathway expression . Shank3+/ΔC and Mecp2R308/Y also showed aberrant response timing and reduced Purkinje-cell dendritic spine density . Overall , our observations are potentially accounted for by defects in instructed learning in the olivocerebellar loop and response representation in the granule cell pathway . Our findings indicate that defects in associative temporal binding of sensory events are widespread in autism mouse models .
In autism spectrum disorder ( ASD; hereafter referred to as autism ) , atypical sensory processing is widely reported starting in infancy ( Leekam et al . , 2007; Markram and Markram , 2010; Dinstein et al . , 2012 ) . In addition to early-life abnormal processing of single sensory modalities ( Leekam et al . , 2007 ) , more complex deficits become apparent as early as 2 years of age , a time when autistic children attend poorly to natural combinations of spoken stimuli and natural visual motion ( Klin et al . , 2009 ) , a circumstance that calls upon the ability to integrate , from moment to moment , information from two sensory modalities , hearing and vision . Abnormalities of sensory responsiveness are strongly correlated with severity of social phenotypes in high-functioning autism patients ( Hilton et al . , 2010 ) . Taken together , these observations suggest that abnormal processing of multiple sensory modalities on subsecond time scales might impede the acquisition of cognitive and affective capacities that are affected in autism . Abnormal sensory processing in autism is likely to arise in part from genetic mutations and variants that predispose for neural circuit dysfunction . To investigate the ability to associate two near-simultaneous sensory inputs , we used delay eyeblink conditioning , a form of learning that is found in multiple mammalian species ( McCormick and Thompson , 1984; Ivarsson and Hesslow , 1994; Boele et al . , 2010; Heiney et al . , 2014 ) . Persons with autism show alterations to delay eyeblink conditioning ( Sears et al . , 1994; Oristaglio et al . , 2013 ) . Delay eyeblink conditioning depends on plasticity in the cerebellum , a common site of anatomical deviation in patients with autism , and cerebellar gross and cellular malformation are common features of autistic brains ( Wang et al . , 2014 ) . These factors led us to search for aberrations in the quantitative parameters of delay eyeblink conditioning . Autism is among the most heritable of neuropsychiatric disorders ( Gaugler et al . , 2014 ) , and hundreds of autism risk loci have been identified ( Abrahams and Geschwind , 2008; Devlin and Scherer , 2012; Stein et al . , 2013 ) . We examined five mouse models that both recapitulate mutations that occur in human idiopathic and syndromic autisms and display phenotypes reminiscent of human autism ( Abrahams and Geschwind , 2008; Banerjee-Basu and Packer , 2010; Abrahams et al . , 2013; http://gene . sfari . org ) . Four of the models incorporate global mutations with strong expression in the cerebellum: Shank3+/ΔC , the C-terminal deletion model of Shank3 associated with Phelan-McDermid syndrome ( Kouser et al . , 2011 , 2013 ) ; Mecp2R308/Y , a mild truncation model of Mecp2 associated with Rett syndrome ( Ben-Shachar et al . , 2009; Shahbazian et al . , 2002a; Moretti et al . , 2006; De Filippis et al . , 2010 ) ; Cntnap2−/− , a knockout of Cntnap2 associated with cortical dysplasia-focal epilepsy syndrome ( Peñagarikano et al . , 2011 ) ; and patDp/+ , a mouse model of the 15q ( 11–13 ) duplication syndrome closely linked to autism ( Nakatani et al . , 2009; Tamada et al . , 2010; Piochon et al . , 2014 ) . A fifth model , a knockout of the tuberous sclerosis protein L7-Tsc1 ( L7/Pcp2Cre::Tsc1flox/+ and L7/Pcp2Cre::Tsc1flox/flox ) , specifically affects cerebellar Purkinje cells ( PCs ) ( Tsai et al . , 2012 ) . Because different circuit defects might have differential effects on the properties of eyeblink conditioning , we analyzed learning deficits quantitatively in terms of two major features of learning: the probability of generating a response , reflecting the learning process itself; and the magnitude and timing of individual responses , reflecting the neural representation of the learned response .
Three mouse models showed deficits in the response probability during training . In L7-Tsc1 mice ( Figure 3A ) , heterozygous mutant mice ( L7/Pcp2Cre::Tsc1flox/+or HET , n = 18 ) reached a response probability of 32 . 0 ± 4 . 3% , significantly lower than the 51 . 5 ± 3 . 5% level reached in control littermates ( n = 16 ) ( last four training sessions; unpaired two-sample t-test , p = 0 . 01; effect size , Cohen's d′ = 1 . 21 ) . Furthermore , homozygous mutant mice ( L7/Pcp2Cre::Tsc1flox/+ or MUT , n = 5 ) completely failed to acquire CRs ( 1 . 4% ± 0 . 7% in L7/Pcp2Cre::Tsc1flox/flox , n = 5; one-way analysis of variance test ( ANOVA ) across all groups , p < 0 . 0001 , F ( 2 , 35 ) = 19 . 82 , with Bonferroni post hoc statistical differences between L7/Pcp2Cre::Tsc1flox/flox and wild-type littermates , p = 3 × 10−9 , Cohen's d′ = 3 . 01 , and L7/Pcp2Cre::Tsc1flox/+ and wild-type littermates , p = 0 . 00002 , d′ = 1 . 41 ) . Further analysis of L7-Tsc1 mice focused on L7/Pcp2Cre::Tsc1flox/+ only . 10 . 7554/eLife . 06085 . 006Figure 3 . Probability defects are present in four mouse models . ( A ) Time course of response probability with acquisition training in L7-Tsc1 model mice . Black: WT . Red: L7/Pcp2Cre::Tsc1flox/+ . ( B ) Time course of response probability with acquisition training in Cntnap2 model mice . Black: Cntnap2+/+ . Red: Cntnap2−/− . Green: Cntnap2+/− . ( C ) Time course of response probability with acquisition training in Shank3ΔC . Black: Shank3+/+ . Red: Shank3+/ΔC . ( D ) Time course of response probability with acquisition training in Mecp2R308 . Black: WT . Red: Mecp2R308/Y . In panels ( A ) through ( D ) , bar plots indicate response probability averaged over the last four training sessions . ( E ) Probability deficits across all groups . Dashed line: normalized wild-type littermate level . In all panels , shading and error bars indicate SEM , and * indicates p < 0 . 05 . n ≥ 10 mice for each group . Figure 3—figure supplement 1 shows response probability in each group of animals during extinction and reacquisition . DOI: http://dx . doi . org/10 . 7554/eLife . 06085 . 00610 . 7554/eLife . 06085 . 007Figure 3—figure supplement 1 . Extinction and reacquisition . Extinction and reacquisition . Tan shading indicates the extinction period . In the L7-Tsc1 plot , the red line indicates the mean of heterozygous mice , while the blue line indicates the mean of homozygous mice . Black lines with gray shading indicate the mean ± SEM for the wild-type littermates for each cohort . CR performance on last day of reacquisition compared to last day of acquisition . n ≥ 10 for each group . DOI: http://dx . doi . org/10 . 7554/eLife . 06085 . 007 In Cntnap2 mice ( Figure 3B ) , homozygous mutant mice ( Cntnap2−/− , n = 12 ) reached a response probability of 35 . 1% ± 6 . 2% , significantly lower than the 57 . 2% ± 2 . 9% level reached in wild-type littermates ( Cntnap2+/+ , n = 13 ) ( last four training sessions; Bonferroni post hoc test after one-way ANOVA , p = 0 . 007 , d′ = 0 . 96 ) . Notably , Cntnap2+/− mice , which show behavioral similarity to Cntnap2+/+ mice ( Peñagarikano et al . , 2011 ) , were likewise statistically indistinguishable in learning or response amplitude from wild-type mice ( n = 14 mice; Bonferroni post hoc tests after one-way ANOVA , p > 0 . 5 ) . In Shank3ΔC mice ( Figure 3C ) , the heterozygous mutant mice ( Shank3+/ΔC , n = 17 ) reached a response probability of 55 . 9% ± 3 . 7% , lower than the 67 . 2% ± 2 . 2% reached in the wild-type littermates ( Shank3+/+ , n = 21 ) ( unpaired two-sample t-test , p = 0 . 015 , d′ = 1 . 10 ) . In all three mouse models , probability deficits were present throughout training ( two-way repeated measures ANOVA , main genotype effect; Cntnap2−/−: F ( 1 , 23 ) = 7 . 72 , p = 0 . 01; L7/Pcp2Cre::Tsc1flox/flox: F ( 1 , 23 ) = 11 . 70 , p = 0 . 002; Shank3+/ΔC: F ( 1 , 25 ) = 4 . 59 , p = 0 . 04 ) . One model did not show differences in learning probability or time course: Mecp2R308/Y heterozygotes ( Figure 3D; 57 . 2% ± 2 . 9% WT vs 57 . 8% ± 3 . 6% Mecp2R308/Y , unpaired two-sample t-test , p = 0 . 9; two-way repeated measures ANOVA: main genotype effect , F ( 1 , 22 ) = 0 . 10 , p = 0 . 7 ) . We also applied our new analysis technique to a data set previously gathered by our group on the 15q duplication model mice ( Piochon et al . , 2014 ) . We detected a significant difference in response probability that was consistent with previously observed impairment . Throughout acquisition training , response probability in patDp/+ mice ( n = 10 ) was smaller than wild-type littermates ( n = 11 ) ( two-way repeated measures ANOVA: main genotype effect , F ( 1 , 19 ) = 19 . 84 , p = 0 . 0003 ) , culminating in a difference at the end of training ( 34 . 2% ± 2 . 9% patDp/+ vs , 49 . 2% ± 2 . 6% WT , unpaired two-sample t-test , p = 0 . 001 , d′ = 1 . 46 ) . In summary , the five models showed a gradient of defects in probability , ranging from L7/Pcp2Cre::Tsc1flox/flox ( no learning ) to Mecp2R308 heterozygotes ( intact learning ) ( Figure 3E ) . To test whether learned blinks were disrupted , we measured their amplitude normalizing to an unconditioned reflex blink amplitude of 1 . After 12 days of acquisition training , three mutant models showed deficits in response amplitude: L7/Pcp2Cre::Tsc1flox/+ , Shank3+/ΔC , and Mecp2R308/Y . L7/Pcp2Cre::Tsc1flox/+ mice generated smaller-amplitude learned blinks throughout training ( two-way repeated measures ANOVA: main genotype effect , F ( 1 , 23 ) = 7 . 71 p = 0 . 01 ) that culminated in a difference in amplitude at the end of training ( last four training sessions: 0 . 28 ± 0 . 03 in L7/Pcp2Cre::Tsc1flox/+ vs 0 . 39 ± 0 . 05 in littermate controls , unpaired two-sample t-test , p = 0 . 02 , d′ = 0 . 86 ) ( Figure 4A , right ) . In Shank3ΔC mice ( Figure 4C ) , response amplitude was similar to wild-type for most of training ( main genotype effect , F ( 1 , 24 ) = 1 . 45 , p = 0 . 2 ) , but culminated in a small reduction by the end of training ( 0 . 31 ± 0 . 02 Shank3+/ΔC vs 0 . 36 ± 0 . 01 Shank3+/+ , p = 0 . 03 , d′ = 0 . 38 ) . Mecp2R308/Y mice ( Figure 4D; n = 11 ) showed consistently smaller learned responses throughout training ( two-way repeated measures ANOVA: main genotype effect: F ( 1 , 22 ) = 12 . 72 , p = 0 . 002 ) , culminating in a difference in amplitude at the end of training ( last four training sessions , 0 . 28 ± 0 . 02 Mecp2R308/Y in vs 0 . 44 ± 0 . 04 WT , unpaired two-sample t-test , p = 0 . 002 , d′ = 1 . 11 ) . CRs in Mecp2R308/Y mice also reached maximum amplitude much earlier in the training period ( Figure 4D ) . 10 . 7554/eLife . 06085 . 008Figure 4 . Amplitude defects are present in three mouse models . ( A ) Time course of response probability with acquisition training in L7-Tsc1 model mice . Black: WT . Red: L7/Pcp2Cre::Tsc1flox/+ . ( B ) Time course of response probability with acquisition training in Cntnap2 model mice . Black: Cntnap2+/+ . Red: Cntnap2−/− . Green: Cntnap2+/− . ( C ) Time course of response probability with acquisition training in Shank3ΔC . Black: Shank3+/+ . Red: Shank3+/ΔC . ( D ) Time course of response probability with acquisition training in Mecp2R308 . Black: WT . Red: Mecp2R308/Y . In panels ( A ) through ( D ) , bar plots indicate response probability averaged over the last four training sessions . ( E ) Probability deficits across all groups . Dashed line: normalized wild-type littermate level . In all panels , shading and error bars indicate SEM , and * indicates p < 0 . 05 . n ≥ 10 mice for each group . DOI: http://dx . doi . org/10 . 7554/eLife . 06085 . 008 We did not observe statistically significant differences in response amplitude or its development in Cntnap2 mice ( two-way repeated measures ANOVA: main genotype effect , F ( 2 , 32 ) = 0 . 15 , p = 0 . 85; 0 . 32 ± 0 . 03 Cntnap2−/− vs 0 . 34 ± 0 . 02 Cntnap2+/+ , Bonferroni post hoc test after one-way ANOVA , p = 0 . 82 ) ( Figure 4B , right ) or in 15q duplication mice ( two-way repeated measures ANOVA: main genotype effect , F ( 1 , 19 ) = 1 . 81 , p = 0 . 2 ) , including at the end of training ( last four training sessions 0 . 31 ± 0 . 02 WT vs 0 . 27 ± 0 . 05 patDp/+ , unpaired two-sample t-test , p = 0 . 4; also see Piochon et al . , 2014 ) . In summary , defects in blink amplitude ranged from large effects exceeding 1 standard deviation ( Mecp2R308/Y ) to no statistically detectable difference ( Cntnap2−/− and patDp/+; Figure 4E ) . We asked whether CR extinction and savings , two learning processes that require prior eyeblink conditioning , were affected in these five mouse lines ( Figure 3—figure supplement 1 ) . After training , 4 days of extinction led to the near-disappearance of CRs in all autism model groups ( CR percentage , day 12 acquisition vs day 4 extinction; paired t-tests , p < 0 . 05 for all comparisons ) except for L7/Pcp2Cre::Tsc1flox/flox , which did not acquire CRs in the first place . The time courses of extinction were not statistically distinguishable between any autism model group and its corresponding wild-type littermates ( p > 0 . 05 for all main genotype effects ) , indicating that perturbation of cerebellar cortex-dependent and other mechanisms that are necessary for initial eyeblink conditioning ( Robleto et al . , 2004 ) did not strongly affect overall extinction in the mouse models . In addition , the mouse models that initially acquired CRs also successfully reacquired CRs after 3 days of retraining ( Figure 3—figure supplement 1B; paired t-tests of day 4 extinction vs day 3 reacquisition , p < 0 . 05 for all comparisons ) , with no appreciable difference in CR percentage between groups ( p > 0 . 05 for all instances ) . The accelerated nature of this reacquisition , a process known as savings , is thought to depend in part on plasticity in the DCN ( Medina et al . , 2001; Ohyama et al . , 2006 ) . In short , learning deficits in the mouse models tested were specific to acquisition and were not observed in extinction or reacquisition . The cerebellum is thought to be critical for task timing , and both patients with cerebellar lesions and autism patients show disrupted timing in cerebellum-dependent tasks , including eyeblink conditioning . We therefore examined the timing of the CRs during unpaired CS trials , for which the entire response time course could be analyzed ( Figure 5 ) . Two groups of mice showed differences in timing: Shank3+/ΔC and Mecp2R308/Y . Learned responses produced by the Shank3+/ΔC mice began at the same time ( onset latency: 148 . 7 ± 4 . 9 ms Shank3+/+ , vs 144 . 6 ± 4 . 4 ms Shank3+/ΔC , p = 0 . 5 ) , rose faster ( rise time: 91 . 8 ± 0 . 5 ms Shank3+/+ vs 79 . 5 ± 0 . 3 ms Shank3+/ΔC , p = 0 . 04 , d′ = 0 . 70 ) , and peaked earlier ( peak latency: 317 . 5 ± 9 . 0 ms Shank3+/+ vs 287 . 7 ± 5 . 8 ms Shank3+/ΔC , p = 0 . 03 , d′ = 1 . 02 ) ( Figure 5A , right ) compared to wild-type littermates . In Mecp2R308/Y animals , learned responses began at the same time ( onset latency: 120 . 9 ± 4 . 0 ms WT vs 117 . 7 ± 4 . 9 ms Mecp2R308/Y , p = 0 . 6 ) , rose more slowly ( rise time: 113 . 2 ± 12 . 4 ms WT vs 158 . 4 ± 15 . 6 ms Mecp2R308/Y , p = 0 . 04 , d′ = 1 . 04 ) , and peaked later ( peak latency: 278 . 3 ± 14 . 9 ms WT vs 328 . 8 ± 16 . 4 ms Mecp2R308/Y , p = 0 . 04 , d′ = 1 . 03 ) compared with wild-type littermates ( Figure 5B , right ) . No alterations in onset latency , peak latency , or rise time could be detected in L7/Pcp2Cre::Tsc1flox/+ ( Figure 5C ) , Cntnap2−/− mice ( Figure 5D ) , or patDp/+ mice ( Piochon et al . , 2014 ) ( p > 0 . 05 for all comparisons; summary of all mouse lines , Figure 5E , F ) . 10 . 7554/eLife . 06085 . 009Figure 5 . Timing defects are present in two mouse models . ( A ) Analysis of Mecp2R308/Y Mecp2R308 response timing ( rise time and peak latency ) . Inset: representative eyelid movement traces . Purple line: CS duration . Scale bars: horizontal , 100 ms; vertical , 20% of unconditioned response ( UR ) amplitude . Arrowheads: peak times . ( B ) Analysis of Shank3ΔC response timing ( rise duration and peak time ) . Inset: representative eyelid movement traces . Purple line: CS duration . Scale bars: horizontal , 100 ms; vertical , 20% of UR amplitude . Arrowheads: peak times . ( C ) Analysis of Cntnap2 response time ( rise time and peak latency ) . ( D ) Analysis of L7-Tsc1 response time ( rise time and peak latency ) ( E ) Peak time deficits across all groups . ( F ) Rise time deficits . In plots ( E ) and ( F ) , dashed lines indicate normalized wild-type littermate level . In all panels , shading and error bars indicate SEM , and * indicates p < 0 . 05 . n ≥ 10 mice for each group . DOI: http://dx . doi . org/10 . 7554/eLife . 06085 . 009 Autism has been suggested to be a general disorder of excessive sensory responsiveness , a concept known as the ‘intense world’ hypothesis ( Markram and Markram , 2010 ) . Potentially , our results in these mouse models could be accounted for by alterations in sensory responsiveness , a common feature of autism . To test this possibility , we measured responses to the US and to the pre-training CS . In US-only trials , we found no differences in unconditioned response ( UR ) latency measured from US onset ( p ≥ 0 . 2 for unpaired comparisons for each cohort ) or UR rise time ( p ≥ 0 . 4 for unpaired two-sample comparisons for each cohort ) ( Table 1 , ‘Unconditioned response’ ) and no differences in the correlation between UR velocity and UR magnitude ( analysis of covariance group × peak interaction , p ≥ 0 . 2 for all cohorts ) . We detected no differences among wild-type cohorts for UR latency ( one-way ANOVA , p = 0 . 5 , F ( 4 , 64 ) = 0 . 92 ) or velocity ( one-way ANOVA , p = 0 . 4 , F ( 4 , 64 ) = 1 . 08 ) . 10 . 7554/eLife . 06085 . 010Table 1 . Normal sensory responsiveness , gross motor function , and non-cerebellar learning and memory in five autism mouse modelsDOI: http://dx . doi . org/10 . 7554/eLife . 06085 . 010L7/Pcp2Cre::Tsc1flox/+Cntnap2−/−patDp/+Shank3+/ΔCMecp2R308/YUnconditioned response N18 , 1612 , 1310 , 1117 , 2111 , 12 UR latency ( ms ) 31 . 0 ± 8 . 632 . 6 ± 3 . 145 . 9 ± 8 . 434 . 5 ± 5 . 942 . 0 ± 8 . 729 . 9 ± 4 . 327 . 4 ± 5 . 030 . 7 ± 7 . 339 . 5 ± 8 . 243 . 1 ± 9 . 3 UR rise time ( ms ) 64 . 9 ± 4 . 765 . 7 ± 5 . 567 . 3 ± 4 . 362 . 8 ± 3 . 660 . 1 ± 5 . 557 . 5 ± 3 . 864 . 8 ± 5 . 572 . 8 ± 6 . 362 . 6 ± 3 . 864 . 8 ± 6 . 5Eyelid opening N18 , 1612 , 1310 , 1117 , 2111 , 12 Amplitude ( % UR amp ) 13 . 9% ± 3 . 9%6 . 4% ± 1 . 2%13 . 4% ± 4 . 8%11 . 8% ± 3 . 1%13 . 4% ± 5 . 5%15 . 6% ± 5 . 0%11 . 1% ± 3 . 0%11 . 8% ± 7 . 8%9 . 3% ± 2 . 6%13 . 7% ± 5 . 8%Gait analysis N6 , 710 , 10–5 , 4– Fore stride ( cm ) 4 . 61 ± 0 . 215 . 01 ± 0 . 21–4 . 82 ± 0 . 31–4 . 35 ± 0 . 145 . 15 ± 0 . 46–4 . 92 ± 0 . 28– Fore stance ( cm ) 1 . 42 ± 0 . 061 . 39 ± 0 . 14–1 . 84 ± 0 . 12–1 . 56 ± 0 . 061 . 41 ± 0 . 07–1 . 64 ± 0 . 12– Hind stride ( cm ) 4 . 85 ± 0 . 275 . 22 ± 0 . 34–4 . 98 ± 0 . 27–4 . 84 ± 0 . 155 . 09 ± 0 . 42–5 . 07 ± 0 . 29– Hind stance ( cm ) 2 . 62 ± 0 . 172 . 20 ± 0 . 16–2 . 37 ± 0 . 12–2 . 69 ± 0 . 162 . 00 ± 0 . 17–2 . 27 ± 0 . 12–Swimming Y-maze acquisition N6 , 710 , 10–5 , 4– Acq . 1 ( % correct trials ) 65 . 7% ± 12 . 9%81 . 5% ± 6 . 3%–65 . 0% ± 8 . 6%–76 . 9% ± 7 . 9%71 . 1% ± 11 . 6%–52 . 0 ± 10 . 0%– Acq . 2 ( % correct trials ) 90 . 0% ± 6 . 8%89 . 0% ± 7 . 4%–61 . 0% ± 17 . 2%–75 . 6% ± 7 . 0%91 . 1% ± 4 . 8%–70 . 0% ± 17 . 3%– Acq . 3 ( % correct trials ) 90 . 0% ± 6 . 8%96 . 0% ± 2 . 7%–90 . 0% ± 10 . 0%–80 . 8% ± 8 . 2%95 . 6% ± 3 . 0%–95 . 0% ± 5 . 0%– Acq . 4 ( % correct trials ) 80 . 0% ± 20 . 0%98 . 0% ± 2 . 0%–100% ± 0%–90 . 0% ± 5 . 7%100% ± 0%–94 . 3% ± 3 . 7%– Test ( % correct trials ) 91 . 3% ± 4 . 2%94 . 8% ± 3 . 1%–87 . 2% ± 7 . 9%–93 . 4% ± 3 . 3%99 . 0% ± 1 . 0%–97 . 2% ± 2 . 8%–Unconditioned response was measured in terms of latency and rise time . Eyelid opening in response to initial CS trials was scaled to the size of the unconditioned response . Gait was measured as stride and stance ( cm ) for both forepaws and hindpaws . Swimming Y-maze acquisition was measured in terms of percentage of correct trials over valid trials for four acquisition periods and a test period . For all cells , top value ( roman text ) indicates the mutant mouse , while bottom value ( italic text ) indicates the control or wild-type littermates . All values mean ± SEM . All paired statistical comparisons yielded p-values greater than 0 . 05 . UR , unconditioned response . As a second measure of sensory processing , on the first training day we observed robust eyelid opening in response to the light CS within 100 ms of CS onset ( Table 1 , ‘Eyelid opening’ ) . Eyelid opening only occurred when animals had not yet begun to produce CRs , indicating that these responses were non-associative in nature . Eyelid opening occurred on a similar fraction of trials in all groups ( p > 0 . 1 for unpaired comparisons between each autism model and wild-type littermates ) . Wild-type groups also did not differ detectably ( one-way ANOVA , p = 0 . 9 , F ( 4 , 64 ) = 0 . 22 ) . In summary , sensory sensitivity was unaltered in any of the mouse models , and thus , deficits in delay eyeblink conditioning were not accompanied by upstream alterations in sensory sensitivity or downstream deficits in blink capability . Motor impairments are common in autism patients ( Fournier et al . , 2010 ) , and cerebellar injury leads to both acute and long-lasting motor deficits . However , past investigations of our mouse models show mild or no motor impairments except for gait alterations in patDp/+ mice ( Piochon et al . , 2014 ) . To extend these measurements , in three mouse models we analyzed gait , a motor function that can proceed without learning . We measured forepaw stance , forepaw stride , hindpaw stance , and hindpaw stride . We observed no differences between mutant and wild-type mice in Cntnap2−/− mice , L7/Pcp2Cre::Tsc1flox/+ mice , and Shank3+/ΔC mice ( two-sample t-test , p > 0 . 05 for all comparisons; Table 1 , ‘Gait analysis’ ) . The L7/Pcp2Cre::Tsc1flox/+ result is consistent with previous reports ( Tsai et al . , 2012 ) . Taken together with past research , our findings indicate that gross motor function in adult ASD mouse models is not a necessary consequence of disruption in cerebellum-dependent learning . Mouse models of autism have been shown to be impaired in fear conditioning and hippocampus-dependent reversal ( Crawley , 2008; Silverman et al . , 2010 ) . To test a second , non-cerebellar form of learning , we subjected three of our models to initial acquisition of a water Y-maze . After four training sessions , we did not observe any statistically detectable difference in the ability to find the platform in Cntnap2−/− mice , L7/Pcp2Cre::Tsc1flox/+ mice , or Shank3+/ΔC mice ( two-sample t-test , p > 0 . 05 for all comparisons; Table 1 , ‘Swimming Y-maze acquisition’ ) . The L7/Pcp2Cre::Tsc1flox/+ finding is consistent with previous reports of normal T-maze acquisition ( Tsai et al . , 2012 ) . Therefore , the eyeblink-conditioning deficits we have observed do not reflect a broad impairment in learning mechanisms . Since eyeblink conditioning depends on the cerebellum , we searched for gross anatomical and cell morphological defects in the cerebella of our mouse models . Using histological methods , in Shank3+/ΔC , Cntnap2−/− , Mecp2R308/Y , and patDp/+ mice , we found no differences between mutant mice and wild-type littermates in PC density , anterior or posterior granule layer thickness , and anterior or posterior molecular layer thickness ( p > 0 . 1 , all comparisons ) . In L7/Pcp2Cre::Tsc1flox/+ mice , for which alterations in PC density have been previously reported ( Tsai et al . , 2012 ) , we found no difference for anterior or posterior granule layer thickness and molecular layer thickness for L7/Pcp2Cre::Tsc1flox/+ ( p > 0 . 1 , all comparisons ) . In summary , with the exception of L7/Pcp2Cre::Tsc1flox/+ mice , these mouse lines do not show gross alterations in granule or PC density . PC arbors are shaped by the cumulative effects of granule cell ( GrC ) input ( Joo et al . , 2014 ) , and therefore , would be potentially altered in their form . We used Sholl analysis to examine the morphology of PC dendritic arbors in Shank3+/ΔC , Cntnap2−/− , Mecp2R308/Y , patDp/+ , and L7/Pcp2Cre::Tsc1flox/+ mice . Only Shank3+/ΔC mice differed from wild type , showing higher complexity of distal dendrites ( two-way repeated measures ANOVA , main genotype effect , F ( 1 , 39 ) = 3 . 50 , p = 0 . 07 ) , with a significant distance × genotype interaction ( F ( 16 , 624 ) = 2 . 77 , p = 0 . 0002; Figure 6A ) . Further analysis of Shank3+/ΔC mice revealed that compared with wild type , the center of mass of the Sholl distribution was farther from the soma ( p = 0 . 03 ) and had a greater total number of crossings at distances farther than 96 μm from the soma ( p = 0 . 01 ) . 10 . 7554/eLife . 06085 . 011Figure 6 . Purkinje cell dendritic arbors show structural defects in Shank3+/ΔC and Mecp2R308/Y mice . ( A ) Purkinje cell ( PC ) dendrite arborization defect is present in Shank3+/ΔC . Left: Sholl analysis example for Shank3+/ΔC . Right: groupwise Sholl analysis for Shank3+/ΔC . Sholl analysis for other four mouse models did not show similar arborization defects , as shown in Figure 6—figure supplement 1 . ( B ) Spine density defects are present in Shank3+/ΔC and Mecp2R308/Y . Left: example image of Shank3+/+ dendritic arbor . Right: spine density for Shank3+/ΔC and Mecp2R308/Y groups . In all panels , shading and error bars indicate SEM , n . s . indicates p > 0 . 05 , and * indicates p < 0 . 05 . n ≥ 12 cells for each group . DOI: http://dx . doi . org/10 . 7554/eLife . 06085 . 01110 . 7554/eLife . 06085 . 012Figure 6—figure supplement 1 . Lack of difference in PC arborization in four ASD mouse models . ( Left to right ) Mecp2R308 , Cntnap2 , patDp/+ ( 15q11-13 ) , and L7-Tsc1 ( wild-type littermate vs heterozygote ) . n ≥ 15 cells for each group . DOI: http://dx . doi . org/10 . 7554/eLife . 06085 . 012 Closer examination of PC arbors ( Figure 6B ) revealed a decrease in the number of visible spines per 10 μm on distal dendrites of Shank3+/ΔC mice ( p = 0 . 04 ) and Mecp2R308/Y mice ( p < 0 . 0001 ) . The remaining three models showed no differences in either PC arbors or spine density ( Figure 6B , Figure 6—figure supplement 1; p > 0 . 3 for main group effect and p > 0 . 5 for space × genotype interactions for all comparisons; p > 0 . 4 for all pairwise comparisons of spine density ) . In summary , differences in dendritic morphology were found specifically in Shank3+/ΔC and Mecp2R308/Y mice , consistent with alterations in GrC input and/or PC dendritic growth mechanisms .
Our conditioning experiments quantified dysfunction in two tasks for which the cerebellum is well-suited: associative learning between multiple senses and the detection of fine timing differences . Two pathways—the olivocerebellar loop ( Figure 7C , red pathway ) and the GrC layer input pathway ( Figure 7C , blue pathway ) —play key roles in the acquisition of learned eyeblink responses in mammals ( McCormick and Thompson , 1984; Yeo and Hesslow , 1998; Garcia et al . , 1999; Attwell et al . , 2001; Longley and Yeo , 2014 ) , including mice ( Koekkoek et al . , 2003 ) . Information about the aversive US is conveyed through the olivocerebellar loop , consisting of PCs in the cerebellar cortex , the inferior olive , and the DCN ( Figure 7C , red pathway ) . This information instructs plasticity in the mossy fiber ( mf ) —GrC—PC pathway , which conveys incoming CS information . The GrC layer pathway undergoes multiple forms of plasticity , including parallel fiber ( PF ) -PC long-term depression ( Hansel et al . , 2001; Carey and Lisberger , 2002; Gao et al . , 2012 ) , and after training . PC output helps to drive a well-timed and well-formed CR ( Choi and Moore , 2003 ) and drive late-stage plasticity in the DCN ( Zheng and Raman , 2010 ) . Thus , defects in the reliable learning and production of CRs might be interpreted as disruption of the olivocerebellar ‘instruction’ system ( Garcia et al . , 1999 ) or the granule cell layer ‘representation’ system ( Arenz et al . , 2009 ) . Activity in the GrC network , which receives direct mf input , is thought to represent key temporal components to drive a well-timed response ( Medina and Mauk , 2000; D'Angelo and De Zeeuw , 2009 ) . Because PC sodium-based simple-spike output acts as an approximately linear readout of synaptic drive ( Walter and Khodakhah , 2006 ) , the time course of CR production might be expected to be constructed from summed patterns of activity in specific combinations of GrCs and inhibitory neurons . Therefore , defects in response timing and amplitude might be interpreted as disruption of synaptic transmission and/or plasticity in the MF pathway . Four mouse models showed decreases in the CR probability: L7-Tsc1 ( L7/Pcp2Cre::Tsc1flox/+ and L7/Pcp2Cre::Tsc1flox/flox ) , patDp/+ , Cntnap2−/− , and Shank3+/ΔC . Upon investigating patterns of gene expression , we found that the disrupted genes in three models ( L7-Tsc1 , patDp/+ , and Cntnap2−/− ) are expressed in PCs , inferior olive , and/or DCN ( Figure 7A , light green and red [regular case] , respectively; Figure 7C , red ) . In L7/Pcp2Cre::Tsc1flox/+ mice , which are PC-specific , early-life loss of Tsc1 leads to increased spine density and decreased excitability in PCs ( Tsai et al . , 2012 ) . This decreased excitability can affect learning by interfering with climbing fiber ( cf ) -based instruction , either by reducing PC dendritic excitability or by making the cerebellar cortex less effective at influencing the DCN , resulting in inhibited IO responsiveness to the US ( Schonewille et al . , 2010 ) . Reduced PC firing would also be expected to reduce response amplitudes , which we have observed . Similarly , patDp/+ mice show cf structural plasticity during development and deregulated PF-PC LTD in adults ( Piochon et al . , 2014 ) , echoing findings in other models ( Koekkoek et al . , 2005; Baudouin et al . , 2012 ) . It should be noted that other forms of cerebellar plasticity can contribute to learning in the absence of PF-PC LTD ( Schonewille et al . , 2011 ) . Taken together , the evidence suggests that cerebellar learning defects in autism mouse models may be strongly shaped by reduced function in the olivocerebellar circuit and associated synaptic plasticity mechanisms . The fourth model that showed a probability defect was Shank3+/ΔC . Shank3 is expressed specifically at postsynaptic densities in the granule cell layer in the mouse cerebellum ( Tu et al . , 1999; Böckers et al . , 2004 , 2005 ) . We observed increased elaboration of the distal dendrites along with decreased spine density ( Figure 6; Figure 7A , light green and red cells [bold case] ) . Neurotrophin-3 ( NT-3 ) from GrCs is required for PC dendritic morphogenesis ( Joo et al . , 2014 ) , suggesting the possibility that the Shank3+/ΔC mutation may disrupt PC dendritic function . We observed both amplitude and timing defects in two mouse models ( Figure 7B , gray circles ) , Shank3+/ΔC and Mecp2R308/Y . These genes are expressed in GrCs ( Figure 7A , yellow and turquoise cells , respectively ) , and Mecp2 is also expressed in Golgi cells ( GoCs ) . Shank3 encodes a scaffolding protein that may influence MF-GrC and GrC-PC synaptic function by reducing glutamatergic transmission and plasticity ( e . g . , Peça et al . , 2011; Yang et al . , 2012; Kouser et al . , 2013 ) , thus , impairing cerebellar learning ( Giza et al . , 2010; Andreescu et al . , 2011 ) . Likewise , Mecp2 expression is dramatically upregulated in GrCs after P21 , a time when MF-GrC and PF-PC synapses are formed and still maturing ( Altman , 1972 ) , suggesting that Mecp2 plays a role in MF-GrC synapse function ( Mullaney et al . , 2004 ) and glutamatergic synaptic transmission and plasticity ( Moretti et al . , 2006 ) . It is notable that despite the fact that Mecp2 is also expressed in PCs ( Mullaney et al . , 2004 ) , Mecp2R308/Y mice showed no defect in probability of learning . We chose these mice for their relatively weak motor dysfunction so that we could characterize eyeblink-conditioning deficits in detail . Other Mecp2 mutants might show more of a probability phenotype . In addition to specific cerebellar substrates , delay eyeblink conditioning also depends on processing outside the cerebellum ( Boele et al . , 2010; Figure 7A , dark gray cells; Figure 7C , dark gray arrows ) . Several genes in our mouse models ( though not Shank3 ) are likely to be expressed in trigeminal nucleus , which transmits sensory information to the pons and mf pathway , as well as the red nucleus and facial nucleus , which ultimately drive the production of the eyeblink ( Figure 7A , dark gray cells; Figure 7C , dark gray arrows; Figure 7—figure supplement 1 ) . The acquisition of delay eyeblink conditioning may also be modulated by the amygdala and hippocampus ( Lee and Kim et al . , 2004; Boele et al . , 2010; Sakamoto and Endo , 2010; Taub and Mintz , 2010 ) , but we did not detect two known consequences of such modulation , learning during the first training session and short-latency alpha responses to the CS . Past investigations of autism ( Sears et al . , 1994; Oristaglio et al . , 2013 ) and Fragile X syndrome ( Koekkoek et al . , 2005; Tobia and Woodruff-Pak , 2009 ) have reported the percentage of CS-responses that exceed a fixed threshold ( ‘% CRs’ ) , as well as CR size averaged across all trials . However , these measures conflate changes in the probability of learning with changes in the amplitude of learned responses . For example , a study that examined de novo ( i . e . , no previous conditioning ) delay eyeblink conditioning ( Sears et al . , 1994 ) found that in high-functioning ( average IQ > 100 ) autistics , the %CR fraction rose more rapidly than in controls , reaching close to a half-maximum after only two blocks of trials . In the direction of loss-of-function , impairments in delay eyeblink conditioning have been observed in Fragile X patients ( Koekkoek et al . , 2005; Tobia and Woodruff-Pak , 2009 ) ; in this case , PC-specific knockout of the Fragile X protein Fmr1 in mice was sufficient to cause eyeblink-conditioning defects , suggesting that learning was specifically perturbed . For comparison with the work reported here , future human eyeblink-conditioning studies would have to distinguish changes in learning from changes in response amplitude . A second promising domain for investigations of ASD patients is eyeblink response kinetics . Variations in response kinetics may depend on the specific genetic background . In idiopathic autism ( Sears et al . , 1994 ) , CRs came approximately 50 ms earlier , as measured using both the time to CR onset and the time to CR peak . Similarly , after two sessions of trace conditioning ( Oristaglio et al . , 2013 ) , delay conditioning initially leads to a decrease in response onset and latency of approximately 50 ms , followed by a convergence toward normal performance as training continues . In contrast , Fragile X patients show no differences in timing in early training sessions ( Koekkoek et al . , 2005; Tobia and Woodruff-Pak , 2009 ) , but after average CR amplitude reaches a plateau , the peak latency to CR decreases by approximately 30 ms ( Tobia and Woodruff-Pak , 2009 ) . Changes of 30–50 ms are comparable in size to the effects we have observed in mice with granule cell pathway perturbation . In addition , in a valproate-based rat model of autism ( Arndt et al . , 2005 ) , prematurely timed eyeblink responses were found for long interstimulus intervals ( Murawski et al . , 2009 ) . In summary , past findings suggest that perturbation of cerebellar granule cell layer activation may be common in both idiopathic and syndromic autism . The general observation of shortened latency is consistent with our findings in Shank3+/ΔC mice , suggesting this line as a model for the timing deficits observed in autistic persons . Finally , although past measurements have been done in older children post-diagnosis , eyeblink conditioning can be assayed in subjects as young as 5 months of age ( Claflin et al . , 2002 ) . The possibility of early testing suggests that delay eyeblink conditioning could be a biomarker ( Reeb-Sutherland and Fox , 2015 ) for identifying pre-diagnosis perturbations in cerebellum-dependent learning . Eyeblink-conditioning defects appear more often in mouse autism models than other non-autism-like phenotypes ( Table 2 ) . This specific dissociation ( i . e . , the absence of correlation with non-cerebellar phenotypes ) suggests that cerebellar plasticity and autism's cognitive deficits might be related in some specific manner . The cerebellum arises repeatedly in the study of autism ( Wang et al . , 2014 ) . In an analysis of gene–phenotype associations ( Meehan et al . , 2011 ) , autism-related genes were found to be associated with a cluster of phenotypes that included social defects , abnormal motor behavior , and cerebellar foliation . A number of ASD genes are co-expressed in the cerebellum ( Menashe et al . , 2013 ) , and ASD patients show differences in many cerebellar cell types ( Bauman and Kemper , 1985; Fatemi et al . , 2002; Whitney et al . , 2008; Wegiel , et al . , 2010 ) as well as gross cerebellar structure , starting at an early age ( Hashimoto et al . , 1995; Abell et al . , 1999; Stanfield et al . , 2008; Courchesne et al . , 2011 ) . Therefore , ASD genes are highly likely to shape cerebellar circuit function . Effects on cerebellar function could even have downstream consequences for function of distal brain regions of known cognitive significance to which the cerebellum supplies information ( Wang et al . , 2014 ) . 10 . 7554/eLife . 06085 . 015Table 2 . Complete table of previously reported autism-like and motor defects in mouse models combined with data from the present studyDOI: http://dx . doi . org/10 . 7554/eLife . 06085 . 015Mouse modelAutism-like behaviorsDelay eyeblink conditioningMovement/strengthOther tasksSocialUltrasonic callsGrooming timeMaze flexibilityEyeblink learning*Eyeblink amplitude*Eyeblink timing*RotarodGaitMaze acquisitionStartle and prepulse inhibitionAnxietyLearned fearShank3 [a]↓↑↑nd↓↓↓↔↔*↔*↓nd↔Cntnap2 [b]↓↓↑↓↓↔↔↑↔*↔*↔↔ndMecp2 [c , d , e]↓↓ndnd↔↓↑↔nd↓nd↑↓L7-Tsc1–mutant [f]↓↓↑↓↓ndnd↓↓↔ndndndL7-Tsc1–het [f]↓↓↑↓ ( ? ) ↓↓↔↔↔↔ndndndpatDp/+ [g , h , i]↓↓nd↓↓↔↔↑↓↔↔↑↑Social , downward arrows indicate reduced performance on three-chamber preference test of mouse vs object , interactions with novel mouse , or play behavior . Ultrasonic vocalizations ( USV ) are used as an assay of communicative behavior . Ultrasonic , downward arrows indicate longer latency or fewer calls ( adult ) , or more distress calls or longer latency to first call by pups . Repetitive or perseverative behaviors are assayed by grooming and flexibility on maze tasks . Eyeblink learning , downward arrows indicate a decrease in response probability . Eyeblink amplitude , downward arrows indicate a decrease in response amplitude . Eyeblink timing , downward arrows indicate earlier shifts in peak latency and decrease in rise time , while upward arrows indicate later shifts in peak latency and increase in rise time . Maze flexibility , downward arrows indicate impairment on T-maze alternation or reversal or flexibility on a Morris water or Barnes maze . Gross motor functions are assayed by rotarod and gait tasks . Rotarod , table entries indicate differences in the time to fall from an accelerating rotarod . Gait , table entries indicate differences in stance or stride parameters . Maze acquisition , downward arrow indicates impairment of acquisition on Morris water maze , Barnes maze , walking T-maze , or swimming T-maze . Anxiety , up arrows indicate increased freezing and closed-arm preference in elevated plus maze , increased light–dark preference , or decreased open-field behavior . Unless otherwise specified , the downward arrow indicates a significant decrease relative to wild-type , the upward arrow indicates a significant increase relative to wild-type , the horizontal arrow indicates no significant difference relative to wild-type , and ‘nd’ indicates unknown . The ‘*’ in row 5 indicates a difference lacking statistical significance . References: [a] Kouser et al . , 2011; [b] Peñagarikano et al . , 2011; [c] Shahbazian et al . , 2002a; [d] Moretti et al . , 2006; [e] De Filippis et al . , 2010; [f] Tsai et al . , 2012; [g] Nakatani et al . , 2009; [h] Tamada et al . , 2010; [i] Piochon et al . , 2014 . However , our results must also be reconciled with a recent study that started not from ASD genes , but from specific perturbations to cerebellar function ( Galliano et al . , 2013 ) . That work revealed little effect on a variety of standard non-motor tasks , including social , navigational , and memory tasks . Those tasks differ from current tests of autism model face validity . For example , the social assay involved consecutive presentation of mouse/object stimuli , as opposed to the simultaneous choice that occurs in the three-chamber test ( Yang et al . , 2011 ) . Likewise , no test was given for perseveration such as maze reversal or grooming duration ( Tsai et al . , 2012 ) . We suggest that rigorous evaluation of cerebellar involvement in non-motor function will require tasks of greater difficulty and complexity than past practice . We have shown that mouse autism models have difficulty in a cerebellum-dependent form of associating sensory stimuli that are spaced closely in time . The integration of closely timed events across sensory modalities could be critical for statistical learning . Statistical learning can encompass the association of an auditory or visual stimulus to predict some other event , a capacity that is likely to be at the core of the acquisition of language ( Ferguson and Lew-Williams , 2014 ) and other cognitive capacities ( Dinstein et al . , 2012 ) . Such learning is commonly assumed to require neocortical plasticity via Hebbian uninstructed learning . In addition , statistical learning from unexpected events is also efficiently supported by instructed plasticity ( Courville et al . , 2006 ) , a phenomenon for which cerebellar circuit architecture is well-suited ( Marr , 1969 ) . Since the neocortex and cerebellum communicate with one another bidirectionally , these two brain systems might play complementary roles in learning from experience . Projections to forebrain are present in early postnatal life ( Diamond , 2000 ) , and early childhood disruption of the cerebellum affects the development of social cognition and language ( Riva and Giorgi , 2000; Steinlin , 2008; Bolduc et al . , 2012 ) . In this context , eyeblink conditioning is an example of learning from the close timing of two events of different sensory modality , and defects in it may reflect broader difficulties in subsecond temporal sensory association . If such difficulties are present in early stages of autism , the cerebellum may be a potential target for early-life therapeutic intervention .
Cntnap2 mice were bred at Princeton University on a heterozygote–heterozygote strategy using breeding pairs obtained from the Geschwind laboratory at the University of California , Los Angeles ( Peñagarikano et al . , 2011 ) . These animals were originally generated by the Peles laboratory ( Weizmann Institute of Science , Israel ) through the replacement of the first exon of Caspr2 ( Cntnap2 ) using gene-targeting techniques in mice with the imprinting control region ( ICR ) background ( Poliak et al . , 2003 ) . The mice were then outbred on the C57BL/6J background for at least 10 generations and characterized behaviorally ( Peñagarikano et al . , 2011 ) . For behavioral experiments , 39 animals from 17 litters were used . Shank3+/ΔC mice were bred at Princeton University on a heterozygote–heterozygote strategy using breeding pairs acquired from the Worley laboratory at Johns Hopkins University . These mice were generated by the conditional deletion of exon 21 of Shank3 to excise its C-terminal domain , including the Homer-binding domain ( Kouser et al . , 2013; http://jaxmice . jax . org/strain/018389 . html ) . The mice were generated on a mixed background and backcrossed on a C57BL/6J background for at least five generations . Only heterozygotes of the C-terminal mutation were used ( Durand et al . , 2007 ) . For behavioral experiments , 38 animals from 16 litters were used . Mecp2R308/Y mice were bred at Princeton University on a heterozygote-wild-type strategy using a breeding pair acquired from Jackson Laboratories ( B6 . 129S-Mecp2tm1Hzo/J , stock no . : 005439 ) . Mice on the 129/SvEv background have a truncating mutation of Mecp2 introduced through the insertion of a premature stop after codon 308 ( Shahbazian et al . , 2002a ) . These mice were backcrossed on the C57BL/6J background for at least 10–12 generations . Because these mice show a regressive phenotype , they were tested at 16–20 weeks , an age at which the mice begin showing cognitive symptoms and minor motor dysfunction ( ‘early symptomatic’ to symptomatic phase: Shahbazian et al . , 2002a; Moretti et al . , 2006; De Filippis et al . , 2010 ) . For behavioral experiments , 28 animals from 11 litters were used . The Tsc1 mice were bred at Princeton University from breeding pairs on a mixed ( C57BL/gJj , 129 SvJae , BALB/cJ ) background acquired from the Sahin laboratory at Boston Children's Hospital , Harvard Medical School ( Tsai et al . , 2012 ) . These mice were originally generated by crossing L7/Pcp2-Cre mice with Tsc1flox/flox mice ( Tsai et al . , 2012 ) . For the present study , the offspring of this cross were crossed to produce the L7/Pcp2Cre::Tsc1flox/+ ( heterozygous ) and L7/Pcp2Cre::Tsc1flox/flox ( homozygous ) animals . Littermate controls were pooled from Tsc1+/+ ( pure wild-type ) , Tsc1flox/+ , L7Cre;Tsc1+/+ ( L7Cre ) , and Tsc1flox/flox ( flox ) mice . For behavioral experiments , 34 animals from 18 litters were used . patDp/+ ( 15q11-13 duplication ) mice were acquired from the Hansel laboratory at the University of Chicago and the Takumi laboratory at Hiroshima University and tested as previously reported ( Nakatani et al . , 2009; Piochon et al . , 2014 ) . Data from the eyeblink conditioning experiments described in Piochon et al . ( 2014 ) are available upon request from the corresponding author . For all experiments , we used 2- to 4-month-old males with matched littermates unless otherwise indicated . To ensure that the ages of the mice did not affect the results , we corrected our statistical tests of average CR performance , response probability , and response amplitude across sessions 9–12 and the CR timing parameters , using analysis of covariance tests with age ( days ) as a covariant with post hoc Tukey's tests ( Piochon et al . , 2014 ) . This analysis produced no changes in statistical significance of the findings reported throughout this paper ( Tukey's test , p > 0 . 05 in all instances ) . Mice were group-housed ( at ≥ 8 weeks of age ) and maintained on a 12-hr reverse light–dark cycle with ad libitum access to food and water . All experiments were performed according to protocols approved by the Princeton University Institutional Animal Care and Use Committee . Each mouse was head-fixed above a stationary , freely rotating foam wheel , which allowed it to locomote throughout the experiment ( Figure 2A ) . In this position , the US ( airpuff ) could be delivered from a consistently to the eye through a blunted 27-gage needle . The eyelid deflection was detected using a Hall effect sensor ( AA004-00 , NVE Corporation , Eden Prairie , MN ) that was mounted above the same eye ( Koekkoek et al . , 2002 ) . Prior to placement in the experimental apparatus , each mouse was briefly anesthetized with isoflurane and a small neodymium magnet ( 3 mm × 1 mm × 1 mm , chrome , item N50 , Supermagnetman , Birmingham , AL ) was attached to the lower eyelid with cyanoacrylate glue ( Krazy Glue , Westerville , OH ) . The sensor provided a readout of eyelid position by linearly converting a change in magnetic field due to the displacement of the magnet relative to the sensor a change in voltage . The CS ( ultraviolet LED ) was also delivered to the ipsilateral eye . The animals were allowed to habituate to this apparatus for at least 195 min over 3–5 days . Following habituation , acquisition training took place over 12 training sessions ( 1 session/day , 6 days/week ) , during which the animals received 22 blocks of 10 trials each . CSs ( ultraviolet light , 280 ms ) were paired with an aversive US ( airpuff delivered by a blunted needle to the cornea , 30–40 psi , 30 ms , co-terminating with the CS ) . Ultraviolet light is in the sensitive range of laboratory mice ( Jacobs et al . , 2001 ) . Each block contained 9 paired US-CS trials and 1 unpaired CS trial , arranged pseudorandomly within the block ( Figure 2B ) . Each trial was separated by an interval of at least 12 s ( see below ) . Following acquisition training , the mice received extinction and reacquisition training . Extinction training took place over 4 sessions ( 1 session/day ) consisting of 22 blocks of 10 trials each . Each block contained five unpaired CS trials and five unpaired US trials , arranged pseudorandomly within the block . Reacquisition training took place over 3 sessions , and the animals received the same training sequence as in acquisition training . Trials were triggered automatically using a custom MATLAB ( Mathworks , Natick , MA ) graphical user interface . Stimuli were triggered by Master-8 ( AMPI , Inc . , Jerusalem , Israel ) via the data acquisition system ( National Instruments , Austin , TX ) . ( Scripts for data collection and analysis along with sample data are available at https://github . com/akloth0325/eyeblink-conditioning . ) The Master-8 controlled the stimulus timing and sent square signals to an ultraviolet LED and a Toohey Pressure System IIe spritzer ( Toohey Co . , Fairfield , NJ ) to generate the CS and US , respectively . The output from the Master-8 was returned to the data acquisition system . The voltage output of the Hall-effect sensor was filtered and amplified ( band-pass filtered from 0 . 01 Hz to 4 kHz , gain adjusted to signal quality ) and sent to the data acquisition system . The beginning of an individual trial was subject to the following criteria . First , at least 12 s must have elapsed since the last trial . Time was added to the interval between any two consecutive trials according to the stability of the eyelid position signal: if the eyelid position signal ( the ‘baseline’ signal ) strayed outside an experimenter-determined range during 1 s prior to the planned delivery of the CS , an additional 1 s was added to the intertrial interval until this criterion was met , after which the trial was initiated . The experimenter used the voltage range of UR ( baseline to peak ) during 3–12 unpaired US trials delivered at the beginning of the session to determine an acceptable voltage range for baseline activity prior to the beginning of each trial; typically , this range was ±10% of the average size of the UR . The data from each trial were normalized prior to analysis . For the paired US-CS trials , the eyelid position was normalized to the range between the baseline and the peak amplitude of the UR during the trial . For the unpaired CS trials , the eyelid position was normalized to the range between the baseline and the UR peak for the most recent US-CS trial . Then , the response probability and response amplitude for a single training session were calculated . This normalization scheme yielded results that were not significantly different from those acquired by normalizing to a sessionwide average UR ( paired t-tests within groups for CR performance , response probability , and response amplitude on session 12 , p > 0 . 05 in all instances ) . The analysis method was inspired by brain slice recording of single-synapse plasticity ( O'Connor et al . , 2007 ) to analyze the full range of detectable responses to a CS ( Figure 2 ) . The peak response size for the period between 100 ms and 280 ms after the onset of the CS was collected for every trial during each session , and a probability distribution was computed from these data . The part of the probability distribution that lay below a peak response size of 0 was considered the ‘non-response distribution’ . This part of the distribution plus a reflection of this distribution for a positive peak response size was subtracted from the original probability distribution . The remaining distribution was the ‘response distribution’ . The response probability for the given session was the area under the response distribution . The response amplitude was computed as the center of mass for the response distribution . Response timing was analyzed from the unpaired CS trials . The normalized response during the CS scored as a CR if it exceeded 0 . 15 between 100 ms and 400 ms after the onset of the CS and remained below 0 . 05 between 0 ms and 99 ms . ( Again , trials for which the responses exceeded 0 . 05 between 0 ms and 99 ms after the onset of the CS were excluded . ) As sensory and motor tests , motor function was analyzed using unpaired US trials from the first session of training . Peak time , rise time , and onset time were calculated on smoothed individual traces as described above , within 75 ms of US onset . Photic eyelid opening was analyzed during the first session of eyeblink conditioning , during which no conditioned eyeblink was generated . Using the normalized individual eyelid deflection traces , deflections that were more than 5% below the baseline 70–250 ms after the CS onset—but not before—were counted . Mice underwent one session of habituation training ( 1 day ) , four sessions of acquisition training ( the next day ) , and two sessions of testing ( the following day ) in a water Y-maze ( custom made: 32 cm arms positioned at 120° from one another , made of semitransparent polycarbonate ) filled with opaque water ( non-toxic white tempera paint was added to achieve opacity ) . On the habituation day , mice were dropped into 10 cm of water in order to measure their swimming ability . The habituation day consisted of three 60-s trials , each trial starting from one arm of the maze . No platform was hidden beneath the surface of the water during this phase of training . During acquisition , the mice were randomly sorted into leftward-going or rightward-groups; this selection determined in which arm the platform would be hidden beneath the surface of the opaque water for each mouse . For five trials per training session , the mice were dropped into the arm closest to the experimenter and were given 40 s to find the platform . On the following day , the animals underwent two more sessions of the same protocol to test memory . The swimming trajectories of the mice were captured on video and were processed by a custom Python script ( available at https://github . com/bensondaled/three-chamber ) to determine whether the animal found the platform on a given trial . Excursions to the wrong arm of the maze were counted as incorrect . Results were reported the fraction of correct trials to valid trials , where valid trials included all trials on which the animal successfully to swam to either the left or the right arms of the Y-maze . Mice videotaped during two runs along a 100-cm track over a plexiglass surface . Each run was initiated with an airpuff to the hindlimb . Runs were videotaped ( iPhone 6 , 40 frames/s ) from below , and light was sourced from below . After being separated using a custom MATLAB scripts , JPEG stacks were analyzed using FIJI Manual Tracker ( LOCI , Madison , WI ) for the centroid of each paw . Stance and stride parameters were calculated from four paw centroid trajectories ( ≥10 strides per run ) for each animal . Mice were fitted with a 1′′ × ½′′ × 1/32′′ custom titanium headplate ( Ozden et al . , 2012; Heiney et al . , 2014 ) . During the surgery , each mouse was anesthetized with isoflurane ( 1–2% in oxygen , 1 l/min , for 15–25 min ) and mounted in a stereotaxic head holder ( David Kopf Instruments , Tujunga , CA ) . The scalp was shaved and cleaned , and an incision was made down the midline of the scalp . The skull was cleaned and the scalp margin was kept open with cyanoacrylate glue ( Krazy Glue ) . The center of the headplate was positioned over bregma and attached to the skull with quick-drying dental cement ( Metabond , Parkell , Edgewood , NY ) . Following the surgery , the mice received a non-steroidal anti-inflammatory drug ( 0 . 1 ml , 50 mg/ml Rimadyl [carprofen , Zoetis , Florham Park , NJ] ) subcutaneously and were allowed to recover for at least 24 hr . Tissue from separate groups of mice for each cohort was used to analyze the morphology of the cerebellum . For Nissl staining and immunohistochemistry , the mice were anesthetized with 0 . 15 ml ketamine-xylazine ( 0 . 12 ml 100 mg/ml ketamine and 0 . 80 ml mg/ml xylazine diluted 5× in saline ) and transcardially perfused with 4% formalin in Delbucco's phosphate buffered saline ( PBS ) . The brain was extracted and stored at 4°C in 4% formalin in PBS overnight . Then , the brains were split into hemispheres . The hemispheres used for Nissl staining were stored in 0 . 1% sodium azide in PBS at 4°C until vibratome sectioning . The hemispheres used for immunohistochemistry were prepared for cryosection . These hemispheres were stored in 10% sucrose in PBS at 4°C overnight and were blocked in a solution of 11% gelatin/10% sucrose . The block was immersed in a mixture of 30% sucrose/10% formalin in PBS for 2 hr and then stored in 10% sucrose in PBS at 4°C for up to 2 weeks . For Golgi-Cox staining , the mice were anesthetized with 0 . 15 ml ketamine-xylazine ( 0 . 12 ml of 100 mg/ml ketamine and 0 . 80 ml mg/ml xylazine diluted 5× in saline ) and decapitated immediately . The brain was removed quickly in ice-cold PBS and processed using the FD Rapid GolgiStain kit ( FD Neurotechnologies , Inc . , Columbia , MD ) , according to the kit instructions . Brain hemispheres used for Nissl staining were blocked sectioned sagittally on a vibratome at a thickness of 70 µm . The sections were mounted on Fisherbrand SuperFrost microscope slides ( Thermo Fisher Scientific , Waltham , MA ) and allowed to dry at room temperature overnight . Then , they were Nissl stained with cresyl violet according to standard procedures and coverslipped with Permount ( Thermo Fisher Scientific , Waltham , MA ) . The sections were imaged at 5× magnification and ‘virtual slices’ were constructed from serial images captured by the MicroBrightField software Stereo Investigator ( MBF Biosciences , Williston , VT ) . The thicknesses of the molecular layer and the granule layer were measured on anterior and posterior portions of vermal sections of the cerebellum at 150-µm intervals using ImageJ ( National Institutes of Health , Bethesda , MD ) . Brains used for Golgi-Cox staining were sectioned sagittally on a vibratome at a thickness of 120 μm . The sections were mounted on slides and allowed to dry in the dark at room temperature overnight . Then , they were processed for Golgi staining according to the instructions for the FD Rapid GolgiStain kit and coverslipped with Permount . The sections were imaged at 20× and 40× and images of Golgi-stained PCs and captured by the MicroBrightField software Stereo Investigator . The cross-sectional area of the soma and the maximum height , maximum width , and the cross-sectional area of the PC dendritic arbor were measured using ImageJ . In addition , the complexity of the PC dendritic arbor was determined using Sholl analysis ( Sholl , 1956 ) using ImageJ; briefly , the number of intersections of the dendritic arbor with concentric circles drawn at 12-μm intervals from the soma was counted ( e . g . , see Figure 5D ) . Spines on the distal dendrites were counted in an unbiased manner from these cells ( e . g . , see Figure 5D ) . The spines on distal dendrites of every fifth branchlet ( random starting point ) were counted and the dendrite length was measured . Brain hemispheres used for immunohistochemistry were sectioned sagittally on a cryotome ( −20°C ) at a thickness of 30 μm and stored in PBS . Sections were immunostained with rabbit anti-calbindin ( 1:2000 , Invitrogen , Waltham , MA ) as the primary antibody and donkey anti-rabbit AlexaFluor 488 ( 1:300 , Invitrogen , Waltham , MA ) . Sections were counterstained with 4' , 6-diamidino-2-phenylindole ( DAPI , 1:100 , Invitrogen , Waltham , MA ) . The sections were mounted on Fisherbrand SuperFrost microscope slides ( Thermo Fisher Scientific , Waltham , MA ) slides and coverslipped with VectaShield without DAPI ( Vector Labs , Burlingame , CA ) . The sections were imaged at 10× magnification on an epifluorescence microscope and ‘virtual slices’ were constructed from serial images taken by the MicroBrightField software Stereo Investigator . PCs were counted and the length of the PC layer was measured for each sample using ImageJ . All data and samples were analyzed with by an experimenter who was blinded to genotype . All pairwise statistical tests were unpaired two-sample t-tests unless otherwise noted . Time course data were analyzed using two-way ANOVAs with repeated measures; main genotype effects were reported regardless of significance , whereas main session effects ( which would indicate a learning effect through time ) are significant and session × genotype interactions are not significant unless otherwise indicated . When comparing a single measurement across more than two groups , one-way analyses of variance were performed with Bonferroni post hoc tests with planned comparisons . Correction for potentially confounding variables ( i . e . , age ) was performed using analysis of covariance tests with the confounding variable as the covariant and followed by Tukey's post hoc tests . Tests were performed using GraphPad Prism 6 ( GraphPad Software , Inc . , La Jolla , CA ) and SPSS 21 ( IBM , Armonk , NY ) . All data are displayed as mean ± standard error of the mean ( SEM ) unless otherwise noted in the text or legend . Where significant differences were discovered with pairwise comparisons , effect sizes are also reported as Cohen's d′ . | On a windy day , hearing the sound of wind makes many individuals squint in anticipation in order to protect their eyes . Linking two sensations that arrive within a split second of one another , such as sound and the feeling of wind , is a type of learning that requires the cerebellum , a region found at the base of the brain . When done in a laboratory setting , this particular form of learning has been dubbed eyeblink conditioning . Individuals with autism tend to have difficulties with appropriate matching of different senses . For example , they have trouble identifying a video that goes with a spoken soundtrack . They also do not learn eyeblink conditioning the same way that other individuals do . However , it is not known which circuits in the brain are responsible for their difficulty . Kloth et al . now investigate this issue by asking whether versions of genes that increase the risk of autism in humans also disrupt eyeblink conditioning in mice . They tested five types of mouse model , each with a different genetic mutation that has previously been linked to autism . All five of these mutations cause defects in different cell types of the cerebellum , and all mice have abnormal social and habitual behaviors , similar to autistic people . The tests involved shining a bright light at the mice , which was followed , a split second later , by a puff of air that always causes the mice to blink . After this had occurred dozens of times , the mice started to blink earlier , as soon as the light appeared , in anticipation of the puff of air . To test whether the mice had successfully learned to respond to just the bright light , the light was also occasionally flashed without a puff of air . Kloth et al . found that the mice generally performed poorly in eyeblink conditioning , although in different ways depending on which cell types of the cerebellum were affected by the genetic mutations . Some mice blinked too soon or too late after the light appeared; others blinked weakly or less frequently; and some did not blink at all . This suggests that autism can affect the processing of sensory information in the cerebellum in different ways . This work is important because it demonstrates that a form of split-second multisensory learning is generally disrupted by autism genes . If defects in cerebellar learning are present early in life , they could keep autistic children from learning about the world around them , and drive their developing brains off track . Hundreds of autism genes have been found . Linking these genes to a single brain region identifies the cerebellum as an important anatomical target for future diagnosis and intervention . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"neuroscience"
] | 2015 | Cerebellar associative sensory learning defects in five mouse autism models |
Many animal groups exhibit rapid , coordinated collective motion . Yet , the evolutionary forces that cause such collective responses to evolve are poorly understood . Here , we develop analytical methods and evolutionary simulations based on experimental data from schooling fish . We use these methods to investigate how populations evolve within unpredictable , time-varying resource environments . We show that populations evolve toward a distinctive regime in behavioral phenotype space , where small responses of individuals to local environmental cues cause spontaneous changes in the collective state of groups . These changes resemble phase transitions in physical systems . Through these transitions , individuals evolve the emergent capacity to sense and respond to resource gradients ( i . e . individuals perceive gradients via social interactions , rather than sensing gradients directly ) , and to allocate themselves among distinct , distant resource patches . Our results yield new insight into how natural selection , acting on selfish individuals , results in the highly effective collective responses evident in nature .
In many highly coordinated animal groups , such as fish schools and bird flocks , the ability of individuals to locate resources and avoid predators depends on the collective behavior of the group . For example , when fish schools are attacked by predators , 'flash expansion' ( Pitcher et al . , 1993 ) and other coordinated collective motions , made possible above a certain group size , reduce individual risk ( Handegard et al . , 2012 ) . Similarly , fish can track dynamic resource patches far more effectively when they are in a group ( Berdahl et al . , 2013 ) . When an individual responds to a change in the environment ( e . g . , predator , resource cue ) , this response propagates swiftly through the group ( Rosenthal et al . , 2015 ) , altering the group’s collective motion . How are such rapid , coordinated responses possible ? These responses may occur , in part , because the nature of social interactions makes animal groups highly sensitive to small changes in the behavior of individual group members; theoretical ( Couzin et al . , 2002; D’Orsogna et al . , 2006; Kolpas et al . , 2007 ) and empirical ( Tunstrøm et al . , 2013; Buhl et al . , 2006 ) studies of collective motion have revealed that minor changes in individual behavior , such as speed ( Tunstrøm et al . , 2013 ) , can cause sudden transitions in group state , reminiscent of similarly sudden phase transitions between collective states in physical systems ( such as the solid-liquid-gas transitions as a function of increasing temperature ) . It has been proposed that individuals may trigger such changes in collective state by responding to the environment , thereby initiating a coordinated response at the group level ( e . g . , Couzin et al . ( 2002 ) ; Kolpas et al . ( 2007 ) ; Couzin and Krause , 2003 ) . This mechanism requires that the behavioral rules of individual animals within a population have evolved in a way that allows groups to transition adaptively among distinct collective states . The evolutionary processes that could lead to this population-level property , however , remain poorly understood . The feedback between the behavioral phenotypes of individuals , the collective behaviors that these phenotypes produce , and individual-level fitness consequences has made it challenging to study how complex collective behaviors evolve ( Torney et al . , 2011 ) . Many species , including fish and birds , form groups in which members have low genetic relatedness , which implies that kin selection alone cannot explain the evolution of collective behavior . Moreover , while natural selection acts on the behavioral phenotypes of selfish individuals , collective behaviors are group-level , or perhaps even population-level , properties rather than heritable individual phenotypes . To understand how collective behaviors evolve , then , one must first understand the mapping between individual phenotypes and collective behavior , and between collective behavior and individual fitness . Here , we take advantage of detailed studies of the social interaction rules and environmental response behaviors of schooling fish ( Berdahl et al . , 2013; Katz et al . , 2011 ) to develop a biologically-motivated evolutionary model of collective responses to the environment . Using analytical methods and evolutionary simulations , we study how individual behavioral rules produce collective behaviors , and how collective behaviors , in turn , govern the fitness and evolution of selfish individuals . To relate individual and collective behaviors to fitness , we consider a fundamental task faced by fish and other motile organisms: finding and exploiting dynamic resources ( Stephens et al . , 2007 ) . In our model , individuals respond to the locations of near neighbors and also to local measurements of resource quality . Each individual achieves a fitness determined by the resource level it experiences over its lifetime . We use this framework to explore the evolution of complex collective responses to the environment , and how such responses are related to transitions in collective state .
We model the movement behaviors of each individual in a population of size N using two experimentally-motivated ( Berdahl et al . , 2013; Katz et al . , 2011 ) behavioral rules: a social response rule and an environmental response rule . The social response rule is motivated by experimental studies of pairwise interactions among golden shiners ( Notemigonus crysoleucas ) ( Katz et al . , 2011 ) . Individual fish avoid others with whom they are in very close proximity . As the distance between individuals increases , however , interactions gradually change from repulsive to attractive , with maximum attraction occurring at a distance of two-four body lengths . For longer distances , individuals still attract one another but the strength of attraction decays in magnitude ( Appendix section 1; Katz et al . , 2011 ) . As found in experimental studies of golden shiners ( Katz et al . , 2011 ) and mosquitofish ( Gambusia holbrooki ) ( Herbert-Read et al . , 2011 ) there need not be an explicit alignment tendency; rather alignment can be an emergent property of motion combined with the tendencies for repulsion and attraction described above . To capture these observed social interactions ( or ‘social forces’ ) , we model the acceleration of individuals using a force-based method ( Katz et al . , 2011 ) . The ith individual responds to its neighbors using the following rule: ( 1 ) Fs , i=−∇∑j∈NiCre−|xi−xj|/lr−Cae−|xi−xj|/la , where Fs , i is the social force on the ith individual , xi is the position of the ith individual , ∇ is the two-dimensional gradient operator , the term in brackets is a social potential , Ca , Cr , la , and lr are constants that dictate the relative strengths and length scales of social attraction and repulsion , and the set Ni is a set of the k nearest neighbors of the ith individual , where a neighbor is an individual within a distance of lmax of the focal individual . Equation 1 does not include explicit alignment with neighbors . A similar model is discussed in D’Orsogna et al . ( 2006 ) . In Equation 1 , lmax determines the length scale over which individuals are influenced by social interactions . If lmax is greater than lr but less than la , individuals repel one another at short distances but do not attract one another . We refer to such individuals as asocial ( Appendix section 1 ) . If lmax is greater than both lr and la , individuals repel one another at short distances and are attracted to one another at intermediate distances as observed by Katz et al . ( 2011 ) . Finite k ensures that individuals can only respond to a limited number of their neighbors in crowded regions of space and provides a simplified model of sensory-based social interactions ( e . g . , Rosenthal et al . ( 2015 ) ; Strandburg-Peshkin et al . ( 2013 ) ) . Finite k also ensures that individuals are limited to finite local density ( Appendix section 3 ) . To model the response of individuals to the environment , we develop an environmental response rule based on experimentally-observed environmental responses of golden shiners ( Berdahl et al . , 2013 ) . In particular , in a dynamic , heterogeneous environment , individual golden shiners respond strongly to local sensory cues by slowing down in favorable regions of the environment , and speeding up in unfavorable regions . In contrast , fish respond only weakly to spatial gradients in environmental quality and instead adjust their headings primarily based on the positions of their near neighbors . Accordingly , we model the ith individual’s environmental response as a function of the level of an environmental cue ( in this case , the level of a resource ) at its current position: ( 2 ) Fa , i=[Ψi ( S ( xi ) ) −η|vi|2]vi|vi| , where Fa , i is the autonomous force the ith individual generates by accelerating or decelerating in response to the environment , Ψ ( ⋅ ) is a monotonically decreasing function of the value of an environmental cue , S ( xi ) is the cue value at the ith individual’s position , η is a damping term that limits individuals to a finite speed , and vi is the ith individual’s velocity . In the absence of social interactions , individuals travel at preferred speed vi*=Ψi/η ( for Ψi>0 ) . Changes in speed are crucial in the schooling behavior of fish ( Tunstrøm et al . , 2013; Berdahl et al . , 2013 ) , and as we show below , are also responsible for generating effective collective response in our model . Following the experimental results in Berdahl et al . ( 2013 ) we assume that individuals do not change their headings in response to the cue . In what follows , we refer to 'cue' and 'resource' interchangeably as we model the case where the cue is the resource itself ( see e . g . , Torney et al . ( 2009 ) ; Hein and McKinley ( 2012 ) for cases where the cue is not a resource ) . Combining social and environmental response rules yields two equations that govern each individual’s movement ( in two dimensions ) : ( 3 ) dxidt=vi , and ( 4 ) mdvidt=Fs , i+Fa , i , where m is mass . D’Orsogna et al . ( 2006 ) explores the behavior of a similar model with Ψi=Ψ constant over the full parameter space . Here we focus on a parameter regime that yields behavioral rules that match the experimental observations of Katz et al . ( 2011 ) and Berdahl et al . ( 2013 ) . We simulate a discretized version of the system described by Equations 3 and 4 . In particular , we choose a time step , τ , within which the acceleration due to social influences ( Equation 1 ) and resource value S ( xi ) are assumed to be constant . Positions , speeds , and accelerations of all individuals at time t+τ are then given by the solutions to Equations 3 and 4 at time t+τ , with the values of S ( xi ) and |xi−xj| determined at time t . A navigational noise vector of small magnitude γ and uniform heading 0 to 2π is added to the velocity of each agent at each time step . Taking the limit as τ goes to zero means that individuals are constantly acquiring information and instantaneously altering their actions in response . In Appendix section 3−6 , we analyze a continuum approximation of this limiting model and below we discuss results of this analysis alongside simulation results . The social interaction rule allows us to build an interaction network for the entire population . Two individuals are socially connected if at least one of them influences the other through Equation 1 . We define a 'group' as a set of individuals that belong to the same connected component in this network . The natural environments in which organisms live are often heterogeneous and dynamic ( Stephens et al . , 2007 ) . Consequently , we simulate populations of individuals in dynamic landscapes , where individuals make decisions in response to local sensory cues ( local measurements of a resource ) and these decisions have fitness consequences for the individuals within the population ( Guttal and Couzin , 2010; Torney et al . , 2011 ) . In keeping with experimental observations ( Berdahl et al . , 2013 ) , we assume individuals follow a simple environmental response function: Ψi=ψ0−ψ1S ( xi ) , where ψ0 dictates the ith individual’s preferred speed when the level of the environmental cue is zero and ψ1 determines how sensitive the ith individual is to the cue value ( Berdahl et al . , 2013 ) . Rather than prescribing values of ψ0 and ψ1 , we use an evolutionary framework similar to that developed by Guttal and Couzin ( 2010 ) to allow these two behavioral traits to evolve along with the maximum interaction length lmax , which determines whether individuals are social ( lmax> length scale of social attraction ) or asocial ( lmax< length scale of social attraction , Appendix section 1 ) . In each generation , N individuals are located in a two-dimensional environment in which each point in space is associated with a resource value that changes over time ( see Materials and methods ) . Individuals move through the environment using the interaction rules described above , and each individual has its own value of the ψ0 , ψ1 , and lmax parameters . At the end of each generation , we compute each individual’s fitness as the mean value of the resource it experienced during that generation . Each individual then reproduces with a probability proportional to its relative fitness within the population . N offspring comprise the next generation where each offspring inherits the traits of its parent modified by a small mutation ( Appendix section 2 ) . For reference , we compare the evolution of populations in which ψ0 , ψ1 , and lmax are allowed to evolve , to the evolution of populations of asocial individuals , for which lmax is set to a constant ( Appendix section 1 ) .
In populations of asocial individuals , the baseline speed parameter and environmental sensitivity increase consistently through evolutionary time ( Figure 1A–B ) . Asocial individuals move through the environment , slowing down in regions where the resource value is high and speeding up when the resource value is low ( Video 1 ) . As one would expect from random walk theory ( Schnitzer , 1993; Gurarie and Ovaskainen , 2013 ) , individuals more rapidly encounter regions of the environment with high resource value when they travel at high preferred speeds ( Equation A65; Gurarie and Ovaskainen , 2013 ) , and the more they reduce speed in regions of the environment with high resource quality , the more time they spend in these regions ( Schnitzer , 1993 ) . Because of these two effects , the fittest asocial individuals have high baseline speeds ( i . e . , high ψ0 ) and accelerate and decelerate rapidly in response to changes in the resource value ( i . e . , high ψ1; Figure 1A–B , Appendix ) . 10 . 7554/eLife . 10955 . 003Figure 1 . Evolution of behavioral rules . ( A , B ) show evolutionary dynamics of populations of asocial individuals ( i . e . , maximum length scale of social interactions lmax fixed; see text ) . ( C-E ) show evolutionary dynamics of individuals in which the maximum length scale of social interactions lmax is allowed to evolve . Brightness of color indicates the frequency of a phenotype in the population . In asocial populations , baseline speed parameter ψ0 ( A ) and environmental sensitivity ψ1 ( B ) increase continually through evolutionary time . When lmax is allowed to evolve ( C ) , individuals quickly become social ( lmax approaches maximum allowable value of 30 ) , and baseline speed parameter ψ0 ( D ) and environmental sensitivity ψ1 ( E ) stabilize at intermediate values . Mean fitness of social populations ( F , red points ) is over five times higher than mean fitness of asocial populations ( F , blue points ) , and the coefficient of variation in fitness is over four times lower in social populations ( F inset ) . Unless otherwise noted , parameter values in all figures are as follows: C=CrCa=1 . 1 , l=lrla=0 . 13 , N=500 , k=25 , γ=0 . 01 , τ=1 , m=1 , ν=1 , ρ=0 . 16 , M=2 , λ0=10 , λ1=20 , α= ( 1 , 0 ) , β=0 . 1 , and τp=1500 . DOI: http://dx . doi . org/10 . 7554/eLife . 10955 . 00310 . 7554/eLife . 10955 . 004Video 1 . Asocial population . Responses of population of asocial individuals ( points ) and dynamic resource peak ( resource value shown in grayscale; dark regions have high resource value , light regions have low resource value ) . Length of tail proportional to speed . Peak centroid moves according to 2D Brownian motion with drift vector α and standard deviation β ( see Materials and methods ) . In Videos 1–4 , view is zoomed in to area surrounding moving resource peak ( field of view is 50lr×50lr , where lr is the length scale of repulsion; full environment is projected onto a torus with edge length 346lr ) . Behavioral parameters as follows: Cr=1 . 1 , Ca=1 , lr=1 , la=7 . 5 , γ=0 . 01 , τ=1 , m=1 , η=1 , ψ0=3 , ψ1=2 . 54 . Environmental parameters in Videos 1–4 are: ρ=0 . 16 , N=300 , M=2 , λ0=10 , λ1=20 , α=[0 . 06 0] , β=0 . 5 . DOI: http://dx . doi . org/10 . 7554/eLife . 10955 . 004 When populations are allowed to evolve sociality , the evolutionary process selects for very different behaviors ( Figure 1C–E ) . Selection quickly favors sociality , and individuals evolve large maximum interaction lengths ( Figure 1C ) . Over evolutionary time , selection removes individuals with high and low values of ψ0 and ψ1 from the population and an evolutionarily stable state ( ESSt; Maynard Smith , 1982 ) emerges that is characterized by a single mode at the dominant value of each trait ( Figure 1D–E; Appendix section 2 ) . The ESSt resulting from selection on ψ0 , ψ1 , and lmax is robust in that it is resistant to invasion by phenotypes near the ESSt , and by invaders with trait values far from the ESSt ( Appendix section 2 ) . Throughout evolution , populations of social individuals achieve mean fitness values that are approximately five times higher than those of asocial populations , and a coefficient of variation in fitness approximately four times lower than that of asocial individuals ( Figure 1F ) . Notably , a single individual drawn from a population at the ESSt can invade a resident population of asocial individuals and the social strategy quickly sweeps through the population ( Appendix section 2 ) . To understand why this invasion occurs , consider a population of asocial individuals that slow down in favorable regions of the environment . If the environment does not change too rapidly , such individuals will accumulate in regions where the resource level is high . This phenomenon has been studied mathematically in the context of position-dependent diffusion ( Schnitzer , 1993 ) , and will occur , in general , when individuals lower their speeds in response to the value of an environmental cue . A social mutant that responds to the environment , and to its neighbors , can take advantage of the correlation between density and resource quality by climbing the gradient in the density of its neighbors ( Equation 1 ) . In this case , the positions of neighbors contain information about the value of resources and social mutants quickly invade asocial populations leading to a rapid increase in mean fitness ( Appendix section 2 ) . The high fitness of the evolved phenotype is due , in part , to a collective resource tracking ability , similar to that found in golden shiners ( Berdahl et al . , 2013 ) . Evolved individuals can find and track resource peaks as they move through the environment ( Figure 2A , Video 2; Materials and methods ) , whereas asocial individuals and social individuals with trait values far from the ESSt cannot ( Videos 1 , 3–4 ) . Tracking occurs via a dynamic process . Individuals near the edge of the peak move rapidly , whereas individuals nearer to the peak center ( where the resource value is high ) move slowly ( Equation 2 ) . As in fish schools ( Berdahl et al . , 2013 ) , individuals turn toward near neighbors ( Equation 1 ) and travel toward the peak center . This collective tracking behavior is particularly important when the resource field changes rapidly over time . As a resource peak moves , individuals at its trailing edge experience a resource value that becomes weaker through time ( Figure 2A ) . As the resource value becomes weaker , these individuals accelerate ( Equation 2 ) , but turn toward neighbors on the peak ( Equation 1 ) and thus travel toward the moving peak ( Figure 2A ) . When the environment contains multiple resource peaks , evolved populations fuse spontaneously to form groups whose sizes correspond to that of the peak they are tracking ( Figure 2B ) , even though no individual is able to assess peak size , or know whether there are multiple peaks in the environment . This behavior is consistent with recent sonar observations of foraging marine fish showing that fish form shoals that match the sizes of dynamic resource patches ( Bertrand et al . , 2008; Bertrand et al . , 2014 ) . Our model demonstrates that collective tracking behavior similar to that observed in real fish schools can evolve through selection on the decision rules of individuals . 10 . 7554/eLife . 10955 . 005Figure 2 . Collective tracking of dynamic resource and length-scale matching . ( A ) Sequence ( left to right , top to bottom ) of individuals interacting with moving resource peak ( resource value in grayscale , darker = higher resource value ) . Peak is drifting to the right ( grey arrow ) . Colors indicate the regime into which each agent falls ( red: Ψ>2 . 95 , blue: 0<Ψ<2 . 95 , green: Ψ<0 ) . Length of tail is proportional to speed . Peak centroid moves according to 2D Brownian motion with drift ( see Materials and methods ) . ( B ) When environments contain multiple resource peaks , evolved populations divide into groups that match peak sizes , e . g . , in a two-peak environment , the size of group on each peak is proportional to peak size . Total size of two peaks is constant so that the larger the first peak ( Peak 1 , x-axis ) , the smaller the second peak . Peak size computed as the integral of the resource value over the entire peak ( see Materials and methods ) . Group size is mean size of the group nearest each peak ( mean taken over the last 2 , 500 time steps of each simulation ) . Points ( and error bars ) represent mean ( ± 2 standard errors ) of 1 , 000 simulations for each combination of peak sizes . Parameters as in Figure 1 with M=2 and values of ψ0 , ψ1 , and lmax taken from a population in the ESSt . DOI: http://dx . doi . org/10 . 7554/eLife . 10955 . 00510 . 7554/eLife . 10955 . 006Video 2 . Population at the evolutionarily stable state ( ESSt ) . Responses of population of individuals evolved for 1500 generations to the ESSt to dynamic resource peaks . Behavioral parameters as in Video 1 with k=25 , ⟨ψ0⟩=3 , ⟨ψ1⟩=2 . 45 , and ⟨lmax⟩=29 , where ⟨⋅⟩ denotes mean over the population . Note rapid accumulation of individuals near peaks and dynamic peak-tracking behavior of groups . DOI: http://dx . doi . org/10 . 7554/eLife . 10955 . 006 That individuals in evolutionarily stable populations have intermediate baseline speeds and intermediate environmental sensitivities ( Figure 1D–E ) raises a question: what determines the evolutionarily stable values of these traits ? It is tempting to conclude that these trait values are determined by the nature of the environment alone . However , the fact that the evolutionary trajectories of social and asocial populations are so different ( Figure 1 ) , suggests that the collective behaviors discussed above strongly influence the outcome of evolution . Analysis of Equations 1–4 reveals that the preferred speed parameter divides the dynamical behavior of populations into distinct collective states ( Figure 3; analysis in Appendix section 5 ) . For Ψ<0 , individuals have a preferred speed of zero and the inter-individual distances are governed by initial conditions . In this state , individuals resist acceleration due to social interactions . For small Ψ>0 , individuals form relatively dense groups that move through the environment as collectives , either milling , swarming , or translating ( D’Orsogna et al . , 2006 ) , the collective motions exhibited by real schooling fish ( Tunstrøm et al . , 2013 ) . Individual speeds are relatively low and inter-individual distances are short . For large Ψ , inter-individual distances are large , and individuals move through the environment quickly . Dynamic changes among theses states are evident in Video 2 . These collective states are also clearly distinguishable in Figure 3 ( 0<Ψ<1 . 6 and Ψ>2 . 9 ) and Appendix Figure 9 ( Ψ<0 ) , and are separated by abrupt changes in the distances between near neighbors ( the inverse of local density , Figure 3 ) or potential energy ( Appendix Figure 9 ) . The location of transitions between states depends on the parameters of the social response rule ( e . g . , number of neighbors an individual pays attention to k; Figure 4 ) . The transitional regimes between these states are reminiscent of the first-order phase transitions that occur in some physical systems , for example at the transition between liquid water and water vapor . As in the liquid-vapor phase transition , transitions in collective state are characterized by strong hysteresis ( Figure 3 ) . If the population begins with large Ψ , mean distance to neighbors remains stable for decreasing Ψ and then decreases abruptly ( Figure 3 , Appendix Figure 9 upper curve ) . If Ψ is then increased , mean distance to neighbors increases but follows a different functional relationship with Ψ ( Figure 3 , lower curve ) . We refer to the collective states as station-keeping ( Ψ<0; see Appendix Figure 9 ) , cohesive ( small Ψ ) , and dispersed ( large Ψ ) . The analogy between transitions in collective state in our system and first order phase transitions in physical systems can be made more precise by analyzing the formation rate of groups when Ψ is in the hysteresis region . In the hysteresis region , the rate at which groups of individuals form spontaneously ( and therefore nucleate a transition from the dispersed to cohesive state ) depends strongly on Ψ; when Ψ is near the upper bound of the hysteresis region , the time required for a group to form spontaneously is very long ( see Appendix section 5 . 4 ) . From a thermodynamic perspective , this makes the spontaneous formation of groups extremely unlikely , which explains why populations that begin in the dispersed state follow the upper branch of the hysteresis curve shown in Figure 3 . 10 . 7554/eLife . 10955 . 007Figure 3 . Hysteresis plot of the distance to 10 nearest neighbors , averaged over the entire population ⟨d10NN⟩ ( points and error bars ) as a function of preferred speed parameter Ψ in a uniform environment . Figure produced by starting with a population with Ψ=4 in a uniform environment . Population is allowed to equilibrate for 5000 time steps and ⟨d10NN⟩ is then computed . Ψ is then lowered . This process is repeated until Ψ=−1 , at which point the same procedure is used to increase Ψ . Upper curve corresponds to decreasing Ψ . Lower curve corresponds to increasing Ψ . Regimes where Ψ~0 and Ψ∈ ( 1 . 6 , 2 . 95 ) correspond to transitions between collective states . Points and ( error bars ) correspond to mean ( ± 2 standard errors ) of 50 replicate simulations . Parameters as in Figure 1 with lmax=30 . DOI: http://dx . doi . org/10 . 7554/eLife . 10955 . 00710 . 7554/eLife . 10955 . 008Figure 4 . Evolved populations are positioned near transitions in collective state . Upper panels show mean distance to 10 nearest neighbors ( ⟨d10NN⟩ , color scale ) from simulated populations . A separate populations is simulated in a uniform environment for each value of the social attraction strength ( Ca ) , number of neighbors an individual reacts to ( k ) , and the decay length of social attraction ( la ) parameters . Red is low density corresponding to dispersed state , and blue is high density corresponding to cohesive state . Points show the mean value of ψ0 of populations in the EESt ( populations evolved for 1 , 000 generations in an environment with dynamic resource peaks ) . Evolved populations are positioned near transition between cohesive and dispersed states . Lower panels are based on analytical calculations and show the predicted regions in which the dispersed state is stable ( white ) and unstable ( black , Appendix section 5 ) . Parameters as in Figure 1 with M=15 , λ0=10 , λ1=1 . 6 , α= ( 1 , 0 ) , β=0 . 1 , and τp=1500 . DOI: http://dx . doi . org/10 . 7554/eLife . 10955 . 008 For a wide variety environmental conditions ( Appendix section 2 ) and social parameters ( Figure 4 ) , the evolutionarily stable trait values have a notable feature: the evolved values of the baseline speed parameter , ψ0 , place individuals in the population slightly above the transition between cohesive and dispersed states when S=0 ( Figure 4 , upper panels , Figure 5; points in both figures show mean ψ0 values of population in the ESSt ) , and the evolved environmental sensitivity , ψ1 , is large enough that locally , groups of individuals cross from the dispersed state through the cohesive and station-keeping states in regions of the environment where the resource value is high ( Figure 2A , colors indicate instantaneous value of Ψ for each individual ) . In other words , the evolved values of ψ0 and ψ1 allow local subpopulations to undergo sudden changes from one collective state to another in the proximity of favorable regions of the environment . Importantly , the approximate location of the transition between cohesive and dispersed states can be predicted by directly analyzing Equations 1–4 without considering details of the environment , or the mapping between behavior and fitness ( Figure 4 compare upper panels [simulation] to lower panels [analytical prediction] ) . While the precise evolutionarily stable values of ψ1 depend on the parameters of the environment ( Appendix section 2 ) , the evolutionarily stable values of ψ0 place the population near the cohesive-dispersed transition in many different kinds of environments ( Appendix Figure 5 ) . As we show below , being near this transition allows groups to respond quickly to changes in the environment . Our results demonstrate , that such locations in behavioral state-space are , in fact , evolutionary attractors . 10 . 7554/eLife . 10955 . 009Figure 5 . Mean distance to nearest neighbors ⟨d10NN⟩ ( curves ) and ESSt value of ψ0 ( points ) as a function of social parameters . Points denote mean ESSt value of ψ0 . Note abrupt transitions in density as function of Ψ , as shown in Figure 3 . In all cases , ESSt value of ψ1 causes populations to cross transition when resource value is high ( i . e . , ψ0−ψ1λ0<0 , where λ0 is maximum resource value of each peak ) . Densities and ESSt values generated as described in Figure 4 . DOI: http://dx . doi . org/10 . 7554/eLife . 10955 . 009 The evolutionary results presented in Figure 1 assume that individuals do not appreciably deplete the resource . We can explore an alternative scenario in which resource peaks are depleted through consumption ( Appendix section 2 . 8 ) . In that case , the ith individual consumes resources at a rate uS ( xi ) per time step . We repeated evolutionary simulations assuming either a high or low rate of resource consumption u . For high consumption rate ( 100 individuals can deplete a peak in roughly five time steps ) , lmax still increases so that individuals are attracted to one another through social interactions , but selection for large lmax is much weaker than the case shown in Figure 1C ( see Appendix Figure 7 ) . Moreover , ψ0 and ψ1 increase continually through evolutionary time . This result is intuitive because when resources are depleted rapidly , the locations of neighbors convey little information about the future location of resources and transitioning from the dispersed to cohesive state may actually be maladaptive . By contrast , when individuals consume the resource at a more moderate rate ( Appendix Figure 7 ) , evolutionary trajectories parallel the trajectory shown in Figure 1C–E; there is strong selection for high lmax , ψ0 reaches a stable value that is situated directly above the hysteresis region shown in Figure 3 , and ψ1 evolves to a stable value that is large enough to allow individuals to cross from dispersed to cohesive , and station-keeping states in regions of the environment where the resource value is high . Why do populations of selfish individuals evolve behavioral rules that place them near the transition between collective states ? Dispersed , cohesive , and station-keeping states are each associated with a characteristic density ( low , intermediate , and high , respectively; Figure 3 , Appendix Figure 9 ) . If individuals enter the cohesive and station-keeping states where the resource level is high , the density of individuals becomes strongly correlated with the resource distribution ( Figure 6A ) . The similarity between the distribution of individuals and the distribution of the resource can be quantified by the Kullback-Leibler divergence ( KL divergence ) , an information-theoretic concept that measures the distance between two distributions ( Figure 6A inset ) . Though individuals cannot sense resource gradients , they can detect gradients in the density of their neighbors ( Equation 1 ) , and can therefore move up the resource gradient . 10 . 7554/eLife . 10955 . 010Figure 6 . Collective computation and social gradient climbing . ( A ) Collective computation of the resource distribution ( grayscale represents resource value , normalized to maximum of 1 ) . Curves show local density of individuals at different distances from the resource peak center ( maximum value also normalized to 1 ) . Note the rapid accumulation of individuals near the peak center . The distribution of individuals becomes increasingly concentrated in the region where the resource level is highest; inset shows that the Kullback-Leibler divergence between the resource distribution and the local density of individuals decreases through time as the two distributions become more similar . ( B ) Number of individuals near peak center ( within one decay length , λ1 , of peak center ) as a function of time . Red and blue points and confidence bands represent means ±1 sd . for 100 replicate simulations . Red points and band is ESSt population and blue points and band is an asocial population with the same parameter values . Curves are analytical predictions based on Equations 3 and 4 ( Appendix section 6 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 10955 . 010 The abrupt transitions in the density of individuals between dispersed and cohesive states ( Figure 3 ) mean that there is a strong density gradient in regions of the environment where individuals in the dispersed state border individuals in the cohesive state ( e . g . , Figure 2A , 6A , Video 2 ) . This suggests that the behavior of an individual in this region can be approximated by considering only its interactions with individuals that are on the resource peak ( i . e . , where density is high ) . Using this assumption , we derive analytically the rate at which new individuals join ( or rejoin ) a group on the resource peak ( Appendix section 6 . 5 ) . Asocial individuals arrive at a resource peak at a rate κa , where κa is a constant ( Figure 6B , blue curves and points; Equation A65 ) . However , social individuals initially arrive at a rate that increases as more individuals reach the peak , such that the number of individuals on the peak , Ns , increases exponentially with time: Ns≈κs , 1+exp ( κs , 2t ) , where κs , 1 and κs , 2 are positive constants ( Figure 6B , red curves and points; Equation A68–A70 ) . Analytical calculations ( Figure 6B , solid lines ) agree well with results of numerical simulations ( Figure 6B , points and confidence bands ) . The rapid accumulation shown in Figure 6 is especially important when the environment changes quickly with time; it allows groups to respond swiftly to changes in the resource field and enables the emergent resource tracking behavior described above . The form of Equations ( 3–4 ) implies that an individual’s behavioral response combines personal information about the environment ( Equation 2 ) with social cues ( Equation 1 ) . In fact , under a time rescaling , our model is equivalent to one in which the relative strength of social forces varies across the environment ( Appendix section 4 ) . The tradeoff between using social information and personal information is inherent in social decision-making ( Couzin et al . , 2005; Couzin , et al . , 2011 ) . This tradeoff means that individuals with large ψ0 and ψ1 are , by default , less responsive to their neighbors . Perturbing the values of ψ0 and ψ1 of individuals in populations at the ESSt show that , in populations with high mean ψ0 , individuals fail to form large groups and are poor at tracking resource peaks ( Appendix section 2 . 6 , Appendix Figure 6 ) . In populations with high mean values of ψ1 , individuals form groups ( Appendix section 2 . 7 ) , but fail to exploit regions with the highest resource quality . Individuals with low values of ψ0 or ψ1 form groups but do not effectively track dynamic resources ( Appendix section 2 . 7 ) .
Our model demonstrates that selection on the behavioral phenotypes of selfish individuals can lead to the rapid evolution of distributed sensing and collective computation . The mechanism that promotes this evolution involves the use of public information: when individuals respond to the environment by slowing down in regions of high resource quality – a behavior that is adaptive even in the absence of social interactions ( Appendix Figure 2 ) – their positions become correlated with the locations of resources . Social individuals can exploit this public information by climbing gradients in the density of their neighbors . As in simple , game-theoretic models of social foraging ( e . g . , Clark and Mangel , 1984 ) , social individuals gain a fitness advantage by using information about the environment gleaned by observing neighbors . Because of this , asocial populations are readily invaded by social mutants and collective behaviors evolve ( Appendix section 2 ) . Evolutionarily stable populations occupy a distinctive location in behavioral state space: one in which small changes in individual behavior cause large changes in collective state ( Figures 4 , 5 ) . When individuals respond to local environmental cues by accelerating or decelerating , local populations transition between the collective states shown in Figure 3 ( e . g . Figure 2A ) . This creates the strong spatial gradient in population density ( Figure 6A ) and allows groups to track dynamic features in the environment rapidly . Perturbations of this evolutionarily stable state cause individuals either to weigh social information too heavily ( i . e . , small ψ0 and/or ψ1 ) , in which case groups fail to explore effectively ( Video 3 , Appendix Figure 7 ) , or to weigh personal information too heavily ( i . e . , large ψ0 and/or ψ1 ) , in which case individuals fail to exploit the social information that enables dynamic resource tracking ( Video 4 , Appendix Figure 7 ) . Because of this , mutants with phenotypes far from the evolutionarily stable state are removed from the population by natural selection . The transitions we observe in collective state bear a resemblance to phase transitions in physical systems , and our results lend credence to the hypothesis that natural selection can result in the evolution of biological systems that are poised near such bifurcation points in parameter space . Importantly , we show that these high-fitness regions of parameter space can be predicted a priori from the structure of individual decision rules , even without knowledge of the environment . 10 . 7554/eLife . 10955 . 011Video 3 . Population with mean ψ0 below the ESSt value . Responses of perturbed ESSt population to dynamic resource peaks . All parameters as in Video 2 except that each individual’s value of ψ0 parameter is lowered so that the population mean ⟨ψ0⟩=0 . 4 . Note swarms of individuals form in regions of the environment that are far from resource peaks . Individuals explore poorly and therefore have low fitnesses . DOI: http://dx . doi . org/10 . 7554/eLife . 10955 . 01110 . 7554/eLife . 10955 . 012Video 4 . Population with mean ψ0 above the ESSt value . Responses of perturbed ESSt population to dynamic resource peaks . All parameters as in Video 2 except that each individual’s value of ψ0 parameter is increased so that the population mean ⟨ψ0⟩=8 . 8 . Note that individuals do not form large groups near resource peaks and fail to track peaks as they move . DOI: http://dx . doi . org/10 . 7554/eLife . 10955 . 012 Collective computation is a notion that has strongly motivated research on animal groups ( Berdahl et al . , 2013; Couzin , 2007; Cvikel , et al . , 2015 ) . In our model , populations perform a collective computation through their social and environmental response rules . When individuals are exposed to a heterogeneous resource environment , their responses to the environment cause a modification of the local population density; individuals aggregate in regions where the resource cue is strong . The population performs a physical computation in the formal sense ( Schnitzer , 2002 ) : physical variables – the positions and relative densities of neighbors – represent mathematical ones – spatially resolved estimates of the quality of resources in the environment . The environments considered in our study bear a strong resemblance to those encountered in dynamic coverage problems in distributed control theory ( Bachmayer and Leonard , 2002 ) , dynamic optimization problems ( Passino , 2002 ) , and Monte Carlo parameter estimation ( McKay , 2003 ) . Combining an evolutionary approach to algorithm design with collective interactions may therefore be a useful starting point for optimization schemes or control algorithms for autonomous vehicles , particularly if the structure of social interactions leads to bifurcation points in behavioral parameter space as in the model studied here . Understanding the feedback loop between individual behavior , collective behavior of populations , and selection on individual fitness is a major challenge in evolutionary theory ( Guttal and Couzin , 2010; Torney et al . , 2011; Pruitt and Goodnight , 2014 ) . Our framework closes this loop and demonstrates how distributed sensing and collective computation can evolve through natural selection on the decision rules of selfish individuals .
Our model of the resource environment incorporates three salient features of the resource environments that schooling fish and other social foragers encounter in nature . These features are: 1 ) spatial variation in resource quality , 2 ) temporal variation in resource quality , and 3 ) characteristic length scales of resource patches ( Stephens et al . , 2007; Bertrand et al . , 2008; Bertrand et al . , 2014 ) . Accordingly , we model a two-dimensional environment in which the resource is distributed as a set of M resource peaks . We assume the boundary of the environment is periodic such that individuals , inter-individual potentials , and resource peaks are all projected onto a torus . Each of the M peaks decays like a Gaussian with increasing distance to the peak center . The value of the resource in a single peak at a location , xi , is given by ( 5 ) S ( x , xs ) =λ0e−|x−xs|2λ12 , where λ0 is a constant that determines the resource value at the peak center and λ1 is a decay length parameter , and xs is the location of the centroid of the peak of interest . The total resource value the ith individual experiences S ( xi ) is the sum over all peaks in the environment . Each peak moves according to Brownian motion with drift vector α and standard deviation β . At each time step , each peak has a probability 1/τp of disappearing and reappearing at a new location , chosen at random from all locations in the environment . | In nature , we see many examples of highly coordinated movements of groups of individuals; think of a flock of birds turning swiftly in unison or a crowd of people filing through the exit of a building . A common feature of these behaviors is that they occur without any centralized control , and that they involve sudden and often dramatic changes in the 'collective state' of the group ( i . e . speed , or the distances between individuals ) . In the past , researchers have likened these transitions in collective behavior to phase transitions in physical systems , for example , the transition between liquid water and water vapor . However , it is not clear how such collective responses could have evolved . Natural selection is an evolutionary process whereby individuals with particularly 'fit' traits produce more offspring than others . Over many generations , these beneficial traits tend to become more common in the population . Hein , Rosenthal , Hagstrom et al . developed a mathematical model to investigate whether the capacity of a population to perform collective motions could evolve through natural selection . The model shows that over many generations , populations consistently evolve a unique collective trait whereby small responses of individuals to an environmental cue can cause spontaneous changes in the collective state of the local population . These transitions in collective state greatly enhance the ability of individuals to locate and exploit resources . Hein , Rosenthal , Hagstrom et al . ’s findings suggest that natural selection acting on the behavior of individuals can cause a population to evolve a distinctive , collective behavior . The next challenge will be to identify a biological system in which the evolution of collective motion can be studied experimentally to test these predictions . | [
"Abstract",
"Introduction",
"Model",
"development",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"ecology",
"evolutionary",
"biology"
] | 2015 | The evolution of distributed sensing and collective computation in animal populations |
Cell survival is one of several processes regulated by the Notch pathway in mammalian cells . Here we report functional outcomes of non-nuclear Notch signaling to activate autophagy , a conserved cellular response to nutrient stress , regulating survival in murine natural T-regulatory cells ( Tregs ) , an immune subset controlling tolerance and inflammation . Induction of autophagy required ligand-dependent , Notch intracellular domain ( NIC ) activity , which controlled mitochondrial organization and survival of activated Tregs . Consistently , NIC immune-precipitated Beclin and Atg14 , constituents of the autophagy initiation complex . Further , ectopic expression of an effector of autophagy ( Atg3 ) or recombinant NIC tagged to a nuclear export signal ( NIC-NES ) , restored autophagy and suppressor function in Notch1-/- Tregs . Furthermore , Notch1 deficiency in the Treg lineage resulted in immune hyperactivity , implicating Notch activity in Treg homeostasis . Notch1 integration with autophagy , revealed in these experiments , holds implications for Notch regulated cell-fate decisions governing differentiation .
Notch signaling is a conserved pathway , directed by ligand-dependent processing , culminating in the release of the Notch intracellular domain ( NIC ) from its membrane-tethered precursor , allowing its translocation to the nucleus to initiate transcription ( Kopan and Ilagan , 2009 ) . Signaling through Notch1 is obligatory for acquisition of T-cell fate ( Radtke et al . , 1999 ) . However , Notch signaling is differentially activated during T-cell development and in mature T-cells , including the differentiation of induced Tregs ( iTregs ) from Foxp3-ve naive T-cell precursors ( Radtke et al . , 1999; Laky et al . , 2015; Maekawa et al . , 2015; Yvon et al . , 2003; Wolfer et al . , 2001 ) . The transcription factor Foxp3 specifies Treg identity and unlike in induced ( i ) Tregs where Notch1 regulates Foxp3 expression in response to activation cues , in naturally arising Tregs , Foxp3 expression is developmentally regulated ( Fontenot et al . , 2003 ) . Interestingly , activated Tregs are also distinguished by non-nuclear Notch1 signaling ( Perumalsamy et al . , 2012 ) . Macro-autophagy ( autophagy ) is a conserved catabolic process , which buffers cells from limiting nutrient conditions ( Nakatogawa et al . , 2009 ) , and is controlled by the evolutionarily conserved Atg/ATG family of proteins ( Nakatogawa et al . , 2009; Mizushima et al . , 2011 ) . Autophagy is implicated in the regulation of diverse aspects of immunity including the differentiation and homeostasis of CD8+ memory T-cells and Treg function amongst other cell types ( Deretic et al . , 2013; Parekh et al . , 2013; Xu et al . , 2014 ) . The mechanism[s] by which autophagy is regulated in T-cell subsets are likely diverse and not characterized . In this study we describe Notch1 signaling to autophagy in the context of activated Treg survival and consequences to Treg function . Through the analysis of Tregs , activated by T-cell receptor ( TCR ) cross-linking in vitro , we show that ( non-nuclear ) Notch1 regulates survival via the activation of an autophagy-signaling cascade . In agreement with their central role in coordinating decisions of cell death and survival , mitochondria were a prominent cellular target , responsive to perturbations of the Notch-autophagy cascade . We validate these interactions by reconstituting the Notch-autophagy cascade in Notch1 deficient Tregs . We also show that Notch1 confers cytokine-independence via the activation of autophagy in T-effectors , which are derived from naïve T-cells . The consequence of Notch activity in Treg physiology is assessed in two contexts: One , the suppression of antigen-induced T-cell proliferation and correction of defective glucose clearance in genetically obese mice using adoptive transfers of test Treg populations . Second , the characterization of inflammation and increased activation of T-cell subsets occurring in mice with Notch1 ablated in the Foxp3 ( Treg ) lineage . Collectively , these findings suggest a hitherto unappreciated role for the ( non-nuclear ) Notch-autophagy axis in the regulation of natural Treg function .
To assess a role for autophagy in Treg survival , activated Tregs are switched to complete medium , which contains serum but is not supplemented with the cytokine IL-2 . Cells are monitored at various time points for induction of autophagy or survival following modulations described in the sections that follow . The recruitment of the microtubule-associated protein LC3 and its smaller lipidated form LC3II , into the autophagosome membrane is a molecular signature and necessary event in the progression of autophagy ( Kabeya et al . , 2000 ) . The change in LC3 can be detected in immunoblots of cell lysates , where the modified isoform is detected at a reduced molecular weight or by immunostaining intact cells when large puncta are marked by antibodies to LC3 . An increase in the LC3II isoform was detected in lysates of Tregs , which had been cultured without IL-2 for 6 hr , relative to the onset of the assay ( T0 ) ( Figure 1A ) . Immunostaining with the same antibody as used for the immunoblots and visualization of intact cells by confocal microscopy , showed that diffuse LC3 staining observed in Tregs at T0 , progressively changed to large , readily visualized puncta by 6 hr , persisting till 15 hr following cytokine-withdrawal ( Figure 1B ) . Quantifiable changes in fluorescence intensity and size of puncta were detected over this period ( Figure 1B and Figure 1—figure supplement 1A ) . It should be noted that Tregs are viable throughout the course of this assay ( Figure 1—figure supplement 1B ) . The protein Atg5 , a molecular indicator of the activation of autophagy ( Mizushima , et al . , 2011 ) , was also increased following cytokine withdrawal as detected by Immunoblots of Tregs cultured without cytokine ( Figure 1C ) . 10 . 7554/eLife . 14023 . 003Figure 1 . Autophagy is activated on cytokine withdrawal in activated Tregs . ( A ) Immunoblots probed for LC3 in lysates of Tregs at onset ( T0 ) and after 6 hr culture without IL-2 . The values below are densitometry analysis of LC3II relative to tubulin . ( B ) Z-projected confocal images of Tregs at onset ( T0 ) and cultured without IL-2 for times indicated and stained for LC3 ( green ) and Hoechst 33342 ( blue ) . Change in fluorescence intensities for LC3 relative to T0 are plotted . ( n=150 cells/time point ) . ( C ) Immunoblot probed for ATG5 in lysates of Tregs cultured as described in A . ( D−F ) Apoptotic damage following 15 hr of IL-2 withdrawal in Tregs cultured in the presence of Bafilomycin ( Baf ) or 3-MA ( D ) or transduced with shRNA specific for VPS34 ( E ) or ATG7 ( F ) or a scrambled control ( Scr ) . Immunoblots of scrambled and shRNA transfected cells are shown below . Data shown are the mean ± SD from at least 3 independent experiments , *p<0 . 03 . Scale bar 5 μm . This figure is accompanied by Figure 1—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 14023 . 00310 . 7554/eLife . 14023 . 004Figure 1—figure supplement 1 . Tregs activate autophagy in response to cytokine deprivation . ( A ) Z-projected confocal field views of activated Tregs cultured without IL-2 for T6 and T15 hr and input populations ( T0 ) fixed , permeabilized and stained for LC3 ( green ) and Hoechst 33342 ( blue ) . ( B ) Apoptotic damage induced activated WT or Notch1-/- Tregs cultured with or without IL-2 . Scale bar 5 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 14023 . 004 To assess if the activation of autophagy was necessary for Treg survival , inhibitors that block induction or progression of autophagy were tested . The inclusion of Bafilomycin A ( Baf ) , which blocks autophagosome-lysosome fusion or 3-Methyladenine ( 3MA ) , at the onset of the cytokine deprivation assay resulted in cell death , when measured at 15 hr , indicating that Treg survival was abrogated ( Figure 1D ) . To validate observations made with chemical inhibitors , we expressed shRNA to Vps34 ( the lipid kinase regulating initiation of autophagy ) or to Atg7 , proteins implicated in autophagy progression , by retroviral infection coupled with antibiotic selection ( as described in methods ) to generate activated Tregs populations ablated for Vps34 or Atg7 protein expression . The loss of either VpS34 or Atg7 in activated Tregs resulted in cell death following cytokine withdrawal ( Figure 1E , F ) , establishing a requirement for these intermediates in Treg survival . We next asked if Notch1 is required for activation of autophagy in activated Tregs in response to cytokine withdrawal . These experiments tested for the involvement of non-nuclear , ligand-dependent Notch1 activity , shown earlier to promote Treg survival ( Perumalsamy et al . , 2012 ) . A pharmacological inhibitor of the enzyme γ-secretase , ( GS inhibitor ) -X , inhibits cleavage ( S3 cleavage post ligand binding ) and release of NIC . Culturing Tregs with GSI-X abrogated survival following cytokine withdrawal ( Figure 2A ) . In GSI treated cells , immunestaining for Notch1 showed that that staining was restricted to the cell membrane , confirming that receptor cleavage is disrupted ( Figure 2B and Figure 2—figure supplement 1A ) . Further , induction of LC3 puncta was not observed in Tregs cultured with GSI during cytokine withdrawal ( Figure 2C , and Figure 2—figure supplement 1B ) , indicating that autophagy is not activated . Earlier work from the laboratory had identified a non-redundant role for the ligand Delta Like Ligand ( DLL ) -1 in Treg survival . Confirming ligand-dependence , ablation of DLL-1 in Tregs , compromised survival and induction of LC3 puncta in cells cultured without cytokine ( Figure 2—figure supplement 1C and D ) . Contrastingly , but in agreement with published analysis of NIC1 signaling in Tregs , both the induction of LC3 puncta or cytokine-independent survival in Tregs was unchanged following shRNA-mediated ablation of RBPJ-κ , a co-factor of NIC transcription ( Figure 2D , E and Figure 2—figure supplement 1E ) . The ablation of RBPJ-κ attenuated expression of Hes1 a target of NIC ( Figure 2F ) confirming efficacy of shRNA ablation . Taken together , these data suggested that ligand dependent Notch1 activity was required for autophagy and regulation was independent of canonical Notch interactions , such as those requiring RBPJ-κ ( Kopan and Ilagan , 2009 ) . 10 . 7554/eLife . 14023 . 005Figure 2 . Notch regulates induction of autophagy in Tregs . ( A ) Apoptotic damage in Tregs cultured for 15 hr without IL-2 , with or without GSI . ( B ) Representative images of Tregs cultured without IL-2 , with or without GSI-X for 6 hr , stained for NIC ( mNIA , red ) and Hoechst 33342 ( blue ) . ( C ) Representative , Z-projected images of Tregs treated as in ( B ) for 6 hr and stained for LC3 and Hoechst 33342 . Plot below indicates the change in fluorescence intensities of LC3 relative to T0 ( mean +/- SD , n=120 cell/ condition ) . ( D ) Tregs transduced with shRNA to RBPJκ or scrambled control , cultured without and with IL-2 for 6 hr and stained for LC3 and Hoechst 33342 . ( E ) DiOC6 uptake in cells transduced with shRNA to RBPJκ or scrambled control and cultured with and without IL-2 for 6 hr . ( F ) Immunoblot analysis for Hes1 and tubulin in cells expressing control or RBPJκ shRNA . ( G ) Z-projected confocal images of wildtype ( WT ) or Notch1-/- Tregs , cultured without IL-2 for 15 hr and stained for LC3 and Hoechst 33342 . Plot below indicates the change in fluorescence intensities of LC3 relative to T0 ( mean +/- SD , n=120 cell/ condition ) . ( H ) Apoptotic damage measured in Notch1-/- Tregs transduced with recombinant ATG3 or pBABE vector control , at onset ( T0 ) or after culture without IL-2 for 15 hr . ( I ) Notch1-/- Tregs transduced with NIC-NES cultured for 15 hr in the conditions indicated and scored for apoptotic damage . ( J ) and ( K ) Immune complexes precipitated from Treg lysates using indicated ( IP ) antibodies . Immunoblots were probed for proteins indicated on the right . Data show the mean ± SD of at least 3 independent experiments . In all micrographs , LC3 staining is in green and Hoechst 33342 in blue . Scale bar: 5 μm; *p≤0 . 03 This figure is accompanied by Figure 2—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 14023 . 00510 . 7554/eLife . 14023 . 006Figure 2—figure supplement 1 . Notch1 signaling from the cytoplasm is needed for autophagy induction in Tregs . ( A ) Montage of confocal images ( central stack ) of Tregs cultured without IL-2 for 6 hr in the absence ( upper panel ) or presence ( lower panel ) of GSI-X and then immune-stained for NIC ( clone mNIA ) and counterstained with Hoechst 33342 ( nucleus ) . ( B ) Z-projected confocal images of Tregs cultured without IL-2 for 6 hr in the absence ( upper panel ) or presence ( lower panel ) of GSI-X . Cells were stained with an antibody to LC3 ( green ) and counterstained with Hoechst 33342 ( blue ) and imaged . Numbers indicate pixel intensities and cells analysed . ( C ) Apoptotic damage at 15 hr following culture without IL-2 in Tregs transduced with scrambled or DLL-1 specific shRNA , followed by antibiotic selection for transfected cells . ( D ) Change in fluorescence intensities for LC3 , relative to T0 ( onset of experiment ) in Tregs transduced with scrambled or DLL-1 specific shRNA measured 6 hr after IL-2 withdrawal . N=120cells/time point , *p</= 0 . 03 . ( E ) Z-projected confocal images of Tregs cultured without IL-2 for 6 hr following retroviral transfection with scrambled ( SCR ) or RBPJ-κ specific shRNA , followed by antibiotic selection for transfected cells . Cells were stained with an antibody to LC3 ( green ) and counterstained with Hoechst 33342 ( blue ) and imaged . Numbers indicate pixel intensities and cells analysed . ( F ) Z-projected confocal field views of images of WT or Notch1-/- Tregs cultured without IL-2 for 15 hr and stained for LC3 ( green ) and Hoechst33342 ( blue ) . Numbers indicate pixel intensities and cells analysed . ( G ) Representative images showing activated Notch1-/- Tregs are retrovirally transduced with recombinant NIC-NES ( upper panel ) or NIC-NLS ( lower panel ) and stained for Notch using the mN1A antibody clone ( red ) . Cells were counterstained with Hoechst 33342 ( blue ) . Scale bar 5 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 14023 . 006 More direct evidence that Notch1 regulates activation of autophagy in response to cytokine withdrawal came from the analysis of activated Notch1-/- Tregs generated from mice with a targeted deletion of Notch1 in the mature T-cell compartment ( Cd4-Cre::Notch1lox/lox mice ) . The induction of LC3 puncta in response to cytokine withdrawal was blunted in Notch1-/- Tregs ( Figure 2G and Figure 2—figure supplement 1F ) , and cell survival compromised in cells cultured without cytokine ( Figure 1—figure supplement 1B ) . In consonance with dependence on autophagy , retroviral transduction of recombinant Atg3 in Notch1-/- Tregs restored survival following cytokine withdrawal ( Figure 2H ) . Atg3 is an E2-ubiquitin-like-conjugating enzyme , which catalyzes the lipidation of the effector protein LC3I ( ATG8 ) ( Mizushma et al . , 1998 ) , a key step in autophagy progression . Atg3 is also implicated in maintaining mitochondrial health under conditions of nutrient stress ( Besteiro et al . , 2011; Radoshevich and Debnath , 2011 ) . From these experiments we concluded that Notch1 activation of autophagy was necessary for Treg survival following cytokine withdrawal in culture . NIC is enriched in the cytoplasm of Tregs ( shown later and Perumalsamy et al . , 2012 ) , and we tested if non-nuclear NIC activity regulated autophagy following cytokine withdrawal . Towards this aim , we employed an approach used by others and us to achieve nuclear export of recombinant NIC ( Shin et al . , 2006 ) . Thus , Notch1-/- Tregs were retrovirally transfected with recombinant NIC tagged to a Nuclear Export Signal ( NIC-NES ) , which prevents nuclear residence . NIC-NES transfected Tregs express a recombinant predominantly localized in the cytoplasm ( Figure 2—figure supplement 1G ) . NIC-NES expression protected Notch1-/- Tregs from apoptosis following cytokine-withdrawal ( Figure 2I ) and this protection is attenuated by the inclusion of Baf or 3MA ( Figure 2I ) . Thus , non-nuclear NIC activity could activate autophagy signaling in Tregs following cytokine withdrawal . We next assessed if proteins in the autophagy cascade immune-precipitated with NIC . In Tregs initiated into cytokine-deprivation , an antibody specific for Notch1 ( mN1A ) but not an IgG ( isotype control ) immune-precipitated Beclin1 and Atg14 but not Vps34 and LC3 ( Figure 2J ) , indicating that NIC formed complexes with specific components of the autophagy pathway . In the reverse analysis , Beclin1 immune-precipitated NIC confirming the specificity of its association with NIC and this complex also included Atg14 and Vps34 ( Figure 2K ) . Thus , Beclin is detected in complexes that include either Vps34 or NIC with both complexes including Atg14 as expected . Vps34 and Beclin are known to form multiple and dynamic cellular complexes , which are regulated by the phosphorylation of the proteins themselves as well their interacting partners ( Kang et al . , 2011; Kim et al . , 2012 ) . Thus the exclusion of Vps34 from the NIC complex may indicates interactions with proteins ( such as Beclin ) whose functions may be regulated by Notch activity . The data do not rule out transient associations between NIC and Vps34 . That NIC formed complexes with proteins regulating autophagy in Tregs , indicated a more direct role for Notch1 in the autophagic cascade , although the molecular regulation and dynamics of this interaction remains to be dissected . Mitochondrial re-modeling and activity control cellular responses to changing bioenergetic needs , which facilitate transitions between nutrient-replete and deficient conditions ( Rambold et al . , 2011 , Hailey et al . , 2010 ) . Since Notch modulation of mitochondrial function has been suggested in earlier work ( Perumalsamy et al . , 2010; Kasahara et al . , 2013 ) , mitochondrial organization and integrity in Tregs were next examined . Mitochondria were analyzed employing a combination of biophysical and imaging methods in live cells and a potentiometric fluorescent probe , DiOC6 , which measures mitochondrial trans-membrane potential ( MTP ) , an indicator of mitochondrial energetic state . Uptake of the dye DiOC6 is a well-established flow-cytometry based measure of MTP in intact cells . Activated Tregs undergo almost no change in MTP following cytokine withdrawal , consistent with their survival ( Figure 3A ) . However , inclusion of GSI compromised mitochondrial function as it triggered a loss of MTP within 8–9 hr of culture ( Figure 3A ) . Thus , mitochondrial activity was regulated by Notch in Tregs . We next asked if Notch activity also controlled mitochondrial organization in Tregs . In live activated Tregs , mitochondria marked with MitoTracker Green and visualized by confocal microscopy appear as interconnected structures , which remained so following cytokine withdrawal ( Figure 3B and Figure 3—figure supplement 1A ) . Mitochondrial organization in cells cultured without cytokine was disrupted by perturbations of Notch activity , which included shRNA-mediated ablation of the Notch ligand DLL-1 ( Figure 3C and Figure 3—figure supplement 1B ) or shRNA mediated ablation of Notch1 ( Figure 3D ) or by the inclusion of GSI-X ( Figure 3E and Figure 3—figure supplement 1Cii ) . Mitochondrial organization was also disrupted by the addition of Baf ( Figure 3F and Figure 3—figure supplement 1Ciii ) or by shRNA-mediated ablation of Atg7 ( Figure 3G , and Figure 3—figure supplement 1C iv , v ) , implicating autophagic signaling in organelle integrity . 10 . 7554/eLife . 14023 . 007Figure 3 . Notch regulates mitochondrial organization via autophagy . ( A ) DiOC6 uptake in activated Tregs in the input population at onset of the assay ( T0 ) or in cells cultured without cytokine for 8 hr with or without 10 μM GSI ( Data are plotted to show Mean ± SD , p**</=0 . 001 ) . ( B-G ) Representative confocal ( Z-projected ) images of mitochondria stained with MitoTracker Green in WT Tregs cultured for 6 hr in the following conditions: with or without IL-2 ( B ) , post retroviral transfection of DLL-1 shRNA +/- IL-2 ( C ) ; or transfection of Notch1 shRNA +/- IL-2 ( D ) ; no IL-2+GSI ( E ) no IL-2 + Baf ( F ) or post retroviral transfection of shRNA to Atg7 +/- IL-2 ( G ) . F and G , percent DiOC6-high ( live ) WT and Notch1-/- Tregs at T0 ( F ) or 15 hr after culture without IL-2 ( G ) . Mean ± SD from 3 separate experiments . ( H ) DiOC6 fluorescence in activated Notch1-/- and control Notch1+/+ Tregs in the input population at the onset of the assay ( T0 ) or in cells cultured without cytokine for 15 hr . ( Mean ± SD , p*</=0 . 03 ) . ( I ) FRAP analysis in MitoTracker Green loaded WT or Notch1-/- Tregs ( n=10 cells/ cell type , scale bar 2 μm ) at T0 . Inset: mitochondria in Notch1-/- Tregs cultured for 6 hr with or without IL-2 . ( J and K ) Representative confocal images ( at 6 hr ) of mitochondria loaded with MitoTracker Green in Notch1-/- Tregs transfected with Atg3 or empty vector ( pBABE ) , cultured with or without IL-2 ( J ) or Notch1-/- Tregs transfected with NIC-NES or empty vector ( pBABE ) in IL-2 ( K ) . Images are representative of n=20 cells per experimental group from 2–3 experiments , scale bar 5 μm . This figure is accompanied by Figure 3—figure supplement 1DOI: http://dx . doi . org/10 . 7554/eLife . 14023 . 00710 . 7554/eLife . 14023 . 008Figure 3—figure supplement 1 . Notch1 regulation of mitochondrial organisation is mediated via autophagy . ( A–C ) Representative confocal ( Z-projected ) images of mitochondria stained with MitoTracker Green in activated WT Tregs cultured in the following conditions: with or without IL-2 for 6 hr ( A ) , with IL-2 for 6 hr post retroviral transfection of scrambled or DLL-1 specific shRNA ( B ) ; untreated or transfected with scrambled shRNA cultured without IL-2 ( i and ii ) , without IL-2+GSI ( iii ) , without IL-2 +Baf ( iv ) or without IL-2 post retroviral transfection of shRNA to Atg7 ( v ) for 6 hr ( C ) . ( D–E ) Montages showing confocal ( Z-projected ) images of mitochondria stained with MitoTracker Green in activated Notch1-/- Tregs cultured in the following conditions: with or without IL-2 ( D ) or transduced with empty vector ( pBABE ) or recombinant ATG3 ( E ) or NIC-NES ( F ) . In E and F cells were maintained in IL-2 . Scale bar 5 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 14023 . 008 Uptake of DiOC6 was significantly lower in Notch1-/- Tregs relative to WT Tregs cultured in IL-2 ( Figure 3H ) . Following cytokine withdrawal , further loss of MTP , indicated compromised mitochondrial function in Notch1-/- Tregs ( Figure 3H ) . We compared mitochondrial contiguity in WT and Notch1-/- Tregs using the technique of fluorescence recovery after photo bleaching ( FRAP ) in live cells loaded with MitoTracker Green . Fluorescence recovery in WT Tregs was high , which is consistent with a connected morphology of the organelle ( Figure 3I ) . Low FRAP in Notch1-/- Tregs is indicative of discontinuous structures ( Figure 3I ) , which was confirmed by microscopy ( Figure 3I inset and Figure 3—figure supplement 1D ) . The punctate morphology of mitochondria in Notch1-/- Tregs was reduced , with the organelle appearing more tubular and connected in cells expressing recombinant Atg3 ( Figure 3J and Figure 3—figure supplement 1E ) or recombinant NIC-NES ( Figure 3K and Figure 3—figure supplement 1F ) relative to cells expressing an empty vector control ( Figure 3J , K and Figure 3—figure supplement 1F ) . Thus a loss of MTP and mitochondrial fragmentation was associated with poor survival outcomes following cytokine withdrawal in Tregs . Our experiments established that NIC and autophagy signaling were critical for the integrity and function of mitochondria in these cells . We next assessed if Notch signaling to autophagy was a more generalized mechanism that can be activated in other cells . For this , we employed T-effectors generated from naïve T-cell precursors , as T-effectors do not survive cytokine withdrawal ( Purushothaman and Sarin , 2009 ) and we had prior evidence that ectopic expression of recombinant NIC protected T-effectors from apoptosis in this context ( Bheeshmachar et al . , 2006 ) . T-effectors - generated as described in methods by TCR stimulation of naïve T-cells - do not increase LC3 puncta or the LC3II isoform following cytokine deprivation ( Figure 4A , B and Figure 4—figure supplement 1A ) . Endogenous NIC is nuclear localized in T-effectors unlike Tregs ( Figure 4—figure supplement 1B ) . As was seen with Tregs , the ectopic expression of NIC-NES protected T-effectors from apoptosis triggered by cytokine withdrawal ( Figure 4C ) . However a recombinant NIC modified by the inclusion of a Nuclear Localization Signal ( NIC-NLS , Figure 4C inset ) , which enforces localization to the nucleus did not confer protection from cell death ( Figure 4C ) . We ruled out a role for endogenous Notch1 in this context , by reproducing the protective effect of NIC-NES in T-effectors derived from Cd4-Cre::Notch1lox/lox naïve T-cells ( Figure 4D ) . Further , protection from cell death was abrogated if inhibitors of autophagy - Baf or 3MA - were included in culture ( Figure 4D ) . Notably , mitochondria in T-effectors do not demonstrate contiguity in FRAP assays ( Figure 4E and Figure 4—figure supplement 1C ) or by confocal image analysis ( Figure 4E ) . Thus , the analysis indicated that the crosstalk between Notch1 and autophagy for the regulation of survival is not restricted to Tregs alone . 10 . 7554/eLife . 14023 . 009Figure 4 . NIC-NES rescues cell death via autophagy in T-effectors . ( A ) Z-projected confocal images of T-effectors stained for LC3 ( green ) and Hoechst 33342 ( blue ) at onset of assay ( T0 ) or following culture without IL-2 for 6 hr . ( B ) Immunoblots probed for LC3 in lysates of T-effectors at T0 and cultured without IL-2 for 6 hr . The values indicate densitometry analysis of LC3II relative to tubulin . ( C ) Apoptotic damage induced by IL-2 withdrawal in T-effectors expressing NIC-NES , NIC-NLS or pBABE cultured in the presence or absence of IL-2 . Inset: Representative image of a T-effector stained with Val1744 antibody to detect endogenous Notch ( red ) and counterstained with Hoechst 33342 ( blue ) . ( D ) Apoptotic damage in Notch1-/- T-effectors expressing NIC-NES or pBABE cultured with IL-2 ( black ) , without IL-2 ( grey ) and the addition of Baf ( light grey ) or 3-MA ( white ) for 15 hr . ( E ) FRAP analysis in Tregs and T-effectors loaded with MitoTracker Green at T0 ( n=10 cells/ cell type , scale bar 2 μm ) . Inset above , representative images of MitoTracker Green loaded cells visualized by confocal microscopy . Data shown are the mean ± SD from 3 independent experiments . Scale bar 5 μm . This figure is accompanied by Figure 4—figure supplement 1DOI: http://dx . doi . org/10 . 7554/eLife . 14023 . 00910 . 7554/eLife . 14023 . 010Figure 4—figure supplement 1 . Non-nuclear Notch1 dependent survival is mediated by autophagy in T-effectors . ( A ) Representative Z-projected confocal field views of T-effectors input populations ( T0 ) or cultured without IL-2 for 6 hr and stained for LC3 ( green ) and Hoechst 33342 ( nucleus , blue ) . ( B ) Representative confocal images ( central stack ) of activated Tregs or T-effectors immune-stained for NIC with Val 1744 and counterstained with Hoechst 33342 ( nucleus ) . ( C ) Montage showing time-lapse images of one Treg and one T-effector used in the FRAP analysis plotted in the main figure . The red arrow indicates the bleach spot , scale bar 2 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 14023 . 010 Survival is necessary for cellular function and an important component of homeostasis . The experiments described thus far had revealed a role for Notch signaling to autophagy for activated Treg survival in culture . We next tested if the NIC-autophagy signaling axis was also of consequence to Treg functions in vivo . The transcription factor Foxp3 , specifies Treg identity and suppressor function and while it can be induced in naïve T-cells by appropriate cytokines ( Samon et al . , 2008; Mota et al . , 2014 ) , Foxp3 expression is developmentally regulated in naturally arising Tregs ( Fontenot et al . , 2003 ) . As reported earlier ( Perumalsamy et al . , 2012 ) , NIC is enriched in the cytoplasm of activated Tregs ( Figure 5A , upper panel and Figure 4—figure supplement 1B ) , which contrasts with the more expected pattern of nuclear localization in T-effectors ( Figure 5A , lower panel and Figure 4—figure supplement 1B ) . We noted a small but consistent elevation in the number of Tregs recovered from Cd4-Cre::Notch1lox/lox ( 2 . 25 ± 0 . 3 ) relative to Cre negative genetic controls ( 1 . 8 ± 0 . 3 ) mice per 100 million spleen cells . However , Foxp3 expression was comparable in activated Tregs generated from Cd4-Cre::Notch1lox/lox mice and the matched genetic controls , including Tregs from C57Bl/6 ( wildtype ) mice ( Figure 5B and Figure 5—figure supplement 1A , lower panel ) . We recapitulated Notch1-dependence in activation-induced expression of Foxp3 in induced Tregs . Thus , in culture conditions that polarize to the generation of iTregs , in contrast to the genetic control ( Cre negative ) littermates , naïve Cd4-Cre::Notch1lox/lox T-cells , cannot be differentiated to express Foxp3 , ( Figure 5—figure supplement 1A upper panel ) . 10 . 7554/eLife . 14023 . 011Figure 5 . Notch activity and autophagy regulate Treg suppressor function . ( A ) Representative , confocal images ( central stack ) of activated Tregs ( upper panel ) , or T-effectors ( lower panel ) immune-stained for NIC ( red ) and Hoechst 33342 ( blue ) . N: 15–20 cells/experiment . ( B ) Representative confocal ( central stack ) field views of activated Tregs immune-stained for Foxp3 ( green ) and Hoechst 33342 ( blue ) . Tregs were derived from C57Bl/6 ( wildtype , WT ) or Cre negative ( Cre-ve ) or Cd4-Cre::Notch1lox/lox ( Cre+ve ) mice . ( C ) Real Time PCR quantification of genes enriched in Tregs , comparing activated Tregs from Cd4-Cre::Notch1lox/lox ( open bars ) and genetic control Cre-ive ( black bars ) mice . 5–6 mice are included in each group being compared . The data plotted is mean+/-SD . p**</=0 . 001 . ( D ) Flowcytometry based expression of molecules ( mean fluorescence intensity , MFI , relative to control isotype antibody shown ) enriched in Treg subsets , compared in activated Tregs generated from Cre+ive ( open bars ) and Cre-ive ( black bars ) mice . 4–6 mice are included in each group . ( E ) Flowcytometry plots indicating dilution of CFSE in CD45 . 2+ ( OT-II ) gated cells isolated from lymph nodes of mice injected with OT-II cells alone ( i ) or , OT-II co-injected with Tregs transduced with scrambled ( ii ) or Notch1 shRNA ( iii ) , three days after antigen challenge . Inset: ( ii ) confocal images of Tregs detecting Foxp3 in scrambled or Notch1 shRNA groups and ( iii ) immunoblot for Notch1 in shRNA treated groups . ( F ) Flowcytometry plots indicating dilution of CFSE in CD45 . 2+ ( OT-II ) gated cells isolated from lymph nodes of mice injected with OT-II ( i ) OT-II + WT Tregs ( ii ) or OT-II + Notch1-/- Tegs ( iii ) three days after antigen challenge . ( G ) CFSE dilutions of OT-II cells co-injected with Notch1-/- Tregs transduced with empty vector ( pBABE ) ( ii ) or recombinant NIC ( iii ) three days after antigen challenge . Data are representative of 2–3 independent experiments with 2–3 mice/ experimental group . Percentage of cells in the CFSE diluted group is indicated in each plot . Tregs activated in vitro are used in all experimetns . ( H ) Proliferation in CD4+OT-II cells alone ( i ) , or co-injected with Tregs transduced with retroviruses expressing shRNA to VPS34 ( iii ) or a scrambled control ( ii ) post antigen challenge . ( I ) Confocal ( merged ) images of Foxp3 ( green ) immunostaining counterstained with Hoechst 33342 ( blue ) . ( J ) immunoblot detecting VPS34 ( H ) in shRNA treated groups as in H . ( K ) Flow cytometry plots indicating CFSE dilution in naïve CD4+T-cells 72 hr post-stimulation with anti-CD3 and APC in vitro . T-cells were either cultured alone ( no Tregs ) or with Tregs transduced with shRNA as described in H . ( L ) Percent CFSE positive , CD45 . 2+ OT-II cells isolated from host mice isolated after antigen challenge . Host mice were injected with OT-II naïve T-cells alone ( no Tregs ) or , naïve cells co-injected with Notch1-/- Tregs retrovirally transduced with empty vector pBABE or recombinant ATG3 . Scale bar: 5 μm . This figure is accompanied by Figure 5—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 14023 . 01110 . 7554/eLife . 14023 . 012Figure 5—figure supplement 1 . Notch1 signaling and autophagy pathway are needed for Treg function . ( A ) Representative field views of induced-Tregs ( upper panel ) or activated Tregs ( lower panel ) , of indicated genetic backgrounds , immune-stained for Foxp3 . Inset table: Percent cells positive for Foxp3 in the different groups from multiple experiments . ( B and C ) , Gene subsets in Notch1-/- Tregs relative to control Cre-negative Tregs , based on a microarray analysis of activated Tregs preparations ( values plotted are the mean fold change ± SD ) . ( D ) Flowcytomety histogram plots for expression of different proteins in activated Tregs of these genotypes . ( E ) CD4+ and CD8+ T-cell subset ( naïve and memory ) analysis in lymph node cells isolated from 6–8 week Cd4-Cre::Notch1lox/lox and genetic ( Cre negative ) mice . Mean+/-SD of analysis from 6–8 mice of each genotype are shown . ( F ) Flow cytometry quadrant plot showing the gating of CD45 . 2+ donor T-cells , in the lymph nodes isolated from a CD45 . 1 host . Host cells are seen in the upper left quadrant . C , CD45 . 2+ CFSE cells isolated from lymph nodes of unimmunized host mice . The peak in the third log indicates donor cells , which are detected at high intensity . The auto-fluorescence in a CFSE unlabelled population is shown for comparison . ( G ) Flow cytometry plot of the analysis of lymph node cells isolated from a CD45 . 1 ( host mouse ) to detect injected CD45 . 2+ donor cells ( upper left quadrant ) and CFSE labelled Tregs ( lower right quadrant ) . ( H ) A representative field view of Foxp3 staining in activated Tregs transfected with scrambled ( SCR ) or Notch1shRNA . ( I ) Plot of total cell recovery from lymph nodes of host mice 3 days post immunization . All data are representative of 2–3 independent experiments with 2–3 mice/ experimental condition . DOI: http://dx . doi . org/10 . 7554/eLife . 14023 . 012 Before we assessed functional capabilities of Tregs of the two genotypes , we compared gene and protein expression of a subset of molecules associated with Treg activation and function . Activated Tregs generated from Cd4-Cre::Notch1lox/lox mice and their genetic ( Cre negative ) controls were compared for molecules enriched in Treg subsets . Gene-expression analysis by microarray was confirmed using RT-PCR analysis and in some instances flow-cytometry analysis for proteins . In this analysis , activated Tregs of the two genotypes showed no striking differences , with loss of Notch1 transcript serving as the positive ( internal ) control ( Figure 5C and Figure 5—figure supplement 1B ) . Similarly , comparison of gene sets enriched for cytokines associated with Tregs revealed no differences in the Cd4-Cre::Notch1lox/lox or genetic ( Cre negative ) control mice ( Figure 5—figure supplement 1C ) . Similar trends were observed in flow-cytometry based analysis of cell surface and intracellular markers implicated in Treg differentiation and function ( Figure 5D and Figure 5—figure supplement 1D ) . These experiments indicated a small difference in the expression of the gene Eos and high-affinity IL-2 receptor expression . A better understanding of the functional significance of these changes awaits analysis of other gene groups using combination of approaches described here . The distribution of naïve and memory T-cell subsets in circulation , an indication of the resting vs . activated state of the immune system was comparable in Tregs derived from Cre negative controls and Cd4-Cre::Notch1lox/lox mice ( Figure 5—figure supplement 1E ) . Based on the analysis of transcripts , protein expression and immune subsets we expected no difference between wildtype and Notch1-/- Tregs in functional assays of suppression . Indeed , in earlier work when tested for suppressor activity in vitro , activated Tregs of the two genotypes were comparable ( Perumalsamy et al 2012 ) . However , suppression is a multi-step process with distinct demands that operate on cells in vivo , hence we tested Tregs from Notch1-/- and genetic controls in assays of immune-suppression in vivo . Established protocols ( Quah et al , 2007; Klein et al . , 2003 ) , involving the transfer of responder and suppressor cells into host ( immune-competent ) mice were followed to measure the ability of Tregs to suppress antigen-induced T-cell proliferation . In this assay , purified naïve ( OT-II ) T-cells loaded with a fluorescent dye – chosen because it partitions into daughter cells following division - were injected into host mice , which are subsequently challenged with ovalbumin as OT-II cells respond to this antigen . The OT-II T-cells can be distinguished from host cells by antibody-mediated detection of the CD45 . 2 molecule on their cell surface ( Figure 5—figure supplement 1F ) . Three days post antigen challenge lymph node cells are isolated from the host , cells are gated on CD45 . 2+ and the CFSE profile scored in this group using two-color flow-cytometry . The distribution of CFSE loaded cells spans multiple dilutions of lower intensity , indicative of proliferation ( Figure 5Ei ) , whereas in mice not challenged with antigen , CFSE will be detected at the highest intensity ( Figure 5—figure supplement 1G , filled histogram ) . Since functional Tregs suppress T-cell proliferation , co-injection of Tregs and CFSE loaded naïve T-cells , results in reduced/no dilution of CFSE ( Figure 5Eii ) following antigen challenge . Thus , the CFSE dilution in antigen-responsive T-cells in a co-injection ( adoptive transfer ) protocol indicates the percentage of T-cells proliferating in the absence or presence of Tregs . Additionally , by loading Tregs with CFSE and tracking transferred OT-II T-cells by the expression of CD45 . 2 , it can be demonstrated that both the subsets localize to the same lymph node , thereby enabling this analysis ( Figure 5—figure supplement 1H ) . Following the principle described above , naïve T-cells isolated from OT-II ( CD45 . 2+ ) mice , were loaded with the fluorescent dye CFSE as described in methods . CFSE loaded cells were injected into congenic hosts , ( CD45 . 1+ , B6SJL mice ) , which differ only in the expression of the CD45 isoform . Naïve T-cells were co-injected without or with in vitro activated ( CD45 . 1+ ) Tregs , which were earlier transfected with either shRNA to Notch1 or a scrambled control . Host mice were challenged with the appropriate antigen ( mOVA ) , 15–18 hr after receiving cells . Three days post challenge , cells were isolated from draining lymph nodes of host mice and analyzed for CFSE dilution in the gated CD45 . 2+ subset donor OT-II T-cells ( Figure 5E , panel i ) . In mice co-injected with Tregs transfected with scrambled shRNA , proliferation of OT-II cells was expectedly blunted and CFSE fluorescence in OT-II cells detected at the highest intensity of CFSE indicating an undivided population ( Figure 5E panel ii ) . On the other hand , in cells isolated from mice injected with Tregs transduced with Notch1 shRNA , OT-II T-cells proliferated robustly ( Figure 5E panel iii ) , and were comparable to proliferation in OT-II cells injected without Tregs ( Figure 5E , panel i ) . This indicates that Notch1 controls the suppressor activity of Tregs . The expression of Foxp3 was not changed by Notch1 ablation ( Figure 5Eii , inset and Figure 5—figure supplement 1I ) . Similarly , suppressor activity was attenuated in Notch1-/- Tregs ( Figure 5F , compare panels , i and iii ) as compared to wildtype Tregs ( TregsWT ) ( Figure 5F panel ii ) . However , reconstitution with recombinant NIC restored suppressor activity of Notch1-/- Tregs ( Figure 5G , compare iii and ii ) . While the data suggest that Notch1 activity can regulate Treg suppressor function , these results are in contrast to an earlier observation from the laboratory wherein Notch1-/- Tregs showed activity comparable to control Tregs in vitro co-culture suppressor assays ( Perumalsamy et al . , 2012 ) , which we can reproduce ( data not shown ) . Thus , the suppressor assay in the culture dish appears to recapitulate a subset of the diverse cues present in vivo . Another possibility is that the accumulation of IL-2 produced by T-effectors in the co-culture assay , allows for the survival and hence suppressor activity of Notch1-/- Tregs in vitro . The in vivo assay was a more sensitive read-out of Notch activity in Tregs and we next tested if modulating autophagy in activated Tregs modified suppressor activity . In agreement with the reported requirement of autophagy in Treg function ( Parekh et al . , 2013 , Wei et al . , 2016 ) , shRNA to Vps34 abrogated TregsWT suppressor activity relative to the cells treated with the scrambled shRNA ( Figure 5H compare iii with ii ) as compared to the control group injected without Tregs ( Figure 5Hi ) . The ablation of Vps34 ( inhibition of autophagy ) did not modulate the expression of Foxp3 ( Figure 5I , J ) . However , in an in vitro assay of Treg function , which measures suppression of conventional T-cell proliferation via stimulation of T-cell receptor in the presence of antigen presenting cells , the ablation of VpS34 in activated Tregs abrogated suppressor activity ( Figure 5K , compare i and ii ) . Responder T-cell proliferation was comparable to proliferation of cells stimulated without Tregs ( Figure 5K , iii ) . Thus , Vps34 was critical for Treg function and its requirement revealed in suppressor assays performed in vitro or in vivo . Furthermore , suppressor activity in assays in vivo was restored in Notch1-/- Tregs transfected with recombinant Atg3 ( Figure 5L and Figure 5—figure supplement 1J ) . Expression of Atg3 had also protected cells from apoptosis , shown in earlier experiments . Since Notch1 modulated Treg function , we tested if this was mediated through non-nuclear Notch activity . In the OT-II T-cell proliferation assay , expression of recombinant NIC-NES restored suppressor activity in transferred Notch1-/- Tregs ( Figure 6Aii ) . Suppressor activity was not restored in Notch1-/- Tregs transfected with NIC-NLS ( Figure 6Aiii ) . All comparisons were made with Notch1-/- Tregs transfected with a vector control group ( Figure 6Ai ) , in which condition suppressor activity is not detected . Next we tested the requirement of non-nuclear Notch and Atg3 activity in Treg function , in another model of immune-inflammation . The assay tracks the clearance of glucose injected into the blood stream of obese mice , and builds on the observation that impaired glucose clearance ( also referred to as glucose intolerance ) is a feature of obesity ( Winer et al . , 2009 ) . In this context , the injection of Tregs has been shown to correct defects in the clearance of glucose , although protection is short-term and persists for 10–15 days ( Cipolletta et al . , 2012 , Feuerer et al . , 2009 ) . The underlying mechanism of correction by Tregs is not understood , as there are no deficiencies in Treg number or function reported in obese ( leptin receptor-deficient , Leprdb/db ) mice ( De Rosa et al . , 2007 ) . Nonetheless , the correction of defective glucose clearance remains a reproducible assay of Treg function . In agreement with published observations we show that the clearance of ( injected ) glucose from blood in fasting Leprdb/db mice , is substantially delayed compared to heterozygous ( Leprdb/+ ) littermates ( Figure 6B ) . However , in Leprdb/db mice injected with TregsWT , and tested 7–8 days later , clearance of blood glucose is comparable to Leprdb/+animals ( Figure 6B ) . Correction is transient and is lost within 2 weeks following adoptive transfer ( not shown ) . These data recapitulate published observations made with Tregs in this model ( Eller et al . , 2011 ) . Hence , we next tested Notch1-/- Tregs in this assay system . Activated Notch1-/- Tregs transfected with the control vector did not improve glucose clearance , however , transferring NIC-NES expressing Notch1-/- Tregs , restored the clearance of glucose to rates comparable to Leprdb/+ mice ( Figure 6C ) , indicating that enrichment of non-nuclear Notch activity restored Notch1-/- Tregs activity in this assay . Similarly , as seen in the T-cell proliferation assay , injecting Notch1-/- Tregs transfected with recombinant Atg3 , also corrected the rate of glucose clearance to a level comparable to that of control , non-obese mice ( Figure 6—figure supplement 1A ) . Collectively , the experiments confirmed an important role for Notch1 activity in Treg function . 10 . 7554/eLife . 14023 . 013Figure 6 . Non-nuclear NIC regulates Treg suppressor activity . ( A ) Proliferation of CFSE loaded CD4+ OT-II cells co-injected with untransfected , or NIC-NES or NIC-NLS transfected Notch1-/- Tregs . The numbers indicate the percentage of the population with CFSE dilution Inset: staining for Notch1 in transfected cells . ( B ) IPGTT in Leprdb/+ ( △ ) or Leprdb/db ( filled ∆ ) mice injected intravenously with PBS ( △ or▲ ) , or WT ( ⬤ ) or Notch1-/- ( □ ) Tregs . ( C ) IPGTT in Leprdb/+ ( △ ) or Leprdb/db mice injected intravenously with Notch1-/- Tregs expressing recombinant NIC-NES ( ■ ) or pBABE ( □ ) . Data are mean ± SD of 2 independent experiments with three animals in each condition . ( D ) Whole lymph nodes ( representative ) from Cre+ and Cre- mice and plotted below , cell recoveries from lymph nodes from mice from different experiments . ( E ) T-cell immune cells subset analysis in lymph nodes isolated from wild type and mutant mice . ( F ) Subset analysis of T-cells in lymph-nodes of mutant mice , which are either not injected or injected with WT Tregs mice seven days prior to analysis . ( G ) Representative confocal images ( field views ) of activated Tregs from Cd4-Cre::Notch1lox/lox and Cre-ve mice fixed and stained for Foxp3 ( green ) and counterstained with Hoechst 33342 ( nucleus , blue ) . *p≤0 . 03; Scale bar: 5 μm This figure is accompanied by Figure 6—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 14023 . 01310 . 7554/eLife . 14023 . 014Figure 6—figure supplement 1 . Treg suppressor function is regulated by Notch1 signaling from the cytoplasm . ( A ) IPGTT analysis in Leprdb/+ ( △ ) or Leprdb/db , mice injected intravenously seven days prior to the assay with activated Notch1-/-Tregs infected to express recombinant ATG3 ( ⬤ , ■ ) or pBABE ( □ ) . ( B ) Analysis of lymphoid subsets isolated from lymph nodes of mice of the indicated genotypes . ( C ) Lymph node cellularity in Foxp3-Cre::Notch1lox/lox mice either untreated or injected with WT-Tregs seven days prior to the analysis . ( D ) Representative images of littermates of genetic ( Cre- ) controls and Foxp3-Cre::Notch1lox/lox ( Cre+ ) mice at 3 weeks of age . ( E ) Representative images showing coarsened and patchy distribution of fur and crusting around eyes in Foxp3-Cre::Notch1lox/lox mice . DOI: http://dx . doi . org/10 . 7554/eLife . 14023 . 014 Another line of evidence implicating Notch1 in Treg homeostasis came from mice with a targeted deletion of Notch1 in the Treg lineage ( i . e . under the control of the Foxp3 promoter ) . Mice with a Notch1 ablation , were born in reduced numbers ( 1:16 as against the expected 1:8 ratio ) , presented with features consistent with dysregulated immune function and some similarities with Scurfy mice , which lack Tregs because of a loss-of-function mutation in the Foxp3 gene ( Fontenot et al . , 2003 ) . Foxp3-Cre::Notch1lox/lox mice ( analyzed from 4–8 weeks of age ) were viable , presented with enlarged lymph nodes and increased cellularity relative to Cre-negative littermate controls ( Figure 6D ) . Increased cellularity resulted from increased numbers of CD4+CD62LlowCD44high ( effector memory ) and CD4+CD62LhighCD44high ( central memory ) T-cell subsets ( Figure 6E ) . Changes in the CD4+CD25high and CD4+CD69+ subsets , which are expressed on Tregs and are early markers of T-cell activation respectively , were inconsistent , showing small increases in number in some mice or remaining comparable to genetic controls in others ( Figure 6E ) . No striking changes were observed in B-cell and myeloid subsets ( Figure 6—figure supplement 1B ) . Further , in Foxp3-Cre::Notch1lox/lox mice , which had been injected with WT Tregs , seven days prior to analysis , lymph node inflammation and the number of activated CD4+ T-cell subsets in lymph nodes was substantially reduced ( Figure 6F and Figure 6—figure supplement 1C ) . This suggested that defects in Treg function might contribute to the observed inflammation in Foxp3-Cre::Notch1lox/lox mice . In preliminary experiments , the adoptive transfer of Notch1-/- Tregs did not reduce lymphoid accumulation ( not shown ) . Notably , several Foxp3-Cre::Notch1lox/lox mice were smaller in size than littermate controls , had shorter , roughened fur and developed crusting of skin around the eyes by 4–5 weeks ( Figure 6—figure supplement 1D , E ) . Nonetheless , Foxp3 expression , assessed by immunostaining in Tregs was comparable in mice of the different genotypes ( Figure 6G ) . The analysis of Notch1-/- Tregs in functional assays of inflammation together with the phenotypes of mice with a deletion of Notch1 in Tregs suggests an important role for the receptor in immune homeostasis .
Tregs control inflammatory responses mounted by the immune system ( Buckner , 2010; Sakaguchi , 2004 ) . In this study we position Notch1 regulated autophagy as an integral controlling element of activated Treg homeostasis . This is inferred from the observations that survival and function are compromised following perturbations of Notch signaling or autophagy in TregsWT and the restoration of autophagy and suppressor activity following reconstitution of Notch1-/- Tregs with the nuclear-excluded recombinant NIC-NES or Atg3 . Of note , the requirement for Notch1 in thymus-derived activated Tregs is distinct from identified roles for Notch-TGFβ interactions in iTregs ( Samon et al . , 2008; Mota et al . , 2014 ) . More generally , our experiments suggest that Notch-induced autophagy is a mechanism of quality control of mitochondrial architecture . This was also indicated by the restoration of organelle connectivity if NIC-NES or Atg3 were expressed in Notch1-/- Tregs or in T-effectors . The interactions between the organelle remodeling machinery observed earlier ( Perumalsamy et al . , 2010 ) ) and the Notch-autophagy signaling axis , underpinning these outcomes remain to be characterized . Cell survival in complex and changing environments associated with inflammation is an important component determining immune cell function . The induction of cell death in Tregs in response to cytokine deprivation appeared to correlate with suppressor function in vivo with one exception . The loss of suppressor activity in Notch1-/- Tregs in vivo , did not agree with outcomes of suppressor assays in vitro , wherein Notch1 appeared to be dispensable for Treg suppression ( Perumalsamy et al . , 2012 ) . Based on the understanding of Notch activity in Tregs we speculate that Notch1-/- Tregs suppress T-cell proliferation in vitro as concentrations of cytokines and growth factors that build up in the dish likely protect Notch1-/- Tregs from death . As Foxp3 levels are not compromised in these cells , this will allow suppressor activity . That the ablation of Vps34 compromised Treg activity in both in vivo and in vitro assays , is consistent with autophagy controlling Treg survival and playing a defining role in maintenance of functional identity ( Wei et al . , 2016 ) , i . e . regulating Foxp3 expression . While we did not detect a loss in Foxp3 expression in the 24–48 hr period following ablation of Vps34 , the eventual loss of Foxp3 expression and compromised survival as shown by Wei , et al . , may underlie defective suppressor activity observed in the 3–4 day duration suppressor assays . The dynamic regulation of Notch and its role as a critical determinant of Treg differentiation , function and homeostasis has emerged from work by another group ( Charbonnier et al . , 2015 ) . Similarly , a more recent report ( Wei et al . , 2016 ) , focused on events at the earlier stages of Treg undergoing activation to define a critical role for autophagy during differentiation of Tregs . Although focusing on different stages of ( natural ) Treg development , taken together , the studies suggest dynamic regulation of Notch activity ( Charbonnier et al . , 2015 ) or autophagy ( Wei et al . , 2016 ) in Treg homeostasis , which support the conclusions from our work albeit with some differences . Unlike the study by Charbonnier et al . , which positions RBPJ-k dependent Notch signaling as a negative regulator of Treg function we find that deletion of Notch1 in the Treg lineage resulted in features of immune-inflammation consistent with a role for Notch1 in Treg function . However , both the studies implicate non-canonical Notch1 activity as necessary for maintenance of Treg identity . We speculate that the increased representation of recently activated and T-memory subsets in Foxp3-Cre::Notch1lox/lox mice , observed in our experiments , is likely a response to antigens experienced in the high barrier – but not SPF1 – conditions , that mice are housed in . Notably , genetic ( littermate ) controls were indistinguishable from wild type mice indicating the absence of overt infection . Interestingly , the phenotypes of autoimmunity reported in Atg7 deficient mice ( Wei et al . , 2016 ) , align with lymphoid proliferation and increased cellularity we observe in Foxp3-Cre::Notch1lox/loxmice . The interaction between Notch and autophagy revealed in our experiments suggests a hitherto unappreciated role for Notch signaling in the regulation of this process . Whether the effects of autophagy is executed through Notch1 signaling at all stages of Treg activation remains to be investigated . Our experiments have focused on activated Tregs , an approach that identified a defining role for mTORC1 in Treg homeostasis ( Zheng et al . , 2013 ) . Multiple studies have confirmed that the deficiency of Notch1 at later stages of T-cell development does not impair lineage commitment of Tregs . We propose however , that Notch1 tunes a late event in the differentiation of Tregs already committed to the suppressor lineage . This is consistent with our observation that low-levels of immune challenge elicit features of inflammation as compared to the Cre negative littermates . Notch is implicated in instructive fate choices in the T-cell lineage , with commitment to T-cell fate the earliest amongst these ( Kopan and Ilagan , 2009 ) . Here we demonstrate a critical role for non-nuclear Notch1 activity , wherein Notch-autophagy interactions follow from immune stimulation and are important determinants of Treg function as evidenced in assays of adoptive transfers . Despite defects in Notch1-/-Treg function , mice with an ablation of Notch1 in mature CD4+T-cells did not present overt inflammatory phenotypes , which likely reflect its role in tuning responses of effector T-cells ( Laky et al . , 2015 ) . However , the targeted deletion of Notch1 in the Treg lineage resulted in inflammatory features suggested an integral role for Notch1 in Treg homeostasis . The possibility that Notch1 activity in Tregs is critical in specific contexts and redundant or non-essential in others cannot be excluded ( Zhou et al . , 2015 ) . It is tempting to speculate that the integration of autophagy and non-nuclear Notch activity may be a conserved mechanism that tunes cell-fate decisions governed by the receptor in other cell types .
The Notch1lox/lox and Cd4-Cre::Notch1lox/lox ( Notch1-/- ) strains were a gift from Freddy Radtke ( Wolfer et al . , 2001 ) . Foxp3tm4 ( YFP/Cre ) Ayr/J , C57BL/6J , B6SJL , OT-II and Leprdb/db strains were obtained from the Jackson Laboratory . Notch1lox/lox and Foxp3tm4 ( YFP/Cre ) Ayr/J strains were crossed to generate Foxp3-Cre::Notch1lox/lox mice . Notch1lox/lox mutant mouse stains were backcrossed with C57BL/6 mice . Except where specified , all experiments used mice within the age group of 8–12 weeks . Mice were housed in controlled temperature and light environments that are maintained in high barrier conditions with specific IVC ( individually ventilated cages ) controlled systems . The housing environment is tested and routinely monitored for the full pathogen panel recommended by FELESA ( Federation of Laboratory Animal Science Associations ) . Breeding colonies were maintained in-house and all experimental protocols were approved by the Institutional Animal Ethics Committee ( NCBS-AEC-AS-6/1/2012; INS-IAE-2016/01[N] ) and are in compliance with the norms of the Committee for the Purpose of Control and Supervision of Experiments on Animals , Govt . of India . To activate cells , CD4+CD25+ natural Tregs were isolated from murine spleens ( using a combination of negative selection to enrich for CD4+ cells and a second step of positive selection for CD25+ cells ) following manufacturers instructions ( R&D Systems or Invitrogen ) . The CD4+CD25+ cells thus isolated were confirmed to be ~95% Foxp3+ by immune-staining followed by confocal microscopy based analysis across multiple experiments . Cells were activated ( ~2x106/ml ) by co-culturing with beads coated with antibodies to CD3 and CD28 ( 20ul/ ml; Invitrogen ) in 24 well plates . After 48 hr , beads were removed by magnetic separation , and activated Tregs continued in culture in 50% conditioned medium and IL-2 ( 2 ng/ml ) ( R&D Systems ) or used in assays . For the generation of induced Tregs , naïve CD4+ T-cells isolated by negative selection as above were activated ( 1x106/ml ) using beads coated with antibodies to CD3 and CD28 ( 20 ul/ml; Invitrogen ) in presence of IL-2 ( 2 ng/ml ) and TGF-β ( 2 ng/ml ) for 72 hr . Generation of T-effectors and retroviral generation and infection of T-cells was as previously described ( Perumalsamy et al . , 2012 ) . Briefly , retroviruses were packaged in HEK 293T using the packing vector pCLEco . The viral supernatant was concentrated and cells were infected after 24 hr stimulation by spinfection in 24 well plates ( 500 g for 90 min at 32°C ) and continued in culture . After 48 hr , cells were harvested and continued in medium supplemented with IL-2 ( 1 ug/ml ) for another 18–24 hr . Culture conditions included IL-2 ( 1 ug/ml ) and the antibiotic Puromycin ( 1 ug/ml ) for 48 hr to enrich for infected cells . Live cells were selected on day 2 by centrifugation in Ficoll , prior to use in functional assays . Knockdowns were assessed by Western blotting analysis of cell lysates ( 0 . 3–0 . 5 x106 cells per lysate ) post antibiotic selection . The plasmids pBABE , pBABE-NIC-NLS , pBABE-NIC-NES were gifts and have been described before ( Perumalsamy et al . , 2010 ) . pBABE-puro-ATG3 was from Addgene ( MA , USA ) . shRNA specific for VPS34 , ATG7 , RBPJ-k , Dll-1 , Notch1 and scrambled control were from Origene . Activated Tregs were cultured with or without IL-2 ( 0 . 3 x 106/ml ) . After 15–18 hr , cells were harvested and tested for the induction of apoptotic damage . Nuclear morphology was scored in 200–300 cells across five fields in coded samples stained with Hoechst 33 , 342 ( 1 μg/ml ) for 3–5 min at ambient temperature . Cells were stained with DiOC6 ( 40nM in PBS ) for 10 min at 37°C , washed to remove excess dye , re-suspended in PBS and mitochondrial transmembrane potential analyzed by flowcytometry . Cell lysates were prepared using 0 . 4 x 106 cells . Briefly , cell pellets were re-suspended ( by vortexing ) in 20–25 ul of SDS lysis buffer ( 2% SDS , glycerol , bromophenol blue , 1 M DTT and 1 M Tris-Cl pH 6 . 8 supplemented with a protease inhibitor cocktail - aprotinin , leupeptin and pepstatin ( 2 μg/ml each ) , 10 uM MG132 , 1 mM PMSF , 1 mM NaF and 1 mM Na3VO4 ) and boiled for 10 min at 100°C . Whole cell lysates were resolved by SDS-PAGE and transferred to nitrocellulose membrane , ( GE Healthcare ) and incubated overnight at 4°C with primary antibodies at concentrations recommended by the manufacturers . The membrane was washed thrice with TBS-Tween20 followed by HRP-conjugated secondary antibody ( CST , 1:1000 dilution ) for 1 hr at ambient temperature . Membranes were developed using an ImageQuant LAS 4000 Biomolecular Imager ( GE Healthcare ) and quantified with Image J software . For immune-precipitation analysis , 4x106 Tregs were lysed for 30 min at 4°C on a rotational cell mixer in 1%NP40 buffer ( 50 mM Tris , 1 mM NaCl , 1 mM EDTA ) supplemented with aprotinin , leupeptin and pepstatin ( 2 μg/ml each ) , 10uM MG132 , 1 mM PMSF , 1 mM NaF and 1 mM Na3VO4 . Debris is removed by centrifugation and the supernatant incubated with primary antibody or IgG control ( 10 µg ) for 1 hr at 4°C on a rotational cell mixer . The Immune complexes were precipitated for 2 hr at 4°C using washed Sepharose G plus beads ( 70 ul ) on a rotational cell mixer . Beads bound to complexes were washed five times with ice-cold PBS by centrifugation at 1700 rpm . Finally , beads were boiled in SDS lysis buffer for 10 min before western blot analysis . Primary antibodies used for western blot analysis include LC3 ( D3U4C ) , Atg7 ( D12B11 ) , VPS34 ( D9A5 ) Atg5 ( D5F5U ) , Beclin-1 ( D40C5 ) and Atg14 were from Cell Signaling Technology ( used at a dilution of 1:500 ) ; NICD ( clone mN1A ) and DLL-1 ( C-20 ) from Santa Cruz Biotechnology ( used at adilution of 1:250 ) ; α-tubulin and actin ( used at a dilution of 1:250 ) from Neomarker and Hes-1 ( used at a dilution of 1:250 ) was from Millipore . T-cells adhered to poly-D-lysine coated cover-slips were stained with 100 nM MitoTracker Green for 20 min at 37°C in complete medium , washed to remove excess dye and imaged with a Zeiss LSM Meta 510 as Z-stacks ( 1 . 0 µm , 3 zoom ) using Plan-Apochromat 63 × NA 1 . 4 oil-immersion objective . Images were de-convoluted using Zeiss LSM software and stacks re-merged with Image J for Z-projections . In all experiments involving confocal microcopy based mitochondrial analysis; images of 20–25 cells per experimental conditions , across 2–3 independent were taken . For FRAP analysis , cells were imaged using Zeiss LSM Meta 510 microscope ( oil immersion objective , 63 × . 0 . 9 NA ) . A confocal system with an integrated FRAP module , collected images every 2 s immediately after photo bleaching ( circular ROIs of 1 . 5–2 . 0 μm diameter ) . Fluorescence recovery was analyzed after correcting for photo bleaching and background noise . FRAP measurement was performed on a minimum of 10 cells/ cell type from 3 independent experiments . T-cells ( 2x106 ) adhered to poly-D-lysine–coated dishes ( 1 ug/ml in PBS coated for 15 min at RT ) were fixed with 2% paraformaldehyde for 20 min at RT , permeabilized with neat methanol for 12 min on dry ice or at -20°C ( LC3 ) ; 0 . 1% Tween-20 or 0 . 2% NP40 for 10 min at RT ( Val1744 or Foxp3 and mNIA clones respectively ) and then blocked with 5% BSA at RT . Samples were stained with primary antibodies overnight at 4°C ( at dilutions of 1:100 for Val1744 , 1:50 for mN1A , 1:100 for Foxp3 ) and secondary antibodies for 1 hr at room temperature . Antibodies were from the following sources: cleaved Notch1 ( Notch-Val1744 ) and LC3 ( D3U4C ) were from Cell Signaling Technology; NICD ( clone mN1A ) from Santa Cruz Biotechnology and Foxp3 ( FJK-16s ) was from eBiosciences . Secondary antibodies were from Invitrogen and used at a dilution of 1:1000 . Samples were imaged on Zeiss LSM Meta 510 as Z-stacks ( 1 . 0 µm , 3 zoom ) ; Plan-Apochromat 63 × NA 1 . 4 oil-immersion objective . The stacks were processed to remove background based on secondary controls and Z-projected using Image J software . For all experiments involving imaging based analysis n = 20 cells/ condition in every experiment , across 3 independent experiments . For immunophenotyping , cells isolated from lymph nodes of mice with the required genotypes were stained with antibodies to indicated cell surface proteins and analyzed by flowcytometry . The analysis of cell surface markers was made on the lymphocyte population gated for size in the Forward scatter ( Fsc ) vs . Side scatter ( Ssc ) plot . RNA was isolated from 4x106 activated Tregs using a TRIzol ( Invitrogen ) -based RNA extraction protocol following the manufacture’s instructions . cDNA was prepared using Superscript II ( Invitrogen ) . Quantitative PCR was performed with SYBR Green ( Thermo Scientific ) in triplicate . Relative expression was calculated using the using the ΔΔ threshold cycle method ( 2-ΔΔCT ) . Primers sequences are as follows: CTLA-4 Fwd:GGACGCAGATTTATGTCATTGATC , CTLA-4 Rev:CCAAGCTAACTGCGACAAGGA TGF-Beta Fwd:CAACGCCATCTATGAGAAAACC , TGF-Beta Rev:AAGCCCTGTATTCCGTCTCC GITR Fwd:AACGGAAGTGGCAACAACAC , GITR Rev:CTTGGGGCACAGAGGAAGA IL-10 Fwd:GAAGACCCTCAGGATGCGG , IL-10 Rev:CCTGCTCCACTGCCTTGCT IL-35 Fwd:CAATCACGCTACCTCCTCTTT , IL-35 Rev:AGTTTTTCTCTGGCCGTC CD103Fwd:CGTGGAGAAGAAGGCAGAGT , CD103 Rev:TCGGGGGTAAAGGTCATAGAT Eos Fwd:CCAAGTCCCTGAGTGGTTGT , Eos Rev:TTATCCAGGAGCCGTTCATC Helios Fwd:ACACCTCAGGACCCATTCTG , Helios Rev:TCCATGCTGACATTCTGGAG Neuropilin-1 Fwd:AGCAAGCGCAAGGCTAAGTC , Neuropilin-1Rev:ATCCTGATGAACCTTGTGGAGAGA Ox40 Fwd: CGAATTCCACCATGTATGTGTGGGTTCAG , Ox40Rev:CGGGATCCTCAGGAGCCACCAAGGTGGG Foxp3 Fwd:CACCTATGCCACCCTTATC , Foxp3 Rev:TCCTCTTCTTGCGAAACTC HPRT Fwd:TCAGTCAACGGGGGACATAAA , HPRT Rev:GGGGCTGTACTGCTTAACCAG Activated Tregs from CD4Cre-ve and CD4Cre+ve mice were also analyzed using Affymetrix Mouse_GXP_8X60K AMADID: 49 , 771 using Gene Spring GX Software . CD4+ naïve T cells were isolated from OTII transgenic mice spleens by negative selection using magnetic bead based separation protocols specified by manufacturers ( MagCellect , R & D System ) . Freshly isolated native T-cells ( 2x106 cells/ml in pre-warmed PBS ) were loaded with 5uM CFSE and incubated for 8 min at 37°C ( water bath ) . Cells were washed with complete medium to remove excess dye before use in suppressor assays . 0 . 5x 106 CFSE loaded CD4+ naïve OT-II T-cells , were co-injected with 0 . 3x106 activated Tregs intravenously ( i . v ) into congenic B6SJL host mice . 24 hr later , host mice were subcutaneously injected with 30 ug maleylated-ovalbumin ( OVA ) in CFA . 3 days later host lymph nodes were analyzed for CSFE dilution in the donor cells by gating on the CD45 . 2+ population by flowcytometry . For the in vitro suppressor assay , 10x104 CFSE loaded naive CD4+T-cells were activated in 96 well flat bottom plate ( in triplicates ) with soluble anti-CD3 ( 250 ng/well ) and Mitomycin C ( 50 ug/ml ) treated APC ( 5 x104 ) in the presence of activated Tregs ( 1:4::activated Treg:T-cells ) . 72 hr post-stimulation cells pooled from replicate wells were stained for CD4 expression and CFSE dilutions assessed in the CD4+ cells by flowcytometry . Rescue of lymphadenopathy in Foxp3-Cre::Notch1lox/lox mice was based on a published protocol ( Fontenot et al . , 2003 ) . Briefly , 1x106 freshly isolated WT Tregs were adoptively transferred ( i . v . ) into 4-week old Foxp3-Cre::Notch1lox/lox mice . Seven days later , single cell suspensions were prepared from the lymph nodes ( axillary and inguinal ) of each of the mice injected with Tregs and matched un-injected controls . Lymph node cells were gated on the lymphocyte population based on size by the Forward scatter ( Fsc ) vs . Side scatter ( Ssc ) plot and analyzed for cell surface markers by flowcytometry using a BD FACSCalibur Cell Analyzer . Mice fasted for 6 hr were injected intra-peritoneally ( ip ) with 150 μl D-Glucose ( 2 g/kg body weight ) in water . D-Glucose ( Fischer scientific ) - 2 grams per kilogram of body weight ( g/kg ) prepared in autoclaved water - was injected intra-peritoneally ( i . p . ) into Leprdb/db and Leprdb/+ mice fasted for 6 hr . Before injecting the glucose solution the base-line blood glucose level was checked . After injecting the glucose solution , blood samples were obtained from the tail tip at the indicated times , and blood glucose concentrations were measured using a handheld glucometer ( Contour TS , Blood glucose meter , Bayer ) . In experiments testing in vivo function of Tregs , 0 . 8x106 activated Tregs ( Notch1-/- or genetic controls ) were adoptively transferred into the Leprdb/db mice by the intravenous route using the tail vein . 7 days after adoptive transfer , IPGTT was performed on these mice . Data shown are the mean ± SD from 3 independent experiments , unless stated otherwise . Statistical significance was calculated with the two-population Student’s t test . Data plots for FRAP assays are mean ± SEM from 3 experiments . Images from confocal microscopy were analyzed using Image J software ( NIC , USA ) and Adobe Photoshop . The confocal image stacks were processed to remove background based on secondary controls and Z-projected using Image J software . The Region of Interest ( ROI ) restricted integrated density was quantified for each cell in Image J software and normalized for area to calculate mean pixel intensity . | A cell must have access to adequate amounts of nutrients if it is to survive and perform its role in the body . Some cells , including immune cells , must travel around the body to sites where nutrients may be in short supply . For example , a type of immune cell called the T-regulatory cell travels to wounds and other hostile places where other types of immune cells are actively fighting infections and/or damaged cells . The T-regulatory cell’s job is to reduce excess or harmful immune activity to prevent the immune system from damaging healthy cells in the body . However , it was not clear how they survive and protect themselves in these tough situations . In 2012 , researchers showed that a protein called the Notch receptor helps the T-regulatory cells survive in conditions where nutrients are scarce . Now , Marcel et al . – including some of the researchers involved in the 2012 study – have investigated how the Notch receptor protects mouse T-regulatory cells . The experiments show that T-regulatory cells grown in nutrient-poor conditions in the laboratory activate a protective mechanism known as autophagy . This mechanism recycles damaged or nonessential structures within the cell into nutrients . The Notch receptor is needed to trigger autophagy at the right time and the loss of this receptor causes the cells to die when they are starved of nutrients . Further experiments found that the Notch receptor is essential for the T-regulatory cells to work properly in mice . Animals born lacking this receptor in T-regulatory cells quickly develop inflammation in response to even minor irritations . More studies are needed to reveal the details of this protective strategy , and to find out if it is used by other cells in the body . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"cell",
"biology",
"immunology",
"and",
"inflammation"
] | 2016 | Notch1 regulated autophagy controls survival and suppressor activity of activated murine T-regulatory cells |
The NLRP1 inflammasome is a multiprotein complex that is a potent activator of inflammation . Mouse NLRP1B can be activated through proteolytic cleavage by the bacterial Lethal Toxin ( LeTx ) protease , resulting in degradation of the N-terminal domains of NLRP1B and liberation of the bioactive C-terminal domain , which includes the caspase activation and recruitment domain ( CARD ) . However , natural pathogen-derived effectors that can activate human NLRP1 have remained unknown . Here , we use an evolutionary model to identify several proteases from diverse picornaviruses that cleave human NLRP1 within a rapidly evolving region of the protein , leading to host-specific and virus-specific activation of the NLRP1 inflammasome . Our work demonstrates that NLRP1 acts as a 'tripwire' to recognize the enzymatic function of a wide range of viral proteases and suggests that host mimicry of viral polyprotein cleavage sites can be an evolutionary strategy to activate a robust inflammatory immune response .
The ability to sense and respond to pathogens is central to the mammalian immune system . However , immune activation needs to be properly calibrated , as an overactive immune response can at times be as pathogenic as the pathogen itself . To ensure accurate discrimination of self and non-self , innate immune sensors detect broadly conserved microbial molecules such as bacterial flagellin or double-stranded RNA ( Janeway , 1989 ) . However , such microbial patterns can be found on harmless and pathogenic microbes alike . More recently , pathogen-specific activities such as toxins or effector enzymes have also been shown to be targets of innate immune recognition ( Jones et al . , 2016; Mitchell et al . , 2019; Vance et al . , 2009 ) . Such a system for detection , termed effector-trigged immunity ( ETI ) , has been well-established in plants ( Cui et al . , 2015; Jones et al . , 2016 ) and is emerging as an important means to allow the immune system to distinguish pathogens from harmless microbes in animals ( Fischer et al . , 2020; Jones et al . , 2016 ) . Complicating the success of host detection systems , innate immune sensors are under constant selective pressure to adapt due to pathogen evasion or antagonism of immune detection . Such evolutionary dynamics , termed host-pathogen arms races , result from genetic conflicts where both host and pathogen are continually driven to adapt to maintain a fitness advantage . The antagonistic nature of these conflicts can be distinguished via signatures of rapid molecular evolution at the exact sites where host and pathogen interact ( Daugherty and Malik , 2012; Meyerson and Sawyer , 2011; Sironi et al . , 2015 ) . Consistent with their role as the first line of cellular defense against incoming pathogens , innate immune sensors of both broad molecular patterns as well as specific pathogen-associated effectors have been shown to be engaged in genetic conflicts with pathogens ( Cagliani et al . , 2014; Chavarría-Smith et al . , 2016; Hancks et al . , 2015; Tenthorey et al . , 2014; Tian et al . , 2009 ) . Inflammasomes are one such group of rapidly evolving cytosolic immune sensor complexes ( Broz and Dixit , 2016; Chavarría-Smith et al . , 2016; Evavold and Kagan , 2019; Rathinam and Fitzgerald , 2016; Tenthorey et al . , 2014; Tian et al . , 2009 ) . Upon detection of microbial molecules or pathogen-encoded activities , inflammasome-forming sensor proteins serve as a platform for the recruitment and activation of proinflammatory caspases including caspase-1 ( CASP1 ) through either a pyrin domain ( PYD ) or a caspase activation and recruitment domain ( CARD ) ( Broz and Dixit , 2016; Rathinam and Fitzgerald , 2016 ) . Active CASP1 mediates the maturation and release of the proinflammatory cytokines interleukin ( IL ) −1β and IL-18 ( Broz and Dixit , 2016; Rathinam et al . , 2012 ) . CASP1 also initiates a form of cell death known as pyroptosis ( Broz and Dixit , 2016; Rathinam et al . , 2012 ) . Together , these outputs provide robust defense against a wide array of eukaryotic , bacterial , and viral pathogens ( Broz and Dixit , 2016; Evavold and Kagan , 2019; Rathinam and Fitzgerald , 2016 ) . The first described inflammasome is scaffolded by the sensor protein NLRP1 , a member of the nucleotide-binding domain ( NBD ) , leucine-rich repeat ( LRR ) -containing ( NLR ) superfamily ( Martinon et al . , 2002; Ting et al . , 2008 ) . NLRP1 has an unusual domain architecture , containing a CARD at its C-terminus rather than the N-terminus like all other inflammasome sensor NLRs , and a function-to-find domain ( FIIND ) , which is located between the LRRs and CARD ( Ting et al . , 2008 ) . The FIIND undergoes a constitutive self-cleavage event , such that NLRP1 exists in its non-activated state as two , noncovalently associated polypeptides ( D'Osualdo et al . , 2011; Finger et al . , 2012; Frew et al . , 2012 ) , the N-terminal domains and the C-terminal CARD-containing fragment . The importance of the unusual domain architecture of NLRP1 for mounting a pathogen-specific inflammasome response has been elucidated over the last several decades ( Evavold and Kagan , 2019; Mitchell et al . , 2019; Taabazuing et al . , 2020 ) . The first hint that NLRP1 does not detect broadly conserved microbial molecules came from the discovery that the Bacillus anthracis Lethal Toxin ( LeTx ) is required to elicit a protective inflammatory response against B . anthracis infection via one of the mouse NLRP1 homologs , NLRP1B ( Boyden and Dietrich , 2006; Greaney et al . , 2020; Moayeri et al . , 2010; Terra et al . , 2010 ) . Paradoxically , inflammasome activation is the result of site-specific cleavage in the N-terminus of mouse NLRP1B by the Lethal Factor ( LF ) protease subunit of LeTx , indicating that protease-mediated cleavage of NLRP1 does not disable its function but instead results in its activation ( Chavarría-Smith and Vance , 2013; Levinsohn et al . , 2012 ) . More recently , the mechanism by which LF-mediated proteolytic cleavage results in direct NLRP1B inflammasome activation has been detailed ( Chui et al . , 2019; Sandstrom et al . , 2019 ) . These studies revealed that proteolysis of mouse NLRP1B by LF results in exposure of a novel N-terminus , which is then targeted for proteasomal degradation by a protein quality control mechanism called the ‘N-end rule’ pathway ( Chui et al . , 2019; Sandstrom et al . , 2019; Wickliffe et al . , 2008; Xu et al . , 2019 ) . Since the proteasome is a processive protease , it progressively degrades the N-terminal domains of NLRP1B but is disengaged upon arriving at the self-cleavage site within the FIIND domain . Degradation of the N-terminal domains thus releases the bioactive C-terminal CARD-containing fragment of NLRP1B from its non-covalent association with the N-terminal domains , which is sufficient to initiate downstream inflammasome activation ( Chui et al . , 2019; Sandstrom et al . , 2019 ) . By directly coupling NLRP1 inflammasome activation to cleavage by a pathogen-encoded protease , NLRP1 can directly sense and respond to the activity of a pathogen effector . Such a model indicates that the N-terminal domains are not required for NLRP1 activation per se , but rather serve a pathogen-sensing function . Interestingly , the N-terminal ‘linker’ region in mouse NLRP1B that is cleaved by LF is rapidly evolving in rodents , and the analogous linker region is likewise rapidly evolving in primate species ( Chavarría-Smith et al . , 2016 ) . These data suggest that selection from pathogens has been driving diversification of this protease target region of NLRP1 , possibly serving to bait diverse pathogenic proteases into cleaving NLRP1 and activating the inflammasome responses . Consistent with the rapid evolution in NLRP1 at the site of proteolytic cleavage , LF neither cleaves nor activates human NLRP1 ( Mitchell et al . , 2019; Moayeri et al . , 2012; Taabazuing et al . , 2020 ) . Despite this , human NLRP1 can also be activated by proteolysis when a tobacco etch virus ( TEV ) protease site is engineered into the rapidly evolving linker region of human NLRP1 that is analogous to the site of LF cleavage in mouse NLRP1B ( Chavarría-Smith et al . , 2016 ) . Thus , like mouse NLRP1B , it has been predicted that human NLRP1 may serve as a ‘tripwire’ sensor for pathogen-encoded proteases ( Mitchell et al . , 2019 ) . Here , we investigate the hypothesis that viral proteases cleave and activate human NLRP1 . We reasoned that human viruses , many of which encode proteases as necessary enzymes for their replication cycle , may be triggers for NLRP1 activation . To pursue this hypothesis , we focused on viruses in the Picornaviridae family , which encompass a diverse set of human enteric and respiratory pathogens including coxsackieviruses , polioviruses , and rhinoviruses ( Zell , 2018 ) . These viruses all translate their genome as a single polyprotein , which is then cleaved into mature proteins in at least six sites in a sequence-specific manner by a virally encoded 3C protease , termed 3Cpro ( Laitinen et al . , 2016; Solomon et al . , 2010; Sun et al . , 2016; Zell , 2018 ) . 3Cpro is also known to proteolytically target numerous host factors , many of which are associated with immune modulation ( Croft et al . , 2018; Huang et al . , 2015; Lei et al . , 2017; Mukherjee et al . , 2011; Qian et al . , 2017; Wang et al . , 2019; Wang et al . , 2012; Wang et al . , 2014; Wang et al . , 2015; Wen et al . , 2019; Xiang et al . , 2014; Xiang et al . , 2016; Zaragoza et al . , 2006 ) . Because 3Cpro are evolutionarily constrained to cleave several specific polyprotein sites and host targets for replicative success , we reasoned that human NLRP1 may have evolved to sense viral infection by mimicking viral polyprotein cleavage sites , leading to NLRP1 cleavage and inflammasome activation . Using an evolution-guided approach , we now show that NLRP1 is cleaved in its rapidly evolving linker region by several 3Cpro from picornaviruses , resulting in inflammasome activation . These results are consistent with the recent discovery that human rhinovirus ( HRV ) 3Cpro cleaves and activates NLRP1 in airway epithelia ( Robinson et al . , 2020 ) . We also find that variation in the cleavage site among primates , and even within the human population , leads to changes in cleavage susceptibility and inflammasome activation . Interestingly , we observe that proteases from multiple genera of viruses cleave and activate human NLRP1 and mouse NLRP1B at different sites , supporting a role for an evolutionary conflict between viral proteases and NLRP1 . Taken together , our work highlights the role of NLRP1 in sensing and responding to diverse viral proteases by evolving cleavage motifs that mimic natural sites of proteolytic cleavage in the viral polyprotein .
Our hypothesis that NLRP1 can sense viral proteases is based on two prior observations . First , both human NLRP1 and mouse NLRP1B can be activated by N-terminal proteolysis ( Chavarría-Smith et al . , 2016 ) . Second , the linker in primate NLRP1 , which is analogous to the N-terminal disordered region of NLRP1B that is cleaved by LF protease , has undergone recurrent positive selection ( Chavarría-Smith et al . , 2016 ) , or an excess of non-synonymous amino acid substitutions over what would be expected by neutral evolution ( Kimura , 1983 ) . We reasoned that this rapid protein sequence evolution may reflect a history of pathogen-driven selection , wherein primate NLRP1 has evolved to sense pathogen-encoded proteases such as those encoded by picornaviruses . To test this hypothesis , we first generated a predictive model for 3Cpro cleavage site specificity . We chose to focus on the enterovirus genus of picornaviruses , as there is a deep and diverse collection of publicly available viral sequences within this genus due to their importance as human pathogens including coxsackieviruses , polioviruses , enterovirus D68 , and HRV ( Blom et al . , 1996; Pickett et al . , 2012 ) . We first compiled complete enterovirus polyprotein sequences from the Viral Pathogen Resource ( ViPR ) database ( Pickett et al . , 2012 ) and extracted and concatenated sequences for each of the cleavage sites within the polyproteins ( Figure 1A and B , Supplementary files 1 and 2 ) . After removing redundant sequences , we used the MEME Suite ( Bailey et al . , 2009 ) to create the following 3Cpro cleavage motif: [A/Φ]XXQGXXX ( where Φ denotes a hydrophobic residue and X denotes any amino acid ) , which is broadly consistent with previous studies that have experimentally profiled the substrate specificity of enterovirus 3Cpros ( Blom et al . , 1996; Fan et al . , 2020; Jagdeo et al . , 2018; O'Donoghue et al . , 2012; Figure 1C ) . We next optimized our 3Cpro cleavage site motif prediction by querying against predicted viral polyprotein and experimentally validated host cleavage sites ( Laitinen et al . , 2016 ) , allowing us to set thresholds for predicting new cleavage sites ( Supplementary files 3 and 4 ) . Due to the low-information content of the polyprotein motif ( Figure 1C ) , such predictions are necessarily a compromise between stringency and capturing the most known cleavage sites . In particular , we wished to make sure that the model was able to capture a majority of experimentally validated human hits ( compiled in Laitinen et al . , 2016 ) in addition to the known sites of polyprotein cleavage ( ‘true positives’ ) , while minimizing the prediction of sites outside of known polyprotein cleavage sites ( ‘false positives’ ) . By adjusting the model to allow greater flexibility for amino acids not sampled in the viral polyprotein ( see Materials and methods and Figure 1—figure supplement 1 and Supplementary file 4 ) , we were able to capture 95% of known viral sites and the majority of the known human hits , while limiting the number of false negative hits within the viral polyprotein ( Figure 1D ) . We next used our refined model to conduct a motif search for 3Cpro cleavage sites in NLRP1 using Find Individual Motif Occurrences ( FIMO ) ( Grant et al . , 2011 ) . We identified three occurrences of the motif across the full-length human NLRP1 protein ( Figure 2A ) . Of these sites , one in particular , 127-GCTQGSER-134 , fell within the previously described rapidly evolving linker ( Chavarría-Smith et al . , 2016 ) and demonstrates the lowest percent conservation across mammalian species at each of the predicted P4-P4’ positions ( Figure 2B ) . To assess if human NLRP1 is cleaved by enteroviral 3Cpro , we co-expressed a N-terminal mCherry-tagged wild-type ( WT ) human NLRP1 with the 3Cpro from the model enterovirus , coxsackievirus B3 ( CVB3 ) in HEK293T cells ( Figure 2C ) . The mCherry tag stabilizes and allows visualization of putative N-terminal cleavage products , similar to prior studies ( Chavarría-Smith et al . , 2016 ) . We observed that the WT but not catalytically inactive ( C147A ) CVB3 3Cpro cleaved NLRP1 , resulting in a cleavage product with a molecular weight consistent with our predicted 3Cpro cleavage at the predicted 127-GCTQGSER-134 site ( 44 kDa ) ( Figure 2D ) . Based on the presence of a single cleavage product , we assume that the other predicted sites are either poor substrates for 3Cpro or less accessible to the protease as would be predicted from their NetSurfP-reported ( Klausen et al . , 2019 ) coil probability within structured domains of the protein ( Figure 2A and Figure 2—source data 1 ) . To determine if the cleavage occurs between residues 130 and 131 , we mutated the P1’ glycine to a proline ( G131P ) , which abolished 3Cpro cleavage of NLRP1 ( Figure 2D ) . CVB3 3Cpro cleavage of NLRP1 resulted in a similarly intense cleavage product when compared to the previously described system in which a TEV protease site was introduced into the linker region of NLRP1 ( Chavarría-Smith et al . , 2016; Figure 2D ) . Taken together , these results indicate that cleavage of WT NLRP1 by a protease from a natural human pathogen is robust and specific . During a viral infection , 3Cpro is generated in the host cell cytoplasm after translation of the viral mRNA to the polyprotein and subsequent processing of the viral polyprotein into constituent pieces ( Laitinen et al . , 2016 ) . To confirm that virally-produced 3Cpro , or the 3 CD precursor that can also carry out proteolytic cleavage during infection ( Laitinen et al . , 2016 ) , is able to cleave NLRP1 , we virally infected cells expressing either WT NLRP1 or the uncleavable ( G131P ) mutant . We observed accumulation of the expected cleavage product beginning at 6 hr post-infection when we infected cells expressing WT NLRP1 and no cleavage product when we infected cells expressing the 131P mutant ( Figure 2E ) . These results validate that CVB3 infection can result in rapid and specific cleavage of human NLRP1 . Previous results with a TEV-cleavable human NLRP1 showed that cleavage by TEV protease was sufficient to activate the human NLRP1 inflammasome in a reconstituted inflammasome assay ( Chavarría-Smith et al . , 2016 ) . Using the same assay , in which plasmids-encoding human NLRP1 , CASP1 , ASC , and IL-1β are transfected into HEK293T cells , we tested if the CVB3 3Cpro activates the NLRP1 inflammasome . We observed that the CVB3 3Cpro results in robust NLRP1 inflammasome activation , as measured by CASP1-dependent processing of pro-IL-1β to the active p17 form ( Figure 3A ) . As expected , CVB3 3Cpro activation of the NLRP1 inflammasome was prevented by introduction of a mutation in the NLRP1 FIIND ( S1213A ) ( D'Osualdo et al . , 2011; Finger et al . , 2012; Frew et al . , 2012; Figure 3—figure supplement 1 – panel A ) , which prevents FIIND auto-processing and the release of the bioactive C-terminal UPA–CARD ( Chui et al . , 2019; Sandstrom et al . , 2019 ) . Consistent with recent results ( Robinson et al . , 2020 ) , we also observed that chemical inhibitors of the proteasome ( MG132 ) or the Cullin-RING E3 ubiquitin ligases that are required for the degradation of proteins with a novel N-terminal glycine ( MLN4924 ) ( Timms et al . , 2019 ) , also blocked CVB3 3Cpro activation of NLRP1 ( Figure 3—figure supplement 1 – panel B ) . To confirm that 3Cpro-induced inflammasome activation resulted in release of bioactive IL-1β from cells , we measured active IL-1β levels in the culture supernatant using cells engineered to express a reporter gene in response to soluble , active IL-1β . When compared to a standard curve ( Figure 3—figure supplement 2 ) , we found that 3Cpro treatment resulted in release of >4 ng/ml of active IL-1β into the culture supernatant ( Figure 3B ) . Importantly , in both western blot and cell culture assays , 3Cpro-induced inflammasome activation was comparable to TEV-induced activation and was ablated when position 131 was mutated , validating that CVB3 3Cpro cleavage at a single site is both necessary and sufficient to activate NLRP1 ( Figure 3A and B ) . Taken together , our results are consistent with CVB3 3Cpro activating the NLRP1 inflammasome via site-specific cleavage and subsequent ‘functional degradation’ ( Chui et al . , 2019; Sandstrom et al . , 2019 ) . We next wished to test whether CVB3 infection , through the site-specific cleavage of NLRP1 by 3Cpro , is able to activate the NLRP1 inflammasome . Consistent with our prediction , recent work has revealed that HRV infection can cleave and activate human NLRP1 in airway epithelia ( Robinson et al . , 2020 ) . However , prior work has also implicated a role for the NLRP3 inflammasome in enterovirus infection ( Kuriakose and Kanneganti , 2019; Xiao et al . , 2019 ) , including activation of the NLRP3 inflammasome during CVB3 infection in mice and human cell lines ( Wang et al . , 2019; Wang et al . , 2018 ) . NLRP1 and NLRP3 have distinct expression patterns ( Robinson et al . , 2020; Zhong et al . , 2016 ) including in epithelial cells , which are important targets of enterovirus infection . NLRP3 is activated in response to various noxious stimuli or damage signals associated with pathogen infection ( Evavold and Kagan , 2019; Spel and Martinon , 2021 ) . In contrast , NLRP1 is activated by direct proteolytic cleavage of its N-terminal ‘tripwire’ region by viral proteases . We therefore wished to confirm that specific 3Cpro cleavage of NLRP1 during CVB3 infection is able to activate the NLRP1 inflammasome . We first virally infected 293 T cells , which do not express either NLRP1 or NLRP3 , that were co-transfected with either WT NLRP1 or the uncleavable ( G131P ) mutant in our reconstituted inflammasome assay and measured active IL-1β in the culture supernatant . Eight hours after infection with CVB3 , we observe robust release of active IL-1β into the culture supernatant when cells were transfected with WT NLRP1 but not the uncleavable mutant NLRP1 ( Figure 3C ) . To test whether CVB3 infection can activate the inflammasome in an NLRP1-dependent fashion in cells that naturally express an intact NLRP1 inflammasome , we took advantage of the fact that NLRP1 has been described as the primary inflammasome in human keratinocytes ( Zhong et al . , 2016 ) . We therefore infected WT , NLRP1 , or CASP1 KO ( Figure 3—figure supplement 3 ) immortalized HaCaT human keratinocytes with CVB3 and measured release of active IL-1β in the culture supernatant . Consistent with our model that CVB3 infection cleaves and activates the NLRP1 inflammasome , we observe a significant increase in supernatant IL-1β after CVB3 infection that is reduced in cells that lack either NLRP1 or CASP1 ( Figure 3D ) . Together , these results indicate that CVB3 infection , through 3Cpro cleavage of the tripwire region of NLRP1 , activates the NLRP1 inflammasome . Our evolutionary model in which NLRP1 is evolving in conflict with 3Cpro suggests that changes in the NLRP1 linker region , both among primates and within the human population ( Figure 4A ) , would confer host-specific differences to NLRP1 cleavage and inflammasome activation . To test this hypothesis , we aligned the linker regions from NLRP1 from diverse mammals and human population sampling and compared the sequences around the site of CVB3 3Cpro cleavage ( Figure 4B and C and Figure 4—figure supplement 1 ) . We noted that while a majority of primate NLRP1s are predicted to be cleaved similarly to the human ortholog , several primate proteins would be predicted to not be cleaved by enteroviral 3Cpro as a result of changes to either the P4 , P1 or P1’ residues . To confirm these predictions , we made the human NLRP1 mutants G127E or G131R , which reflect the Old World monkey or marmoset residues at each position , respectively . As predicted , both primate NLRP1 variants prevented 3Cpro cleavage of NLRP1 ( Figure 4D ) . These results indicate that multiple viral 3Cpro activate host NLRP1 in a host specific manner and suggest that single changes within a short linear motif can substantially alter cleavage susceptibility and inflammasome activation . We further observed that this cleavage site is largely absent in non-primate species ( Figure 4—figure supplement 1 ) , suggesting that a 3Cpro cleavage site mimic emerged in simian primates 30–40 million years ago . While many other mammalian species have a region that is alignable to the primate linker , we noted that this region is unalignable to any sequence in the linker region of NLRP1 proteins from rodents or bats ( Figure 4B and Figure 4—figure supplement 1 ) . Despite this , we found that there was weak cleavage of mouse NLRP1B at a site closer to the N-terminus than the 127-GCTQGSER-134 site found in human NLRP1 ( Figure 4D and Figure 6A ) , suggesting that an independent cleavage site could have arisen elsewhere in mouse NLRP1B . These data suggest that NLRP1 in other mammals may have convergently evolved cleavage sites in the linker region despite not having a cleavable sequence in the precise position that human NLRP1 is cleaved . Differential host susceptibility to NLRP1 cleavage and activation extends to the human population level . Using GnomAD ( Karczewski et al . , 2020 ) , we sampled the alternative alleles within the direct cleavage site ( Figure 4C ) . While this region does not appear to be highly polymorphic in humans , we note that one alternative allele ( rs150929926 ) results in a Q130R mutation and is present in >1 in every 1000 African alleles sampled . Introducing this mutation into NLRP1 , we find the Q130R mutation eliminates NLRP1 cleavage susceptibility to CVB3 3Cpro ( Figure 4E ) . In the case of primate and human diversity alleles at the site of 3Cpro cleavage , we also find that loss of cleavage susceptibility results in a loss of inflammasome activation in response to 3Cpro ( Figure 4F ) , supporting the aforementioned notion that single changes in the linker region can have drastic impacts on the ability of different hosts to respond to the presence of cytoplasmic 3Cpro . Our evolutionary model predicted that NLRP1 would be cleaved by a broad range of 3Cpro from viruses in the enterovirus genus ( Figure 1B ) . To test this hypothesis , we cloned 3Cpro from representative viruses from four additional major species of human enteroviruses: enterovirus 71 ( EV71 , species: Enterovirus A ) , poliovirus 1 ( PV1 , species: Enterovirus C ) , enterovirus D68 ( EV68 , species: Enterovirus D ) , human rhinovirus A ( HRV , species: Rhinovirus A ) , in order to compare them to the 3Cpro from CVB3 ( species: Enterovirus B ) ( Figure 5A ) . Despite <50% amino acid identity between some of these proteases ( Figure 5—figure supplement 1; Figure 5—figure supplement 1—source data 1 ) , the overall structures of these proteases are similar ( Figure 5—figure supplement 2 ) and the cleavage motifs are closely related ( Figure 5A ) . Consistent with this predicted target similarity and prior data with HRV ( Robinson et al . , 2020 ) , we found that every tested member of enterovirus 3Cpro was able to cleave NLRP1 between residues 130 and 131 ( Figure 5B ) . Moreover , expression of every tested enterovirus 3Cpro resulted in activation of the inflammasome in a manner that was dependent on cleavage at the 127-GCTQGSER-134 site ( Figure 5C ) . Enteroviruses are only one genus within the broad Picornaviridae family of viruses . We next asked if viruses in other Picornaviridae genera that infect humans are also able to cleave and activate human NLRP1 . We were unable to generate a robust sequence motif for every genera of picornavirus due to lower depth of publicly available sequences . Instead , we cloned a 3Cpro from a representative of every genus of picornavirus that are known to infect humans: encephalomyocarditis virus ( EMCV , genus: Cardiovirus ) , parechovirus A virus ( ParA , genus: Parechovirus ) , Aichi virus ( Aichi , genus: Kobuvirus ) , hepatitis A virus ( HepA , genus: Hepatovirus ) , salivirus A virus ( SaliA , genus: Salivirus ) , and rosavirus A2 ( Rosa2 , genus: Rosavirus ) . Each of these viral proteases is <20% identical to CVB3 3Cpro ( Figure 5—figure supplement 1—source data 1 ) . Despite this , the sequence motif built from cleavage sites within the polyprotein of these individual viruses is broadly consistent with the motif seen in enteroviruses ( Figure 5A ) , reflective of the strong evolutionary constraint on evolution of the sequence specificity of these proteases and overall structural conservation of the active sites of these proteases ( Figure 5—figure supplement 2 ) . Interestingly , we found that there was substantial variation in NLRP1 cleavage sites across these diverse 3Cpro even though most picornavirus proteases cleaved human NLRP1 to some degree ( Figure 5B ) . For instance , while 3Cpro from EMCV and ParA did not cleave NLRP1 , we observed distinct cleavage sites for 3Cpro from Aichi , HepA , SaliA and Rosa2 ( Figure 5B ) , all of which have at least one cleavage site predicted to occur in the linker region ( expected size between 40 kDa and 67 kDa ) . Confirming that these proteases cleave at a site that is distinct from that of enteroviruses , the G131P NLRP1 mutant is still cleaved by the non-enteroviral proteases ( Figure 5B ) . Surprisingly , when we interrogated NLRP1 inflammasome activation by 3Cpros from Aichi , HepA , SaliA , and Rosa2 , all of which robustly cleave NLRP1 at a site in the linker region , we found that only Rosa2 was able to activate the NLRP1 inflammasome ( Figure 5C ) . While it is possible that NLRP1 cleavage by 3Cpro from these other viruses is too weak or in a region that may be inconsistent with activation , we also noted that there are obvious cleavage sites in NLRP1 that are outside of the linker region and closer to the FIIND autocleavage site . Cleavage at these sites in NLRP1 , or cleavage of other host genes , may interfere with activation that may have otherwise been induced by 3Cpro cleavage in the linker region . Indeed , we find that co-expression of 3Cpro from Aichi , HepA , SaliA can attenuate NLRP1 activation by TEV protease ( Figure 5—figure supplement 3 ) , consistent with the idea that these three proteases can actively block NLRP1 activation . Further investigation will be needed to determine the exact mechanism by which this occurs . Nevertheless , our data demonstrate that non-enteroviral 3Cpros can cleave NLRP1 at independent sites in the rapidly evolving linker region and can , in at least one case , activate the human NLRP1 inflammasome . To further confirm that 3Cpro cleavage ( or lack thereof ) of NLRP1 is reflective of 3Cpro during viral infection , we infected cells expressing WT or 131P NLRP1 with EMCV . Consistent with our co-transfection experiments , we see no cleavage of NLRP1 when we infect with EMCV , despite seeing robust cleavage when we infect with CVB3 ( Figure 5D ) . Likewise , we see no IL-1β release when we infect either inflammasome-reconstituted HEK293T cells or inflammasome-competent HaCaT cells with EMCV ( Figure 5—figure supplement 4 ) . These data indicate that evolution of viral 3Cpro cleavage specificity alters whether a virus can be sensed by the NLRP1 tripwire or not . Two bacterial pathogen effectors are known to activate mouse NLRP1B , the LF protease from B . anthracis ( Boyden and Dietrich , 2006; Greaney et al . , 2020; Moayeri et al . , 2010; Terra et al . , 2010 ) ; ( Chavarría-Smith and Vance , 2013; Levinsohn et al . , 2012 ) and the IpaH7 . 8 E3 ubiquitin ligase from Shigella flexneri ( Sandstrom et al . , 2019 ) . Interestingly , in both of these cases , activation is specific to the 129 allele of mouse NLRP1B , whereas the B6 allele of NLRP1B is not activated by these pathogenic effectors . Given the power of mouse models for understanding inflammasome biology , we wished to determine if 3Cpros cleave and activate mouse NLRP1B . Strikingly , when we co-transfected NLRP1B from either the 129 or the B6 strains with diverse enterovirus 3Cpros , we observed allele-specific cleavage products ( Figure 6A ) . Consistent with data in Figure 4D , we observed weak cleavage of 129 NLRP1B by CVB3 3Cpro . In addition , we found that 3Cpro from other enteroviruses varied substantially in their ability to cleave 129 NLRP1B , including no detectable cleavage with EV71 3Cpro and a different dominant position of cleavage by HRV 3Cpro . Despite this variation , we only observed weak cleavage ( Figure 6A , left ) and little to no inflammasome activation ( Figure 6B , left ) by any enterovirus 3Cpros tested against 129 NLRP1B . In contrast , enterovirus 3Cpro cleavage of B6 NLRP1B resulted in a consistent-sized cleavage product across all enterovirus 3Cpros that ranged in intensities between the different viral proteases , more similar to our observations with human NLRP1 . Most interestingly , we observed that co-transfection with HRV 3Cpro resulted in the appearance of a very strong cleavage product ( Figure 6A , right ) , almost complete loss of full length B6 NLRP1B ( Figure 6A and B , right ) and very strong activation of the inflammasome ( Figure 6B , right ) . These data indicate that mouse NLRP1B can also be cleaved and activated by viral proteases , which suggests that the evolution of the N-terminus of NLRP1B between closely related mouse strains ( Figure 6—figure supplement 1 ) is not only shaping susceptibility to tripwire cleavage by the bacterial LF protease , but also impacts tripwire cleavage by viral 3Cpros . Taken together , these data further support the model in which both host and viral evolution , even within closely related host and viral species , shape the outcome of the interaction between NLRP1 and 3Cpro .
Pathogens and their hosts are locked in a continual evolutionary conflict in which each side is attempting to exploit the others’ weakness . One particularly successful strategy that pathogens have adopted is to exploit host processes that are highly constrained , leaving the host little room to evolutionarily adapt to overcome the pathogen . For instance , molecular mimicry of host proteins is commonly deployed by pathogens to antagonize host defenses , as it limits the evolutionary options for the host to counter-evolve ( Elde and Malik , 2009 ) . Beyond mimicry of entire proteins or protein domains , pathogens can also mimic so-called ‘short linear motifs’ ( SLIMs ) through evolution of only a small number of amino acids to hijack highly conserved host processes such as post-translational modifications or binding by small protein domains ( Chemes et al . , 2015; Hagai et al . , 2014 ) . Although these strategies are generally described as taking advantage of host evolutionary constraint , pathogens also have potential weak points of evolutionary constraint . In particular , proteases from positive-sense RNA viruses , such as picornaviruses , need to specifically cleave numerous sites within the viral polyprotein in order to reproduce . Thus , changing protease specificity requires concomitant changes to several independent cleavage sites , which is difficult to accomplish in a single evolutionary step . On top of that , protease cleavage motifs often only span a small number of amino acids ( Schechter and Berger , 1967 ) , potentially facilitating the independent evolution of these SLIMs in host proteins . Here , we show that the inflammasome protein , NLRP1 , serves as a sensor for diverse proteases from the Picornaviridae family of human pathogens by mimicking the highly conserved protease cleavage sites found within the viral polyproteins . By exploiting a constrained feature of viral evolution and tying it to a pro-inflammatory immune response , such a system allows the immune system to recognize and respond to a wide range of viral proteases expressed in the host cytoplasm . NLRP1 represents one of the few known cases of mammalian ETI ( Cui et al . , 2015; Fischer et al . , 2020; Jones et al . , 2016 ) , where pathogen-mediated cleavage of NLRP1 promotes its activation . By holding the small C-terminal CARD-containing fragment in a non-covalent association with the larger N-terminal fragment , the majority of the protein can serve as a sensor for pathogen-encoded effectors ( Mitchell et al . , 2019; Taabazuing et al . , 2020 ) . This presents an opportunity to allow NLRP1 to evolve to be recognized by pathogenic effectors , ultimately leading to degradation of the N-terminal fragment . Indeed , mouse NLRP1B has been shown to be specifically cleaved by the protease-containing secreted effector from B . anthracis ( LF ) as well as being ubiquitylated by an E3-ubiquitin ligase from S . flexneri ( IpaH7 . 8 ) ( Sandstrom et al . , 2019 ) . While these two examples provide evidence that the mouse NLRP1B inflammasome operates by a ‘functional degradation’ model , a direct pathogen-encoded activator of human NLRP1 had remained elusive . We now show , using an evolution-guided approach , that proteases from diverse picornaviruses , including human pathogens such as coxsackievirus B3 ( CVB3 ) , human rhinovirus A ( HRV ) , enterovirus D68 ( EV68 ) and poliovirus 1 ( PV1 ) and rosavirus A2 ( Rosa2 ) , specifically cleave several independently evolved sites in human NLRP1 , leading to activation of the NLRP1 inflammasome and release of pro-inflammatory cytokines such as IL-1β . Together with recent findings ( Robinson et al . , 2020 ) , our work has thus identified proteases from a diverse range of picornaviruses as pathogen-encoded activators of human NLRP1 . We previously speculated that the unique domain architecture of NLRP1 would allow the N-terminal linker of human NLRP1 to freely evolve to be recognized by pathogenic effectors . Indeed , by harvesting publicly available enterovirus polyprotein sequences for known 3Cpro cleavage sites , we created a 3Cpro cleavage motif that was used to successfully predict the site of enterovirus 3Cpro cleavage at position 130–131 within the rapidly-evolving linker NLRP1 . Additionally , our finding that numerous enteroviruses also cleave at the Q130-G131 site and activate pro-inflammatory cytokine release suggests that human NLRP1 serves as a general enteroviral protease sensor by encoding a polyprotein cleavage site mimic . Our phylogenetic assessment of the Q130-G131 3Cpro cleavage site in NLRP1 suggests that NLRP1 sensing of enteroviruses at this specific site is an innovation in the primate lineage , and is largely absent in all other mammalian lineages with exception of a possible independent acquisition by members within the Caprinae subfamily of mammals ( e . g . goats , sheep ) ( Figure 3—figure supplement 1 ) . Interestingly , even within the primate lineage and a small fraction of the human population , some primate orthologs and human variants are cleavage-resistant and therefore do not activate the inflammasome upon cytoplasmic expression of 3Cpro . Such data may hint at three different possible explanations for these changes . First , evolutionary drift in the absence of pressure from pathogenic enteroviruses may account for loss of enterovirus 3Cpro responsiveness in these genes . Second , selection to sense another viral protease may shape the same region of the linker . Finally , while the ETI model of NLRP1 suggests that enteroviral cleavage of NLRP1 has evolved to activate a beneficial immune response in certain contexts , the effects of NLRP1 overactivation may be detrimental in other contexts . In human skin keratinocytes , where NLRP1 is regarded as the key inflammasome , all components of the NLRP1 inflammasome are basally expressed and thus poised to elicit an inflammatory response ( Zhong et al . , 2016 ) . Here , germline mutations in NLRP1 that result in overactivation can cause growth of warts in the upper airway in a condition known as recurrent respiratory papillomatosis ( JRRP ) ( Drutman et al . , 2019 ) and an increase in skin cancer susceptibility and skin disorders such as multiple self-healing palmoplantar carcinoma ( MSPC ) , familial keratosis lichenoides chronica ( FKLC ) and auto-inflammation with arthritis and dyskeratosis ( AIADK ) ( Grandemange et al . , 2017; Herlin et al . , 2020; Soler et al . , 2013; Zhong et al . , 2016; Zhong et al . , 2018 ) . Additional recent work has indicated that dsRNA can also activate the NLRP1 inflammasome in human keratinocytes ( Bauernfried et al . , 2020 ) , adding to the role that NLRP1 may play in the inflammatory response . Beyond the skin , NLRP1 is also basally expressed in tissues such as the gut and brain ( D'Osualdo et al . , 2015; Kaushal et al . , 2015; Kummer et al . , 2007 ) , which are sites of picornavirus replication where overactivation upon infection may result in immunopathology . Further in vivo studies will help determine the role of NLRP1 in antiviral immunity and/or immunopathology during viral infection . Facilitating these studies , our discovery that 3Cpro from HRV potently cleaves and activates NLRP1B from B6 but not 129 mice suggests that rhinovirus infection of B6 mice may be a good model for studying the in vivo consequences of viral-mediated NLRP1 inflammasome activation . Intriguingly , 3Cpros from nearly every genus of human-infecting picornavirus can cleave NLRP1 somewhere in the rapidly evolving linker region between the PYD and NLR domain , although only enteroviruses cleave at the specific site between position 130 and 131 . These data suggest that this extended linker , which we previously found showed widespread signatures of positive selection ( Chavarría-Smith et al . , 2016 ) , may be convergently evolving to mimic cleavage sites from a diverse range of viruses at multiple independent sites . Supporting that model , we observe a similar phenomenon in mouse NLRP1B , where multiple viral proteases cleave at different sites within NLRP1 in a strain-specific manner . These data highlight the important functional differences in cleavage specificity between even closely related 3Cpro that are not accounted for by predictive models . Further studies will be required to understand the precise relationships between sites within NLRP1 and individual protease specificity . Intriguingly , not all these cleavage events lead to inflammasome activation in the same way that enteroviral cleavage does , and we find evidence for antagonism of NLRP1 activation by some 3Cpros , suggesting that additional activities of 3Cpro may be the next step in the arms race , serving to prevent inflammasome activation even after the tripwire has been tripped . Taken together , our work suggests that host mimicry of viral polyprotein cleavage motifs could be an important evolutionary strategy in the ongoing arms race between host and viruses . Indeed , one explanation for the somewhat surprising observation that the specificity of viral proteases changes at all within a viral family such as the picornaviruses is that there is evolutionary pressure from the host to evolve cleavage sites and protease specificity . Prior work has highlighted the roles that viral proteases can play in antagonizing host immune factors and driving host evolution to avoid being cleaved ( Patel et al . , 2012; Stabell et al . , 2018 ) . In that case , the viral proteases would evolve to antagonize new factors while maintaining polyprotein cleavage . However , mimicry coupled with cleavage-activating immunity as seen with NLRP1 could be an even stronger pressure to shape the protease specificity . By turning the tables , these host processes may drive the type of functional diversification of viral protease specificity that we observe in order to avoid cleaving NLRP1 and other similar ETI factors . We expect that this work may lead to the discovery that such an evolutionary strategy may be more broadly deployed at other sites of host-pathogen conflicts .
To build the motif , 2658 nonredundant enteroviral polyprotein sequences were collected from the Viral Pathogen Resource ( ViPR ) and aligned with 20 well-annotated reference enteroviral polyprotein sequences from RefSeq ( Supplementary file 1 ) . P1 and P1’ of the eight annotated cleavage sites across the RefSeq sequences served as reference points for putative cleavage sites across the 2658 ViPR sequences , with the exception of enterovirus D polyproteins . The 3Cpro cleavage site for VP3-VP1 within polyproteins from the clade of enterovirus D have been described to be undetectable and have thus been removed ( Tan et al . , 2013 ) . Four amino acyl residues upstream ( P4-P1 ) and downstream ( P1’-P4’ ) of each cleavage site were extracted from every MAFFT-aligned polyprotein sequence , resulting in 2678 sets of cleavage sites ( RefSeq sites included ) . Each set of cleavage sites representative of each polyprotein was then concatenated . Next , 1884 duplicates were removed from the 2678 concatenated cleavage sites . The remaining 796 nonredundant , concatenated cleavage sites were then split into individual 8-mer cleavage sites and the 6333 8-mers were aligned using MAFFT to generate Geneious-defined sequence logo information at each aligned position . Pseudo-counts to the position-specific scoring matrix were adjusted by total information content within each position relative to the two most information-dense position P1 and P1’ ( pseudocount = 0 ) and the least information-dense position P3 ( pseudocount = 1 ) . The 0 . 002 p-value threshold for FIMO motif searching against human NLRP1 was determined to optimize the capture of 95% of initial input cleavage sites within the set of 2678 whole enteroviral polyproteins and a majority sites within a previously described dataset of enteroviral 3Cpro targets ( Laitinen et al . , 2016 ) . Prediction of the coil probability across human NLRP1 ( NCBI accession NP_127497 . 1 ) was conducted using the protein FASTA as the input for the NetSurfP web server ( http://www . cbs . dtu . dk/services/NetSurfP/ ) . Complete polyprotein sequences from 796 picornaviruses with non-redundant 3Cpro cleavage sites ( see ‘Motif generation and search’ section above ) were downloaded from ViPR . Sequences were aligned using MAFFT ( Katoh and Standley , 2013 ) and a neighbor-joining phylogenetic tree was generated using Geneious software ( Kearse et al . , 2012 ) . An alignment and phylogenetic tree of all the 3Cpro sequences used in this study was generated similarly . To identify mammalian NLRP1 homologs , and species that lack NLRP1 , the human NLRP1 protein sequence was used to query the RefSeq protein sequence database , a curated collection of the most well-assembled genomes , using BLASTp ( Altschul et al . , 1997 ) . Sequences were downloaded and aligned using MAFFT implemented in Geneious software . Consensus sequence logos shown were generated using Geneious software . We determined that NLRP1 was ‘absent’ from a clade of species using the following criteria: ( 1 ) when searching with human NLRP1 , we found an obvious homolog of another NLRP protein ( generally NLRP3 , NLRP12 or NLRP14 ) but no complete or partial homolog of NLRP1 and ( 2 ) this absence was apparent in every member of the clade of species ( >2 species ) in the RefSeq database . For NLRP1 cleavage assays , the coding sequences of human NLRP1 WT ( NCBI accession NP_127497 . 1 ) , human NLRP1 mutants ( G131P , G131R , Q130R , G127E ) , human NLRP1 TEV or mouse NLRP1B ( 129 allele , NCBI accession AAZ40510 . 1; B6 allele , NCBI accession XM_017314698 . 2 ) were cloned into the pcDNA5/FRT/TO backbone ( Invitrogen , Carlsbad , CA ) with an N-terminal 3xFlag and mCherry tag . For NLRP1 activation , the same sequences were cloned into the pQCXIP vector backbone ( Takara Bio , Mountain View , CA ) with a C-terminal Myc tag ( human NLRP1 sequences ) or N-terminal EGFP and C-terminal HA ( mouse NLRP1B sequences ) . Vectors containing the coding sequences of human NLRP1 TEV ( NLRP1-TEV2 ) , ASC , human and mouse CASP1 , human IL-1β-V5 , mouse IL-1β , and TEV protease ( Chavarría-Smith et al . , 2016 ) were generous gifts from Dr . Russell Vance , UC Berkeley . Single point mutations were made using overlapping stitch PCR . A list of primers used to generate the wild-type and mutant NLRP1 constructs are described in Supplementary file 5 . CVB3 3Cpro and EMCV 3Cpro were cloned from CVB3-Nancy and EMCV-Mengo plasmids ( generous gifts from Dr . Julie Pfeiffer , UT Southwestern ) . Remaining 3Cpro sequences were ordered as gBlocks ( Integrated DNA Technologies , San Diego , CA ) . Each 3Cpro was cloned with an N-terminal HA tag into the QCXIP vector backbone . Catalytic mutations were made using overlapping stitch PCR . A list of primers and gBlocks used to generate the protease constructs are described in Supplementary file 5 . Following cloning , all plasmid stocks were sequenced across the entire inserted region to verify that no mutations were introduced during the cloning process . All cell lines ( HEK293T , HEK-Blue-IL-1β , and HaCaT ) are routinely tested for mycoplasma by PCR kit ( ATCC , Manassas , VA ) and kept a low passage number to maintain less than one year since purchase , acquisition or generation . HEK293T cells were obtained from ATCC ( catalog # CRL-3216 ) , HEK-Blue-IL-1β cells were obtained from Invivogen ( catalog # hkb-il1b ) and HaCaT cells were obtained from the UC Berkeley Cell Culture Facility ( https://bds . berkeley . edu/facilities/cell-culture ) and all lines were verified by those sources . All cells were grown in complete media containing DMEM ( Gibco , Carlsbad , CA ) , 10% FBS ( Peak Serum , Wellington , CO ) , and appropriate antibiotics ( Gibco , Carlsbad , CA ) . For transient transfections , HEK293T cells were seeded the day prior to transfection in a 24-well plate ( Genesee , El Cajon , CA ) with 500 µl complete media . Cells were transiently transfected with 500 ng of total DNA and 1 . 5 µl of Transit X2 ( Mirus Bio , Madison , WI ) following the manufacturer’s protocol . HEK-Blue IL-1β reporter cells ( Invivogen , San Diego , CA ) were grown and assayed in 96-well plates ( Genesee , El Cajon , CA ) . To establish NLRP1 and CASP1 knockouts in human immortalized keratinocyte HaCaT cells , lentivirus-like particles were made by transfecting HEK293T cells with the plasmids psPAX2 ( gift from Didier Trono , Addgene plasmid # 12260 ) and pMD2 . G ( gift from Didier Trono , Addgene plasmid # 12259 ) and either the pLB-Cas9 ( gift from Feng Zhang , Addgene plasmid # 52962 ) ( Sanjana et al . , 2014 ) or a plentiGuide-Puro , which was adapted for ligation-independent cloning ( kindly gifted by Moritz Gaidt ) ( Schmidt et al . , 2015 ) . Guide sequences are shown in Supplementary file 5 . Conditioned supernatant was harvested 48 and 72 hr post-transfection and used for spinfection of HaCaT cells at 1200 x g for 90 min at 32°C . Forty-eight hours post-transduction , cells with stable expression of Cas9 were selected in media containing 100 µg/ml blasticidin . Blasticidin-resistant cells were then transduced with sgRNA-encoding lentivirus-like particles , and selected in media containing 1 . 3 µg/ml puromycin . Cells resistant to blasticidin and puromycin were single cell cloned by limiting dilution in 96-well plates , and confirmed as knockouts by Sanger sequencing ( Figure 3—figure supplement 3 ) . 100 ng of epitope-tagged human NLRP1 WT , human NLRP1 mutants ( G131P , G131R , Q130R , G127E ) , human NLRP1 TEV or mouse NLRP1B was co-transfected with 250 ng of HA-tagged protease-producing constructs . Twenty-four hours post-transfection , the cells were harvested , lysed in 1x NuPAGE LDS sample buffer ( Invitrogen , Carlsbad , CA ) containing 5% β-mercaptoethanol ( Fisher Scientific , Pittsburg , PA ) and immunoblotted with antibodies described below . For human NLRP1 activation assays , 5 ng of ASC , 100 ng of CASP1 , 50 ng of IL-1β-V5 , and 100 ng of various protease-producing constructs were co-transfected with 4 ng of either pQCXIP empty vector , wild-type or mutant pQCXIP-NLRP1-Myc constructs . For inhibitor treatments , cells were treated with either 0 . 5 µM MG132 or 1 . 0 µM MLN4924 18 hr after transfection . Twenty-four hours post-transfection , cells were harvested and lysed in 1x NuPAGE LDS sample buffer containing 5% β-mercaptoethanol or in 1x RIPA lysis buffer with protease inhibitor cocktail ( Roche ) and immunoblotted with antibodies described below or culture media was harvested for quantification of IL-1β levels by HEK-Blue assays ( see below ) . For mouse NLRP1B activation assays , 50 ng of mouse CASP1 , 50 ng of mouse IL-1β , and 100 ng of various protease-producing constructs were co-transfected with either 4 ng of 129 NLRP1B or 2 . 5 ng B6 NLRP1B constructs . Twenty-four hours post-transfection , cells were harvested in 1x RIPA lysis buffer with protease inhibitor cocktail ( Roche ) and immunoblotted with antibodies described below . CVB3 and EMCV viral stocks were generated by co-transfection of CVB3-Nancy or EMCV-Mengo infectious clone plasmids with a plasmid expressing T7 RNA polymerase ( generous gifts from Dr . Julie Pfeiffer , UT Southwestern ) as previously described ( McCune et al . , 2020 ) . Supernatant was harvested , quantified by plaque assay or TCID50 on HEK293T cells , and frozen in aliquots at −80°C . For viral infections of HEK293T cells , cells were transfected in 24-well plates and infected with 250 , 000 PFU ( MOI = ~1 ) CVB3 or EMCV or mock infected for the indicated times . For cleavage assays , cells were transfected with 100 ng of the indicated NLRP1 construct and , 24 hr after transfection , cells were harvested and lysed in 1x NuPAGE LDS sample buffer containing 5% β-mercaptoethanol and immunoblotted with antibodies described below . For activation assays , cells were transfected with 4 ng of the indicated NLRP1 construct and 5 ng of ASC , 100 ng of CASP1 , 50 ng of IL-1β-V5 . Twenty-four hours after transfections , cells were infected with virus ( or mock infected ) and culture supernatant was collected 8 hr later ( 32 hr post-transfection ) . Culture supernatant was filtered through a 100 , 000 MWCO centrifugal spin filter ( MilliporeSigma , Burlington , MA ) for 10 min at 12 , 000xg to remove infectious virus and IL-1β levels were quantified by HEK-Blue assays ( see below ) . For viral infections of HaCaT cells , cells were plated in 24-well plates . The next day , cells were infected with 100 , 000 PFU/well ( MOI = ~0 . 4 ) CVB3 or EMCV or mock infected . Forty-eight hours after infection , culture supernatant was collected , spin filtered as described above to remove infectious virus , and IL-1β levels were quantified by HEK-Blue assays ( see below ) . To quantify the levels of bioactive IL-1β released from cells , we employed HEK-Blue IL-1β reporter cells ( Invivogen , San Diego , CA ) . In these cells , binding to IL-1β to the surface receptor IL-1R1 results in the downstream activation of NF-kB and subsequent production of secreted embryonic alkaline phosphatase ( SEAP ) in a dose-dependent manner ( Figure 3—figure supplement 1 ) . SEAP levels are detected using a colorimetric substrate assay , QUANTI-Blue ( Invivogen , San Diego , CA ) by measuring an increase in absorbance at OD655 . Culture supernatant from inflammasome-reconstituted HEK293T cells or HaCaT cells that had been transfected with 3Cpro or virally infected ( see above ) was added to HEK-Blue IL-1β reporter cells plated in 96-well format in a total volume of 200 µl per well . On the same plate , serial dilutions of recombinant human IL-1β ( Invivogen , San Diego , CA ) were added in order to generate a standard curve for each assay . Twenty-four hours later , SEAP levels were assayed by taking 20 µl of the supernatant from HEK-Blue IL-1β reporter cells and adding to 180 µl of QUANTI-Blue colorimetric substrate following the manufacturer’s protocol . After incubation at 37°C for 30–60 min , absorbance at OD655 was measured on a BioTek Cytation five plate reader ( BioTek Instruments , Winooski , VT ) and absolute levels of IL-1β were calculated relative to the standard curve . All assays , beginning with independent transfections or infections , were performed in triplicate . Harvested cell pellets were washed with 1X PBS , and lysed with 1x NuPAGE LDS sample buffer containing 5% β-mercaptoethanol at 98C for 10 min . The lysed samples were spun down at 15000 RPM for two minutes , followed by loading into a 4–12% Bis-Tris SDS-PAGE gel ( Life Technologies , San Diego , CA ) with 1X MOPS buffer ( Life Technologies , San Diego , CA ) and wet transfer onto a nitrocellulose membrane ( Life Technologies , San Diego , CA ) . Membranes were blocked with PBS-T containing 5% bovine serum albumin ( BSA ) ( Spectrum , New Brunswick , NJ ) , followed by incubation with primary antibodies for V5 ( IL-1β ) , FLAG ( mCherry-fused NLRP1 for protease assays ) , Myc ( NLRP1-Myc for activation assays ) , HA ( viral protease or mouse NLRP1B ) , β-tubulin , or GAPDH . Membranes were rinsed three times in PBS-T then incubated with the appropriate HRP-conjugated secondary antibodies . Membranes were rinsed again three times in PBS-T and developed with SuperSignal West Pico PLUS Chemiluminescent Substrate ( Thermo Fisher Scientific , Carlsbad , CA ) . The specifications , source , and clone info for antibodies are described in Supplementary file 6 . | The immune system recognizes disease-causing microbes , such as bacteria and viruses , and removes them from the body before they can cause harm . When the immune system first detects these foreign invaders , a multi-part structure known as the inflammasome launches an inflammatory response to help fight the microbes off . Several sensor proteins can activate the inflammasome , including one in mice called NLRP1B . This protein has evolved a specialized site that can be cut by a bacterial toxin . Once cleaved , this region acts like a biological tripwire and sparks NLRP1B into action , allowing the sensor to activate the inflammasome system . Humans have a similar protein called NLRP1 , but it is unclear whether this protein has also evolved a tripwire region that can sense microbial proteins . To answer this question , Tsu , Beierschmitt et al . set out to find whether NLRP1 can be activated by viruses in the Picornaviridae family , which are responsible for diseases like polio , hepatitis A , and the common cold . This revealed that NLRP1 contains a cleavage site for enzymes produced by some , but not all , of the viruses in the picornavirus family . Further experiments confirmed that when a picornavirus enzyme cuts through this region during a viral infection , it triggers NLRP1 to activate the inflammasome and initiate an immune response . The enzymes from different viruses were also found to cleave human NLRP1 at different sites , and the protein’s susceptibility to cleavage varied between different animal species . For instance , Tsu , Beierschmitt et al . discovered that NLRP1B in mice is also able to sense picornaviruses , and that different enzymes activate and cleave NLRP1B and NLRP1 to varying degrees: this affected how well the two proteins are expected to be able to sense specific viral infections . This variation suggests that there is an ongoing evolutionary arms-race between viral proteins and the immune system: as viral proteins change and new ones emerge , NLRP1 rapidly evolves new tripwire sites that allow it to sense the infection and launch an inflammatory response . What happens when NLRP1B activates the inflammasome during a viral infection is still an open question . The discovery that mouse NLRP1B shares features with human NLRP1 could allow the development of animal models to study the role of the tripwire in antiviral defenses and the overactive inflammation associated with some viral infections . Understanding the types of viruses that activate the NLRP1 inflammasome , and the outcomes of the resulting immune response , may have implications for future treatments of viral infections . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"microbiology",
"and",
"infectious",
"disease",
"immunology",
"and",
"inflammation"
] | 2021 | Diverse viral proteases activate the NLRP1 inflammasome |
Public health programs are starting to recognize the need to move beyond a one-size-fits-all approach in demand generation , and instead tailor interventions to the heterogeneity underlying human decision making . Currently , however , there is a lack of methods to enable such targeting . We describe a novel hybrid behavioral-psychographic segmentation approach to segment stakeholders on potential barriers to a target behavior . We then apply the method in a case study of demand generation for voluntary medical male circumcision ( VMMC ) among 15–29 year-old males in Zambia and Zimbabwe . Canonical correlations and hierarchical clustering techniques were applied on representative samples of men in each country who were differentiated by their underlying reasons for their propensity to get circumcised . We characterized six distinct segments of men in Zimbabwe , and seven segments in Zambia , according to their needs , perceptions , attitudes and behaviors towards VMMC , thus highlighting distinct reasons for a failure to engage in the desired behavior .
Voluntary Medical Male Circumcision ( VMMC ) is a critical strategy for HIV prevention in countries with high HIV and low circumcision prevalence ( Auvert et al . , 2005; Bailey et al . , 2007; Gray et al . , 2007; World Health Organization and Joint United Nations Programme on HIV/AIDS , 2007; Njeuhmeli et al . , 2011; World Health Organization and UNAIDS , 2011 ) . The VMMC program has been scaling up rapidly across 14 eastern and southern African countries and close to 12 million circumcisions , against the 20 million target among 15–49 year old males , have been achieved to date ( World Health Organization , 2016 ) . However , given that resources for VMMC scale-up are finite , achieving the desired impact of VMMC on reducing HIV incidence will require prioritizing sub-populations of males and generating demand for VMMC amongst them ( Sgaier et al . , 2015; Sgaier et al . , 2014 ) . There is a need for a greater evidence base , including the use of new approaches adapted from the private sector , to inform the design of demand generation approaches for VMMC ( Sgaier et al . , 2015 ) . In particular , taking a ‘market segmentation’ approach has been highlighted as an important strategy to implement and enable more efficient use of resources . With ‘market segmentation’ , the target population is sub-divided into groups where members of the group share elements in common ( Thomas , 2016; Wedel and Kamakura , 2006 ) . Segmenting the target populations along demographics , such as age or geographic location , has been commonly used in public health . In the private sector , the use of behavioral or psychographic segments has been prevalent ( Bhatnagar and Ghose , 2004; Wade and Eagles , 2003; Desarbo et al . , 1995; Gloy and Akridge , 1999 ) . These segments are constructed based on factors including shared attitudes , values , emotions , perceptions , beliefs and behaviors . Given that these factors play a critical role in driving decisions , often more so than age or other demographic factors ( Carpenter , 2010; Velicer et al . , 2007 ) , segmenting the population along psychographic and behavioral parameters should provide a more powerful tool to target them effectively for behavior change . The application of segmentation has been such a successful strategy to marketing of products and services to consumers that it is now often executed to inform the design of most marketing campaigns in the private sector . We are not aware of any public health or development program that demonstrated the application of behavioral psychographic segmentation . On the other hand , many academic models of behavior , such as the Health Belief Model , Self-determination Theory , and the Reasoned Action Approach , have used differences between individuals in behavioral variables such as ( risk ) perception , beliefs , and motivations to explain why a behavior has or has not occurred ( Carpenter , 2010; Armitage and Conner , 2001; Fishbein and Ajzen , 2010; Ryan and Deci , 2000 ) . However , segmentation techniques are not well-established in the application of these theories to design interventions . We designed and implemented a hybrid behavioral-psychographic segmentation study in Zambia and Zimbabwe . This quantitative study was conducted among males 15–29 years old , given previous evidence that identified this as the most efficient and impactful age for the VMMC programs in both countries to target ( Awad et al . , 2015a; Awad et al . , 2015b ) . In an age-structured mathematical model , Awad et al . ( Awad et al . , 2015a; Awad et al . , 2015b ) assessed the impact of prioritizing different age groups for VMMC in Zimbabwe ( Awad et al . , 2015a ) and Zambia ( Awad et al . , 2015b ) . The model took multiple factors into account , including VMMC effectiveness , cost-effectiveness , reduction in HIV incidence , program cost , and programmatic feasibility . Overall , the analysis found that targeting young males , especially the 15–29 age group , combined some of the largest reductions in HIV incidence with some of the highest program efficiency gains . Therefore , we focused our study on this age group . Both psychographic and behavioral factors were investigated to reveal specific groups of men based on their needs , perceptions , attitudes and behaviors towards VMMC . This research was built on findings from an integrated qualitative study that combined consumer journey mapping , a market research approach that maps the context and sequence of experiences and interactions along the path to decisions from the perspective of the individual , and a behavioral economics game that simulates the real world context of participants to identify context , emotions and mental models that underlie behaviors ( Eletskaya M , Sgaier SK , Kretschmer S , Prasad R , Mulhausen J , Vaish A . 2017 . Employing consumer journey research and behavioral science to understand decision making for Voluntary Medical Male Circumcision in Zimbabwe and Zambia , in preparation ) . In both countries , this integrated qualitative study identified the men’s journey to getting circumcised , the intention-to-action gap that stalled the journey of many to undergoing circumcision , and the underlying factors that inhibited or facilitated the progression of different men to complete their journey . We also explored how the different segments differed in their perceived and actual risk towards HIV/AIDS infection . Finally , we developed a typing algorithm to help programs classify men into the different segments and profile men on risk . Collectively , the insights and tools enabled the development of more effective demand generation activities targeted to specific segments . To our knowledge , this is the first study that uses a hybrid behavioral-psychographic segmentation approach in a public health program at national scale . Our approach , analysis and findings will not only help the national VMMC programs in Zambia and Zimbabwe , but also lay the groundwork for segmentation based on behavioral variables to be used in other development programs .
Figure 1—figure supplement 1 shows the demographic and cultural characteristics of the population sample in Zambia and Zimbabwe . Figure 1—figure supplement 2 characterizes the social acceptability of VMMC , as well as the perceived risk of HIV and other sexually-transmitted infections ( STIs ) in both countries’ samples . The two populations are broadly comparable on the measured parameters , such as age , education level , religion , working status , the perception whether many men had already been circumcised in the community , reasons for circumcision ( for circumcised men only ) , and the perceived risk for HIV/STIs . The starkest difference was that in Zambia , a much greater share of the population was only educated to primary-school level , and a smaller percentage was employed than in Zimbabwe . In order to explain the differences between segments , the results of canonical correlations were analyzed and five key roots influencing intention to choose VMMC were identified . Three of them were common in both countries , while two had some country specifics . The last two roots in Zambia were eliminated from segments profiling due to relatively low contribution into variance between segments . Motivation to go for VMMC , which includes intention to go/advocate for circumcision combined with positive attitudes and beliefs about VMMC , was the most important factor ( similar in both countries ) . It was followed by rejection due to cognitive dissonance in Zimbabwe and control over cognitive dissonance in Zambia , which explains rejection of VMMC and self-efficacy in going for the procedure ( this factor combines , on the one hand , high concern about HIV , and , on the other hand , doubts in VMMC efficacy in regard to protection from HIV along with signs of cognitive dissonance , such as interest to circumcision along with hesitation and uncertainty in its need ) . These two factors were followed by perceived lack of ability ( for both countries it includes perceived level of knowledge about VMMC and desire to have more information/hesitation due to lack of information about the procedure ) and acceptance of social support ( in Zimbabwe ) /perceived self-efficacy against social pressure ( in Zambia ) . The latter includes self-efficacy to go for circumcision even if it is not accepted by people around ( in Zambia ) or readiness to provide social support for others and accept it from others ( in Zimbabwe ) combined with perceived level of social support ( from family and wider community ) . The last factor was personal constraints ( such as fears of pain and embarrassment during the procedure and concerns about the surgery and healing process ) which also was common in both countries . Based on these factors , six segments in Zimbabwe and seven segments in Zambia were identified ( Table 1A and B ) . Each segment was profiled overall based on three levels ( low , medium , or high expression ) of each factor , explaining the characteristics of and differences among the segments ( Table 1A and B ) . For instance , a segment in Zimbabwe characterized by strong levels of motivation , neutral levels of rejection due to cognitive dissonance , an average perceived lack of ability , high acceptance of social support , and moderate levels of fear of the procedure was called Enthusiasts ( see Table 1A and B for a characterization of all segments , and Figure 1 for an overview of the constructs investigated ) . There was almost an even distribution of the population across the different segments in each country ( Figure 2A and B ) The segments can be prioritized by the program for targeting using a number of criteria . Below we illustrate two: ease of converting to higher levels of VMMC uptake and estimated behavioral risk for acquiring HIV/AIDS . The segments are strongly differentiated in terms of the proportion of men who have already been circumcised , which shows that attitudes , motivations and belief correlate with circumcision behavior ( Table 2 ) . Most of the men in the Friends-driven Hesitant ( 86% ) , Scared Rejecters ( 90% ) and Indifferent Resisters ( 94% ) in Zambia and Neophytes ( 94% ) , Rejecters ( 97% ) , Embarrassed Rejecters ( 67% ) , and Highly Resistant ( 99% ) in Zimbabwe had not been circumcised . The highest levels of circumcision are found amongst the Self-reliant Believers ( 71% ) and Traditional Believers ( 71% ) in Zambia and Champions ( 76% ) in Zimbabwe . However , given that the sizes of the segments are different , the relative distribution of uncircumcised men by segment is different ( Figure 2A–D ) . Whilst the Champions represent 17% of uncircumcised men in Zimbabwe , only 6% of uncircumcised men are found in this segment and therefore represent low potential for targeting . Similarly , in Zambia , 19% of the men are Self-reliant Believers , but only 9% of uncircumcised men are found within this segment . Given that males are at different stages of their journey towards VMMC ( Eletskaya M et al . , in preparation ) , targeting uncircumcised males who are in the ‘committed’ stage could be one strategy that the program could employ to achieve its targets ( Figure 2—figure supplement 1 ) . We see considerable differences among segments on proportion of males who are uncircumcised and committed ( Table 2 ) . Proportions of uncircumcised , committed males to VMMC also vary from segment to segment , providing the program for opportunities to target the ‘low-hanging fruit’ ( Figure 2E and F ) . In Zimbabwe , the Enthusiasts represent the biggest opportunity for the program . Slightly more than half of the men in this segment are uncircumcised , but at the same time are strongly committed to circumcision ( Table 2 ) . They represent 38% of the males in the population who are uncircumcised and committed to choose VMMC ( Figure 2E ) . These men want to get circumcised , but need some additional support to cope with their fears and cognitive dissonance . These men also have the highest potential to advocate for VMMC once circumcised ( Figure 2—figure supplement 1 ) . The second level of potential is within the Embarrassed Rejecters and Neophytes ( Figure 2E ) . Neophytes can be converted more easily by addressing their knowledge gaps and dissonance . Embarrassed Rejecters will benefit from having more circumcised men around them , as they are highly influenced by a sense of social inclusion and acceptance . While the Highly Resistant represent 21% of the uncircumcised males , they are very difficult to convert ( only 3% of the uncircumcised committed males are within this segment ) . By just targeting three segments ( Enthusiasts , Embarrassed Rejecters and Neophytes ) , Zimbabwe can circumcise close to 50% of the uncircumcised men in the population . The Friends-driven Hesitants and Enthusiasts represent the biggest opportunity in Zambia . Both have a large proportion of uncircumcised and committed men ( Table 2 ) . While the largest proportion of uncircumcised men in Zambia are within Indifferent Resistants , there are few who are committed . They require a lot of education and support from their social networks . The program could deprioritize the Traditional Believers as most of them will go for circumcision for traditional reasons eventually , anyway . The remaining segments are more or less the same in terms of ease of conversion and targeting each segment will require addressing specific needs ( for example the safety concern of ‘Knowledgeable Hesitants’ or strong fear of surgical procedures of Scared Rejecters ) . On the other hand , the program may want to prioritize those segments that are at highest risk for acquiring HIV to have a more effective impact on the epidemic . In this case , the priority of segment targeting will change , as the estimated risk of acquiring HIV , as measured by the index we created , does not align with the likelihood to already be circumcised nor intent to get circumcised , by segment . In Zimbabwe , the Highly Resistant segment is least likely to be circumcised or intend to get circumcised across all six segments in the country . However , the Highly Resistant segment is assessed as most at risk of acquiring HIV based on sexual behaviour , with 78% having moderate or high risk on the estimated risk index ( see Figure 3 ) . In Zimbabwe , then , if prioritization of segments is made by risk for acquiring HIV , the priorities would be the following: 1 ) Highly Resistant , 2 ) Embarrassed Rejecters ( 73% moderate to high risk ) , 3 ) Scared Rejecters ( 66% moderate to high risk ) , and then Neophytes , Enthusiasts and Champions ( 57% , 57% and 56% respectively ) . In Zambia , estimated risk is less differentiated across the segments overall , with the priority being the following: ( 1 ) Scared Rejecters ( 67% moderate to high risk ) , ( 2 ) Indifferent Resistants , Knowledgeable Hesitants and Traditional Believers ( 65% , 64% and 64% respectively ) , ( 3 ) Self-reliant Believers and Friends-driven Hesitants ( 62% and 63% respectively ) and then 4 ) Socially-supported Believers ( 57% ) ( see Figure 3 ) . Further assessing the segments based on their self-perceived risk is additionally helpful in understanding their approach to VMMC . For example , when evaluating the perceived vs . estimated risk profiles in Zambia , segments having a higher self-perception of being at risk of HIV infection already have higher rates of circumcision . An exception are Traditional Believers , who circumcise for traditional reasons rather than for health reasons; interestingly , their perception vs . estimated risk profile is like that of the Indifferent Resistants ( see Figure 3 ) . In contrast , in Zimbabwe , the perception vs . risk profiles do not align with circumcision rates as clearly . Neophytes , Enthusiasts and Embarrassed Rejecters have the highest rates of moderate and high perceived risk ( 76% , 73% and 71% respectively ) , but Champions have by far the highest rate of circumcisions . The segment identification algorithm for Zimbabwe resulted in 9 rating questions organized in a decision tree . In Zimbabwe , any given man ( aged 15–29 years ) needs to answer only 2 , 3 or 4 of the questions , based on his path through the decision tree , to be classified into his appropriate segment ( Figure 4 ) . In Zambia , the algorithm similarly resulted in 8 questions , of which only 2 to 4 need to be asked to classify a man into his segment ( Figure 4—figure supplement 1 ) .
Adapting the use of ‘market segmentation’ from the private sector , this research demonstrated the application of a hybrid behavioral-psychographic segmentation for men to investigate types of obstacles to demand for VMMC in Zambia and Zimbabwe . Importantly , the research design was built using prior evidence that identified underlying factors – beliefs , barriers , influences – that inhibit or facilitate men to mentally commit to and then take action to getting circumcised ( Montaño et al . , 2014; Price et al . , 2014 ) . Characterizing males aged 15–29 , the resulting segmentation solutions in each country identified segments which are strongly differentiated in their levels of circumcised men , levels of commitment to VMMC among uncircumcised men and , crucially , the underlying factors , e . g . , beliefs , fears , social influences , important for influencing them to take action to getting circumcised . It is worthwhile to compare this behavioral-psychographic segmentation approach to a popular categorization provided by the Diffusion of Innovation Theory ( Rogers , 2003 ) . This framework specifies that in a given population , a novel product or innovation is first taken up by Innovators . It then spreads to Early Adopters , Late Adopters , and finally - after reaching critical mass - to Laggards . This typically happens in a sigmoidal ( ‘S-shaped’ ) fashion ( Dearing , 2009 ) . Diffusion of Innovations Theory emphasizes that different segments can be targeted in different ways . For instance , Innovators can be partners along the way to generate support for a new idea . In contrast , early Adopters are a testing ground and need close support , but Late Adopters are more likely to follow social norms and the fear of being left behind . This classification system is useful if behavioral-psychographic segmentation is not possible , but does not take the specific beliefs , attitudes , and emotions into account that place individuals in either of these categories . Without knowing the underlying reasons for why someone is an adopter or not , little can be done to anticipate and shift behaviors in a specific case . In sum , while Diffusion of Innovations Theory provides a generalized and parsimonious classification , the segmentation approach we outline here provides a generalizable method . The segments that result from the behavioral-psychographic technique will be situation-specific , and therefore provide a foundation for more targeted interventions , than the fixed segments of the Diffusion of Innovations Theory . For instance , Diffusion of Innovations Theory outlines that Laggards need a great amount of familiarity with an innovation before they use it ( Rogers , 2003 ) . However , the characteristics of some segments we found that would most closely fit into the ‘Laggards’ category , such Embarrassed Rejecters and Highly Resistants in Zimbabwe , suggest that lack of familiarity is not at their root of resistance to VMMC . The segment classification algorithms provide the ability to accurately classify any man in the field into his segment in order to provide messaging and interventions most appropriate to influencing him individually to take action to getting circumcised . For example , interpersonal communicators can use the segment classification tool to identify to which segment a man belongs and use a pre-scripted and/or ad hoc approach to counseling that man targeting the issues specific to that segment . For instance , if a man in Zimbabwe is identified as uncircumcised and belonging to the Scared Rejecters segment , the focus of the counseling for that man can quickly address the need to provide clarity on the process of getting circumcised , offer reassurance around the pain that will be felt ( for example , by providing anchors about how much and when in the process of the procedure and healing period pain can be expected ) and how the pain will be able to be managed . In addition , well-crafted mass media communications can be used , which focus on specific critical and differentiating factors for different segments . Men in the segments targeted will then self-selectively pay attention to those communications based on the interest the segmentation results have identified they have for them ( for example , a Scared Rejecter will be more likely to respond to reassurance about pain than would an Embarrassed Rejecter ) . Adding two additional questions to the end of the segment classification tool asking the self-reported number of times a man has sexual intercourse in a typical month and the self-reported number of different partners with whom a man has sex in a typical month would provide the ability to specifically identify and counsel men with regard to their personal estimated risk of HIV infection . Having identified the relative proportions of each segment still uncircumcised , and the proportions of these men committed to getting circumcised or not , the segments in Zambia and Zimbabwe each have been prioritized for targeting . Highest-priority segments can be those that have the most ‘low-hanging fruit’ – uncircumcised men already mentally committed to getting circumcised ( Table 3 ) . Prioritizing these ‘low-hanging fruit’ is especially key in resource-constrained settings , where mass targeting can be replaced with smaller-scale , but focused , communications and interventions which specifically focus on the factors determining action for each targeted segment . Even mass media communication campaigns can be designed to create self-selected attention to them by target segments by using messaging especially relevant for each target segment . For example , a sports car commercial and a truck commercial will draw variable attention by those more interested in owning and driving each type of vehicle . Additionally , specific strategies for targeting each segment , based on the factors which can most influence men in that segment to take action , have been summarized . These strategies include key messages to use , appropriate use of mass media and interpersonal communicators , use of the men as advocates for VMMC , and perhaps the potential offering of device options for getting circumcised . In Zimbabwe , for example , the highest priority segment for targeting is the Enthusiasts , who represent 21% of uncircumcised men , with 85% already committed to getting circumcised . They have relatively higher-risk sexual behavior , but are more likely to advocate for VMMC to other men , once circumcised , because of their strong beliefs about the benefits of VMMC for themselves and their community . For greatest influence , messaging for this segment should focus on detailed information on the procedure , the healing process and pain management to reduce uncertainty , and the potential for improved relationship with their partner , but they should also be counseled on their risky behavior . These communications would be best accomplished through interpersonal communicators . Once circumcised , these men should be actively engaged as VMMC advocates to promote it to other men of their peer age groups . The relative sizes of segments within provinces differ at the province level , and are likely to differ at the district level . Understanding these differences provides programs with the ability to develop more localized programming . For example , districts with higher proportions of segments of men who are more fearful of surgery can prioritize particular interventions designed for these men , such as communications about the safety of the procedure and pain management during the procedure . Targeted strategies using the segmentation results will require adoption of these results and recommended strategies into local VMMC program planning and tactical implementation efforts . While this will initially require additional efforts and potential reprogramming of some strategies , the potential to target and convert men to action for VMMC could increase demand , while doing so with greater efficiency . Similar segmentation solutions could be created and used in all countries where VMMC targets are far from achievement . To date , the segmentation tool has been implemented in pilot tests in both Zimbabwe and Zambia . This ongoing work is conducted in partnership with two non-governmental organizations with extensive experience in VMMC program delivery , Population Services International in Zimbabwe and Society for Family Health in Zambia . An evaluation of initial data is currently underway . Unpublished preliminary findings suggest that training community health workers to use the segmentation tool has measurable effects on overall conversion rates to the VMMC procedure in both countries . Policy makers in both Zimbabwe and Zambia have expressed interest in scaling up the segmentation approach once pilot data has been published . The study has several limitations . The segmentation study was conducted with males aged 15–29 ( core target audience ) , although the VMMC program is targeted towards men up to the age of 49 . Thus , accuracy of the results among the older men requires additional evidence . HIV positive men also were out of scope of this research . For this research , data was collected only once and changes in the population over time will be measured in each country after implementation of the strategies in order to monitor the validity of the results . Any self-report design will also be subject to potential biases , such as social desirability bias . For instance , men may be reluctant to disclose unflattering attitudes to researchers . While self-report biases cannot be fully avoided , they are also pervasive among respondents in all contexts ( Nederhof , 1985 ) , and so unlikely to differ starkly between segments . In addition , the economic decision game asked men to select the option they thought other men would choose ( rather than the one they themselves would select ) , thereby reducing perceived judgment on their own attitudes . Further studies could estimate the extent of existing biases by comparing face-to-face self-report with self-administered , or forced-choice designs . Besides the limitations of the survey , there are some challenges related to translation of these results into action and their proper implementation . The crucial factors in regard to implementation of the results include proper prioritization of the segments , assessment of investments needed for implementation of interventions and strategies , created based on the results and effectiveness of interventions , and strategies need to be evaluated to ensure that they resonate with chosen segments . In conclusion , behavioral-psychographic segmentation is a viable method to identify the diversity of drivers or barriers to a behavior that may exist within a group of healthcare beneficiaries . In this case study , we focused on revealing the different factors that prevented men in Zimbabwe and Zambia from taking up a crucial HIV prevention intervention . In the field , respondents can be allocated to segments with substantial accuracy , using simple decision trees . On the policy-making level , we then provided strategies for targeting the different segments with different messages and channels . In any segmentation application , the identification of ‘low-hanging fruit’ segments will be crucial to maximize the impact of an intervention . Future public health strategies should therefore heed both the diversity of messaging and channels required to target different segments , but also consider prioritizing some segments over others depending on the likely impact and ease of conversion . If interventions are targeted to resonate with segments found through accurate field-based typing , HIV transmission rates in Zimbabwe and Zambia could decrease as the spread of HIV in men is reduced . Beyond the HIV application introduced here , behavioral-psychographic segmentation is likely to be a valuable tool whenever a group of stakeholders is diversified in their beliefs , emotions , and attitudes towards a target behavior .
Responses to a questionnaire , which formed the basis of behavioral segmentation , were collected in 2015 via face-to-face personal interviews among men in Zambia and Zimbabwe using structured quantitative surveys programmed on mobile devices . Surveys were conducted by male , local interviewers who were contracted by the market research company Ipsos in Zambia , and by Ipsos sub-contractors in Zimbabwe . The design of the questionnaire utilized a framework based on the Integrated Behavior Model ( IBM ) ( Montano and Kasprzyk , 1990; Yzer , 2012 ) and was guided by results from a qualitative stage of the research program , which indicated interest and information-seeking about VMMC , uncertainty about the need for VMMC and anxiety felt by a man about getting circumcised were key indicators of cognitive dissonance experienced by men as the main barrier to taking action to get circumcised ( Eletskaya M et al . , in preparation ) . Qualitative data was generated from two sources: journey mapping , and a decision-making game with subsequent hot-state interviews . From journey mapping , we obtained the temporal milestones in the process towards making a decision , and the proportion of men at each milestone . This method also uncovered the beliefs and attitudes for and against circumcision , as well as communication channels , that were relevant to men at each temporal stage . For example , mass communication was more relevant to men in earlier stages , whereas friends gained influence in later stages , and healthcare providers were most influential in very late stages of the decision-making process ( Eletskaya M et al . , in preparation ) . The decision-making game consisted of scenarios that simulated the real-world contexts of the participants . Men were presented with several hypothetical options ( decisions ) in response to a scenario , and were asked to select the one they thought a majority of other men would choose ( Eletskaya M et al . , in preparation ) . This was done in order to reduce the men’s pressure towards carefully-deliberated and socially-desirable answers . Through the game , and subsequent hot-state interviews , we obtained additional information about beliefs , emotions , biases , and contextual factors , as well as triggers to act to get circumcised . As an example of a contextual factor , close male friends were found to be more influential on men’s beliefs and attitudes than female partners ( Eletskaya M et al . , in preparation ) . Qualitative data was collected for a variety of strata of men , whether they were already circumcised or not; or if not , whether they already intended to undergo the procedure or not . This was done in order to obtain a broad picture of prevalent beliefs , biases , emotions , and contextual factors . The qualitative data were then used to inform which beliefs , emotions , attitudes , and contextual factors the survey questions should address . In turn , the survey formed the basis of the key differentiating variables for quantitative segmentation . Montaño used the IBM to quantitatively identify key beliefs about male circumcision in Zimbabwe ( Giles et al . , 2005; Rhodes et al . , 2007; Montaño et al . , 2014 ) . Based on the IBM , intention is a key driver of behavior . Intention is driven by an individual’s beliefs and perceived norms about the behavior and self-perceived control over being able to act on the behavior ( Fishbein and Ajzen , 2010 ) . Many of the key circumcision-related beliefs identified by Montaño in Zimbabwe were also identified as relevant beliefs in Zambia in other studies ( Price et al . , 2014 ) ; thus , the model was viewed as applicable for both countries . For the current research , the IBM was employed and modified to identify specific groups of men , differentiated from each other in terms of constructs underlying behavior . The beliefs identified by Montaño as influencing motivation for getting circumcised were used to inform and build these constructs . The core constructs that lead to an intention to get circumcised include attitudes about the act of getting circumcised ( defined by both emotional feelings and functional beliefs about the behavior ) , perceived norms ( defined by both beliefs about others’ expectations and others’ behaviors ) and personal agency to get circumcised ( defined by both beliefs about personal control and personal efficacy ) . These determine intent , but other factors around perceptions of HIV risk and ability to control risk of infection , as well as structural factors such as availability of information and service for MC and service quality , are either barriers to or facilitators for taking-action . The qualitative phase of this research program also revealed that in some men , the absence of action in going for VMMC can be driven by additional motivational barriers rooted in cognitive dissonance ( Eletskaya M et al . , in preparation ) . Thus , measurement of the presence of cognitive dissonance ( through 3 key components identified as important by the prior integrated qualitative study: interest and information-seeking about VMMC , uncertainty about the need for VMMC and anxiety felt by a man about getting circumcised ) was implemented and combined with measurement of IBM theory constructs to provide a single design framework ( Figure 1 ) . In the structured questionnaire , each construct was measured through a presentation of a number of questions for which respondents were asked to give an answer using a 7-point rating scale . The questionnaire was specifically designed to differentiate among men in their answers to the questions regarding their needs , perceptions , attitudes , beliefs and behaviors toward VMMC such that subsequent data analysis could partition the men into distinct segments , so that similarities are maximized within each segment and dissimilarities are maximized between segments . A categorical variable was used to determine where in the process men were in getting circumcised: ( 1 ) not aware of male circumcision as method for HIV prevention , ( 2 ) aware , but do not believe in benefits of male circumcision’ , ( 3 ) believe in the benefits , but not yet committed to getting circumcised , ( 4 ) committed to getting circumcised , but not yet scheduled it , ( 5 ) scheduled it , but not yet circumcised , ( 6 ) circumcised , but not advocating to other men to get circumcised , ( 7 ) circumcised and advocating to other men to get circumcised . The sample consisted of 4001 men ( both circumcised and uncircumcised ) , aged 15–29 years: 2001 men in Zimbabwe and 2000 in Zambia . Circumcised men were also included to be able to identify the full set of factors that lead to the actual decisions of seeking VMMC ( not only intention to go for circumcision ) and factors that influence attitudes and behaviors post-VMMC . For practical fieldwork cost and logistics purposes , the research targeted the districts with the highest concentrations of uncircumcised men in each country , cumulatively accounting for 80% of the uncircumcised populations in each country . Districts were first sorted from high to low by their populations of uncircumcised men . Then , the cumulative percentage of all uncircumcised men was calculated . Around 50% of districts were below the 80% cut-off point , such that the research was carried out in 38 of 72 districts in Zambia , and 35 of 61 districts in Zimbabwe . Country-level sample sizes ( n = 2000 or 2001 men in each country ) were determined based on experience with cluster segment sizes and the need for minimum sample size in the smallest resulting segment to be large enough for significance testing for differences across segments . Typically , cluster segmentations on consumers result in 4 to 8 segments , with the smallest segment representing as low as 5–10% of the total sample . The minimum desired sample for significance testing was judged as n = 100 . Consequently , if this sample represents the smallest segment with a size of 5% of the total sample , the resulting total sample size should be n = 2000 ( n = 100 * 20 ) . Samples were distributed by age in proportion to the population size for each age group in each district . Households were randomly sampled in the selected districts and a male was approached in each household for the interview . If more than one eligible male lived in the household , selection among these males was made by random selection using a table of random numbers ( Kish , 1949 ) . Once a quota for an age group was reached in a district , only males who met open quota criteria were interviewed . If the household’s selected male was not available or ineligible , the next household was approached . A small incentive was provided to compensate respondents for their time and refusal rates were low ( <5% ) and mostly due to men not having time to participate – consistent with rates experienced for other similar research . All respondent data was used in analysis of the results; none were excluded . This study used data obtained from human participants . The dataset ( anonymized survey responses ) is owned by the governments of Zimbabwe and Zambia , and the authors have requested the respective governments to make the data publicly available . This request is currently subject to government approval . Until the data are publicly available , the data are made available upon reasonable request ( criteria for access may apply subject to assessment by the respective governments ) . Requests for access to the data can be made to the following: Zimbabwe Ministry of Health and Childcare Box CY1122 , Causeway , Harare , Zimbabwe Tel:+263 4 290 1210 Zambia Ministry of Community Development , Mother and Child Health Community House , Sadzu Road , Lusaka , Zambia Tel:+260 211 225 327 | Companies invest a significant amount of time and money into market research that helps them to understand the behaviors , beliefs and motivations of their potential customers . By then “segmenting” people into groups according to these characteristics , marketing messages can be produced that target specific groups more effectively . Most public health efforts are either mass communication campaigns or target particular age groups . However , some public health organizations are starting to study whether the segmenting tactics used by companies could also help to promote healthy behaviors . For example , male circumcision has been shown to reduce the transmission of HIV in Africa . Identifying the beliefs , emotions , motivations or other barriers that stop men from getting circumcised and then targeting specific messages to different groups could help to increase the number of men who opt for circumcision . Sgaier et al . now present evidence that suggests that segmentation could help to promote circumcision in Zimbabwe and Zambia . 4 , 000 men from these countries answered a survey that had been designed based on previous research that investigated how men make the decision whether to be circumcised . Analyzing the results using k-means clustering , a machine learning algorithm , enabled Sgaier et al . to identify six distinct segments in the men from Zimbabwe and seven in the men from Zambia . Further analyses found that the risk of contracting HIV also varied by segment . Sgaier et al . then demonstrated that field workers could use a series of questions to allocate men to each of the groups with an accuracy of over 60% . The segmentation method therefore looks like a promising tool that could be applied to a wide range of public health campaigns . As well as targeting specific groups of people with messaging that resonates specifically with them , segmentation could also highlight those people who are likely to be most easily convinced by a particular health intervention . More research is now needed to improve the usability of the tools that field workers can use to segment their audience . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"epidemiology",
"and",
"global",
"health",
"tools",
"and",
"resources"
] | 2017 | A case study for a psychographic-behavioral segmentation approach for targeted demand generation in voluntary medical male circumcision |
EGFR-mutant NSCLCs frequently respond to EGFR tyrosine kinase inhibitors ( TKIs ) . However , the responses are not durable , and the magnitude of tumor regression is variable , suggesting the existence of genetic modifiers of EGFR dependency . Here , we applied a genome-wide CRISPR-Cas9 screening to identify genetic determinants of EGFR TKI sensitivity and uncovered putative candidates . We show that knockout of RIC8A , essential for G-alpha protein activation , enhanced EGFR TKI-induced cell death . Mechanistically , we demonstrate that RIC8A is a positive regulator of YAP signaling , activation of which rescued the EGFR TKI sensitizing phenotype resulting from RIC8A knockout . We also show that knockout of ARIH2 , or other components in the Cullin-5 E3 complex , conferred resistance to EGFR inhibition , in part by promoting nascent protein synthesis through METAP2 . Together , these data uncover a spectrum of previously unidentified regulators of EGFR TKI sensitivity in EGFR-mutant human NSCLC , providing insights into the heterogeneity of EGFR TKI treatment responses .
Lung cancer is the leading cause of cancer-related mortality worldwide , with non-small-cell lung cancer ( NSCLC ) being the most common subtype ( Bray et al . , 2018; Herbst et al . , 2018 ) . Activating mutations in the kinase domain of epidermal growth factor receptor ( EGFR ) are present in about 10% to 40% of NSCLC patients , most frequently in-frame deletions in exon 19 ( ex19 del ) and a missense arginine-to-leucine mutation at codon 858 ( L858R ) ( Sharma et al . , 2007; Pao and Chmielecki , 2010 ) . The approval and use of EGFR tyrosine kinase inhibitors ( TKIs ) , such as erlotinib and gefitinib , have provided therapeutic breakthrough and achieved clinical success ( Kobayashi et al . , 2005; Rosell et al . , 2012 ) . However , the responses of EGFR-mutant NSCLC patients to EGFR TKIs are rarely complete with variable duration ( Kobayashi et al . , 2005; Rosell et al . , 2012 ) , suggesting that other factors could modulate the dependency of mutant EGFR and thus influencing EGFR TKI efficacy . Moreover , acquired resistance inevitably develops , leading to disease progression in almost all patients ( Pao and Chmielecki , 2010 ) . A growing number of studies have been focusing on the understanding of mechanisms underlying EGFR TKI resistance , involving a variety of genetic and non-genetic alterations in signaling pathways and cell state ( Sequist et al . , 2011 ) . Secondary EGFR on-target mutations , most frequently T790M mutation , account for about half of relapsed tumours with acquired resistance ( Kobayashi et al . , 2005; Pao et al . , 2005; Sequist et al . , 2011 ) . Resistance can also result from reactivation of key downstream signaling pathways originally inhibited by EGFR TKIs , such as PI3K-AKT and RAS-MEK-ERK pathways ( Chong and Jänne , 2013; Niederst and Engelman , 2013; Camidge et al . , 2014; Rotow and Bivona , 2017 ) . For instance , resistance has been associated with amplification or activation of MET , HER2 , FGFR and AXL , as well as PIK3CA mutation or loss of PTEN ( Engelman et al . , 2006; Engelman et al . , 2007; Sos et al . , 2009; Takezawa et al . , 2012; Zhang et al . , 2012; Ware et al . , 2013 ) . Moreover , activation of NF-kB and YAP signaling pathways also confers resistance to EGFR TKIs ( Bivona et al . , 2011; Hsu et al . , 2016; Chaib et al . , 2017 ) . Non-genetic alterations , such as histological transformation from NSCLC to small cell lung cancer ( SCLC ) and epithelial-to-mesenchymal transition ( EMT ) , have also been reported in relapsed tumors ( Sequist et al . , 2011 ) . Furthermore , accumulating evidence has suggested that small subpopulations of cancer cells can evade lethal drug treatment by entering a drug tolerant ‘persister’ ( DTP ) state and serve as a founder for acquiring heterogeneous drug-resistance mechanisms upon long-term drug treatment ( Sharma et al . , 2010; Ramirez et al . , 2016 ) . Importantly , the mechanisms of EGFR TKI resistance elucidated to date can encompass multiple mechanisms simultaneously in one patient or even tumour , creating significant obstacles for designing better treatment strategies for patients ( Pao and Chmielecki , 2010; Chong and Jänne , 2013 ) . Moreover , a number of EGFR TKI-resistant tumours lack known resistance mechanisms ( Sequist et al . , 2011 ) , suggesting the existence of previously unrecognized mediators of EGFR TKI efficacy . In order to better understand the incomplete response of EGFR-mutant NSCLC to EGFR TKI as well as identify prospective mechanisms of resistance , we conducted a genome-scale CRISPR-Cas9 genetic screen in a human NSCLC cell line harbouring activating EGFR mutation . By generating EGFR TKI resistance , we identified a number of genes , when deleted , enhanced or reduced EGFR TKI sensitivity and consequently prevented or accelerated development of EGFR TKI resistance , respectively .
HCC827 , a very commonly used EGFR-mutant NSCLC cell line , harbors ex19 del in EGFR and is highly sensitive to EGFR TKI treatment . Generation of a dose-response curve upon 3 days’ erlotinib exposure yielded the IC50 25 . 6 ± 3 . 6 nM ( Figure 1A ) . However , even with high dose of erlotinib ( 1 µM and above ) , about 30% of the cells could still survive this initial pulse ( Figure 1A ) . Examination of the cell proliferation in the presence of DMSO control or 1 µM erlotinib over a 30 day period using IncuCyte demonstrated that HCC827 cells were still able to proliferate at a low rate when exposed to high dose of erlotinib ( Figure 1B ) . Additionally , colony formation assay also confirmed the existence of a small fraction of viable cells after 9 days’ erlotinib treatment , referred to as ‘drug tolerant persister’ ( DTP ) cells ( Figure 1C ) . Thereafter , the DTP cells commenced cell proliferation in the presence of erlotinib , yielding colonies of cells referred to as ‘drug tolerant expanded persister’ ( DTEP ) cells or drug resistant cells ( Figure 1C ) . These data suggested that EGFR inhibition in the cultured cells mimics clinical observations of the incomplete response and/or innate resistance to EGFR TKI treatment , allowing the assay window to screen for mediators of EGFR TKI sensitivity . We aimed to systematically identify genetic modifiers that regulate the response of EGFR-mutant NSCLC cells to EGFR TKI treatment by applying a genome-scale CRISPR-Cas9 loss-of-function screening approach ( Figure 1D ) . We introduced a pooled lentiviral single guide RNA ( sgRNA ) library targeting 18 , 360 genes ( five sgRNAs per gene ) into HCC827 cells with constitutive Cas9 expression ( HCC827-Cas9 ) and treated these cells with DMSO or erlotinib . We intentionally applied high dose of erlotinib ( 1 µM ) to allow the survival of a small subpopulation of DTP cells and development of drug resistance in the long-term . This strategy ensured that genes whose deletion synergize with or confer resistance to erlotinib could be negatively or positively selected from the screen , respectively , following erlotinib treatment compared to DMSO treatment . Three weeks post-treatment , cells were harvested and subjected to next-generation sequencing ( NGS ) to identify differential sgRNA representation between DMSO and erlotinib treated populations ( Figure 1D and Figure 1—figure supplement 1A–C ) . Differential sgRNA representation was evaluated in the form of log2 fold change between the erlotinib- and DMSO-treated samples . A robust z score was calculated using the median and mean-absolute deviation for the calculated fold changes across the entire sgRNA library . To provide a qualitative assessment of the screen performance , we plotted the P values calculated by the redundant small interfering RNA ( siRNA ) activity ( RSA ) test , representing the probability of a gene hit based on the collective activities of multiple sgRNAs per gene , against Q1- and Q3-based z scores ( Figure 1E–F ) . To identify high-confidence negatively selected hits , we used a stringent RSA ≤ −4 and Q1 z-score ≤ −1 threshold . This analysis identified 35 genes , the inactivation of which caused sensitization to erlotinib in HCC827 cell line ( Figure 1E , Figure 1—figure supplement 1D , Supplementary file 1 and 2 ) . FGFR1 and YAP1 , two known mediators of EGFR TKI resistance in EGFR-mutant NSCLC ( Ware et al . , 2013; Hsu et al . , 2016; Chaib et al . , 2017; Ghiso et al . , 2017 ) , were among these hits . Mapping the 35-gene set on the STRING protein-protein interaction network and the Reactome database generated clusters consisting of heparan sulfate metabolism , GPCR/G-protein signaling , Hippo-YAP signaling pathway , as well as components of the SAGA/transcriptional complex ( Figure 1G and Figure 1—figure supplement 1F ) , providing potential synthetic lethal partners with EGFR TKI in EGFR-mutant NSCLC . A less stringent threshold ( RSA ≤ −3 and Q1 z-score ≤ −1 ) generated a larger list of 122 genes whose loss sensitized HCC827 cells to erlotinib treatment ( Supplementary file 1 and 2 ) . Similarly , we applied a stringent RSA ≤ −4 and Q3 z-score ≥1 threshold to identify 47 genes whose inactivation conferred resistance to erlotinib in HCC827 cell line ( Figure 1F , Figure 1—figure supplement 1E , Supplementary file 1 and 2 ) . A less stringent threshold ( RSA ≤ −3 and Q3 z-score ≥1 ) expanded the list to 171 genes ( Supplementary file 1 and 2 ) , including negative regulators of YAP signaling such as NF2/Merlin and SAV1 , further warranting the importance of YAP signaling in mediating EGFR TKI resistance in lung cancer . Protein network and pathway analyses of the 47 positively selected hits enriched for PI3K-mTOR signaling ( such as PTEN , TSC1 and TSC2 ) and RAS-MAPK signaling ( such as NF1 , SPRY2 and LZTR1 ) pathways ( Figure 1H and Figure 1—figure supplement 1G ) , two well-established modes of resistance to EGFR TKI in EGFR-mutant NSCLC ( Pao and Chmielecki , 2010; Chong and Jänne , 2013; Niederst and Engelman , 2013; Rotow and Bivona , 2017 ) , validating the good performance of our CRISPR screen . Moreover , several genes encoding proteins involved in protein ubiquitination and degradation pathway were positively selected from the screen ( Figure 1F and H ) . Among them , we identified KEAP1 and FBXW7 , the loss of which have previously been shown to confer resistance to EGFR TKI treatment in EGFR-mutant NSCLC cells ( Krall et al . , 2017; Ye et al . , 2017 ) . Interestingly , components of the Cullin 5 ( CUL5 ) -RING E3 ligase ( CRL5 ) complex ( such as CUL5 , RNF7 and UBE2F ) as well as ARIH2 , an Ariadne family RING-in-Between-RING ( RBR ) E3 ligase working together with CRL5 , were amongst the positively selected hits , suggesting their previously unrecognized role in mediating EGFR TKI resistance in lung cancer . Together , our genome-wide CRISPR-Cas9 loss-of-function genetic screen successfully revealed both known and potential synthetic lethal vulnerabilities with EGFR TKI as well as modes of resistance in EGFR-mutant NSCLC . Next , we set out to validate a number of selected hits from the primary screen . We first focused on the synergy hits including YAP1 , USP22 and GPCR/G-protein signaling-related LPAR2 , GNB2 , PKN2 and RIC8A , by selecting one sgRNA for each gene and monitoring the gene deficiency and erlotinib efficacy by western blots ( Figure 2A ) . Long-term colony formation assays confirmed that all sgRNAs targeting these genes strongly sensitized HCC827 cells to both erlotinib and gefitinib treatment ( Figure 2B–C ) . The synthetic lethal phenotype was similarly observed in another EGFR-mutant NSCLC cell line NCI-H3255 that expresses EGFR-L858R ( Figure 2—figure supplement 1A–C ) . For validation of resistance hits , we intentionally selected individual sgRNAs targeting novel mediators of erlotinib resistance ( ARIH2 , CUL5 , RNF7 , KCTD5 , PDCD10 and BTAF1 ) ( Figure 2D ) and monitored their ability to confer resistance to EGFR TKIs by performing long-term colony formation assays . Our results confirmed that all sgRNAs demonstrated significant resistance to both erlotinib and gefitinib treatment in HCC827 cells ( Figure 2E–F ) as well as in NCI-H3255 cells ( Figure 2—figure supplement 1D–F ) . Together , the validation study corroborated our CRISPR screen findings , providing confidence to investigate additional hits generated from our CRISPR screen ( Supplementary file 1 and 2 ) . LPAR2 , one of our strongest synergy hits , is a member of the lysophosphatidic acid receptor ( LPAR ) family . LPARs , consisting of LPAR1-6 , are membrane-bound G-protein-coupled receptors ( GPCRs ) ( Yung et al . , 2014 ) . Inspection of sgRNAs targeting all six LPAR members demonstrated that deletion of LPAR1/2/3 tends to have synergistic effect with erlotinib treatment in HCC827 cells , although only LPAR2 passed our stringent hit selection criteria ( Figure 2—figure supplement 2A ) . We then asked whether pharmacologically targeting LPARs with LPAR antagonists could synergize with EGFR TKI to prevent drug resistance . Indeed , colony formation assays in HCC827 and NCI-H3255 cells demonstrated that LPAR antagonists , such as Compound 2 ( LPAR1 selective ) ( Qian et al . , 2012 ) , AM095 ( LPAR1 selective ) ( Swaney et al . , 2011 ) and Ki16425 ( targeting LPAR1-3 ) ( Ohta et al . , 2003 ) , synergized with erlotinib to inhibit clonogenic cell growth ( Figure 2—figure supplement 2B–C ) . These results nominate LPAR antagonists might , in principle , be combined with EGFR inhibition to delay resistance occurrence . Among the novel synergy hits , we further focused on the most prominent hit , RIC8A , and decided to confirm its role in cell survival upon EGFR inhibition in more details . RIC8A functions as a biosynthetic chaperone and guanine nucleotide exchange factor ( GEF ) for a subset of G protein α subunits ( Nishimura et al . , 2006; Gabay et al . , 2011; Chan et al . , 2013; Kant et al . , 2016 ) . Previous study has shown that RIC8A deficiency leads to loss of Gα subunits , and consequently , Gβγ dimers dissociated from Gα subunits could be recognized by KCTD5 for degradation ( Chishiki et al . , 2013; Boularan et al . , 2015; Brockmann et al . , 2017 ) . First , we deleted RIC8A in HCC827 cells using two independent sgRNAs . Western blot analysis confirmed RIC8A knockout efficiency and the consequent effect on the protein abundance of Gαq ( GNAQ ) and Gβ2 ( GNB2 ) subunits as well as the on-target inhibition of EGFR and its downstream signaling upon erlotinib treatment ( Figure 3A ) . We then examined whether loss of RIC8A affects the overall EGFR TKI response in HCC827 cells by generating a dose-response curve . Cells deficient in RIC8A exhibited enhanced sensitivity to both erlotinib and gefitinib ( Figure 3B and Figure 3—figure supplement 1A ) . Consistently , RIC8A depletion dramatically accelerated apoptosis induction by erlotinib , represented by the increased caspase 3/7 activity ( Figure 3C ) . To further explore the role of RIC8A in a long-term drug treatment associated with acquired drug resistance , we performed the colony formation assays in the absence or presence of erlotinib and gefitinib . Our results demonstrated that loss of RIC8A had no effect on the basal proliferation of HCC827 cells while dramatically suppressed erlotinib- and gefitinib-resistant colony formation ( Figure 3D–E ) , which could be rescued by overexpression of CRISPR/Cas9-resistant RIC8A ( Figure 3—figure supplement 1B ) . Similar effects were also observed in two additional NSCLC cell lines with the same EGFR ex19 del , HCC4006 and PC9 ( Figure 3—figure supplement 1E–L ) . Moreover , we also showed that loss of RIC8A sensitized cells to EGFR TKI treatment and efficiently prevented erlotinib- and gefitinib-resistant colony formation in NCI-H3255 cell line harboring EGFR-L858R mutation ( Figure 3F–3J and Figure 3—figure supplement 1C ) . EGFR T790M ‘gatekeeper’ mutation is a major resistance mechanism in EGFR-mutant NSCLC patients in response to first-generation EGFR TKI treatment ( Pao et al . , 2005; Sequist et al . , 2011 ) . Third-generation EGFR TKIs , such as EGF816 ( Jia et al . , 2016; Lelais et al . , 2016 ) and the FDA-approved agent osimertinib ( Jänne et al . , 2015 ) , can bind to and inhibit mutant EGFR with and without the T790M mutation . However , the third-generation EGFR TKIs also produce a partial response followed by progression and acquired resistance ( Rotow and Bivona , 2017 ) . To investigate the potential role of RIC8A in mediating response to third-generation EGFR TKI , we deleted RIC8A in NCI-H1975 cells , which harbor EGFR-L858R and T790M mutations and thus insensitive to erlotinib ( Figure 3K and Figure 3—figure supplement 1D ) . Strikingly , loss of RIC8A in NCI-H1975 cells dramatically enhanced the growth suppressive effect ( Figure 3L ) , promoted apoptosis ( Figure 3M ) , and attenuated development of resistance ( Figure 3N and O ) upon EGF816 treatment . Moreover , loss of RIC8A exhibited no effect on the EGFR TKI sensitivity in EGFR wild-type ( EGFR-WT ) NSCLC cell lines ( A549 , NCI-1299 and NCI-H460 ) or normal human bronchial epithelial cell line BEAS-2B ( Figure 3—figure supplement 2A–P ) . Taken together , these data suggest that loss of RIC8A is synthetic lethal with EGFR inhibition across a panel of NSCLC cell lines with various EGFR mutations , representing a general mechanism . Next , we aimed to understand the molecular mechanism through which RIC8A loss synergizes with EGFR inhibition . Gα protein coupled GPCR signaling has been well characterized to modulate activities of Rho/Rac GTPase , which in turn lead to actin cytoskeleton remodeling , consequently regulating YAP signaling via both LATS1/2-dependent and LATS1/2-independent mechanisms ( Yu et al . , 2012; Ma et al . , 2019 ) . Given the important roles of RIC8A in mediating Gα activation , we hypothesized that RIC8A is a positive regulator of YAP signaling . To test this hypothesis , we first deleted RIC8A in HEK293A cells using two independent sgRNAs ( Figure 4A ) and verified that loss of RIC8A significantly decreased YAP reporter ( GTIIC-GFP ) activity ( Figure 4B and Figure 4—figure supplement 1A ) , reduced expression of classical YAP target genes ( ANKRD1 , CTGF , and CYR61 ) ( Figure 4C ) , and suppressed YAP-dependent growth of HEK293A cells ( Figure 4—figure supplement 1D ) . Furthermore , these defects resulting from RIC8A loss can be overcome by overexpression of the constitutively active YAP ( YAP-5SA ) ( Figure 4—figure supplement 1B–D ) . Together , these data strongly support the role of RIC8A in positively regulating YAP signaling in HEK293A cells . We then asked whether RIC8A is also a potent positive regulator of YAP signaling in EGFR-mutant NSCLC cells . Indeed , across a panel of EGFR-mutant NSCLC cell lines , loss of RIC8A resulted in an increase in YAP phosphorylation at Ser127 site , an indicator of YAP inactivation ( Figure 4D ) . In line with this finding , loss of RIC8A significantly suppressed the expression of YAP target genes in all tested cell lines ( Figure 4E ) . Together , these data demonstrate that RIC8A positively regulates YAP signaling in EGFR-mutant NSCLC cells . Moreover , overexpression of YAP-5SA in representative HCC827 cell line blocked the effect of RIC8A loss on YAP signaling ( Figure 4F and G ) . Importantly , YAP activation itself conferred resistance to EGFR inhibition and blocked RIC8A loss-induced synthetic lethality with EGFR inhibition ( Figure 4H–4J ) . Collectively , these results suggest that loss of RIC8A synergizes with EGFR inhibition by attenuating YAP signaling in lung cancer . To dive deeper into the connection between RIC8A and YAP signaling , we speculated that RIC8A positively regulates YAP signaling via Gα-Rho/Rac axis . We first tested whether RHOA inhibition could confer synthetic lethality with EGFR inhibition . Indeed , loss of RHOA increased YAP phosphorylation at Ser127 site ( Figure 4—figure supplement 2A ) , decreased YAP-target gene expression ( Figure 4—figure supplement 2B ) , and induced synthetic lethality with EGFR inhibition in HCC827 cells ( Figure 4—figure supplement 2C ) . Then we examined whether RHOA activity is reduced by RIC8A loss . Unfortunately , after many attempts , we observed little decrease in the active RHOA signal upon RIC8A loss using a RHOA G-LISA Activation Assay Kit ( Figure 4—figure supplement 2D–E ) . However , this could be due to the fast dynamics of RHOA activation-inactivation cycle , making it difficult to capture the real-time changes in RHOA activity upon RIC8A loss . In addition , ARHGAP29 , encoding a Rho GTPase activating protein , was previously reported to be a YAP target gene ( Qiao et al . , 2017 ) . We observed that RIC8A loss caused significant decrease in the expression of ARHGAP29 ( Figure 4—figure supplement 2F ) , which could provide a negative-feedback mechanism to alleviate the decrease of RHOA activity resulting from RIC8A loss . Therefore , we believe our inability of detecting RHOA activity changes is most likely due to both technical reasons and the negative-feedback mechanism . Consistently , we observed a morphological alteration in HCC827 cells upon RIC8A loss ( Figure 4—figure supplement 2G ) and the decrease of Cofilin phosphorylation that is downstream of the RHOA-ROCK signaling ( Figure 4—figure supplement 2H ) , indicating that loss of RIC8A indeed negatively impacted the output from RHOA activation . Moreover , treatment with Y-27632 , the inhibitor of Rho-associated kinase ROCK , also induced synthetic lethality with EGFR inhibition in HCC827 cells ( Figure 4—figure supplement 2I–J ) . Taken together , these data suggested the RIC8A-Gα-RHOA-YAP signaling axis is involved in the regulation of EGFR TKI sensitivity in EGFR-mutant NSCLC cells . Admittedly , RIC8A might regulate YAP signaling through other effectors , and comprehensive understanding of the signaling between RIC8A and YAP warrants future characterizations . To further validate the possible role of ARIH2 in EGFR TKI resistance , we introduced into HCC827 cells two independent sgRNAs targeting ARIH2 and confirmed the ARIH2 knockout efficiency by western blot analysis ( Figure 5A ) . Loss of ARIH2 in HCC827 cells decreased sensitivity to erlotinib ( Figure 5B ) and reduced apoptosis induction upon erlotinib treatment ( Figure 5C ) . Accumulating evidence suggests that small subpopulations of cancer cells can survive strong EGFR inhibition by entering a DTP state , which could allow the emergence of heterogeneous EGFR TKI resistance mechanisms in EGFR-mutant lung cancers ( Sharma et al . , 2010; Ramirez et al . , 2016 ) . To examine whether loss of ARIH2 could increase the reservoir of DTP cells , we treated control ( sgAAVS ) or ARIH2-deficient ( sgARIH2 ) HCC827 cells with 1 µM of erlotinib or gefitinib for three rounds with each treatment lasting for 72 hr . Our results demonstrated that loss of ARIH2 facilitated more cells to enter into the DTP state ( Figure 5D–E ) . To test long-term effects of ARIH2 loss , we first performed colony formation assays and confirmed that ARIH2 loss substantially enhanced clonogenic cell survival upon EGFR inhibition ( Figure 5F–G ) , which could be rescued by overexpression of CRISPR/Cas9-resistant ARIH2 ( Figure 5—figure supplement 1 ) . Next , we carried out a cell competition assay , in which unlabeled parental HCC827 cells were mixed in a ratio of 100:1 with RFP-labeled control or ARIH2-deficient cells and maintained in culture in the absence or presence of EGFR TKIs for three weeks ( Figure 5H ) . Analysis of RFP-positive cells showed substantial outgrowth of the ARIH2-deficient cells under the pressure of EGFR inhibition ( Figure 5I–J ) . The percentage of RFP-positive cells remained similarly in the absence of EGFR TKI ( Figure 5I–J ) , together with no differential colony formation between control and ARIH2-deficient cells in the absence of EGFR TKI ( Figure 5F–G ) , suggesting that loss of ARIH2 has little or no effect on the basal proliferation of HCC827 cells . Likewise , drug resistance phenotype associated with ARIH2 loss was also observed in NCI-H3255 cell line ( Figure 5—figure supplement 2A–F ) . On the contrast , ARIH2 loss had no effect on the EGFR TKI sensitivity in EGFR-WT NSCLC cells or normal cells ( Figure 5—figure supplement 3A–P ) . Thus , our data provide a chain of evidence to demonstrate that ARIH2 loss reduces the sensitivity of EGFR-mutant NSCLC cells to EGFR inhibition and promotes acquired resistance . We continued to assess the resistant effect of ARIH2 loss to erlotinib treatment in vivo . We established xenografts of control or ARIH2-deficient HCC827 cells in nude mice . Mice were enrolled in the study once tumors had reached approximately 200 mm3 in size , and were randomly assigned to receive either vehicle or 10 mg/kg erlotinib once daily for the duration of the study ( Figure 5K–L and Figure 5—figure supplement 4 ) . As shown in Figure 5K , loss of ARIH2 alone had little effect on tumor growth in the vehicle treatment group . Erlotinib treatment efficiently suppressed tumor growth ( compared to vehicle treatment ) , with resistance emerging after approximately 90 days of continuous erlotinib treatment in control xenograft tumors ( Figure 5K ) . Importantly , loss of ARIH2 significantly accelerated the development of resistant tumors ( Figure 5K–L ) . Taken together , these data further support the notion that loss of ARIH2 confers resistance to EGFR inhibition in EGFR-mutant NSCLC . To gain mechanistic insights into how ARIH2 loss confers resistance to EGFR inhibition , we compared the global protein changes between control and ARIH2-deficient HCC827 cells by quantitative mass spectrometry analysis ( Figure 6A ) . Using a stringent criteria ( |Log2 fold change| ≥ 0 . 9 and P value ≤ 0 . 001 ) , we observed that 46 proteins significantly increased and 13 proteins decreased upon ARIH2 loss ( Figure 6A ) . We surmised that loss of ARIH2 might increase the abundance of certain essential proteins to survive EGFR inhibition . Therefore , we searched the 46 proteins with increased abundance upon ARIH2 loss and were particularly interested in a few hits for further characterization , including METAP2 , ALDOA and PSAT1 . METAP2 , also known as methionine aminopeptidase 2 , is a eukaryotic initiation factor 2 ( eIF2 ) -associated glycoprotein which possesses dual functions of regulating global protein synthesis rate and co-translationally removing the N-terminal methionine from nascent proteins ( Datta , 2000 ) . ALDOA ( fructose-bisphosphate aldolase A ) and PSAT1 ( phosphoserine aminotransferase ) are important enzymes involved in glycolysis and serine biosynthesis pathways , respectively . We first validated by immunoblotting analysis that loss of ARIH2 indeed increased protein levels of METAP2 , ALDOA and PSAT1 in HCC827 cells ( Figure 6B ) , without affecting their corresponding mRNA levels ( Figure 6C ) , suggesting a possible post-transcriptional regulation mechanism . Loss of ARIH2 also led to increase of METAP2 and ALDOA proteins , but not PSAT1 protein , in NCI-H3255 cells ( Figure 6—figure supplement 1A–B ) . Moreover , loss of ARIH2 or CUL5 also dramatically increased protein abundance of METAP2 , ALDOA and PSAT1 in HEK293T cells ( Figure 6—figure supplement 1C ) , suggesting a general mechanism of regulation of these proteins by ARIH2 and CUL5 complex . We continued to examine whether increased abundance of these proteins could confer resistance to EGFR inhibition in EGFR-mutant NSCLC cells . We first focused on METAP2 , the abundance of which increased most upon ARIH2 loss , and ectopically overexpressed a hemagglutinin ( HA ) tagged METAP2 ( HA-METAP2 ) in HCC827 cells ( Figure 6D ) . A long-term colony formation assay of the vector- and HA-METAP2-expressing HCC827 cells demonstrated that METAP2 overexpression confers resistance to EGFR inhibition ( Figure 6E–F ) . Given its role in regulating global protein synthesis , we sought to assess de novo protein synthesis in both vector- and METAP2-overexpressing cells by L-azido-homoalanine ( AHA ) labeling ( Figure 6G ) . EGFR inhibition by erlotinib treatment drastically inhibited nascent protein synthesis ( Figure 6G ) , in line with its effect on overall growth inhibition . Importantly , cells with METAP2 overexpression exhibited increased protein synthesis , compared to vehicle-expressing cells , when challenged by erlotinib treatment ( Figure 6G ) . Furthermore , loss of ARIH2 , which is able to increase METAP2 protein level , also increased nascent protein synthesis upon EGFR inhibition ( Figure 6H ) . Together , these data suggest that ARIH2 loss confers resistance to EGFR inhibition , at least in part , by maintaining higher protein synthesis rate through METAP2 . Next , we investigated how ARIH2 regulates METAP2 protein level . EGFR inhibition did not alter METAP2 protein level in both control and ARIH2-deficient HCC827 cells ( Figure 6—figure supplement 1D–F ) , ruling out the possibility that METAP2 is a downstream effector of EGFR which could be regulated by ARIH2 . Proteasome inhibitor ( Bortezomib ) treatment increased METAP2 protein abundance in HCC827 cells ( Figure 6I ) , suggesting that METAP2 protein level is regulated by the proteasome-dependent degradation pathway . Moreover , bortezomib induced METAP2 protein level increase was only observed in control cells but not in ARIH2-deficient cells ( Figure 6J ) , indicating that proteasome-mediated METAP2 protein degradation is dependent on ARIH2 . We further attempted to examine METAP2 ubiquitination in HCC827 cells . Unfortunately , we were unable to detect endogenous METAP2 protein ubiquitination ( Figure 6—figure supplement 1G ) , likely due to that the METAP2 ubiquitination level is too low to be detected or ARIH2 regulates METAP2 degradation indirectly . Next , we also wondered whether ARIH2 regulates METAP2 protein translation/synthesis . Specifically , we assessed de novo METAP2 protein synthesis in control or ARIH2-deficient HCC827 cells by AHA labeling followed by streptavidin pulldown and showed that loss of ARIH2 increased nascent METAP2 protein synthesis ( Figure 6K ) , suggesting that ARIH2 indeed regulates METAP2 protein translation . As a control , EGFR protein synthesis remained unchanged upon ARIH2 loss ( Figure 6K ) . Taken together , these data demonstrate that ARIH2 is capable of regulating both protein translation and protein degradation of METAP2 . Future studies are required to gain a comprehensive picture of the regulation of METAP2 protein level by ARIH2 . Lastly , we ectopically overexpressed HA-ALDOA or HA-PSAT1 in HCC827 cells as well ( Figure 6L ) . Long-term colony formation assays showed that overexpression of ALDOA or PSAT1 is capable to promote resistance to EGFR inhibition ( Figure 6M–N ) . Thus , we concluded that loss of ARIH2 confers resistance to EGFR inhibition in EGFR-mutant NSCLC cells by integrating multiple mechanisms .
Understanding and overcoming resistance to EGFR TKIs remain a major challenge in NSCLC . Several studies have previously employed loss- or gain-of-function screens to profile genetic interactions with mutant EGFR in NSCLC cells to look for synthetic lethality ( Bivona et al . , 2011; de Bruin et al . , 2014; Sharifnia et al . , 2014; Lantermann et al . , 2015; Liao et al . , 2017 ) . However , they utilized either small interfering RNA ( siRNA ) or short hairpin RNA ( shRNA ) approach or focused on small subsets of genes ( such as tumor suppressors , oncogenes and genes encoding kinases ) , and identified distinct genetic modifiers of EGFR TKI sensitivity . Herein in this study , we presented an unbiased genome-scale CRISPR-Cas9 screening strategy to systematically capture the breadth of genetic modifiers of mutant EGFR dependency in NSCLC . We purposely selected HCC827 , a cell line with high sensitivity to EGFR TKI , and generated resistance by applying clinically relevant concentration of erlotinib for a sustained period , allowing the identification of both negative and positive regulators of EGFR TKI sensitivity simultaneously . Our genome-wide CRISPR screen successfully revealed a number of known causative genes and signaling pathways associated with EGFR TKI resistance , such as YAP signaling , PI3K-mTOR signaling and RAS-MAPK signaling pathways , reinforcing their significance in mediating mutant EGFR dependency in NSCLC . More importantly , we uncovered a list of previously unrecognized genes whose deletion caused synthetic lethality with or conferred resistance to EGFR TKI treatment , broadening our understanding of EGFR signaling regulation . Importantly , we determined the copy number of these genes by searching the database ( https://cansar . icr . ac . uk/cansar/cell-lines/HCC-827/copy_number_variation/no%20signal/ ) and did not find gain of copy numbers for any of the genes studied herein , ruling out the false positive possibility of artifacts from targeting genes in amplicons . Among those newly identified mutant EGFR dependencies , GPCR/G-protein signaling module stood out as a strong vulnerability with EGFR inhibition . Particularly , genetic ablation as well as pharmacological antagonism of the LPARs strongly sensitized EGFR-mutant NSCLC cells to EGFR inhibition . Interestingly , LPAR signaling has been shown to contribute to malignancy and chemotherapy resistance in various tumors ( Yung et al . , 2014; Hashimoto et al . , 2016 ) . Molecularly , GPCRs initiate downstream signaling cascades through activating heterotrimeric G proteins consisting of Gα subunit ( G12/13 , Gq/11 , Gi/o , or Gs ) and Gβγ heterodimer in concert with other effector proteins ( Dorsam and Gutkind , 2007 ) . Importantly , inactivation of GPCR related effectors , such as RIC8A , PKN2 , GNB2 ( encoding Gβ2 subunit ) or GNA13 ( encoding Gα13 subunit ) , strongly synergized with EGFR inhibition in EGFR-mutant NSCLC cells , revealed by our CRISPR screen . Furthermore , loss of KCTD5 , a previously reported negative regulator of Gβ protein stability ( Brockmann et al . , 2017 ) , promoted resistance to EGFR inhibition . Together , these findings suggest that EGFR-mutant NSCLC depends on parallel GPCR signaling to modulate mutant EGFR addiction . RIC8A , the most prominent synergy hit from our screen , possesses dual functions as a molecular chaperone required for nascent Gαq/i/13 protein folding and initial membrane association as well as a guanine nucleotide exchange factor ( GEF ) for Gαq/i/13 ( Nishimura et al . , 2006; Gabay et al . , 2011; Chan et al . , 2013; Kant et al . , 2016 ) . Interestingly , LPAR signaling coupled with G12/13 has been shown to activate YAP pathway in HEK293A cells ( Yu et al . , 2012 ) . Moreover , majority of human uveal melanomas , driven by activating mutations in Gαq/11 proteins , depend on YAP signaling for tumorigenesis ( Yu et al . , 2014 ) . Ric-8A gene deletion significantly suppressed tumorigenesis in a mouse model of oncogenic Gαq-driven melanoma ( Patel and Tall , 2016 ) , implying its regulation of YAP signaling pathway . Here , we extended the regulation of YAP pathway by GPCR and G proteins to RIC8A , and demonstrated , for the first time , that RIC8A positively regulates YAP signaling to modulate EGFR TKI sensitivity in EGFR-mutant NSCLC cells . Therefore , targeting RIC8A might be promising to prevent EGFR TKI resistance in lung cancer . YAP signaling pathway has also been shown to mediate resistance to other targeted therapies , such as RAF and MEK inhibition ( Lin et al . , 2015 ) . Whether targeting RIC8A could inhibit YAP signaling and consequently enhance treatment response in those circumstances is of great interest and awaits future investigation . On the other hand , reduced NF1 ( neurofibromin ) expression ( de Bruin et al . , 2014 ) or loss of PTEN ( Sos et al . , 2009 ) has been observed in clinical specimens with acquired resistance to EGFR TKI treatment . Both of them strongly scored as resistance hits in our CRISPR screen . Among other positively selected hits , ARIH2 , CUL5 , RNF7 and UBE2F , falling into an ubiquitination related functional module , significantly stood out . Cullin-RING ligases ( CRLs ) , comprising the largest subfamily of E3 ubiquitin ligases , are activated by ligation of the ubiquitin-like protein NEDD8 to a conserved cullin lysine ( Lamsoul et al . , 2016 ) . The NEDD8-conjugating enzyme UBE2F , via specific interaction with E3 ligase RNF7 ( also known as RBX2 ) , specifically neddylates and consequently activates cullin-5 ( CUL5 ) , but not other cullin proteins ( Huang et al . , 2009 ) . ARIH2 ( also known as TRIAD1 ) , a member of the RING-in-between-RING ( RBR ) E3 ubiquitin ligase family ( Dove and Klevit , 2017 ) , binds to and is activated specifically by activated CRL5 complex ( Kelsall et al . , 2013 ) . ARIH2 is essential for embryogenesis and has been shown to modulate inflammatory responses ( Lin et al . , 2013; Kawashima et al . , 2017 ) . The association of CRL5 complex with cancer biology has started to emerge . For instance , CUL5 deficiency has been shown to promote SCLC metastasis by stabilizing integrin β1 ( Zhao et al . , 2019 ) . Additionally , CRL5 complex also plays important roles in modulating multiple aspects of the cellular response to heat shock protein 90 ( HSP90 ) inhibition ( Samant et al . , 2014 ) . Here , for the first time , we emphasized the importance of ARIH2-CRL5 complex in mediating EGFR TKI resistance in NSCLC . Through a proteomics study , we identified METAP2 ( methionine aminopeptidase 2 ) , a protein that has been shown to be involved in NSCLC ( Shimizu et al . , 2016 ) , the abundance of which was dramatically increased upon ARIH2 loss . We further demonstrated that increased METAP2 level was , at least in part , responsible for the drug resistance phenotype resulting from ARIH2 loss . However , identification of direct substrates of ARIH2-CRL5 complex as well as elucidating their involvement in EGFR TKI resistance remains to be studied more systematically in the future . In summary , while clinical studies are necessary to confirm these newly revealed dependencies of mutant EGFR in NSCLC , our genome-wide CRISPR-Cas9 genetic screen together with validation and mechanistic studies expanded the understanding of the heterogeneity of EGFR TKI responses in lung cancer .
Human cancer cell lines were originated from the CCLE ( Barretina et al . , 2012 ) , banked at Novartis Cell Bank , authenticated by single-nucleotide polymorphism analysis and routinely tested as mycoplasma-free . All cell lines were maintained at 37°C with 5% CO2 . HEK293T , HEK293A and A549 cells were cultured in DMEM medium ( Gibco #11995–040 ) supplemented with 10% fetal bovine serum ( FBS ) ( Seradigm #1500–500 ) and penicillin ( 100 units/ml ) -streptomycin ( 100 µg/ml ) ( Gibco #15140–122 ) . HCC827 , HCC4006 , NCI-H3255 , PC9 , NCI-H1975 , NCI-H1299 and NCI-H460 cells were cultured in RPMI medium 1640 ( Gibco #22400–071 ) supplemented with 10% FBS and penicillin-streptomycin . BEAS-2B cells were cultured in BEGM bronchial epithelial cell growth medium bulletkit ( Lonza #CC-3170 ) . Erlotinib ( S1023 ) , Gefitinib ( S1025 ) and Bortezomib ( PS-341 ) were obtained from Selleckchem . NVP-EGF816 and LPAR antagonists ( Compound two and AM095 ) were synthesized at Novartis Institutes for Biomedical Research . Ki16425 ( Cat #5056 ) was purchased from Tocris . Y-27632 ( Cat# Y0503 ) was purchased from Sigma-Aldrich . Drugs for in vitro studies were dissolved in DMSO ( ATCC 4-X ) to yield 10 mM stock solutions and stored at −20°C . For CRISPR-Cas9 mediated gene ablation , the pNGx-LV-c004-3xFlag-Cas9 construct and pNGx-LV-g003 lentiviral backbone for sgRNA cloning were described previously ( DeJesus et al . , 2016 ) . For each sgRNA expression clone , spacer-encoding sense and anti-sense oligonucleotides with appropriate overhangs were synthesized ( IDT ) , annealed , cloned into the BbsI restriction site of the pNGx-LV-g003 backbone . Insertion was verified by DNA sequencing . For cDNA overexpression , pLVX-EF1α-IRES-Puro vector was purchased from TaKaRa ( cat# 631988 ) . Amino-terminally HA-tagged cDNAs were amplified by PCR using Q5 High-Fidelity 2X Master Mix ( NEB Inc , #M0492S ) with the following thermocycling conditions: 30 s at 98°C , 30 cycles of 10 s at 98°C , 20 s at 55°C and then 90 s at 72°C , followed by 2 min at 72°C and holding at 4°C . PCR products were purified with QIAquick PCR Purification Kit ( Qiagen #28104 ) following manufacturer’s instructions . Purified PCR products and empty pLVX-EF1α-IRES-Puro vector were digested by EcoRI-HF ( NEB Inc , #R3101S ) and XbaI ( NEB Inc , #R0145S ) in 1X CutSmart Buffer ( NEB Inc , #B7204S ) at 37°C for overnight . Digested fragments were run in agarose gel and purified with QIAquick Gel Extraction Kit ( Qiagen #28704 ) following manufacturer’s instructions . Ligation was performed ( 3:1 , insert: vector molar ratio ) with T4 DNA Ligase ( NEB Inc , #M0202 ) in 1X T4 DNA Ligase Reaction Buffer ( NEB Inc , #B0202S ) at room temperature for 4 hr . For bacterial transformation , One Shot Stbl3 Chemically Competent E . coli ( Thermo Fisher Scientific , #C737303 ) was used according to the manufacturer’s instructions . Plasmid isolation was performed using QIAprep Spin Miniprep Kit ( Qiagen #27104 ) following manufacturer’s instructions . Insertion was verified by DNA sequencing . AAVS_g1: 5’−3’ GTTAATGTGGCTCTGGTTCT; GNB2_g1: 5’−3’ TCTTTGCCAGGTGCCCACGG; LPAR2_g1: 5’−3’ GCCCGCGAAGAGGTCAGCCG; PKN2_g1: 5’−3’ TCTGCAAATAAAGTACCCTG; RIC8A_g1: 5’−3’ GGAGTGCCGTTAGCAGGAAG; RIC8A_g2: 5’−3’ GGAGCCGCAAGTCAAAGAAC; USP22_g1: 5’−3’ GCATATTCACGAGCATGCGA; YAP1_g1: 5’−3’ ACATCGATCAGACAACAACA; ARIH2_g1: 5’−3’ ATATCTCTGAAACTTGCCAG; ARIH2_g2: 5’−3’ AGTGCTGCTCCCAGCAGCTG; CUL5_g1: 5’−3’ AGCTTGTTTACATAATCCGC; RNF7_g1: 5’−3’ CCTCAAGAAGTGGAACGCGG; KCTD5_g1: 5’−3’ AAGTGGGTCCGACTCAACGT; PDCD10_g1: 5’−3’ CAACTCACCTCATTAAACAC; BTAF1_g1: 5’−3’ GTGAAGTGGATCCTAAAGAG; For single guide RNA ( sgRNA ) lentivirus production , 1 µg of sgRNA construct was co-transfected into HEK293T cells with approximately 80% confluence in a well of the 6-well tissue culture plate ( Corning #353046 ) along with 1 µg packaging ( Δ8 . 9 ) and 0 . 25 µg envelope ( VSV-G ) expression plasmids using 6 . 75 µl FuGENE 6 Transfection Reagent ( Promega #E2692 ) according to the manufacturer’s instructions . Cell culture medium was replaced at 16 hr after transfection , and lentivirus-containing supernatant was harvested at 48 hr and 72 hr post-transfection . Viral supernatant was filtered through a 0 . 45 µm cellulose acetate filter ( Thermo Fisher Scientific #723–9945 ) , aliquoted and stored at −80°C for later use . For cDNA ectopic expression lentivirus production , 6 µg of Ready-to-Use Lentiviral Packaging Plasmid Mix ( Cellecta #CPCP-K2A ) and 4 . 8 µg of cDNA expression construct were co-transfected into HEK293T cells with approximately 80% confluence in a 100 mm cell culture dish ( Corning #430167 ) using 32 . 4 µl FuGENE 6 Transfection Reagent ( Promega #E2692 ) according to the manufacturer’s instructions . Cell culture medium was replaced at 16 hr after transfection , and lentivirus-containing supernatant was harvested at 48 hr and 72 hr post-transfection . Viral supernatant was filtered through a 0 . 45 µm cellulose acetate filter , mixed with 1/3 vol of Lenti-X Concentrator ( Clontech #631232 ) and incubated at 4°C for overnight . Viral particles were pelleted by centrifugation at 1 , 500 g for 45 min at 4°C . The pellet was then gently resuspended with 1 ml of complete DMEM cell culture medium , appropriately aliquoted and stored at −80°C for later use . Human cancer cell lines with constitutive Cas9 expression were generated by lentiviral infection and antibiotic selection . Cas9 expression was confirmed by immunoblotting and gene editing efficiency was tested as follows . Cas9-expressing cells were infected at a low ( ~0 . 5 ) multiplicity of infection ( MOI ) with lentivirus expressing either a control AAVS sgRNA or sgRNA targeting essential genes PSMD1 and PSMA3 and then selected with puromycin . Cells were subsequently seeded in 6-well tissue culture plates , cell culture medium was exchanged 3 days later and the experiment was terminated at day 7 . Cells were trypsinized , resuspended in cell culture medium and the live cell count was determined by trypan blue exclusion on a ViCELL instrument ( Beckman Coulter ) . To generate YAP reporter cell line , HEK293A cells were infected by lentiviruses expressing Cas9 and GTIIC-GFP reporter , and clonal cells were selected and experimentally validated . To knockout specific genes , Cas9-expressing cells in 6-well tissue culture plate were infected by lentivirus expressing sgRNA targeting gene of interest in the presence of complete cell culture medium supplemented with 8 µg/ml polybrene ( AmericanBio #AB01643-00001 ) . Following infection for 18 hr , cell culture medium was replaced by complete cell culture medium . Following another 24 hr , cell culture medium was replaced by complete cell culture medium containing 1 µg/ml puromycin ( Mediatech #MT-61–385-RA ) and mutant cell pools stably expressing the sgRNA were selected . Stable cell lines with ectopic cDNA overexpression were generated in the same manner . Gene silencing efficiency or cDNA overexpression was determined by immunoblotting assay . The genome-wide sgRNA library targeting 18 , 360 protein-coding genes ( ~5 sgRNAs per gene ) was adapted from published sequences ( Sanjana et al . , 2014 ) . For genes lacking published sgRNA sequence information , new sgRNAs were designed for these targets that contained an NGG PAM motif , filtering for GC content >40% and<80% , eliminating homopolymer stretches > 4 , and removing any guides with off-target locations having <4 mismatches across the genome . The sgRNA library was constructed using chip-based oligonucleotide synthesis to generate spacer-tracrRNA-encoding fragments that were PCR-amplified and cloned as a pool into the BpiI site of the pRSI16 lentiviral plasmid ( Cellecta ) . Sequencing of the plasmid pool showed robust normalization , with >90% clones present at a representation of ±5 fold from the median counts in the pool . Library packaging was performed as described previously ( Zeng et al . , 2018 ) . The sgRNA libraries were packaged into lentiviral particles using HEK293T cells . Packaging was scaled up by growing cells in cell stacks ( Corning ) . For each cell stack , 210 million cells were transfected 24 hr after plating using 510 . 3 µl TransIT Transfection Reagent ( Mirus Bio , #MIR2700 ) diluted in 18 . 4 ml Opti-MEM ( Gibco #11058021 ) that was combined with 75 . 6 µg of the sgRNA library and 94 . 5 µg of Ready-to-Use Lentiviral Packaging Plasmid Mix ( Cellecta #CPCP-K2A ) . The next day , the transfected cells received fresh medium . 72 hr post-transfection , lentivirus was collected , filtered , aliquoted , and frozen at −80°C . Viral titer was determined using the Lenti-X qRT-PCR Titration Kit ( Clontech #631235 ) and was typically in the range of 5 × 106 transforming units/ml . For genome-wide screens , sgRNA libraries were transduced at a multiplicity of infection ( MOI ) of 0 . 5 , aiming for coverage of , on average , 1 , 000 cells per sgRNA reagent . MOI was determined by using a 12-point dose-response ranging from 0 to 400 µl of viral supernatants in the presence of 5 µg/ml polybrene and measuring infection rate by FACS as a percentage of red fluorescent protein ( RFP ) -positive cells . Selection was optimized by determining the puromycin dose required to achieve >95% cell killing in 72 hr . Cell viability was measured for a 6-point dose-response ranging from 0 to 20 µg/ml puromycin using the CellTiter-Glo assay ( Promega ) . HCC827-Cas9 cells were seeded into cell stacks ( Corning ) . 24 hr after plating , the culture medium was replaced with fresh medium containing 5 µg/ml polybrene and lentiviral sgRNA library at an MOI of 0 . 5 . 24 hr after infection , the culture medium was replaced with fresh medium containing 2 µg/ml puromycin . 72 hr after puromycin selection , cells were trypsinized and an aliquot of cells was analyzed by FACS to confirm infection and selection efficiency , and the percentage of RFP-positive cells was >90% . 100 million cells , representing the baseline of sgRNA representation , were harvested and snap-frozen using liquid nitrogen . The remaining cells were plated into new cell stacks at 110 million cells per cell stack . The next day , cells were treated with DMSO or 1 µM erlotinib , respectively . Cell culture medium containing DMSO or 1 µM erlotinib was replenished every 3 days . Cells were maintained in culture and split as needed to ensure confluence did not exceed 90% . After 3 weeks of treatment , 100 million cells from each condition were harvested . Genomic DNA was isolated using the QIAamp DNA Blood Maxi Kit ( Qiagen #51194 ) as directed by manufacturer and quantified using PicoGreen ( Invitrogen ) . Illumina sequencing libraries were generated using PCR amplification with primers specific to the genome integrated lentiviral vector backbone sequence . The resulting Illumina libraries were purified using 1 . 8x SPRI AMPure XL beads ( Beckman Coulter ) following the manufacturer’s instructions and qPCR quantified using primers specific to the Illumina sequences using standard methods . Illumina sequencing libraries were pooled and sequenced with a HiSeq 2500 instrument ( Illumina ) . The number of reads was adjusted to cover each sgRNA with approximately 1000 reads . Raw sequencing reads were aligned to the appropriate library using Bowtie ( Langmead et al . , 2009 ) , allowing for no mismatches , and counts were generated . The R software package DESeq2 ( Love et al . , 2014 ) was used to evaluate differential sgRNA representation in the form of log2 fold change between the erlotinib-treated and DMSO-treated samples for EGFR-TKI screen . A robust z-score was calculated using the median and mean-absolute deviation for the calculated fold changes across the entire sgRNA library . For gene-based hit calling , the sgRNAs were ranked by the robust z-score , and the statistical significances for each gene enriched toward higher rank ( RSA up ) and the lower rank ( RSA down ) were evaluated using the Redundant siRNA Activity ( RSA ) algorithm ( König et al . , 2007 ) . The RSA score is a statistical score ( log10 ( P value ) ) representing the probability of a gene hit based on the collective activities of multiple sgRNAs per gene . It is a measure of how significantly the rank order of sgRNAs against a given gene differs from the population of other sgRNAs in the library . Selected hits were searched against STRING database version 11 . 0 for mapping protein interaction network and Reactome database version 67 for pathway analysis . Cells were seeded in 384-well microplates ( Corning #3570 ) at a density of 1 , 000 cells in 30 µl of complete cell culture medium per well and allowed to adhere overnight . Cells were treated in quadruplicate with 6 µl of serial three-fold dilutions of compound in complete cell culture medium ( final DMSO concentration = 0 . 1% ) . Following drug exposure for 72 hr , 25 µl of CellTiter-Glo reagent ( Promega #G7572 ) per well was added and plates were incubated at room temperature for 20 min . Luminescence was read in an EnVision Multimode Plate Reader ( PerkinElmer ) . Assay data were normalized to DMSO values and plotted using a four-parameter concentration-response model in GraphPad Prism 7 . The figures show the mean ± standard deviation of quadruplicate values from representative experiments . Cells were seeded in 384-well microplates ( Corning #3570 ) at a density of 2 , 000 cells per well and allowed to adhere overnight . Cells were treated in quadruplicate with serial three-fold dilutions of compound in complete cell culture medium ( final DMSO concentration = 0 . 1% ) . Following drug exposure for 24 hr , caspase 3/7 activity was measured using the Caspase-Glo 3/7 Assay System ( Promega #G8092 ) according to the supplier’s instructions . Luminescence was read in an EnVision Multimode Plate Reader ( PerkinElmer ) . Cells were seeded in 6-well tissue culture plates at a density of 1 × 105 cells per well and allowed to adhere overnight . Cells were then treated in triplicate with DMSO control or 1 µM erlotinib . Photomicrographs ( 36 images per well ) were taken every 6 hr using an IncuCyte live cell imager ( Essence BioSciences ) and confluence of the cultures was measured using IncuCyte software ( Essence BioSciences ) . DTP cells were generated according to protocols described previously ( Sharma et al . , 2010 ) . In brief , cells were treated with 1 µM of erlotinib or gefitinib for three rounds , with each treatment lasting 72 hr . Viable cells remaining attached on the dish at the end of the 9 d drug treatment were considered to be DTPs and were collected for analysis . The live cell count was determined by trypan blue exclusion on a ViCELL instrument ( Beckman Coulter ) . Percentage of DTPs was calculated by comparing the number of DTPs to the number of cells at the end of the 9 d DMSO treatment . Cells were seeded into 6-well tissue culture plates ( 5 × 104 to 1 × 105 cells per well , depending on the growth rate ) and allowed to adhere overnight in complete cell culture medium . The next day , medium was replaced by complete cell culture medium containing appropriate drugs or DMSO as vehicle control . Cells were exposed to vehicle for 7–10 days or indicated drug for 3–5 weeks , with medium change and fresh drug added twice a week . At the end of treatment , remaining cells were gently washed with PBS , fixed/stained with 0 . 2% crystal violet ( Fisher Scientific #C581-25 ) in 4% paraformaldehyde and incubated at room temperature for 30 min . Cells were washed three times with water to remove excessive dye and allowed to air dry . Pictures of stained cells were taken using an EPSON Perfection V600 scanner . Colony formation was quantified using the ColonyArea ImageJ plugin which provides information about the intensity percentage taking into consideration not only the area covered by the colonies , but also the intensity of staining as a direct relation to the number of cells in a colony ( Guzmán et al . , 2014 ) . For competitive proliferation assay , HCC827 parental cells ( RFP negative ) were mixed with HCC827-sgAAVS/RFP or HCC827-sgARIH2/RFP cells ( 100:1 ) and cultured in the presence of DMSO control or EGFR-TKI ( erlotinib or gefitinib; 1 µM ) for 3 weeks , with medium change and fresh drug added twice a week . Cells were washed with PBS , trypsinized , and the relative percentage of RFP+ cells was determined by flow cytometry analysis using a CytoFLEX S flow cytometer ( Beckman Coulter ) . Data were visualized with FlowJo software ( FlowJo ) . HEK293A-GTIIC-GFP-Cas9 cells or corresponding gene edited cell derivatives were seeded in 6-well tissue culture plates at the same density and allowed to adhere overnight . The following day , cells were trypsinized , collected in cell culture medium and subjected to flow cytometry analysis using a CytoFLEX S flow cytometer ( Beckman Coulter ) . Data were analyzed by FlowJo software ( FlowJo ) . Cells were seeded in 6-well tissue culture plates at the same density and allowed to grow until about 50% of confluence . Cells were then lysed for RHOA G-LISA activation assay according to the manufacturer’s instructions ( Cytoskeleton Cat# BK124 ) . Immunoblotting analysis was performed to assess the total RHOA protein level in the whole cell lysate . Cells were lysed in RIPA buffer ( 25 mM Tris-HCl pH 7 . 6 , 150 mM NaCl , 1% NP-40 , 1% sodium deoxycholate , 0 . 1% SDS ) ( Thermo Fisher Scientific #89901 ) supplemented with 100 x Protease Inhibitor Cocktail ( Sigma #P8340 ) , 100 x Phosphatase Inhibitor Cocktail ( Thermo Fisher Scientific #78427 ) and 25 units/ml Benzonase Nuclease ( Sigma #E8263 ) . Lysate was sonicated using a Diagenode Bioruptor 300 ( High setting , 30 s on , 30 s off , 10 cycles ) , followed by centrifugation at 13 , 000 rpm , 4°C , 10 min . Protein concentration was determined using the DC Protein Assay Kit ( Bio-Rad #5000112 ) according to the manufacturer’s instructions . Equal amount of proteins were resolved by SDS-PAGE and transferred to nitrocellulose membranes ( Bio-Rad #1704159EDU ) using a Trans-Blot Turbo Transfer System ( Bio-Rad #1704150EDU ) according to the manufacturer’s instructions . Membranes were blocked for 1 hr at room temperature with 5% Blotting-Grade Blocker ( Bio-Rad #1706404 ) in Tris Buffered Saline with Tween 20 , pH 8 . 0 ( TBST ) ( Sigma #T9039 ) and then incubated overnight at 4°C with primary antibodies diluted in 5% BSA ( Akron Biotechnology #AK8917-0100 ) . Membranes were washed with TBST , followed by incubation with horseradish peroxidase ( HRP ) conjugated secondary antibody diluted in 5% Blotting-Grade Blocker and visualization with Amersham ECL Western Blotting Detection Reagents ( GE Healthcare #RPN2106 ) or SuperSignal West Pico PLUS Chemiluminescent Substrate ( Thermo Fisher Scientific #34580 ) and Amersham Hyperfilm ECL ( GE Healthcare #28906839 ) . The following antibodies were used in this study ( with dilution factor for immunoblotting ) : anti-PKN2 ( #2612 , 1:1 , 000 ) , anti-YAP ( #14074 , 1:1 , 000 ) , anti-YAP/TAZ ( #8418 , 1:1 , 000 ) , anti-phospho-YAP ( Ser127 ) ( #4911 , 1:1 , 000 ) , anti-phospho-EGFR ( Tyr1068 ) ( #3777 , 1:1 , 000 ) , anti-EGFR ( #4267 , 1:20 , 000 ) , anti-phospho-AKT ( Ser473 ) ( #4058 , 1:1 , 000 ) , anti-AKT ( #9272 , 1:5 , 000 ) , anti-phospho-ERK1/2 ( Thr202/Tyr204 ) ( #9101 , 1:1 , 000 ) , anti-ERK1/2 ( #9102 , 1:5 , 000 ) , anti-ARIH2/TRIAD1 ( #13689 , 1:1 , 000 ) , anti-BTAF1 ( #2637 , 1:1 , 000 ) , anti-GNAQ ( #14373 , 1:1 , 000 ) , anti-HSP90 ( #4877 , 1:5 , 000 ) , anti-ALDOA ( #8060 , 1:5 , 000 ) , anti-METAP2 ( #12547 , 1:1 , 000 ) , anti-GAPDH ( #2118 , 1:5 , 000 ) , anti-RHOA ( #2117 , 1:1 , 000 ) , anti-Cofilin ( #5175 , 1:1 , 000 ) , anti-phospho-Cofilin ( Ser3 ) ( #3313 , 1:1 , 000 ) , from Cell Signaling Technology; anti-α-Tubulin ( T6074 , 1:20 , 000 ) and anti-β-Actin ( A1978 , 1:20 , 000 ) from Sigma; anti-GNB2 ( ab81272 , 1:1 , 000 ) , anti-RIC8A ( ab97808 , 1:1 , 000 ) , anti-USP22 ( ab195289 , 1:1 , 000 ) , anti-CUL5 ( ab184177 , 1:1 , 000 ) , anti-RNF7 ( ab181986 , 1:1 , 000 ) , anti-PDCD10 ( ab180706 , 1:1 , 000 ) , from Abcam; anti-KCTD5 ( #15553–1-AP , 1:1 , 000 ) , anti-PSAT1 ( #10501–1-AP , 1:5 , 000 ) , from Proteintech; Goat Anti-Rabbit IgG Antibody , ( H+L ) HRP conjugate ( #AP307P , 1:5 , 000 ) , Goat Anti-Mouse IgG Antibody , ( H+L ) HRP conjugate ( #AP308P , 1:5 , 000 ) , from Millipore Sigma . Isogenic cell lines used to assess nascent protein synthetic rate were grown in 60 mm dishes until ~70% confluent . Cells were treated with DMSO or 1 µM Erlotinib for 24 hr . Prior to labeling , cells were washed with PBS and then incubated at 37°C with methionine-free media ( Life Technologies , A1451701 ) containing 2% FBS supplemented with DMSO or 1 µM Erlotinib for 1 hr . Medium was then replaced by methionine-free media containing 2% FBS supplemented with 50 µM Click-iT AHA ( L-azidohomoalanine ) ( Life Technologies , C10102 ) and the labeling was performed by incubation at 37°C for 1 hr . Cells were washed three times with PBS and immediately lysed with 200 µl lysis buffer ( 50 mM Tris-HCl , pH 8 . 0 , 1% SDS , supplemented with protease and phosphatase inhibitors and Benzonase at appropriate concentrations ) . Cells were scraped off the dish and transferred into 1 . 5 ml microcentrifuge tubes . Complete cell lysis was achieved by sonication ( Diagenode Biorupter 300: high setting , 30 s on , 30 s off , 10 cycles ) , followed by centrifugation at 13 , 000 rpm , 4°C , 5 min . Protein concentration was determined using the DC Protein Assay Kit ( Bio-Rad #5000112 ) according to the manufacturer’s instructions . 80–100 µg of protein lysates were then subjected to Click-iT reaction for switching azido-modified nascent proteins to alkyne-biotin ( Life Technologies , B10185 ) using the Click-iT Protein Reaction Buffer Kit ( Life Technologies , C10276 ) , followed by protein precipitation according to the manufacturer’s protocol . Air dried protein samples were re-solubilized in 60 µl of 1% SDS in water with vortex followed by heating the samples at 99°C for 10 min . Solubilized protein samples were cleared by centrifugation at 13 , 000 rpm for 1 min to remove any insoluble material . Protein concentration was determined using the DC Protein Assay Kit , and 15 µg of proteins were resolved by Tris-Glycine SDS-PAGE gel . Biotinylated nascent proteins were subjected to immunoblotting using Streptavidin-HRP ( Cell Signaling Technology #3999 , 1:5 , 000 ) . To assess METAP2 protein synthesis , cells were treated as above with the exceptions that AHA labeling is for 3 hr and 150 µg of protein lysates were used for Click-iT reaction . After Click-iT reaction and protein precipitation , air dried protein samples were re-solubilized in 50 µl of 1% SDS in Triton-X lysis buffer ( 50 mM Tris-HCl , pH 7 . 5 , 150 mM NaCl , 10% glycerol , 1% Triton X-100 ) with vortex followed by heating the samples at 99°C for 10 min . Solubilized protein samples were cleared by centrifugation at 13 , 000 rpm for 1 min to remove any insoluble material . Protein concentration was determined using the DC Protein Assay Kit . Samples were diluted with Triton-X lysis buffer to reduce the amount of SDS and same amount of proteins were then used for streptavidin magnetic beads ( Thermo Fisher Scientific Cat# 88817 ) pulldown with incubation at 4°C for overnight . The following day , beads were collected with a magnetic stand and washed three times with Triton-X lysis buffer . After washing , beads were collected and re-suspended with 25 µl of SDS-PAGE reducing sample buffer followed by heating at 99°C for 10 min . The beads were then magnetically separated and the supernatant was saved for western blotting analysis . In vivo poly-ubiquitination of METAP2 was evaluated by the TUBE assay . Briefly , cells were treated with 0 . 1 µM of bortezomib for overnight to enrich poly-ubiquitination of target protein . Cell lysates were prepared and equal amount of cell lysates were incubated with agarose-TUBE beads ( LifeSensors Cat# UM402 ) per the manufacturer’s instructions . Then the beads were washed three times with lysis buffer , and bound proteins were eluted in SDS-PAGE reducing sample buffer for immunoblotting analysis . Cell pellet was lysed with urea buffer: 8 M urea ( Sigma , #U1250 ) , 1% SDS ( Promega , #V6551 ) , 50 mM Tris ( Sigma , #10708976001 ) pH 8 . 5 , phosphatase inhibitor cocktail tablet PhosSTOP ( Roche , #4906837001 ) . Proteins were reduced with 5 mM dithiothreitol ( DTT , Sigma , #D9779 ) 1 hr at room temperature and alkylated with 15 mM iodoacetamide ( IAA , Sigma , #I6125 ) for 1 hr at room temperature in the dark . Proteins were then precipitated with chloroform/methanol to remove salt and detergent . After dissolving the dry protein pellet with 8 M urea , 50 mM Tris pH 8 . 5; the proteins were digested overnight with trypsin ( Promega , #V5072 ) after dilution to 2 M urea . The peptides were acidified to 1% TFA , desalted on SepPak C18 cartridges and eluted with 60% acetonitrile , 0 . 1% TFA . Dried peptides were resuspended in 0 . 1 M TEAB buffer , pH 8 . 5 and then labeled with TMT reagent ( 1:4; peptide:TMT label ) ( Thermo Fisher Scientific ) . The reaction was quenched with 0 . 5% TFA and the six samples were combined to a 1:1 ratio . Mixed and labeled peptides were subjected to high-pH reversed-phase HPLC fractionation on an Agilent X-bridge C18 column ( 3 . 5 µm particles , 2 . 1 mm i . d . , and 15 cm in length ) . Using an Agilent 1200 LC system , a 60 min linear gradient from 10% to 40% acetonitrile in 10 mM ammonium formate separated the peptide mixture into a total of 96 fractions , which were then consolidated into 24 fractions . The dried 24 fractions were reconstituted in 0 . 1% formic acid for LC-MS3 analysis . Labeled peptides were loaded onto a 15 cm column packed in-house with ReproSil-Pur 120 C18-AQ 1 . 9 µM ( 75 µm inner diameter ) in an EASY-nLC 1200 system . The peptides were separated using a 120 min gradient from 3% to 30% buffer B ( 80% acetonitrile in 0 . 1% formic acid ) equilibrated with buffer A ( 0 . 1% formic acid ) at a flow rate of 250 nl/min . Eluted TMT peptides were analyzed on an Orbitrap Fusion Lumos mass spectrometer ( Thermo Fisher Scientific ) . MS1 scans were acquired at resolution 120 , 000 with 350–1500 m/z scan range , AGC target 2 × 105 , maximum injection time 50 ms . Then , MS2 precursors were isolated using the quadrupole ( 0 . 7 m/z window ) with AGC 1 × 104 and maximum injection time 50 ms . Precursors were fragmented by CID at a normalized collision energy ( NCE ) of 35% and analyzed in the ion trap . Following MS2 , synchronous precursor selection ( SPS ) MS3 scans were collected by using high energy collision-induced dissociation ( HCD ) and fragments were analyzed using the Orbitrap ( NCE 65% , AGC target 1 × 105 , maximum injection time 120 ms , resolution 60 , 000 ) . Protein identification and quantification were performed using Proteome Discoverer 2 . 1 . 0 . 81 with the SEQUEST algorithm and Uniprot human database ( 2014-01-31 , 21568 protein sequences ) . Mass tolerance was set at 10 ppm for precursors and at 0 . 6 Da for fragment . Maximum of 3 missed cleavages were allowed . Methionine oxidation was set as dynamic modification; while TMT tags on peptide N termini/lysine residues and cysteine alkylation ( +57 . 02146 ) were set as static modifications . The list of identified peptide spectrum matches ( PSMs ) was filtered to respect a 1% False Discovery Rate ( FDR ) after excluding PSMs with an average TMT reporter ion signal-to-noise value lower than 10 and a precursor interference level value higher 50% . The Student’s test was applied to identify significantly changed protein abundances and adjusted p-values were calculated according to Benjamin and Hochberg . The final list of identified proteins was filtered to achieve a 5% protein FDR . Total RNA was extracted using the RNeasy Plus Mini Kit ( Qiagen #74134 ) and reverse transcribed with TaqMan Reverse Transcription Reagents ( Applied Biosystems #N8080234 ) according to the manufacturer’s instructions . The resulting cDNA products were diluted and subjected to quantitative real-time PCR ( qPCR ) reactions using TaqMan Gene Expression Assays ( Applied Biosystems ) . Specifically , qPCR was performed in 10 µl reactions consisting of 0 . 5 µl TaqMan probe ( Applied Biosystems ) , 5 µl TaqMan Fast Advanced Master Mix ( Applied Biosystems #4444557 ) and 4 . 5 µl diluted cDNA template . Experiments were run on a ViiA 7 Real-Time PCR System ( Applied Biosystems ) . The thermocycling conditions used were 20 s at 95°C , followed by 40 cycles of 1 s at 95°C and 20 s at 60°C . The threshold crossing value ( Ct ) was determined for each transcript and normalized to the housekeeping gene transcript ( GUSB ) . The relative quantification of each mRNA species was assessed using the comparative ΔΔCt method . TaqMan probes used in this study are as follows: GUSB ( Hs00939627_m1 ) , ANKRD1 ( Hs00173317_m1 ) , CTGF ( Hs00170014_m1 ) , CYR61 ( Hs00155479_m1 ) , METAP2 ( Hs00199152_m1 ) , ALDOA ( Hs00605108_g1 ) , PSAT1 ( Hs00795278_mH ) . All animal work was performed in accordance with Novartis Animal Care and Use Committee ( ACUC ) regulations and guidelines . All animals were allowed to acclimate in the Novartis animal facility with access to food and water ad libitum for 3 days prior to manipulation . All cell lines were confirmed as mycoplasma- and rodent pathogens-negative ( IMPACT VIII PCR Profile , IDEXX ) before implantation . Female athymic nude mice ( nu/nu , Charles River Laboratories ) , 6–8 weeks old , were inoculated subcutaneously with 20 million cells suspended in 50% Hank’s balanced salt solution + 50% phenol red-free Matrigel ( BD Biosciences ) . Mice were enrolled in the study once tumors had reached approximately 200 mm3 in size ( day 13 post-implantation ) , and were randomly assigned to receive either vehicle or erlotinib ( LC Laboratories; 10 mg/kg ) ( compound formulation 50% Dexolve-7 ( Generic SBECD ) and 50% of 0 . 1 M Tartaric Acid ) once daily by oral gavage for the duration of the study . Animal body weights were recorded and tumors were measured twice weekly by calipering in two dimensions . Tumor volume was calculated using the following formula: tumor volume ( mm3 ) = L x W2/2 , where L is the longest length of the tumor and W is the length of the tumor perpendicular to L . Unless otherwise noted , data are presented as the means ± SD . For all graphs , data are presented relative to their respective controls . Statistical analyses were performed using GraphPad Prism 8 ( Graphpad software Inc ) . Significance was determined by a two-tailed Student’s t test or ordinary two-way ANOVA , denoted within each figure panel and respective figure legends . p<0 . 05 was considered to be statistically significant . *p<0 . 05 , **p<0 . 01 , ***p<0 . 001 , ****p<0 . 0001; ns , not significant . The mass spectrometry proteomics data have been deposited to the ProteomeXchange Consortium via the PRIDE partner repository with the dataset identifier PXD014198 . CRISPR-Cas9 screen data were summarized in Supplementary file 1 and Supplementary file 2 . | Cancer is caused by cells growing and dividing uncontrollably as a result of mutations in certain genes . Many human lung cancers have a mutation in the gene that makes the protein EGFR . In healthy cells , EGFR allows a cell to respond to chemical signals that encourage healthy growth . In cancer , the altered EGFR is always on , which allows the cell to rapidly grow without any control , resulting in cancer . One approach to treating these cancers is with drugs that block the activity of mutant EGFR . Although these drugs have been very successful , they do not always succeed in completely treating the cancer . This is because over time the cancer cells can become resistant to the drug and start forming new tumors . One way that this can happen is if random mutations lead to changes in other proteins that make the drug less effective or stop it from accessing the EGFR proteins . However , it is unclear how other proteins in cancer cells affect the response to these EGFR inhibiting drugs . Now , Zeng et al . have used gene editing to systematically remove every protein from human lung cancer cells grown in the laboratory to see how this affects resistance to EGFR inhibitor treatment . This revealed that a number of different proteins could change how cancer cells responded to the drug . For instance , cells lacking the protein RIC8A were more sensitive to EGFR inhibitors and less likely to develop resistance . This is because loss of RIC8A turns down a key cell survival pathway in cancer cells . Whereas , cancer cells lacking the ARIH2 protein were able to produce more proteins that are needed for cancer cell growth , which resulted in them having increased resistance to EGFR inhibitors . The proteins identified in this study could be used to develop new drugs that improve the effectiveness of EGFR inhibitors . Understanding how cancer cells respond to EGFR inhibitor treatment could help determine how likely a patient is to develop resistance to these drugs . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"cell",
"biology",
"cancer",
"biology"
] | 2019 | Genome-wide CRISPR screening reveals genetic modifiers of mutant EGFR dependence in human NSCLC |
Heterochromatin is enriched for specific epigenetic factors including Heterochromatin Protein 1a ( HP1a ) , and is essential for many organismal functions . To elucidate heterochromatin organization and regulation , we purified Drosophila melanogaster HP1a interactors , and performed a genome-wide RNAi screen to identify genes that impact HP1a levels or localization . The majority of the over four hundred putative HP1a interactors and regulators identified were previously unknown . We found that 13 of 16 tested candidates ( 83% ) are required for gene silencing , providing a substantial increase in the number of identified components that impact heterochromatin properties . Surprisingly , image analysis revealed that although some HP1a interactors and regulators are broadly distributed within the heterochromatin domain , most localize to discrete subdomains that display dynamic localization patterns during the cell cycle . We conclude that heterochromatin composition and architecture is more spatially complex and dynamic than previously suggested , and propose that a network of subdomains regulates diverse heterochromatin functions .
Eukaryotic genomes are composed of cytologically and functionally distinct chromatin domains called heterochromatin and euchromatin ( Heitz , 1928 ) . Although heterochromatin is primarily comprised of simple repetitive DNA sequences ( Peacock et al . , 1978 ) and transposons ( Carlson and Brutlag , 1978 ) , this domain is necessary for organismal functions , including pericentromeric sister chromatid cohesion ( Bernard et al . , 2001 ) , achiasmate chromosome pairing and segregation in male and female meiosis ( Dernburg et al . , 1996; Karpen et al . , 1996; McKee and Karpen , 1990 ) , and genome integrity ( Peng and Karpen , 2009 ) . Heterochromatin is defined molecularly by H3K9me2/3 ( deposited by the Su ( var ) 3–9 histone methyltransferase [Schotta et al . , 2002] ) and its highly conserved and essential ( Eissenberg et al . , 1992; Aucott et al . , 2008 ) binding partner Heterochromatin Protein 1 ( Grewal and Jia , 2007 ) ( HP1 ) . To understand how HP1 is able to regulate diverse cellular and organismal functions ( Grewal and Jia , 2007 ) , researchers have affinity purified HP1 in human tissue culture lines ( Rosnoblet et al . , 2011; Lechner et al . , 2005 ) , S . pombe ( Motamedi et al . , 2008 ) and D . melanogaster ( Ryu et al . , 2014; Alekseyenko et al . , 2014 ) and identified >100 putative HP1 interacting proteins ( HPips ) by mass spectrometry . However , the overlap between identified HPips in these studies is minimal . Possible explanations include overexpression of the bait , or isolation of different populations of HP1 . Indeed , biochemical ( i . e . salt fractionation and size exclusion chromatography ) and cytological ( i . e . fluorescence correlation spectroscopy and fluorescence recovery after photobleaching ) experiments suggest the presence of distinct HP1 complexes ( Rosnoblet et al . , 2011; Müller et al . , 2009; Schmiedeberg et al . , 2004; Huang et al . , 1998; Kellum et al . , 1995 ) . Regardless , studies in Drosophila have shown that the founding HP1 ortholog ( HP1a ) physically interacts with chromatin ( Bannister et al . , 2001; Lachner et al . , 2001; Lu , 2013 ) , replication components ( Pak et al . , 1997; Murzina et al . , 1999; Pindyurin et al . , 2008 ) , chromatin modifying proteins ( Schotta et al . , 2002; Delattre et al . , 2000; Ito et al . , 2012 ) , mRNA processing proteins ( Piacentini et al . , 2009 ) , telomere protection proteins ( Shareef et al . , 2001; Raffa et al . , 2009; Cenci et al . , 2003 ) and components of small RNAi pathways ( Brower-Toland et al . , 2007; Yin and Lin , 2007 ) . Despite extensive information about HPips , most have not been demonstrated to directly regulate heterochromatin organization or functions , and it is unclear how HPips are organized and regulated within the heterochromatin domain . Historically , polytenized salivary gland chromosomes have been used to determine the localization of chromatin-bound proteins in Drosophila . However , the size of the heterochromatin domain in these terminally differentiated cells is reduced due to severe underreplication of heterochromatic repeats ( Rudkin , 1969 ) , which limits the resolution of HPip localization patterns within heterochromatin . Nevertheless , ATF-2 ( Seong et al . , 2011 ) and PIWI ( Brower-Toland et al . , 2007 ) were shown to occupy restricted regions or subdomains within the entire heterochromatin domain ( hereafter ‘holodomain’ ) in polytene nuclei , suggesting that heterochromatin may be compartmentalized into functional units . However , the generality of subdomain organization for heterochromatin proteins is unknown , especially in mitotically dividing diploid cells . One known function of heterochromatin domains is epigenetic transcriptional silencing of repeated DNAs ( Sienski et al . , 2012 ) and developmentally-regulated protein-coding genes ( Clowney et al . , 2012 ) . Position effect variegation ( PEV ) describes the mosaic expression of euchromatic genes relocated or inserted in or near heterochromatin , which results from spreading of repressive heterochromatic components and clonal inheritance of the silenced state ( reviewed in [Wakimoto , 1998] ) . Modification of PEV has been used as a sensitive assay to identify gene products that regulate heterochromatin structure and function ( Lewis , 1950 ) . For example , loss-of-function mutations in HP1a act as dominant suppressors of PEV ( Su ( var ) ) ( Eissenberg et al . , 1990; Sinclair et al . , 1983 ) , resulting in reduced repression , whereas increased HP1a levels result in enhancement of PEV ( E ( var ) , increased repression ) ( Eissenberg et al . , 1992 ) . Forward genetic screens in Drosophila have identified ~500 dominant mutations ( estimated to map to 150 genes ) that can modify PEV , however only ~30 have been mapped to specific genes thus far ( Elgin and Reuter , 2013 ) . To gain further insight into the organization and function of heterochromatin ( Figure 1 ) , we implemented two approaches: 1 ) a biochemical purification of HP1a to identify novel binding partners , and 2 ) an image-based genome-wide RNAi screen to identify new regulators of HP1a levels and organization . Image analysis of a subset of candidates from both screens identified 30 proteins that localize to heterochromatin . The majority of these suppressed PEV when mutated or depleted by RNA interference ( RNAi ) , demonstrating impact on heterochromatin-mediated gene silencing . Most importantly , more detailed imaging studies showed that both novel and previously known heterochromatin proteins are predominantly localized to restricted subdomains within heterochromatin , and display diverse , dynamic localization patterns during the cell cycle . In addition to greatly expanding our understanding of the number and types of heterochromatin proteins and regulators , these findings lead us to propose that heterochromatin is composed of a dynamic network of subdomains that regulates different heterochromatin functions . 10 . 7554/eLife . 16096 . 003Figure 1 . Workflow to identify novel heterochromatin components and regulators . We devised an unbiased strategy to identify novel components of heterochromatin . First , we identified candidates by performing HP1a immunoprecipitation followed by mass spectrometry ( IP-MS ) and a genome-wide RNAi screen . Candidates that localized to heterochromatin were assayed for effects on PEV . Finally , we investigated their spatial and temporal localization with respect to heterochromatin . DOI: http://dx . doi . org/10 . 7554/eLife . 16096 . 003
To gain a better understanding of the composition of heterochromatin we purified HP1a six independent times , from S2 cells stably expressing HP1a tagged with 3X-FLAG and StrepII ( FS-HP1a ) at ~20% of endogenous HP1a levels ( data not shown ) . Purified samples were analyzed by liquid chromatography-tandem mass spectrometry ( LC-MS/MS ) . MS results identified 135 proteins that were significantly enriched in at least two of the six IP-MS experiments ( Table 1 and Table 1—source data 1 and 2 ) ( hereafter HPips ) . To investigate the potential biological functions of these proteins we used the Database for Annotation , Visualization and Integrated Discovery ( DAVID ) v6 . 7 ( Huang et al . , 2008 , 2009 ) toolset to identify enriched gene ontology ( GO ) terms . Consistent with expectations for heterochromatic proteins , these HPips were enriched for GO categories that include 'chromocenter' , 'chromatin organization' , 'chromatin assembly or disassembly' and 'post-transcriptional regulation of gene expression' ( Supplementary file 1 ) . Initial validation of the approach comes from the observation that the 135 candidate HP1a interactors included 17 of the ~33 previously characterized HPips ( ~52% ) , such as HP2 ( Shaffer et al . , 2002 ) , Lhr ( Greil et al . , 2007 ) , HP4 ( Greil et al . , 2007 ) , HP5 ( Greil et al . , 2007 ) , Su ( var ) 3–9 ( Schotta et al . , 2002 ) and Kdm4a ( Lin et al . , 2008 ) ( Table 1—source data 1 and 2 ) . 10 . 7554/eLife . 16096 . 004Table 1 . HP1a interactors ranked by frequency of detection . The most common HP1a interacting proteins are listed according to the frequency in which they were detected in HP1a IP-MS experiments ( out of six ) . References that link a protein to HP1a by IP , yeast-two-hybrid or immunofluorescence are listed in the third column . Asterisk indicates that the protein has been shown to modulate PEV . See Table 1—source data 1 and 2 for a complete list of hits and Table 1—source data 3 for a silver-stained gel of the IP . DOI: http://dx . doi . org/10 . 7554/eLife . 16096 . 00410 . 7554/eLife . 16096 . 005Table 1—source data 1 . 2-Step HP1a IP-MS . HP1a was purified in the absence of ionizing radiation ( IR ) ( A ) , and 10 min ( B ) and 60 ( C ) minutes after 10 Gy exposure . Number of unique peptides per protein are listed . The HPips identified did not change significantly with respect to irradiation , therefore we used all purifications to identify candidate hits . DOI: http://dx . doi . org/10 . 7554/eLife . 16096 . 00510 . 7554/eLife . 16096 . 006Table 1—source data 2 . 1-Step HP1a IP-MS . HP1a was purified in the absence of IR ( A , Mock and FS-HP1a ) , and 10 ( B , FS-HP1a ) and 60 ( C , FS-HP1a ) minutes after 10 Gy exposure . Number of unique peptides per protein is listed . The HPips identified did not change significantly with respect to irradiation , therefore we used all purifications to identify candidate hits . DOI: http://dx . doi . org/10 . 7554/eLife . 16096 . 00610 . 7554/eLife . 16096 . 007Table 1—source data 3 . HP1a interacts with a large set of proteins . Silver-stained gel of a single step purification from S2 cells stably expressing FS-HP1a ( lanes 1–3 ) or WT ( lane 4 ) S2 cells . HP1a was purified in the absence of IR ( lane 1 ) , and 10 ( lane 2 ) and 60 ( lane 3 ) minutes after 10 Gy exposure . The HPips identified did not change significantly with respect to irradiation , therefore all purifications were used to identify candidate hits . DOI: http://dx . doi . org/10 . 7554/eLife . 16096 . 007Flybase Gene Name# of experiments enriched inLiterature Linking the Gene to HP1aADD1*6Alekseyenko et al . , 2014CG81086Alekseyenko et al . , 2014; Guruharsha et al . , 2011HP5*6Greil et al . , 2007; Alekseyenko et al . , 2014Su ( var ) 3-9*6Schotta et al . , 2002; Alekseyenko et al . , 2014Su ( var ) 2-HP2*6Shaffer et al . , 2002; Alekseyenko et al . , 2014tsr6Hsc70-45Alekseyenko et al . , 2014Kdm4A*5Lin et al . , 2008; Alekseyenko et al . , 2014; Colmenares et al . , unpublishedOdj ( CG7357 ) 5van Bemmel et al . , 2013smt34Alekseyenko et al . , 2014Lhr4Greil et al . , 2007; Alekseyenko et al . , 2014Act5C4Hsc70-34betaTub56D4Chd644Hsp834CG76923Alekseyenko et al . , 2014HP4*3Greil et al . , 2007Tudor-SN3His2B:CG338723eIF-4a3FK506-bp13CG71723CG82583EF23eIF-4B3Hsc70-53Hsp603qm3sta3 Most importantly , 118 of the HPips isolated here ( 89% ) were not previously identified as Drosophila HP1a interactors . Five of these new HPips were previously identified as Su ( var ) s , demonstrating their functional importance to heterochromatin ( Nap1 [Stephens et al . , 2006] , Hel25E [Eberl et al . , 1997] , His2Av [Swaminathan et al . , 2005] , Pp1-87B [Reuter et al . , 1990] , and RpLP0 [Frolov and Birchler , 1998]; Table 1 and Table 1—source data 1 and 2 ) . The remaining 113 HPips were not previously shown to impact heterochromatin functions or associate with HP1a , and potentially represent a large collection of novel heterochromatin components . In order to identify factors that regulate heterochromatin independent of HP1a binding , we performed an image-based genome-wide RNAi screen ( Figure 2A ) for gene depletions that altered heterochromatin architecture ( e . g . HP1a levels or localization ) . Nuclei were identified based on DAPI staining , and analyzed for 33 different imaging features ( e . g . nuclear size , nuclear shape , channel-specific intensity/distribution metrics: see Supplementary file 2 ) . To address known issues associated with genome-wide screens ( e . g . biological noise , transfection efficiency , image quality ) we employed positive ( HP1a dsRNA ) and negative ( GFP dsRNA ) controls , performed the screen in duplicate and utilized Rank Product normalization ( Breitling et al . , 2004 ) , which incorporates replicate consistency and provides an estimated p-value for observed differences . We utilized three different candidate identification methods ( rank lists of individual features of interest , supervised and unsupervised clustering , see ‘Materials and methods’ ) to maximize the number of true positive hits . 10 . 7554/eLife . 16096 . 008Figure 2 . A genome-wide image-based RNAi screen identifies HP1a regulators . Drosophila Kc cells transfected with dsRNA were analyzed for HP1a localization by IF , and DNA was counterstained with DAPI . Cells were visualized using high-throughput fluorescent microscopy and imaging features were extracted using custom Matlab scripts . Wells were normalized and checked for replicate consistency using the Rank Product test and a p-value was calculated . Putative candidates involved in HP1a recruitment/maintenance were selected by identifying amplicons that lowered HP1a intensity , or clustered with HP1a depletions after hierarchical clustering or Support Vector Machine ( SVM ) analysis . ( B ) Genes that clustered using unsupervised hierarchical clustering with either HP1a or Su ( var ) 3–9 positive control depletions are represented by the yellow circle . Supervised machine learning models ( SVMs ) were trained to identify genes that disrupt HP1a staining ( blue circle ) using HP1a depletion controls . HP1a intensity measures ( mean , maximum , relative maximum and kurtosis ) were used to identify another set of candidate genes ( red circle ) . Genes identified by multiple methods are indicated by color below the Venn diagram . See Figure 2—source data 1 for a list of all genes identified in the RNAi screen and the method used to identify them . DOI: http://dx . doi . org/10 . 7554/eLife . 16096 . 00810 . 7554/eLife . 16096 . 009Figure 2—source data 1 . 374 genes putatively regulate heterochromatin . Genes identified from the RNAi screen whose depletion results in reduced HP1a levels or phenocopies HP1a or Su ( var ) 3–9 depletions ( HP1a positive regulators , HPprs ) and the method used to identify them are listed . DOI: http://dx . doi . org/10 . 7554/eLife . 16096 . 00910 . 7554/eLife . 16096 . 010Figure 2—figure supplement 1 . The rank product test is more effective than the robust Z-Score at identifying HP1a knockdowns . HP1a mean intensity was normalized using the robust z-score ( A ) or the rank product test ( B ) . The normalized value ( or p-value ) is plotted versus a ranked list of the amplicons , with a value of one indicating the strongest hit . HP1a RNAi ( positive controls ) are noted in red and the percentage for the highest ranked positive control is indicated with an arrow . DOI: http://dx . doi . org/10 . 7554/eLife . 16096 . 010 First , we focused on the identification of genes whose absence results in reduced HP1a fluorescence , or phenocopies HP1a depletion ( hereafter HP1a positive regulators , HPprs ) . Of the 374 genes identified as putative hits ( Figure 2B and Figure 2—source data 1 ) , 22 were previously implicated in regulating HP1a localization or heterochromatin properties ( e . g . PEV ) ( Table 2 ) . Notably , of the 374 HPprs , only three genes besides HP1a ( Tudor-SN , RpL8 and mRpL3 ) were also identified in the HP1a IP-MS . This suggests that the majority of HPprs are not tightly bound to HP1a , as they do not co-isolate , and may indirectly influence HP1a establishment or maintenance . Second , we identified 564 genes ( including 8 that were identified in the HP1a IP-MS ) that negatively regulate HP1a fluorescence intensity ( i . e . HP1a fluorescence is increased in their absence ) ( Supplementary file 3 ) . We speculate that some of these 564 genes may normally be required for removal/turnover of HP1a , but are not investigated further here . 10 . 7554/eLife . 16096 . 011Table 2 . RNAi screen hits with previously known connections to heterochromatin . Identified hits from the RNAi screen with previously known connections to heterochromatin are listed according to the method of identification ( Hierarchical Clustering , HP1a Intensity or Support Vector Machine [SVM] ) . Whether a gene clustered with HP1a or Su ( var ) 3–9 depletion controls after Hierarchical Clustering is indicated in parentheses . DOI: http://dx . doi . org/10 . 7554/eLife . 16096 . 011Flybase Name or SymbolMethod of IdentificationLink to HeterochromatinReferenceSsrpHierarchical Clustering ( HP1a , Su ( var ) 3-9 ) Part of FACT complexOrphanides et al . , 1999MBD-likeHierarchical Clustering ( HP1a ) Repressive , localizes to chromocenter , part of NuRD complexBallestar et al . , 2001; Marhold et al . , 2004stellateHierarchical Clustering ( HP1a ) Subunit of Casein kinase IIBozzetti et al . , 1995kismetHierarchical Clustering ( HP1a ) Su ( var ) , regulates heterochromatic silencingSchneiderman et al . , 2009 , 2010Spt20Hierarchical Clustering ( HP1a ) Part of SAGA complexWeake et al . , 2009Su ( var ) 205Hierarchical Clustering ( HP1a ) , HP1a Intensity , SVMEncodes HP1al ( 3 ) neo38Hierarchical Clustering ( HP1a ) , SVMRegulates heterochromatic silencingSchneiderman et al . , 2010Hdac3Hierarchical Clustering ( Su ( var ) 3-9 ) Ortholog regulates HP1beta levelsBhaskara et al . , 2010Rm62 ( lip , p68 ) Hierarchical Clustering ( Su ( var ) 3-9 ) Su ( var ) , binds and putatively targets Su ( var ) 3-9 , binds blanks , binds AGO2 , regulates heterochromatic silencingCsink et al . , 1994; Boeke et al . 2011; Gerbasi et al . , 2011; Ishizuka et al . , 2002; Schneiderman et al . , 2010jumuHierarchical Clustering ( Su ( var ) 3-9 ) Localizes to chromocenter , modifier of variegationHofmann et al . , 2010 , 2009MTA1-likeHierarchical Clustering ( Su ( var ) 3-9 ) Part of NuRD complexMarhold et al . , 2004AGO2Hierarchical Clustering ( Su ( var ) 3-9 ) Heterochromatin targeting , Su ( var ) Noma et al . , 2004; Deshpande et al . , 2005moiHierarchical Clustering ( Su ( var ) 3-9 ) Protects telomeresRaffa et al . , 2009AdarHP1a IntensityE ( var ) on the 4th chromosome , edits RNA , silences Hoppel's transposaseSavva et al . , 2013ParpHP1a IntensityE ( var ) , promotes chromatin condensation and represses retrotransposonsTulin and Spradling , 2003Ino80HP1a IntensityOrtholog in mice complexed with YY1 which regulates HP1gamma , regulates heterochromatic silencingWu et al . , 2009; Schneiderman et al . , 2010roX1HP1a IntensitySu ( var ) Deng et al . , 2009moduloSVMLocalizes to chromocenter , Su ( var ) Perrin et al . , 1998; Garzino et al . , 1992blanksSVMRegulates heterochromatic silencingSchneiderman et al . , 2010crolSVMRegulates heterochromatic silencingSchneiderman et al . , 2010SamuelSVMRegulates heterochromatic silencingSchneiderman et al . , 2010WaplSVMSu ( var ) Verni et al . , 2000 Consistent with our expectations , GO terms analysis of all HPprs identified enrichment for genes associated with the chromocenter , chromatin , RNA interference , RNA-binding and sequence-specific DNA binding ( Supplementary file 4 ) . The identification of genes associated with GTP binding , the proteasome , response to heat and glutathione metabolism is unexpected and may represent noise . However , the correct identification of 22 ( Table 2 ) known regulators and the high accuracy in identifying positive controls ( Figure 2—figure supplement 1 ) suggests these categories may represent novel modes of regulating HP1a protein levels and/or distribution in the nucleus . We validated the heterochromatin association of HPips and HPprs by determining if the proteins colocalize with HP1a in S2 tissue culture cells . IP-MS candidates were selected for imaging if they had at least two unique peptides and a 3-fold enrichment over control . Common contaminants were eliminated ( e . g . ribosomal and tubulin proteins [Mellacheruvu et al . , 2013] ) as were proteins previously known to colocalize with HP1a ( e . g . KDM4A [Lin et al . , 2008] , HP4 [Greil et al . , 2007] ) . RNAi screen candidates were chosen based on identification by more than one method ( HP1a intensity metrics , supervised clustering [Support Vector Machine or SVM] , or unsupervised clustering [hierarchical] , or GO terms enrichment [sequence-specific DNA binding , RNA-binding , RNA interference , response to heat , chromatin organization] ) . The candidate list was further refined based on the availability of clones from the Berkeley Drosophila Genome Project ( Yu et al . , 2011 ) . For genes with multiple isoforms , the gene isoform predominantly expressed in S2 cells according to published stranded RNA-seq data ( Brown et al . , 2014 ) was chosen . Based on these criteria , we subcloned 89 unique protein-coding open reading frames ( ORFs ) ( 44 identified by HP1a IP-MS only , 44 by RNAi screen only , 1 from both HP1a IP-MS and RNAi screen ) into a GFP expression vector and analyzed colocalization with mCherry-HP1a ( Figure 3 ) by calculating the Pearson correlation coefficient ( Costes et al . , 2004 ) ( PCC ) . Low-resolution/high-throughput imaging identified 30 candidates ( 34% of the 89 ) that colocalized with HP1a ( see ‘Materials and methods’ ) , 9 of which were identified by HP1a IP-MS ( 9/44=20% ) and 21 from the RNAi screen ( 21/44=48% ) ( Figure 3 and Figure 3—source data 1 ) . We conclude that ~1/3 of the tested candidates are likely to be physically associated with the heterochromatin domain , and are analyzed in more detail below . The remainder were not localized to heterochromatin due to technical reasons ( e . g . poor expression or produced non-functional proteins ) , or could regulate HP1a/heterochromatin indirectly or represent noise from the screens , and were not studied further . 10 . 7554/eLife . 16096 . 012Figure 3 . Identification of candidates that co-localize with HP1a . Proteins were selected from the HP1a IP-MS ( red circles ) or the RNAi screen ( blue circles ) , tagged with GFP ( green ) , and analyzed for localization with respect to mCherry-tagged HP1a ( red ) . GFP-tagged HP1a was used as a positive control ( gray circles ) . The Pearson correlation coefficient ( PCC ) between mCherry-HP1a and GFP-tagged proteins left of the dashed line was significantly higher than the PCC between mCherry-HP1a and GFP-mod ( green triangle ) , using the two-sided unpaired Mann-Whitney test ( p-value<0 . 05 ) . Numbers on graph correspond to representative images ( right panel ) . Scale bar is 5 µm . See Figure 3—source data 1 for the PCC of all proteins tested . DOI: http://dx . doi . org/10 . 7554/eLife . 16096 . 01210 . 7554/eLife . 16096 . 013Figure 3—source data 1 . Identification of candidates that co-localize with HP1a . Proteins were selected from the HP1a IP-MS or the RNAi screen , tagged with GFP and analyzed for localization with respect to mCherry-tagged HP1a . GFP-tagged HP1a was used as a positive control . Cells shaded red were found to have a significant increase in the Pearson correlation coefficient ( PCC ) between GFP-mod ( dark green shading ) and GFP-ORF PCC using the two-sided unpaired Mann-Whitney test ( p-value<0 . 05 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 16096 . 013 Transcriptional silencing is a defining feature of heterochromatin and screens for dominant modifiers of PEV have identified many key heterochromatin components . To determine if proteins that colocalized significantly with HP1a ( Figure 3 and Figure 3—source data 1 ) affect heterochromatin properties in the fly , we assayed publicly available fly mutants or RNAi lines for modification of PEV ( using either yellow+or white+ reporter genes ) ( see ‘Materials and methods’ for details ) . We tested 11 different HPips or HPprs for mutant or depletion effects on PEV , and found that 8 suppress PEV ( Figure 4 and Figure 4—source data 1 ) . The high success rate at identifying modifiers of PEV ( 8/11 tested = 73% ) suggests that most candidates that colocalize with HP1a ( Figure 3 and Figure 3—source data 1 ) are also likely to impact heterochromatin properties . We extended the PEV analysis to 5 other candidates identified as HPips ( CG7357 , Lhr/HP3 ) or HPprs ( MTA1-like , Khc and Hsc70-4 ) ( Table 1 and Alekseyenko et al . , 2014; Greil et al . , 2007; van Bemmel et al . , 2013 ) , whose colocalization with HP1a was not determined ( above ) . Mutant alleles/RNAi lines for all 5 candidates produced a Su ( var ) phenotype ( Figure 4 and Figure 4—source data 1 ) . Altogether , fly mutants or RNAi lines targeting 13 of 16 tested candidates ( 81% ) produced a Su ( var ) phenotype ( Figure 4 and Figure 4—source data 1 ) . We conclude that the multi-pronged experimental approach ( Figure 1 ) was very efficient at identifying functional heterochromatin components . In addition , since ~30 modifiers of variegation were previously mapped to specific genes ( Elgin and Reuter , 2013 ) , this represents an ~50% increase in the number of known proteins that regulate PEV . Given that CG7357 localizes to heterochromatin and is required for silencing , we propose naming the gene ‘Oddjob’ ( Odj ) . 10 . 7554/eLife . 16096 . 014Figure 4 . HPips and RNAi screen candidates are suppressors of variegation . ( A ) Color Inspector 3D in ImageJ was used to determine the RGB values of 'red' pixels ( indicating loss of suppression ) . The percent of the eye composed of red pixels was then calculated . ( B ) Fly mutants and RNAi lines were tested for impact on white variegation in y , w , KV108 males , and are organized by p-value . Mutations were tested for dominant effects if they were recessive lethal , otherwise homozygotes were analyzed . CG7357[f00521] was scored for variegation using the yellow reporter gene , since the line harbors a mini-white reporter that precludes assessment of white variegation . The p-values were calculated using a 2-tailed , 2-sample unequal variance t-test for white variegation and a 2-sample Kolmogorov-Smirnov test for yellow variegation . Positive and negative controls were performed and are listed in the Figure 4—source data 1 along with the genotypes of all the fly lines used . CG2129 , Ssrp and Ref1 could not be tested for effects on variegation using RNAi lines , due to lethality . DOI: http://dx . doi . org/10 . 7554/eLife . 16096 . 01410 . 7554/eLife . 16096 . 015Figure 4—source data 1 . HPips and RNAi screen candidates are suppressors of variegation . Fly mutants and RNAi lines were tested for white variegation in y , w , KV108 males and are grouped with the appropriate positive and negative control ( s ) . Mutants were only tested for dominant effects if they were recessive lethal . See legend of Figure 4 for additional details . DOI: http://dx . doi . org/10 . 7554/eLife . 16096 . 015 Indiscriminate binding of HPips or HPprs to HP1a predicts that these proteins should be broadly distributed across the entire heterochromatin holodomain . However , HPips or HPprs could selectively bind to different HP1a populations , or directly bind specific repeated sequences , resulting in localization to restricted compartments or subdomains of the heterochromatin holodomain . Additionally , we suspected that implementation of the PCC to define colocalization with HP1a may have led to the identification of false-positives . Therefore , to validate and more precisely determine candidate protein localization patterns in heterochromatin , we analyzed a subset ( 19 ) of the top HP1a colocalization hits ( 30 ) using higher-resolution microscopy and manual curation of the higher-resolution localization patterns ( see ‘Materials and methods’ ) . We successfully validated heterochromatin localization for 7 of the 9 strongest colocalizers ( PCC > 0 . 79 ) , and 12/19 total ( Table 3 and Figure 3—source data 1 ) . Surprisingly , we found that most HPips displayed restricted patterns within heterochromatin . Four general patterns were observed ( Figure 5 and Table 3 ) : 1 ) broad – near-complete overlap with HP1a ( e . g . HP4 ) ; 2 ) narrow – significantly less overlap with HP1a , especially at the periphery of HP1a domains ( e . g . FK506-bp1 ) ; 3 ) focal – one or a few highly restricted foci embedded in or adjacent to HP1a ( e . g . crol ) ; and 4 ) at the heterochromatin boundary – partial overlap with the edge of the HP1a domain ( e . g . Hrb87F , Tlk; Figure 5—figure supplement 1 ) . Hereafter we refer to the narrow , focal and boundary classes as subdomain-forming proteins . 10 . 7554/eLife . 16096 . 016Table 3 . Localization patterns of known heterochromatin components , IP-MS and RNAi screen hits . Top candidates from the localization screen and proteins with a previously known connection to HP1a were imaged at higher resolution and grouped into four categories of heterochromatin localization , based on live imaging in the presence of fluorescently tagged HP1a: broad , narrow , focal , or at the heterochromatin boundary . Localization outside of heterochromatin is also noted . Proteins are sorted by their observed localization patterns . HC = heterochromatin , NR = nucleolar , EC = euchromatin , CP = cytoplasmic . DOI: http://dx . doi . org/10 . 7554/eLife . 16096 . 016Heterochromatic LocalizationOther Localization NotesGene NameIsoformReason InvestigatedBroadNarrowFocalAt HC BoundaryPan NuclearOtherPrevious Published LocalizationEffect on VariegationHeterochromatin protein 4HP4-RAHP1a IP-MSXKc chromocenter ( Greil et al . , 2007 ) Su ( var ) ( Greil et al . , 2007 ) Heterochromatin protein 5HP5-RA¶HP1a IP-MSXKc chromocenter ( Greil et al . , 2007 ) Su ( var ) ( Greil et al . , 2007 ) Lysine ( K ) -specific demethylase 4AKdm4A-RA¶HP1a IP-MSXKc , S2 and BG3 chromocenter ( Colmenares et al . , unpublished ) Su ( var ) ( Colmenares et al . , unpublished ) Suppressor of variegation 3-9Su ( var ) 3-9-RA‡ , ¶HP1a IP-MSXpolytene chromocenter ( Schotta et al . , 2002 ) Su ( var ) ( Reuter et al . , 1986 ) Suppressor of variegation 3-7Su ( var ) 3-7-RB† , ¶literatureXpolytene chromocenter , HC in embryos ( Cleard et al . , 1997 ) Su ( var ) ( Reuter et al . , 1990 ) Lethal hybrid rescueLhr-RA/HP3-RA¶HP1a IP-MSXXcentromeric ( Thomae et al . , 2013 ) ; polytene chromocenter ( Brideau et al . , 2006 ) ; Kc chromocenter ( Greil et al . , 2007 ) Su ( var ) ( this study ) Heterochromatin protein 6HP6-RA¶literatureXSlight narrow HC enrichmentKc chromocenter ( Greil et al . , 2007 ) ; polytene chromocenter ( Joppich et al . , 2009 ) ; Kc cells - centromeric ( Ross et al . , 2013 ) Not a mod ( var ) ( Greil et al . , 2007 ) ; deficiency spanning gene is a Su ( var ) ( Doheny et al . , 2008 ) Oddjob ( CG7357 ) Odj-RA¶HP1a IP-MSXXX-Su ( var ) ( this study ) Su ( var ) 2-HP2Su ( var ) 2-HP2-RBHP1a IP-MSXXXpolytene chromocenter ( Shaffer et al . , 2002 ) Su ( var ) ( Shaffer et al . , 2002 ) blanksblanks-RA*RNAi screenXXXFoci outside HCpan-nuclear ( structured ) ( Gerbasi et al . , 2011 ) Su ( var ) ( this study ) ; OE mod ( var ) ( Schneiderman et al . , 2010 ) CG2129CG2129-RA*RNAi screenXXFoci outside HC-RNAi lines were lethalFK506-binding protein 1FK506-bp1-RAHP1a IP-MSXFoci outside HCnucleolar based on DAPI-staining ( Edlich-Muth et al . , 2015 ) Non-mod ( var ) ( this study ) XNPXNP-RA¶literatureXXactive genes and satellite DNA near HC in polytenes and imaginal discs ( Schneiderman et al . , 2009 ) ; Broad HC in polytenes ( Bassett et al . , 2008 ) ; Beta-heterochromatin of the X chromosome in polytenes ( Emelyanov et al . , 2010 ) OE mod ( var ) ( Schneiderman et al . , 2009 ) ; Su ( var ) ( Bassett et al . , 2008 ) , ( Emelyanov et al . , 2010 ) Suppressor of Under-ReplicationSuUR-RA¶literatureXXpolytene chromocenter ( Makunin et al . , 2002 ) mutation is Su ( var ) , extra copy is E ( var ) : ( Belyaeva et al . , 2003 ) Hormone receptor 83Hr83-RA* , §RNAi screenXXNR--D1 chromosomal proteinD1-RA¶literatureXSlight narrow HC enrichmentHC ( SATI and SATIII ) in embryos ( Aulner et al . , 2002 ) Su ( var ) ( Aulner et al . , 2002 ) lethal ( 3 ) neo38l ( 3 ) neo38-RBRNAi screenXFOCI-Non-mod ( var ) ( this study ) ; OE mod ( var ) ( Schneiderman et al . , 2010 ) crooked legscrol-RDRNAi screenXFOCInuclear ( Mitchell et al . , 2008 ) Su ( var ) ( this study ) ; OE mod ( var ) ( Schneiderman et al . , 2010 ) ADD domain-containing protein 1ADD1-RBHP1a IP-MSXXWeak broad HC enrichmentpolytene chromocenter ( Alekseyenko et al . , 2014 ) Su ( var ) ( Alekseyenko et al . , 2014 ) proliferation disrupterprod-RA¶literatureXXAATAACATAG in 3rd instar larvae brains ( Platero et al . , 1998 ) -Heterogeneous nuclear ribonucleoprotein at 87FHrb87F-RA§RNAi screenXpolytene chromocenter ( Piacentini et al . , 2009 ) Su ( var ) ( Piacentini et al . , 2009 ) Tousled-like kinaseTlk-RFRNAi screenX1-2 foci per nuc . Often 1 focus is abutting HP1anuclear , but not chromatin bound ( Carrera et al . , 2003 ) Su ( var ) ( this study ) RNA and export factor binding protein 1Ref1-RA#HP1a IP-MSXSlight HC enrichmentnuclear membrane and nucleoplasm ( Buszczak and Spradling , 2006 ) -sans fillesnf-RARNAi screenExcept nucleolusnuclear ( Flickinger and Salz , 1994 ) Su ( var ) ( this study ) Hepatocyte nuclear factor 4Hnf4-RARNAi screenExcept nucleolusnuclear ( Palanker et al . , 2009; Gutzwiller et al . , 2010 ) Su ( var ) ( this study ) bicoid-interacting protein 3bin3-RARNAi screenX-Su ( var ) ( this study ) Cullin 4Cul4-RA¶literatureX--female lethal dfl ( 2 ) d-RAHP1a IP-MSXnon-uniform in nucleus ( Penn et al . , 2008 ) Su ( var ) ( this study ) jumeaujumu-RA§RNAi screenXpolytene chromocenter ( Strödicke et al . , 2000 ) Su ( var ) ( Strödicke et al . , 2000 ) La autoantigen-likeLa-RA#HP1a IP-MSECnuclear ( Yoo and Wolin , 1994 ) Non-mod ( var ) ( this study ) Structure specific recognition proteinSsrp-RARNAi screenNRnucleolar ( Hsu , et al . , 1993 ) -*Protein localization is dependent on which terminus of the gene is GFP-tagged and/or cell-type†Stable tagged Kc cell line‡Transient transfection of BG3 cells§Less than 1% of cells expressed the construct#Proteins were only found enriched in one HP1a IP-MS¶Proteins were not tested for colocalization with HP1a in the low-resolution colocalization screen10 . 7554/eLife . 16096 . 017Figure 5 . Heterochromatic proteins display diverse localization patterns . HP4 and HP5 broadly overlap with HP1a . SuUR and FK506-bp1 overlap with the interior of HP1a ( narrow ) . Crol and l ( 3 ) neo38 form a focus within the HP1a domain ( focal ) . Focal proteins are presented as slices , broad and narrow proteins are projections . mCherry-tagged HP1a is in red , GFP-tagged ORF is in green . Scale bar is 5 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 16096 . 01710 . 7554/eLife . 16096 . 018Figure 5—figure supplement 1 . Heterochromatic proteins display diverse localization patterns . mCherry-tagged HP1a is in red , GFP-tagged ORF is in green . Localization patterns are grouped into 7 categories: 'Broad' - almost complete overlap with HP1a; 'Narrow' - only partial overlap with HP1a; 'Focal' - forms a foci or focus that overlap with HP1a; 'Pan-nuclear' - everywhere in the nucleus; 'At HC boundary' - enriched at the periphery of heterochromatin; 'Foci outside of HC' - forms foci outside of heterochromatin; and 'Nucleolus' - localizes to the nucleolus . Proteins labelled with + indicates that in a population of S2 cells they display patterns that fit in more than one category . * indicates that the localization pattern is dependent on cell type or location ( N- or C-term ) of the tag . Proteins labelled with ^ indicate a slice , otherwise images are projections . Prod was tagged with mCherry and image was false-colored; red indicates GFP-tagged HP1a and green indicates mCherry-tagged prod . & indicates experiment done in Kc cells . Hr83: narrow when N-terminal tagged in Kc cells ( stable ) , pan-nuclear when N-term tagged in S2 cells ( transient ) , nucleolar when C-term tagged in S2 cells ( transient ) ; CG2129: N-terminally GFP tagged construct ( transient Kc and S2 ) shows nucleolar localization in lowly expressing cells and foci appear in highly expressing cells; blanks when N-term tagged in stable Kc cells broadly co-localizes with HP1a , N-term tagged in transient S2 has foci next to the HP1a domain , C-term tagged in transient S2 or Kc cells is pan-nuclear , antibody staining is also pan-nuclear with some structure throughout the nucleus . Su ( var ) 3–9 is tagged with mClover . BioTAP tagged ADD1-RA ( red ) was visualized using a Peroxidase antibody followed by immunofluorescence . DOI: http://dx . doi . org/10 . 7554/eLife . 16096 . 018 To better understand the spatial organization of the heterochromatin domain and evaluate the generality of subdomain architecture , we localized 12 previously identified HPips and repeated-DNA binding proteins at high-resolution in S2 cells ( see Table 3 for summary; see Figure 5 and Figure 5—figure supplement 1 for images ) . Broad colocalization with HP1a was verified for HP5 ( Greil et al . , 2007 ) , Kdm4a ( Colmenares et al . , unpublished ) , Su ( var ) 3–7 ( Cléard et al . , 1997 ) , and Su ( var ) 3–9 ( Schotta et al . , 2002 ) . However , other proteins displayed more complex patterns within the heterochromatin of S2 cells than were previously reported using cells containing polytenized chromosomes ( Alekseyenko et al . , 2014; Pindyurin et al . , 2008; Shaffer et al . , 2002 , 2006 ) . For example , SuUR occupies a narrow subdomain within the holodomain , and HP2 forms a narrow subdomain enriched at the domain boundary . Interestingly , ADD1 isoform A forms a narrow subdomain within the holodomain , while ADD1 isoform B ( ADD1-PB ) occupies a focal subdomain at the domain boundary . Overall , 7 of the 12 previously identified HPips and repeated-DNA binding proteins were classified as forming subdomains ( Table 3 ) . We conclude that protein localization within heterochromatin is more complex and diverse than previously recognized . We observed proteins that exhibited broad ( 10 proteins ) , narrow ( 7 ) , focal ( 11 ) and heterochromatin boundary ( 8 ) patterns ( Table 3 ) , which were not mutually exclusive ( see below ) . The identification of a large number of subdomain-forming HPips ( 17/22 heterochromatin proteins tested , 77% ) shows that binding to HP1a is predominantly restricted within the heterochromatin , and not indiscriminate . We hypothesize that an unknown mechanism restricts HPip localization within the HP1a/heterochromatin holodomain ( see ‘Discussion’ ) . HP1a displays dynamic behavior during the cell cycle , which is essential for error-free mitosis ( Hirota et al . , 2005; Mateescu et al . , 2004 ) and replication of heterochromatin ( Quivy et al . , 2008 ) . HP1a is largely removed from chromatin during mitotic prophase , reloads starting at anaphase/telophase and remains a discrete domain associated with chromosomes throughout interphase ( Kellum et al . , 1995 ) . Additionally , the localization of proteins that bind specific satellite repeats ( Prod and GAGA factor ) is cell cycle regulated ( Platero et al . , 1998 ) . Therefore , we used time-lapse microscopy to analyze cell cycle changes in the localization of 7 fluorescently-tagged HP1a interactors/regulators that exhibited multiple patterns in the previous analyses ( Table 3 and Figure 5—figure supplement 1 ) , relative to the heterochromatin domain ( HP1a-GFP ) . The localization patterns were surprisingly dynamic , and in some cases suggest potential biological functions . For example , HP2 and SuUR both colocalize with PCNA foci ( replication [Moldovan et al . , 2007] ) during early , mid and late S-phase ( HP2 , Figure 6B and Video 1; SuUR , Figure 7—figure supplement 1 , Video 2 and Nordman et al . , 2014 ) , suggesting links to replication . Indeed , SuUR prevents polytenization of heterochromatic sequences ( Belyaeva et al . , 1998 ) and inhibits fork progression ( Nordman et al . , 2014 ) . Intriguingly , we observed a similar pattern for HP2 during S phase; euchromatic HP2 foci appear during S-phase ( Figure 6A , 11h 40' ) and completely overlap with PCNA ( Figure 6B ) . Determining if HP2 also functions during replication will require further investigation . Interestingly , HP2 and SuUR localization patterns and dynamics differ during G1 and G2 , perhaps reflecting different biological roles during these cell cycle phases . Just prior to mitosis ( in G2 , Figure 6A , 0' ) , HP2 forms a narrow subdomain within the HP1a domain and is at the HP1a boundary in the same nucleus . During mitosis , HP2 is largely removed from chromosomes ( Figure 6A , 3h ) until anaphase/telophase , when HP2 is recruited shortly after the HP1a domain reforms ( Figure 6A , 3h 20’ ) . Then in G1 , HP2 and HP1a broadly colocalize ( Figure 6A , 3h 40’ ) with the brightest HP2 signal at the HP1a domain boundary ( Figure 6A , 4h 40’ ) . In contrast , SuUR is not as dynamic as HP2; it forms a narrow subdomain within the HP1a domain during both G1 and G2 , but is also released from heterochromatin during mitosis ( Figure 7—figure supplement 1 ) . 10 . 7554/eLife . 16096 . 019Figure 6 . HP2 time-lapse imaging reveals dynamic regulation and overlap with PCNA throughout S-phase . HP2 partially overlaps and is enriched at the boundary of HP1a in G2 , released from chromatin during mitosis and broadly colocalized with HP1a during G1 . Mitosis is used to discriminate G1 from G2 . Dotted lines indicate the cell periphery during mitosis . mCherry-tagged HP1a is in red , GFP-tagged HP2 is in green . Scale bar is 10 µm . ( B ) HP2 overlaps with PCNA foci in early , mid and late S-phase . Representative images of early , mid and late S-phase are shown . mCherry-tagged PCNA is in red , GFP-tagged HP2 is in green . Scale bar is 5 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 16096 . 01910 . 7554/eLife . 16096 . 020Video 1 . HP2 time-lapse imaging reveals dynamic regulation throughout the cell cycle . HP2 partially overlaps and is enriched at the boundary of HP1a in G2 , released from chromatin during mitosis and broadly colocalized with HP1a during G1 . Mitosis is used to discriminate G1 from G2 . mCherry-tagged HP1a is in red , GFP-tagged HP2 is in green . Scale bar is 10 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 16096 . 02010 . 7554/eLife . 16096 . 021Video 2 . Combined SuUR , HP1a and PCNA time-lapse imaging reveals dynamic regulation . SuUR colocalizes with HP1a during G2 and G1 , and colocalizes with PCNA during S-phase . Mitosis is used to discriminate G1 from G2 , while PCNA foci indicate S-phase . Cerulean-tagged HP1a is in blue , YFP-tagged SuUR is in green , mCherry-tagged PCNA is in red . Scale bar is 5 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 16096 . 021 ADD1-PB and Odj also display different dynamic heterochromatin localization patterns . ADD1-PB forms bright focal subdomains with weaker broad enrichment in bulk heterochromatin , whereas Odj forms focal subdomains in G1 that broadly colocalizes with HP1a by the end of G2 . A striking observation is that the intensities of both proteins are lower in G1 compared to G2 , suggesting progressive heterochromatin loading of these proteins during interphase ( Figure 7—figure supplements 2 and 3 , and Videos 3 and 4 ) . Interestingly , even though most ADD1 and all detectable HP1a are removed during mitosis , some ADD1 signal remains attached to chromosomes , distinguishing it from all other HPips analyzed here . Another striking example is FK506-bp1 , which displays a narrow localization pattern as well as a ring around the nucleolus throughout much of the cell cycle ( Figure 7—figure supplement 4 and Video 5 ) . Interestingly , FK506-bp1 accumulates foci outside of heterochromatin during G2 , which do not colocalize with markers for replication ( PCNA ) , centromeres ( CID ) or telomeres ( HOAP ) ( data not shown ) . 10 . 7554/eLife . 16096 . 022Video 3 . ADD1-PB time-lapse imaging reveals dynamic regulation . ADD1-PB forms focal subdomains that abut and overlap HP1a , b not overlap with the centromeric or telomeric markers CID and HOAP ( data not shown ) , respectively . In G2 ADD1-PB is predominantly focal at the heterochromatin boundary . A small amount of discrete signal remains on chromatin during mitosis and persists at low levels into G1 , before eventually increasing in intensity , which suggests loading at the end of G1 or during S-phase . Mitosis is used to discriminate G1 from G2 . mCherry-tagged HP1a is in red , GFP-tagged ADD1-PB is in green . Scale bar is 5 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 16096 . 02210 . 7554/eLife . 16096 . 023Video 4 . Oddjob time-lapse imaging reveals dynamic regulation . Odj broadly co-localizes with HP1a at the end of G2 and disperses from chromosomes during mitosis . It reforms as a focal subdomain after mitosis that gradually increases in size , until it broadly overlaps HP1a again . Mitosis is used to discriminate G1 from G2 . mCherry-tagged HP1a is in red , GFP-tagged Odj is in green . Scale bar is 5 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 16096 . 02310 . 7554/eLife . 16096 . 024Video 5 . FK506-bp1 time-lapse imaging reveals dynamic regulation . FK506-bp1 narrowly co-localizes with HP1a throughout much of the cell cycle and loses co-localization with HP1a 20 min to 1 hr before HP1a is released from chromosomes ( prophase ) . After mitosis , the narrow co-localization pattern of FK506-bp1 is restored , with a weak ring around the nucleolus , which is located adjacent to the HP1a domain . FK506-bp1 foci then begin to accumulate outside of heterochromatin until just before prophase , when they disappear prior to HP1a removal . Foci do not track with PCNA ( replication ) , CID ( centromeres ) or HOAP ( telomeres ) foci ( data not shown ) . Mitosis is used to discriminate G1 from G2 . mCherry-tagged HP1a is in red , GFP-tagged FK506-bp1 is in green . Scale bar is 10 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 16096 . 024 Supporting the validity of our approach , Lhr displayed its previously described localization pattern ( Brideau et al . , 2006 ) ( broad and centromeric; Table 3 , Figure 7—figure supplement 5 and Video 6 ) . However , in contrast to a previous report that XNP/ATRX is broadly enriched at polytene chromocenters ( Bassett et al . , 2008 ) , we observe that XNP exhibits narrow and focal localization patterns in S2 cells ( Figure 7—figure supplement 6 and Video 7 ) . This is consistent with XNP’s observed enrichment at active genes , satellite DNA and heterochromatin of the X chromosome in imaginal discs and polytene chromosomes ( Schneiderman et al . , 2009; Emelyanov et al . , 2010 ) . Emphasizing the complexity of subdomain architecture , we detected some Lhr and XNP foci that colocalize within the same nucleus , while others do not ( Figure 7—figure supplement 5C ) . 10 . 7554/eLife . 16096 . 025Video 6 . Lhr time-lapse imaging reveals dynamic regulation . Lhr broadly co-localizes with HP1a and is released from chromatin during mitosis . Mitosis is used to discriminate G1 from G2 . mCherry-tagged HP1a is in red , GFP-tagged Lhr is in green . Scale bar is 5 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 16096 . 02510 . 7554/eLife . 16096 . 026Video 7 . XNP time-lapse imaging reveals dynamic regulation . XNP colocalizes with a portion of HP1a in G2 . The majority of XNP is removed during mitosis , however 1-2 foci remain chromatin-bound . In G1 XNP is focal within the HP1a domain but gradually accumulates in size and colocalizes with more HP1a . Mitosis is used to discriminate G1 from G2 . mCherry-tagged HP1a is in red , GFP-tagged XNP is in green . Scale bar is 5 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 16096 . 026 We conclude that the localization patterns for 5 of the 7 HPips studied with time-lapse imaging are dynamic throughout the cell cycle ( Figure 7 ) . Further analysis is required to determine if the changing distributions throughout the cell cycle reflects biological functions . For instance , the prevalence of ADD1-PB foci localized at the heterochromatin boundary could indicate a role in maintaining the border between the heterochromatin and euchromatin domains . 10 . 7554/eLife . 16096 . 027Figure 7 . Time-lapse imaging reveals a variety of dynamic localization patterns within heterochromatin . A graphical representation of the localization patterns of heterochromatic proteins throughout the cell cycle is shown . HP1a is depicted in red , the heterochromatin protein ( HPip ) in green and overlap between the two in yellow . A dotted circle indicates that FK506-bp1 forms a ring around the nucleolus . * indicates foci overlap completely with PCNA during S-phase . DOI: http://dx . doi . org/10 . 7554/eLife . 16096 . 02710 . 7554/eLife . 16096 . 028Figure 7—figure supplement 1 . Combined SuUR , HP1a and PCNA time-lapse imaging reveals dynamic regulation . SuUR colocalizes with HP1a during G2 and G1 , and colocalizes with PCNA during S-phase . Dotted lines indicate the cell periphery during mitosis . Mitosis is used to discriminate G1 from G2 , while PCNA foci indicates S-phase . Cerulean tagged HP1a is in blue , YFP tagged SuUR is in green , mCherry tagged PCNA is in red . Scale bar is 10 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 16096 . 02810 . 7554/eLife . 16096 . 029Figure 7—figure supplement 2 . ADD1-PB time-lapse imaging reveals dynamic regulation . ADD1-PB forms focal subdomains that abut and overlap HP1a , and does not overlap with the centromeric or telomeric markers CID and HOAP ( data not shown ) , respectively . In G2 ADD1-PB is predominantly focal at the heterochromatin boundary . A small amount of discrete signal remains on chromatin during mitosis and persists at low levels into G1 , before eventually increasing in intensity , which suggests loading at the end of G1 or during S-phase . Mitosis is used to discriminate G1 from G2 . Dotted lines indicate the cell periphery during mitosis as it divides into two daughter cells ( G1 ) . mCherry tagged HP1a is in red , GFP tagged ADD1-PB is in green . Scale bar is 10 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 16096 . 02910 . 7554/eLife . 16096 . 030Figure 7—figure supplement 3 . Oddjob time-lapse imaging reveals dynamic regulation . Odj broadly co-localizes with HP1a at the end of G2 and disperses from chromosomes during mitosis . It reforms as a focal subdomain after mitosis that gradually increases in size , until it broadly overlaps HP1a again . Mitosis is used to discriminate G1 from G2 . Dotted lines indicate the cell periphery during mitosis as it divides into two daughter cells ( G1 ) . mCherry tagged HP1a is in red , GFP tagged Odj is in green . Scale bar is 10 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 16096 . 03010 . 7554/eLife . 16096 . 031Figure 7—figure supplement 4 . FK506-bp1 time-lapse imaging reveals dynamic regulation . FK506-bp1 narrowly co-localizes with HP1a throughout much of the cell cycle and loses co-localization with HP1a 20 min to 1 hr before HP1a is released from chromosomes ( prophase ) . After mitosis , the narrow co-localization pattern of FK506-bp1 is restored , with a weak ring around the nucleolus , which is located adjacent to the HP1a domain . FK506-bp1 foci then begin to accumulate outside of heterochromatin until just before prophase , when they disappear prior to HP1a removal . Foci do not track with PCNA ( replication ) , CID ( centromeres ) or HOAP ( telomeres ) foci ( data not shown ) . Mitosis is used to discriminate G1 from G2 . Dotted lines indicate the cell periphery during mitosis as it divides into two daughter cells ( G1 ) . mCherry tagged HP1a is in red , GFP tagged FK506-bp1 is in green . Scale bar is 10 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 16096 . 03110 . 7554/eLife . 16096 . 032Figure 7—figure supplement 5 . Lhr time-lapse imaging reveals dynamic regulation . Lhr broadly co-localizes with HP1a and is released from chromatin during mitosis . Mitosis is used to discriminate G1 from G2 . Dotted lines indicate the cell periphery during mitosis as it divides into two daughter cells ( G1 ) . mCherry tagged HP1a is in red , GFP tagged Lhr is in green . Scale bar is 10 µm . ( B ) Lhr partially overlaps centromeres . mCherry tagged CID is in red , GFP tagged Lhr is in green . Scale bar is 5 µm . ( C ) Some Lhr and XNP foci overlap , but others do not ( arrows ) . mCherry tagged Lhr is in red , GFP tagged XNP is in green . Scale bar is 5 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 16096 . 03210 . 7554/eLife . 16096 . 033Figure 7—figure supplement 6 . XNP time-lapse imaging reveals dynamic regulation . XNP colocalizes with a portion of HP1a in G2 . The majority of XNP is removed during mitosis , however 1–2 foci remain chromatin-bound . In G1 XNP is focal within the HP1a domain while gradually accumulating and colocalizing with more HP1a . Mitosis is used to discriminate G1 from G2 . Dotted lines indicate the cell periphery during mitosis as it divides into two daughter cells ( G1 ) . mCherry-tagged HP1a is in red , GFP-tagged XNP is in green . Scale bar is 10 µm . ( B ) XNP and CID partially overlap . mCherry-tagged CID is in red , GFP-tagged XNP is in green . Scale bar is 5 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 16096 . 033
The heterochromatin domain is defined molecularly by enrichment for HP1 , which binds many different proteins and has been implicated in diverse and sometimes contradictory functions , including repression of transposons ( Lundberg et al . , 2013 ) and genes , and promotion of gene expression ( Piacentini et al . , 2009 ) . How HP1a mediates a wide variety of heterochromatin functions and maintains interactions with multiple binding partners is currently unknown . To better understand this important nuclear domain , we performed two complementary screens to identify novel structural and functional components of Drosophila heterochromatin . We immunoprecipitated HP1a under stringent conditions and performed LC-MS/MS to identify core heterochromatin components . In addition , an unbiased genome-wide RNAi screen was used to identify regulators of heterochromatin organization , maintenance and establishment , independent of their ability to bind HP1a . These screens identified 118 novel putative HP1a interactors and 374 putative regulators of HP1a . Candidate hits were further analyzed for heterochromatin localization , and 34% ( 30/89 tested ) strongly colocalized with HP1a in low-resolution imaging . Higher-resolution microscopy and time-lapse analysis revealed that many heterochromatin proteins occupy subdomains within the holodomain , and display dynamic localization patterns throughout the cell cycle . We identified at least 13 previously unknown components required for heterochromatin-mediated gene silencing ( PEV ) , and demonstrated that the organization and composition of heterochromatin is more complex and dynamic than suggested by previous studies ( see Supplementary file 5 for a summary of the results from our study ) . Although IP-MS was previously used to identify HP1a interacting proteins ( Rosnoblet et al . , 2011; Lechner et al . , 2005; Motamedi et al . , 2008; Ryu et al . , 2014; Alekseyenko et al . , 2014 ) , our results demonstrate that many new heterochromatin components can still be isolated using this approach . Here , we optimized purification conditions by expressing tagged-HP1a at low levels , using high salt concentrations and removing DNA and RNA , which may have released different subpopulations of HP1a complexes than previous studies . Regardless , this approach was validated by identifying at least 17 previously known pericentromeric heterochromatin structural components ( e . g . HP2 , HP3/Lhr , HP4 , HP5 , Su ( var ) 3–9 , Table 1—source data 1 and 2 ) . Most importantly , we isolated 118 HPips that were not previously associated with heterochromatin . The relevance of these novel HPips to heterochromatin structure and function was demonstrated by cytological and phenotypic analyses . First , 2 of these novel HPips ( FK506-bp1 and Odj ) colocalized with HP1a using high-resolution imaging . Second , mutations in 3 of 5 novel HPips tested ( Hsc70-4 , fl ( 2 ) d and Odj ) act as Su ( var ) s ( Figure 4 and Figure 4—source data 1 ) , demonstrating relevance to transcriptional silencing , a well-established heterochromatin function . In contrast , a genome-wide RNAi screen for regulators of heterochromatin architecture has not been reported previously . We utilized multiple methods to identify candidates that disrupted HP1a levels or localization , including ranking gene depletions by changes in HP1a intensity , as well as supervised clustering ( trained on results from HP1a depletions ) and unsupervised clustering ( hits selected based on similarity to HP1a or Su ( var ) 3–9 depletions ) . All three methods identified known regulators of heterochromatin ( Table 2 ) , yet there was surprisingly little overlap between the different approaches . This suggests that more than one method of hit identification may be needed for high-content screening of potentially subtle cytological phenotypes . Regardless , we identified 374 candidate regulators whose loss mimicked HP1a depletion , including 355 with no previously known connection to heterochromatin . Importantly , 7 of 12 tested candidates ( 58% ) colocalized with HP1a at high-resolution ( Table 3 ) , and mutations in 9 of 10 tested candidates acted as suppressors of PEV ( Figure 4 and Figure 4—source data 1 ) . It is important to note that 564 gene disruptions led to increased HP1a intensity; future analyses of these interesting candidates may reveal new factors that inhibit heterochromatin formation and/or are required for removal of HP1a . Additionally , since we extracted 32 imaging features per nucleus and imaged ~400 nuclei per gene depletion , further mining of this extremely rich dataset , in addition to cytological and phenotypic analyses of the remaining candidates , are likely to identify additional proteins that impact heterochromatin , as well as regulators of other processes ( e . g . apoptosis and the cell cycle ) . We conclude that the RNAi screen successfully identified novel heterochromatin components and regulators . Interestingly , only HP1a and three other proteins were identified in both the RNAi and the HP1a IP-MS screens . A potential reason for poor overlap is that the RNAi screen enriched for hits upstream of HP1a deposition , while the IP-MS enriched for genes acting downstream of HP1a loading onto chromatin . This hypothesis is supported by studies demonstrating that loss of HPips does not cause visible defects in HP1a domain organization ( e . g . SMC5/6 [Chiolo et al . , 2011] , KDM4A [Colmenares et al . , unpublished] ) . Therefore , we propose that the complementary approaches utilized in this study enabled identification of different classes of heterochromatin proteins . Finally , the majority of heterochromatin-localized hits are required for transcriptional silencing ( 8/11 = 73% ) , suggesting that further analysis of the RNAi screen hits will identify more heterochromatin regulators . It is important to note that the absence of either HP1a colocalization or transcriptional silencing effects does not eliminate candidates from having important roles in heterochromatin structure or function . First , proteins that localize to euchromatin can regulate heterochromatin; for example , the euchromatic JIL-1 kinase restricts heterochromatin spreading , and JIL-1 mutants modify silencing phenotypes ( Zhang et al . , 2006 ) . Second , the modification of silencing assays only one of many known heterochromatin properties and functions ( Bernard et al . , 2001; Dernburg et al . , 1996; Karpen et al . , 1996; McKee and Karpen , 1990; Peng and Karpen , 2009; Sienski et al . , 2012; Clowney et al . , 2012 ) . Third , heterochromatin proteins that form subdomains may only affect PEV of genes inserted in their local environment . Thus , to exhaustively identify proteins involved in heterochromatin structure and function , more of the candidates identified in our screens need to be analyzed for colocalization with HP1a , and for impact on other heterochromatin functions , such as DNA repair ( Chiolo et al . , 2011 ) and chromosome segregation ( Dernburg et al . , 1996; Karpen et al . , 1996; McKee and Karpen , 1990 ) . Furthermore , inclusion of reporters located in other chromosomes , in addition to the Y chromosome PEV reporter utilized here , will determine if subdomain proteins exert local versus widespread PEV effects . In addition to identifying novel heterochromatin components and regulators , analysis of localization patterns revealed that heterochromatin organization is complex and dynamic . Previous studies using polytenized chromosomes showed that the majority of HPips are broadly distributed across the heterochromatin domain , and that a few heterochromatin proteins localized to sub-regions within heterochromatin ( e . g . piwi [Brower-Toland et al . , 2007] , ATF-2 [Seong et al . , 2011] ) , in some cases due to binding to specific repeated DNAs ( e . g . prod [Platero et al . , 1998] , D1 [Aulner et al . , 2002] ) . However , using cells without polytene chromosomes , we showed that the majority of heterochromatin proteins analyzed ( 17 of 22 ) form subdomains within heterochromatin , and that 5 of the 7 proteins analyzed by live imaging display highly dynamic localization patterns throughout the cell cycle . Importantly , the localization patterns for GFP-tagged HPips recapitulated published results ( e . g . HP4 , HP5 , Lhr ) ( Greil et al . , 2007 ) , suggesting that GFP-tagging per se was not responsible for the diverse patterns observed here . Additionally , we showed that proteins that localize broadly to the underreplicated heterochromatin in non-cycling nuclei with polytenized chromosomes ( e . g . SuUR [Makunin et al . , 2002] , HP2 [Shaffer et al . , 2002] ) can form subdomains in S2 cells . Previous studies showed that many but not all HP1a binding partners contain a conserved PxVxL-like motif ( Smothers and Henikoff , 2000; Nozawa et al . , 2010; Huang et al . , 2006 ) , and that the HP1a chromo shadow domain and C-terminal extension , as well as residues near the PxVxL , determine the strength of HP1a:HPip interactions ( Mendez et al . , 2011 ) . However , the prevalence of subdomain localization patterns within heterochromatin demonstrates that binding to HP1a is not indiscriminate , and must require other , currently unknown mechanisms . We consider three possibilities for subdomain formation that are not mutually exclusive ( Figure 8 ) : 1 ) sequence-specific binding , 2 ) binding to specific HP1a posttranslational modifications , and 3 ) cooperative binding between HP1a and a HPip . Further studies are required to determine if these or other mechanisms are responsible for establishing or maintaining the specific and diverse localization patterns observed for HPips and other heterochromatin proteins . One key question is whether subdomains form at the same genomic locations in every cell , or are initiated and grown at random genomic sites . 10 . 7554/eLife . 16096 . 034Figure 8 . Models for subdomain formation within heterochromatin . We propose three non-mutually exclusive models for subdomain formation of HP1a interacting proteins ( HPips ) within the HP1a ( teal ) heterochromatin holodomain . ( A ) The HPip ( red ) may be recruited to a specific sequence and seeds the formation of a subdomain ( as observed for D1 [Aulner et al . , 2002] and GAGA [Raff et al . , 1994] factor ) . ( B ) HP1a and its orthologs are extensively post-translationally modified by SUMOylation , acetylation , methylation , formylation , ubiquitination and poly ( ADP-ribosyl ) ation ( Alekseyenko et al . , 2014; Lomberk et al . , 2006; LeRoy et al . , 2009 ) . An HPip could have an increased binding affinity for a specific HP1a PTM ( yellow ) . Thus , HP1a PTMs may regulate HP1a complex formation and spatially restrict HPip recruitment . Consistent with the PTM model , HP2 and PIWI have been shown to have higher binding affinities for HP1a proteins containing phospho-mimic mutations in the HP1a chromo shadow domain ( Mendez et al . , 2011 ) . ( C ) Subdomains could form by a cooperative binding mechanism ( Bray and Duke , 2004; Bai et al . , 2010 ) . HP1a can oligomerize at least up to tetramers ( Wang et al . , 2000; Zhao et al . , 2000; Canzio et al . , 2011 ) , forming a multivalent platform for HPip binding ( i . e . more than one HPip binding site per HP1a oligomer ) . Thus , initial binding by an HPip could induce a higher binding affinity between a neighboring HP1a molecule and the HPip . The dotted arrow indicates potential self-interactions between HPips and solid arrows indicate hypothetical HPip on/off rates . DOI: http://dx . doi . org/10 . 7554/eLife . 16096 . 034 This study reveals unexpected complexity within heterochromatin , in terms of both the number of new structural and functional components identified , and their localization to discrete , dynamic subdomains . We speculate that broadly distributed proteins could encode structural components important for universal aspects of heterochromatin architecture and function ( e . g . nucleosomal ordering , variable accessibility of exogenous proteins , and coalescence of heterochromatin domains; reviewed in [Elgin and Reuter , 2013] ) . In contrast , subdomains may regulate specific functions or localized , dynamic structural changes ( e . g . decreased compaction to enable transcription , histone exchange ) . Heterochromatin may be just as structurally and functionally dynamic and diverse as euchromatin , and increased understanding of its organization will likely yield important insights into the nuclear architecture and genome biology . Thus , it will be important to determine the mechanisms responsible for subdomain formation , and how they contribute to specific heterochromatin functions .
2 × 109 S2 cells stably expressing FS-HP1a ( 3X-FLAG-Myc-StrepII-PP-HP1a [where PP is a PreScission Protease cut site] ) under control of the copia promoter ( plasmid construction as described in [Chiolo et al . , 2011] ) were exposed or mock-exposed to 10 Gy of X-rays using a 160 kV source . Cells were allowed to recover 10 or 60 min , harvested at 600 r . c . f for 5 min and flash frozen in liquid nitrogen prior to resuspension/lysis in Buffer A ( 0 . 05% NP-40 , 50 mM Hepes pH 7 . 6 , 10 mM KCl , 3 mM MgCl2 , 10% glycerol , 5 mM NaF , 5 mM β-glycerophosphate , 1 mM Benzamidine , 1X protease inhibitor cocktail [Roche , Basel , Switzerland: 11 836 170 001] , 1 mM PMSF , 25mM NEM , 1:1000 Phosphatase Inhibitor Cocktail 2 [Sigma-Aldrich , St . Louis , MO: P5726] , 1:1000 Phosphatase Inhibitor Cocktail 3 [Sigma P0044] , 1:1000 Protease Inhibitor Cocktail [Sigma P8340] ) in 500 µl/2 × 108 cells . Cell extracts were treated with 10 units benzonase ( EMD Millipore , Hayward , CA: 80601–766 ) per 37 µg of chromatin ( estimated by A260 reading ) at 4°C with mixing for 30 min . Nuclease digest was stopped with 0 . 5 mM EDTA , and HP1a was extracted on ice with 300mM NaOAc for 1 hr with mixing . Cell extracts were cleared by centrifugation at 16 , 100 r . c . f . for 10 min at 4°C . Supernatant was transferred to a new tube and mixed with anti-3XFLAG M2 beads ( Sigma ) O/N at 4°C . Bound material was washed four times with Buffer A at 4°C while mixing , eluted with 3XFLAG peptide and concentrated using Amicon Ultra-0 . 5 Centrifugal Filter Unit with Ultracel-3 membrane . LC-MS/MS was performed at Zentrallabor für Proteinanalytik ( Protein analysis Unit , Medical School of Ludwig-Maximilians University of Munich , Germany ) . Protein material similarly isolated from S2 cells lacking FS-HP1a expression were also analyzed as a negative control . Tandem affinity immunopurification was performed essentially as described for single-step immunopurification except FS-HP1a was incubated with anti-3XFLAG M2 beads ( Sigma ) for 2 hr at 4°C and then bound to Strep-Tactin Superflow beads ( IBA , Goettingen , Germany ) O/N at 4°C and washed and eluted per manufacturer’s instructions . LC-MS/MS was performed at the Scripps Center for Metabolomics and Mass Spectrometry . All MS/MS samples were analyzed using Mascot ( Matrix Science , London , UK; version 2 . 3 . 02 ) and X ! Tandem ( The GPM , thegpm . org; version CYCLONE ( 2010 . 12 . 01 . 1 ) ) . Mascot was set up to search the Drosophila NCBI protein database ( downloaded 2010; 14 , 335 entries ) . X ! Tandem was set up to search a subset of the Drosophila NCBI protein database assuming the digestion enzyme trypsin . Mascot and X ! Tandem were searched with a fragment ion mass tolerance of 0 . 50 Da and a parent ion tolerance of 10 . 0 PPM for single-step immunopurification . Mascot and X ! Tandem were searched with a fragment ion mass tolerance of 0 . 80 Da and a parent ion tolerance of 2 . 0 Da for tandem-step immunopurification . Iodoacetamide derivative of cysteine was specified in Mascot and X ! Tandem as a fixed modification . Methylation of lysine , oxidation of methionine and phosphorylation of serine , threonine and tyrosine were specified in X ! Tandem as variable modifications . Methylation of lysine , oxidation of methionine , acetaldehyde +28 of lysine , formylation of lysine , acetylation of lysine , tri-methylation and di-methylation of lysine and phosphorylation of serine , threonine and tyrosine were specified in Mascot as variable modifications . Variable modifications were accepted if they could be established at greater than 95 . 0% probability by Mascot . Scaffold ( version Scaffold_4 . 0 . 7 , Proteome Software Inc . , Portland , OR ) was used to validate MS/MS-based peptide and protein identifications . Peptide identifications were accepted if they could be established at greater than 95 . 0% probability by the Peptide Prophet algorithm ( Keller et al . , 2002 ) . Protein identifications were accepted if they could be established at greater than 95 . 0% probability and contained at least 2 identified peptides . Protein probabilities were assigned by the Protein Prophet algorithm ( Nesvizhskii et al . , 2003 ) . Proteins that contained similar peptides and could not be differentiated based on MS/MS analysis alone were grouped to satisfy the principles of parsimony . 10 μL of logarithmically growing Drosophila melanogaster Kc embryonic tissue culture cells were seeded at a density of 1 × 106 cells/mL in serum-free Schneider’s medium ( Invitrogen , Carlsbad , CA ) on 384-well plates ( Corning , Corning , NY: #3712 ) containing 0 . 25 µg dsRNA per well . Cells were incubated with dsRNA at room temperature for 30 min . 30 μl of Schneider’s medium ( Invitrogen ) with 1× antibiotics ( Invitrogen ) , and 10% FCS was added to each well . Plates were incubated for 4 days at 25°C in a humid chamber . Cells were exposed to 5 Gy of X-rays using a Faxitron TRX5200 operated at 130 kV and allowed to recover for 60’ prior to fixation ( the results of the radiation aspect of the screen are not reported here ) . Cells were fixed for 5 min with 3 . 7% paraformaldehyde and washed 3X for 5 min in PBS with 0 . 5% Triton X-100 ( PBST ) . Cells were treated for 30 min with blocking solution ( PBST containing 5% FCS ) , followed by overnight 4°C incubation in 10 μl of blocking solution containing 1:500 mouse anti-HP1a antibody ( Developmental Studies Hybridoma Bank , University of Iowa , Iowa City , Iowa: C1A9c ) and 1:1000 rabbit anti-γH2Av ( Rockland/VWR , Limerick , PA: VWR #600-401-914 ) . Cells were then washed 3X for 5 min with PBST , incubated with 10 μl of blocking solution containing secondary antibodies ( Alexa 488-conjugated anti-mouse and Alexa 546-conjugated anti-rabbit antibody at 1:500 dilutions [Invitrogen] ) for 1 hr at room temperature , washed 2X with PBST and 1X with PBS . DNA was stained with 10 μl of 0 . 2 μg/ml of DAPI in PBS for 5 min at room temperature and washed with PBS . Cell plating was performed using a CombiDrop and IF protocol was performed using a V11 Bravo at the Berkeley Screening Center . Plates were imaged using a Zeiss Axio Observer Z1 automated microscope ( Carl Zeiss , Jena , Germany ) , with a Zeiss EC Plan-Neoflaur 40X objective ( N . A . of 0 . 75 ) . All image manipulations and analyses were done with Matlab ( MathWorks , Inc . , Natick , MA ) and DIPimage ( image processing toolbox for Matlab , Delft University of Technology , The Netherlands ) . The Matlab code is available at https://github . com/svcostes/Elife_Pearson_Script . Nuclear segmentation was performed as previously described ( Costes et al . , 2006 ) . Briefly , background heterogeneity was corrected by subtracting the original image blurred by a Gaussian filter of the appropriate size . A constant threshold was then used to identify all nuclei independently of their varying intensities . Touching nuclei were separated using watershed approaches . Briefly , the distance transform of a binary mask encompassing more than one nucleus typically yields multiple bright spots representing the center of each nucleus . These maxima were used as seeds and expanded to fill the binary mask , allowing the separation of each individual nucleus . We used the DIPimage object measurement function to extract a large array of imaging features for DAPI , HP1a and γH2Av intensity , as well as pairwise correlations ( Costes et al . , 2004 ) between HP1a , γH2Av and DAPI . The nuclei were numbered and their boundaries demarcated on a large field of view to enable visual verification of the automatically generated data set . Data processing was carried out in the R Environment ( R Core Team , 2013 ) , and Rank Product analysis ( Breitling et al . , 2004 ) was performed using the Bioconductor package ( Gentleman et al . , 2004 ) to normalize the data and obtain a p-value estimation ( with 100 permutations used to calculate the null density and subsequent p-value estimation ) . Note that all hits identified below were manually inspected to ensure that the images were in focus . Rank Product estimated p-value cut-offs to identify hits for decreased relative HP1a maximum ( i . e . maximum/mean ) , increased HP1a Kurtosis , decreased HP1a average and decreased HP1a maximum intensity ( collectively 'HP1a metrics' ) were chosen based on maximal inclusion of HP1a positive controls , and correspond to 1 . 5E-03 , 5 . 0E-04 , 6 . 3E-04 and 3 . 8E-04 , respectively . Hits displaying increased cell death were eliminated based on the nuclear morphology and the number of nuclei per field . Hits displaying increased HP1a intensity were chosen by taking the overlap of hits with increased HP1a mean intensity ( p-value<0 . 05 ) and increased HP1a maximum intensity ( p-value<0 . 05 ) , and discarding hits that lead to a decreased cell number ( p-value<0 . 05 ) . Two SVMs , using polynomial kernels , were trained based on positive controls ( HP1a RNAi ) and negative controls ( mock RNAi , GFP RNAi , Rho1 RNAi [produces binucleate cells] , Thread RNAi [induces cell death] ) using Rank Product ranks . The classifier was then applied to the entire dataset and the identification of HP1a knockdowns withheld from the training set was used to optimize the SVM . The SVM utilized either all imaging features or all imaging features except those associated with γH2Av ( denoted “SVM - HP1a only features” in Figure 2—source data 1 ) . SVM analysis was performed using the R package svmpath ( Hastie et al . , 2004 ) with a ridge value of 1E-08 and a kernel parameter of 0 . 8 for all imaging features , or 0 . 4 for HP1a features . Rankings from Rank Product analysis using all imaging features or HP1a only imaging features were used to calculate ( using Matlab ) the pair-wise distance between every sample using multiple distance measures ( Spearman , Mahalanobis , Euclidean and Pearson ) . The data were then randomized and the distances of the randomized data were measured repeatedly . An estimated p-value was derived by specifying that the average distance found at the 1 percentile corresponded to a p-value of 0 . 01 . We used a p-value cut-off of 5E-07 to determine significant distances from HP1a or Su ( var ) 3–9 RNAi-treated cells . Next , we identified genes that were pair-wise close to at least five HP1a RNAi-treated samples by more than one distance metric . Finally , Matlab’s dendrogram function was applied to the HP1a Pearson correlation coefficient distance matrix and used to cluster the data . Hits clustering with HP1a were visually identified using Matlab’s clustergram function . Database for Annotation , Visualization and Integrated Discovery ( DAVID ) v6 . 7 ( Huang et al . , 2008 , 2009 ) was used to identify enriched GO terms . Functionally similar annotations as determined by Annotation Clustering in DAVID were not reported unless otherwise indicated . The pCopia-LAP-loxP acceptor plasmids were obtained by insertion of PCR-amplified loxP site , prokaryotic promoter and splice acceptor from pMK33-CTAP with AscI and PacI overhangs into AscI/PacI digested pCopia-LAP ( Cheeseman and Desai , 2005 ) . BS clones were subcloned into pCopia-LAP-loxP plasmids as in ( Yu et al . , 2011 ) and named pCopia-LAP-loxP-ORF-loxP . pCopia-LAP-loxP-ORF-loxP plasmids were used for the high-throughput low-resolution screen using InCell6000 imaging . All other live imaging was done using pCopia-LAP-ORF or pCopia-ORF-LAP plasmids . pCopia-ORF-LAP was generated by removal of the 5’ LAP tag from pCopia-LAP-ORF and introduction of a 19 amino acid polylinker using Gibson cloning and LAP tag 3’ of an ORF insertion site . ORFs were PCR-amplified from pCopia-LAP-loxP-ORF-loxP plasmids and cloned into AscI/PacI digested pCopia-LAP-ORF , or XbaI/PacI or NheI/PacI digested pCopia-ORF-LAP . Primers are listed in Supplementary file 6 . S2 cells were transiently transfected with pCopia-mCherry-HP1a and pCopia-GFP-loxP-ORF-loxP using TransIT-2020 ( MIR 5400; Mirus Bio , Madison , WI ) . Cells were imaged 3 days post-transfection using an InCell 6000 ( GE healthcare Bio-Sciences , Pittsburgh , PA , USA ) in open aperture mode . We captured a single z-slice in 9 fields/well with a 20X-objective ( 0 . 75 NA ) . Nuclei were segmented as previously described ( Costes et al . , 2006 ) using mCherry-HP1a and selected for roundness using a metric based on the perimeter square over the area . Nuclei with average intensity in background range ( for GFP <4 , 000 AU , for mCherry <3 , 000 AU ) were discarded . Nuclei whose average intensity saturated the 16-bit camera were also discarded ( less than 0 . 01% of nuclei ) , leaving ~200 nuclei on average/well ( wells with less than 10 nuclei were discarded ) . The Pearson correlation coefficient ( PCC ) was calculated per nucleus . To determine the significance of the correlation between mCherry-HP1a and GFP-ORF , we compared the PCC of GFP-ORF and mCherry-HP1a to the PCC of GFP-modulo and mCherry-HP1a using a two-sided unpaired Mann-Whitney test . If a construct was transfected in duplicate then the highest scoring well was used . Top scoring proteins from the colocalization screen were assayed for silencing effects if they were previously unknown to modify PEV , and if fly mutant alleles or RNAi lines were available and genotypes did not preclude scoring white variegation ( i . e . constructs not marked with white+; see Figure 4 and Figure 4—source data 1 for list of fly stocks ) . Mutant and RNAi fly stocks were all obtained from the Bloomington Stock center , except for Ago2[51B] which was a kind gift from F . B . Gao ( Xu et al . , 2004 ) . Flies were first crossed into a y , w background with appropriate balancers , then females containing mutations were then crossed with y , w , KV108 males . All stocks used are listed in Figure 4—source data 1 . The KV108 line contains a SUPor-P construct with y+ and w+ reporter genes inserted in the heterochromatin of the Y chromosome , resulting in variegating eye and abdomen pigmentation ( Konev et al . , 2003 ) . Female RNAi flies were crossed with y , w , KV108 males harboring Act::GAL4 . Adult male progeny from these crosses were aged 3–5 days , frozen and imaged for either white variegation in eyes or yellow variegation in the abdomen . Imaging was conducted on homozygous mutants when viable , otherwise heterozygous mutants were imaged . We detected very strong PEV suppression by TM3 balancer chromosomes , and therefore imaged only heterozygous mutants lacking this balancer . Mutant effects on PEV were compared with wildtype flies in a y , w background , whereas RNAi fly effects were compared with a mCherry RNAi fly stock . To quantify white variegation , Color Inspector 3D ( Kai Uwe Barthel , Berlin , Germany ) in Fiji ( Schindelin , 2012; Schneider et al . , 2012 , 2015 ) was used to determine the RGB values of 'red' pixels ( indicating loss of suppression ) ( 0–255 , 0–90 , 0–20 ) . The definition of 'red' was used uniformly across all samples to create a binary mask of the 'red' pixels in each eye . The area of the eye composed of 'red' pixels was then calculated ( Figure 4A ) . The p-values were calculated with a 2-tailed , 2-sample unequal variance t-test using appropriate negative controls for each group ( Figure 4 and Figure 4—source data 1 ) . Code is available at https://github . com/jmswenson/variegation . Yellow variegation was quantified , in a double-blind manner , by manually counting the number of dark spots ( i . e . where yellow is expressed ) on the abdomen , and a p-value was calculated with the two-sample Kolmogorov-Smirnov test . Images were taken using a 60X oil immersion objective ( NA 1 . 40 ) on a Deltavision Spectris microscope ( GE Healthcare ) and images were deconvolved using SoftWoRx ( Applied Precision , LLC ) . Time-lapse images were acquired once every 15–20 min . BioTAP-tagged ADD1 was colocalized with HP1a by performing IF with rabbit anti-peroxidase antibody ( Sigma P1291 ) ( 1:100 ) and mouse anti-HP1a antibody ( C1A9; Developmental Studies Hybridoma Bank ) ( 1:500 ) in fixed S2 cells . Cells were fixed ( 4% PFA for 5 min ) three days after transient transfection ( TransIT-2020 MIR 5400; MirusBio ) . Slides were blocked in 1% milk in PBS with 0 . 4% Triton-X 100 ( PBST ) for 30 min . Primary antibodies were incubated in 1% milk in PBST overnight at 4°C . Secondary antibodies ( goat anti-mouse Alexa 488 and donkey anti-rabbit Alexa 546; Invitrogen A-21121 and A10040 , respectively ) were incubated in 1% milk in PBST for 1 hr . For manual curation , images from at least two independent experiments were analyzed blindly and independently by two investigators , and classified into four non-mutually exclusive categories ( broad , narrow , focal and at the heterochromatin boundary ) based on the predominant localization patterns within a population of cells . RNAi screen data are available at the Drosophila RNAi Screening center ( http://www . flyrnai . org/cgi-bin/DRSC_screen_csv . pl ? project_id=151 ) and the PubChem BioAssay Database , AID= 1159615 ( https://pubchem . ncbi . nlm . nih . gov/assay/assay . cgi ? aid=1159615 ) . The mass spectrometry proteomics data have been deposited to the ProteomeXchange Consortium via the PRIDE ( Vizcaíno et al . , 2016 ) partner repository with the dataset identifier PXD003780 and 10 . 6019/PXD003780 . | If the DNA in a single human cell is stretched from end to end it is about two meters long , yet it all fits into a space that is just six thousandths of a millimeter across . This feat is possible because protein complexes package the cell’s DNA into a form called chromatin to make it more compact . One type of chromatin – called “heterochromatin” – is needed to ensure that the DNA is positioned properly inside the cell’s nucleus and segregated correctly when the cell divides . Heterochromatin contains many repeated DNA sequences that are repressed or ‘silenced’ , as well as some active genes . Though heterochromatin accounts for about 25% of the human genome , little is known about the basic molecular processes that occur in this type of chromatin . This is in part because it is not clear which proteins are present in heterochromatin or how these proteins contribute to its structure and roles within the cell . Swenson , Colmenares et al . have now combined two different approaches to search for proteins that are present in heterochromatin and genes that are needed to regulate heterochromatin’s structure . These searches were conducted using fruit fly cells grown in the laboratory , and identified 118 candidate proteins and 374 candidate genes . Next , Swenson , Colmenares et al . looked more closely at 89 of the proteins and confirmed that 30 did indeed localize to heterochromatin . Unexpectedly , more detailed imaging studies showed that these proteins were often localized to restricted regions within heterochromatin ( referred to as subdomains ) . This closer look also revealed that many of the subdomains are dynamic , because the proteins change where they are localized as the cells grow and divide . Finally , many of the candidate proteins were shown to alter the ability of heterochromatin to silence genes . These findings identify a host of new proteins and genes that bind and regulate heterochromatin . More importantly , Swenson , Colmenares et al . reveal that heterochromatin is structurally complex and contains many dynamic , smaller subdomains . The next critical challenges are to find the molecular mechanisms responsible for this unusual organization and to explore the roles of individual heterochromatin proteins or subdomains . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"chromosomes",
"and",
"gene",
"expression",
"cell",
"biology",
"tools",
"and",
"resources"
] | 2016 | The composition and organization of Drosophila heterochromatin are heterogeneous and dynamic |
CRM1 is a highly conserved , RanGTPase-driven exportin that carries proteins and RNPs from the nucleus to the cytoplasm . We now explored the cargo-spectrum of CRM1 in depth and identified surprisingly large numbers , namely >700 export substrates from the yeast S . cerevisiae , ≈1000 from Xenopus oocytes and >1050 from human cells . In addition , we quantified the partitioning of ≈5000 unique proteins between nucleus and cytoplasm of Xenopus oocytes . The data suggest new CRM1 functions in spatial control of vesicle coat-assembly , centrosomes , autophagy , peroxisome biogenesis , cytoskeleton , ribosome maturation , translation , mRNA degradation , and more generally in precluding a potentially detrimental action of cytoplasmic pathways within the nuclear interior . There are also numerous new instances where CRM1 appears to act in regulatory circuits . Altogether , our dataset allows unprecedented insights into the nucleocytoplasmic organisation of eukaryotic cells , into the contributions of an exceedingly promiscuous exportin and it provides a new basis for NES prediction .
The nuclear envelope ( NE ) separates the cell nucleus from the cytoplasm . Although its lipid bilayers are impermeable for macromolecules , embedded nuclear pore complexes ( NPCs ) allow an exchange of material between these compartments ( Feldherr , 1962 ) . The NPC permeability barrier controls this exchange . It grants small molecules a free passage , but becomes increasingly restrictive as the size of the mobile species approaches or exceeds a diameter of ≈ 5 nm ( Mohr et al . , 2009 ) . Shuttling nuclear transport receptors ( NTRs ) are not bound by this restriction ( for review see: Kimura and Imamoto , 2014 ) . They can traverse NPCs by facilitated translocation and have the capacity to ferry even large cargoes , such as newly assembled ribosomal subunits , across the barrier . Active transport of cargoes against concentration gradients requires an intact NE and NPC-barrier for retaining already transported cargoes in the destination compartment . In addition , it requires an input of metabolic energy , typically by the RanGTPase system . The corresponding duty cycles include not only a switch of Ran between its GDP- and GTP-bound states , but also one round of Ran-shuttling between nucleus and cytoplasm . The energetic coupling occurs through a primary RanGTP-gradient , which is generated through NTF2-mediated import of RanGDP , nucleotide exchange by the nuclear RanGEF ( RCC1 ) and cytoplasmic RanGTP-depletion by RanGAP and RanBP1/RanBP2 . This gradient then directly fuels importin- and exportin-mediated transport cycles . Exportins ( reviewed in Güttler and Görlich , 2011 ) bind their cargo molecules cooperatively with RanGTP inside nuclei , carry them as trimeric RanGTP·exportin·cargo complexes to the cytoplasm , where GTP-hydrolysis triggers release of cargo and Ran . The free exportin can then re-enter nuclei and export the next cargo . CRM1 , also called exportin 1 or Xpo1 , is the most conserved nuclear export receptor ( Adachi and Yanagida , 1989; Fornerod et al . , 1997a; Fukuda et al . , 1997; Stade et al . , 1997 ) . It is structurally well characterised , whereby structures of free CRM1 , of certain cargo•CRM1•RanGTP complexes as well as their assembly- and disassembly-intermediates have been solved ( Dong et al . , 2009; Monecke et al . , 2009; Güttler et al . , 2010; Koyama and Matsuura , 2010; Monecke et al . , 2013; Saito and Matsuura , 2013; Koyama et al . , 2014 ) . CRM1 recognises its cargoes through short linear nuclear export signals ( NESs ) , which comprise 4–5 critical hydrophobic ( Φ ) residues with characteristic spacings ( Wen et al . , 1994; Fischer et al . , 1995; Dong et al . , 2009; Monecke et al . , 2009; Güttler et al . , 2010 ) . CRM1 is essential for viability and is the target of the potentially lethal bacterial toxin leptomycin B ( Nishi et al . , 1994 ) , which blocks NES-binding by covalently modifying a conserved cysteine residue within the NES-binding site ( Kudo et al . , 1999; Dong et al . , 2009; Monecke et al . , 2009; Sun et al . , 2013 ) . CRM1 is known to carry multiple cargoes , including newly assembled 40S and 60S ribosomal subunits , the signal recognition particle SRP , U snRNAs , or the genomic RNA of HIV1 via the Rev protein ( Moy and Silver , 1999; Ciufo and Brown , 2000; Ho et al . , 2000; Ohno et al . , 2000; Gadal et al . , 2001; Thomas and Kutay , 2003; Rouquette et al . , 2005 ) . It is also known to regulate key cellular events by conditional export of transcription factors or cell cycle regulators from the nucleus ( see e . g . Hagting et al . , 1998; Yan et al . , 1998; Yang et al . , 1998 ) . Cell nucleus and cytoplasm are prime examples for the division of labour in a eukaryotic cell . The cytoplasm hosts the machineries of the secretory pathway , many metabolic activities as well as the cytoskeletal structures that account for cell motility and long-range transport . It has also specialised in protein synthesis and de novo protein folding . The nucleus lacks protein synthesis and thus depends on protein import from the cytoplasm . It has specialised in DNA replication and repair , nucleosome assembly , transcription , ribosome assembly , as well as in mRNA splicing and processing . Such specialisation critically relies on a spatial separation of interfering activities: Intranuclear protein synthesis , for example , would be a particularly wasteful process , because ribosomes would then also translate unspliced or incompletely spliced mRNAs , consequently read into introns , add inappropriate residues to the nascent chains , eventually encounter premature stop codons and thus produce truncated protein fragments . Such aberrant translation products would not only be non-functional , but probably also toxic , because they fail to fold , or act in a dominant-negative fashion . It is thus not very surprising that eukaryotic cells have implemented several lines of defence against intranuclear translation , whereby the NE acts as a primary barrier to keep cytoplasmic translation activity out of nuclei . Likewise , even though the 40S and 60S ribosomal subunits assemble inside the nucleus , they gain full translation competence only following late maturation steps in the cytoplasm ( reviewed in Panse and Johnson , 2010; Thomson et al . , 2013 ) . A very general problem is , however , that the NPC barrier is not perfect . Instead , also objects larger than the nominal exclusion limit can leak—albeit slowly—into the nucleus ( Bonner , 1975; Mohr et al . , 2009 ) . Such slow mixing of nuclear and cytoplasmic contents would become a problem if the leaked-in proteins would interfere with nuclear functions or dysregulate cellular activities . Countermeasures might be selective degradation or inhibition in the inappropriate compartment , or , when mis-localised to the nucleus , recognition by an exportin and subsequent retrieval to the cytoplasm . Indeed , precedents for such exportin-mediated back-sorting of cytoplasmic proteins from the nucleus are already known . Animal Xpo6 , for example , keeps actin out of the nucleus ( Stüven et al . , 2003 ) , while Xpo4 and Xpo5 do the same for the translation elongation factors eIF5a ( Lipowsky et al . , 2000 ) and eEF1A respectively ( Bohnsack et al . , 2002; Calado et al . , 2002 ) . CRM1 was shown to expel several cytoplasmic factors from the nuclear compartment , including the RanGTPase system components RanBP1 ( Plafker and Macara , 2000 ) and RanGAP ( Feng et al . , 1999 ) as well as the translation factor subunits eIF2β , eIF5 , eIF2Bε and eRF1 ( Bohnsack et al . , 2002 ) . The full extent of active cytoplasmic confinement has , however , not yet been assessed . We report here global scale analyses of nucleocytoplasmic partitioning in Xenopus oocytes and of CRM1-mediated nuclear export . According to stringent criteria , we identified ≈ 1000 potential CRM1 cargoes from Xenopus laevis oocytes , ≈ 1050 , from human HeLa cells , and ≈ 700 from the yeast S . cerevisiae . We tested a subset of cargo candidates for CRM1-dependent nuclear export in cultured human cells and thereby validated a majority of them as true cargoes . For a subset , we also confirmed direct CRM1-interaction and mapped the corresponding NESs , some of which turned out to have unusual features . The majority of identified CRM1 cargoes are proteins and protein complexes with a very strong bias towards a cytoplasmic localisation , suggesting that their active back-sorting from the nucleus is a major cellular activity . This applies to nearly all translation factors ( including the largest translation factor complexes like eIF2B ) , to factors involved in final ribosomal maturation steps , which might prevent ribosomes from acquiring translation competence already in the nuclear compartment , as well as to regulatory proteins , autophagy-linked factors , peroxisome biogenesis factors and to centrosomal proteins , in both , humans and frogs . Another major functional group of CRM1 cargoes with perfect nuclear exclusion is represented by vesicle coat proteins , which points to a strong evolutionary pressure to preclude the budding of vesicles from the inner nuclear membrane . We also identified numerous new instances , where CRM1 appears to act in regulatory circuits . More generally , these data represent a very rich resource for other researchers seeking information about nucleocytoplasmic distribution and CRM1-controlled localisation .
We were interested in a global view of how soluble proteins and protein complexes partition between the nucleus and the cytoplasm . In order to tackle this question , we applied a deep proteome analysis to the isolated compartments . A problem for such endeavour is that standard cell fractionation procedures rely on shearing forces , often combined with hypotonic lysis or even treatment with detergents ( see e . g . Blobel and Potter , 1966; Dignam et al . , 1983 ) . All these treatments compromise the integrity of the NE . Nuclear proteins , which are not firmly associated with solid structures like chromatin , will then leak out and contaminate the cytoplasmic fraction—just as the nuclear fraction will be contaminated by cytoplasmic components . In order to avoid these problems , we turned to Xenopus laevis stage VI oocytes ( Dumont , 1972 ) . These cells measure ≈1 . 3 mm in diameter and have nuclei of ≈450 µm . Such very large dimensions allow for a manual oocyte dissection into nuclear and cytoplasmic fractions with exceptionally little cross-contamination ( see e . g . De Robertis et al . , 1978 ) . These oocyte nuclei are also special with their volume being 100 , 000 times larger than that of average-sized cells with a G2 DNA contents . The chromatin should therefore make no more than a negligible contribution to nuclear retention of proteins . Instead , the nucleocytoplasmic distribution of a given protein or protein complex in these cells should be solely determined by its passive diffusion properties and by their potential to access active nuclear import and/or export pathways . In addition , oocytes are very long-lived cells that grow over months to their final size , which implies that even slow partitioning processes are likely to have reached a steady state . As a standard experiment , we dissected 60 oocytes , cleared the cytoplasmic fractions off yolk , normalised the nuclear and cytoplasmic fractions to their respective volumes , and identified proteins in three biological replicates by mass spectrometry . Proteins of two replicates were separated by SDS-PAGE ( Figure 1A and 1B ) and in-gel digested with trypsin . As a complementary approach , proteins of the third replicate were digested in solution . Resulting peptides were separated by reverse phase chromatography at pH 10 and obtained fractions analysed by LC-MS/MS . The raw data were searched against a comprehensive Xenopus laevis database ( Wühr et al . , 2014 ) . In this way , a total of 9573 proteoforms ( Smith et al . , 2013 ) were identified , 7015 in isolated nuclei and 7036 in the cytoplasmic fractions . The intersection set comprised 4478 proteoforms ( Figure 1C ) . 10 . 7554/eLife . 11466 . 003Figure 1 . Spatial proteomics of Xenopus laevis oocytes . ( A ) Workflow for mass spectrometric analysis of cytosolic and nuclear proteins . For details , see Materials and methods and main text . ( B ) Analysis of obtained cytosolic and nuclear fractions by SDS-PAGE and Coomassie-staining . The loads correspond to 750 nanolitres of either yolk-free cytoplasm or nuclear contents . ( C ) Venn diagram of proteoforms ( including all allelic variants of a given gene product ) that have been identified in the manually isolated cytoplasms and nuclei . ( D ) Venn diagram is similar to ( C ) , but proteoforms corresponding to a given gene ( foremost allelic variants ) have been merged down to ‘unique proteins’ . Also , proteins were subtracted that just co-purified with nuclei , but do not represent intranuclear proteins; this applied to constituents of the nuclear envelope ( ER ) and the nucleus-associated mitochondrial cloud . DOI: http://dx . doi . org/10 . 7554/eLife . 11466 . 00310 . 7554/eLife . 11466 . 004Figure 1—figure supplement 1 . Estimation for the accuracy of mass spectrometric protein quantitation . ( A ) Log-log regression between amounts of added UPS2 standard proteins and measured iBAQ intensities . ( B ) Correlation of measured amounts of UPS2 standard proteins added to either the nuclear or cytoplasmic fraction . DOI: http://dx . doi . org/10 . 7554/eLife . 11466 . 004 Xenopus laevis is pseudotetraploid; thus it shows greater allelic diversity than other species ( Hellsten et al . , 2007 ) , which represents a challenge for peptide-based protein quantification . We therefore treated allelic isoforms not as separate proteins , but mapped all recognizable allelic forms down to unique protein species . This ‘mapping’ was guided also by comparisons to the human and the ( diploid ) Xenopus tropicalis proteomes ( see Materials and methods ) . We aimed at a high-quality dataset for those proteins that can actually pass through NPCs and partition between the nucleus and cytoplasm . We therefore tried to minimize the number of ‘contaminants’ in the list , which represent , for example , the endoplasmic reticulum ( which co-purifies with nuclei in the form of the NE ) or the mitochondrial cloud ( Heasman et al . , 1984 ) that is tightly associated with the outside of these nuclei . We subsequently tested which proteins disappeared from the nuclear fraction when the NE was manually removed . Such proteins were discarded from the list if their sequence features qualified them also as integral membrane , ER-luminal or mitochondrial proteins . This left us with a total of 5006 unique proteins . Of these , 1126 were identified only in the nuclear fraction , 1059 only in cytosolic fraction and 2821 in both compartments ( Figure 1D , and Supplementary file 1 ) . In order to quantify each of these proteins in the nuclear and cytoplasmic compartments , we employed the iBAQ strategy ( Schwanhäusser et al . , 2011 ) combined with an internal universal protein standard ( UPS2; see Materials and methods ) . We estimate that our quantitation was reliable over a range of 5 orders of magnitude in abundance and accurate within a factor of 2 . 3 ( see Figure 1—figure supplement 1 ) . To validate the quality of the obtained nuclear and cytoplasmic fractions , we first evaluated the partitioning of the previously described marker proteins nucleophosmin ( also called B23 or NO38 ) and gelsolin , which are localised exclusively to nucleus ( Schmidt-Zachmann et al . , 1987 ) or cytoplasm ( Yin and Stossel , 1979; Samwer et al . , 2013 ) , respectively . We chose highly abundant proteins , because these yield a larger number of unique , quantifiable peptides and thus allow for a more precise quantification . For nucleophosmin ( a histone chaperone ) , we measured a nuclear concentration of ≈ 2 . 3 µM , a cytoplasmic concentration of 8 nM , and hence a nucleocytoplasmic ( N:C ) partition coefficient of 300:1 . For gelsolin ( a factor that stabilizes cytoplasmic actin in its G-form ) we measured a nuclear concentration of ≈ 0 . 01 µM , a cytoplasmic of ≈ 4 µM , and thus an N:C ratio of 1:400 . This suggests that the obtained nuclear and cytoplasmic fractions show only very limited cross-contamination and also suggested what range of partition coefficients should be expected also for other proteins . Supplementary file 1 contains complete and simplified data sets for the nucleocytoplasmic distribution of 5006 individual proteins and 9573 proteoforms . It lists the measured concentrations in nucleus and cytoplasm as well as ratios on a log10 scale . To make the data as accessible as possible to other researchers , we included not only unique identifiers for each protein hit , but also clickable annotation links to the corresponding UniProt entries for the identified Xenopus proteins ( X . laevis if available , otherwise X . tropicalis ) as well as to the human orthologues , where annotations as of now are more complete . When broken down to functional groups , it becomes evident that the various cellular processes represent quite characteristic N:C distribution patterns . As these patterns are in many cases tightly linked to the activity of CRM1 , we will discuss these after describing our approach of mapping the CRM1-dependent nuclear exportome . An apparently CRM1-independent aspect relates to the energy supply of the giant oocyte cell nucleus . Efficient duty cycles of numerous enzymes , such as Ran , require a high NTP:NDP ratio , which seems hard to maintain inside these nuclei if ATP were produced only outside ( i . e . by mitochondria or cytosolic glycolysis ) . The problem arises from the short half-life of ATP in living cells ( ≈ 1 min ) , which is less time than an ATP molecule would typically need to diffuse into and across such large nucleus . We now found evidence in the oocyte for two parallel solutions to this problem . The first is a ‘creatine-creatine phosphate energy shuttle’ , which uses diffusion of creatine phosphate ( CP ) instead of ATP for a long-range transport of energy equivalents ( reviewed by Bessman and Carpenter , 1985 ) . It exploits the ≈160 times higher phosphorylation potential of CP as compared to ATP , and operates by creatine kinase ( CK ) synthesising CP from ATP near ATP sources and the reverse reaction at sites of ATP consumption . Oocytes enable such CP shuttle by maintaining high CK concentrations in both , cytoplasm and nucleus ( 3 µM and 1 . 5 µM , respectively; Supplementary file 1 ) . The second solution concerns the glycolysis pathway . The oocyte nucleus contains all glycolytic enzymes , except for hexokinase and phosphofructokinase that catalyse the ATP-consuming preparatory steps . The hexokinase level is generally very low in oocytes ( because they produce glucose 6-phosphate by other means; see Nutt et al . , 2005 and Supplementary file 1 ) , while 6-phosphofructokinase ( PFK ) is confined to the cytoplasm ( N:C ≈ 1:2000; Supplementary file 1 ) . This enzyme distribution predicts a directed flux of fructose 1 , 6 bisphosphate into the nucleus . It further suggests that oocyte nuclei produce the high-energy compounds phosphoenolpyruvate and 1 , 3 bisphoglycerate locally and use them for synthesizing ATP and other NTPs . The exportin CRM1 is known to be essential for viability and to account for nuclear export of numerous targets ( see e . g . Xu et al . , 2012b; Thakar et al . , 2013 for a comprehensive summary of so far identified substrates ) . However , it has been unclear how broad the cargo spectrum really is and which set of cellular processes are directly or indirectly controlled by CRM1-dependent nuclear export . In order to close this gap , we set out to identify cargoes in an unbiased manner from three different species and types of cells , namely: the already-mentioned Xenopus laevis oocytes , human HeLa cells and the yeast S . cerevisiae . We first prepared cellular extracts , used immobilised CRM1 as an affinity matrix and asked which proteins and protein complexes would bind in a RanGTP-dependent manner . As detailed below , CRM1 has many interaction partners of widely different abundance and affinity , which implies that the majority of them bind to the immobilized exportin in a highly sub-stoichiometric manner . The challenge thus is to cleanly distinguish such highly sub-stoichiometric bands from non-specific interactors . To this end , we optimised several parameters , such as the way of immobilisation , the exportin:extract ratios ( to minimise competition between cargoes ) , buffer conditions , incubation time , etc . , in order to maximise the signal-to-noise ratio of the affinity chromatography ( for details see Materials and methods ) . Figure 2A shows the starting extracts and CRM1-bound fractions of such affinity chromatographies , and it documents indeed a very large number of protein species that bound to CRM1 in a RanGTP-dependent manner . Using the mass-spectrometric approaches mentioned above , we identified in the starting extracts ≈2800 ( Xenopus ) , ≈3900 ( human ) and ≈2600 ( yeast ) proteins ( Figure 2B ) . In the ‘CRM1+RanGTP’ samples , ≈2300 , 3000 and ≈2000 proteins were identified . About 14% of these had not been detected in the total extracts and these probably represent low-abundance proteins that had been highly enriched during CRM1 affinity chromatography . 10 . 7554/eLife . 11466 . 005Figure 2 . Identification of potential CRM1 cargoes from 3 species . ( A ) Mouse ( mm ) or yeast ( sc ) CRM1 were immobilised through a biotinylated Avi-tag to streptavidin agarose , and incubated with indicated extracts ( 1 ml ) , without or with the addition of 5 μM RanQ69L5-180GTP . The beads were thoroughly washed and subsequently eluted at 45°C with SDS sample buffer ( which leaves the biotin-streptavidin interaction largely intact ) . Analysis of indicated samples was by SDS-PAGE and Coomassie-staining . 1/200 of the starting extracts and 1/5 of eluates were loaded . ( B ) Starting extracts , CRM1 w/o Ran , and CRM1+RanGTP samples were analysed by mass spectrometry . Venn diagrams represent identified unique proteins . Numbers in parenthesis include also proteins that were not identified in a total Xenopus extract or the ‘CRM1+RanGTP’ sample , but in the isolated nuclear and cytoplasmic fractions; these proteins extend the list of ‘CRM1-non-binders’ . DOI: http://dx . doi . org/10 . 7554/eLife . 11466 . 005 The lists of proteins identified in the CRM1+RanGTP-bound fractions include not only true CRM1 binders and CRM1 export cargoes , but , for sure , also false-positive ones . To classify a given hit as a promising candidate , we therefore relied not only on its mere identification in this fraction . We also considered to which extent it became enriched from the input extract ( ‘input enrichment’ ) , to which extent its CRM1-binding had been stimulated by RanGTP ( ‘RanGTP-stimulation’ ) , as well as its absolute abundance in the ‘CRM1+RanGTP’-bound fraction ( ‘Molar fractions in CRM1+RanGTP’ ) ( Figure 3 ) , which affects the accuracy of quantification and thus the reliability of the first two numbers . As a resource for other researchers , we organised the quantitative data in Excel sheets ( Supplementary files 2–4 ) , which contain not only the just mentioned numbers , but also clickable links to the corresponding UniProt entries , as well as cellular localisations derived either from databases ( human and yeast ) or measured directly ( Xenopus ) . 10 . 7554/eLife . 11466 . 006Figure 3 . Categories of CRM1-binders from HeLa cells . For each identified CRM1-binder , we calculated or estimated three parameters from measured iBAQ intensities: its abundance ( molar fraction ) within the ‘CRM1+RanGTP’-bound sample , the RanGTP-stimulation of its CRM1-binding , and how strongly it became enriched by the ‘CRM1+RanGTP’-affinity chromatography . These numbers where then used to group binders into distinct categories , ranging from ‘A’ ( the most probable cargoes ) to ‘non-binders’ . ( A ) Venn diagrams representing the indicated cargo classes with respect to their identification in the starting extract , ‘CRM1 w/o Ran’- and/or ‘CRM1+RanGTP’-bound samples . ( B ) Scatter plot representing ‘CRM1+RanGTP’-binders from HeLa cells , using the parameters ‘RanGTP-stimulation’ and ‘input-enrichment’ as coordinates . Colouring is according to classification . Most ‘non-binders’ had not been identified in the ‘CRM1+RanGTP’ sample; they are therefore also not plotted . Measurement of the parameters ‘RanGTP-stimulation’ and ‘input-enrichment’ required the identification a given candidate in input , ‘CRM1 w/o Ran’ , as well as in the ‘CRM1+RanGTP’- sample . If undetected in either ‘input’ or ‘CRM1 w/o Ran’ , then the missing parameter was estimated as a lower bound ( based on the detection sensitivity of our MS setup ) . Candidates detected only in ‘CRM1+RanGTP’ were not plotted ( because for them , both parameters would have to be estimated ) . ( C ) Scatter plot is as in ( B ) , but colour code is used to indicate the abundance in the ‘CRM1+RanGTP’-bound sample . DOI: http://dx . doi . org/10 . 7554/eLife . 11466 . 006 We divided the identified proteins into several categories , using as a criterion species-specific thresholds ( Supplementary files 2–4 , and Figure 3 ) . These thresholds had been adjusted to best match the behaviour of proteins , whose specific CRM1-interaction is already established beyond doubt , as well as of proteins that are known not to interact with the exportin ( see Materials and methods ) . CRM1-binders of the category A1 not only had to pass the most stringent thresholds in terms of RanGTP-stimulation of CRM1-binding ( ≥500-fold in the case of Xenopus ) and enrichment from the input extract ( ≥3-fold ) , but also belong to the most abundant proteins in the ‘CRM1+RanGTP’-bound fraction ( Note that the threshold of a 500 times stronger signal than in the negative ( minus RanGTP ) control is far more stringent than the standard of ‘2-fold’ , which is used in most proteomics studies ) . A2 had an even stricter threshold for ‘input enrichment’ ( ≥100 ) , but a relaxed one for the apparent RanGTP-stimulation . It includes cargoes like snurportin , which bind CRM1 so strongly that the affinity is still high even in the absence of RanGTP ( Paraskeva et al . , 1999 ) . In category B , one of the three criteria was relaxed , while the category ‘low abundant’ includes proteins that were only detected in the ‘CRM1+RanGTP’-bound fraction , but were not sufficiently abundant to qualify for category A . We assume that most identified proteins from the categories A , B and ‘low abundant’ are direct CRM1 interactors that either carry a functional NES or occur in stable complexes with NES-containing proteins . Thus , we identified ≈1000 probable cargoes each in Xenopus and human as well as ≈700 in yeast . This is far more than identified for any other NTR ( see e . g . Kimura et al . , 2013 ) , and represents ≈20% of all detectable nuclear or cytosolic proteins , only a small fraction of which ( ≈ 10% ) had been proposed to be associated with CRM1 before ( see Supplementary file 5 ) . This suggests that CRM1 serves a far larger number of cargoes than previously assumed . To estimate our positive discovery rate of direct CRM1 cargos , we tested a subset of these candidates below . Of course , nucleoporins , and FG Nups in particular , were also identified as CRM1 ligands . We consider them , however , as part of the transport machinery and not as cargoes . Some of them bind CRM1 very strongly , for example , the human or Xenopus Nup214•88•62 complex or Nup358 ( see Figure 3b , and e . g . Fornerod et al . , 1997b; Engelsma et al . , 2004 ) . On the other end of the distribution , we identified ≈700 proteins in Xenopus , ≈900 in human and ≈600 in yeast , which were strongly selected against in the ‘CRM1+RanGTP’-bound fractions ( Supplementary files 2–4 , Figure 3 ) . These represent proteins , whose nucleocytoplasmic partitioning is probably not directly affected by CRM1 . Metabolic enzymes ( of e . g . glycolysis , the pentose phosphate pathway etc . ) , protein folding factors , and exclusively nuclear proteins are over-represented in this ‘non-binder’ category . In between , we found a broad zone of ‘ambiguous’ proteins , which actually represent a continuum . Some of them bound still very specifically to the export-form of CRM1 ( according to the ‘minus RanGTP’-control ) ; yet , the binding was weak . We assume that these proteins become only transiently CRM1 cargoes ( e . g . in response to the addition or removal of a post-translational modification ) or that they only transiently associate with bona fide CRM1-cargoes . On the other end of the ‘ambiguous’ category , there are proteins , which appear to be ‘CRM1 non-binders’ . However , their low abundance in the starting extract precluded any reliable judgement of how strongly they were selected against in the ‘CRM1+RanGTP’-bound fraction . The data set contains most of the previously well-validated CRM1 cargoes , such as the nuclear import adapter snurportin , or the nuclear export adapters NMD3 ( Ho et al . , 2000 ) and PHAX ( Kitao et al . , 2008 ) . The vast majority of hits ( ≥ 90% ) , however , were so far not linked to CRM1-mediated nuclear export . We therefore decided to verify a sample of candidates according to a common scheme . The first was the Xenopus translation termination factor eRF3a . Its GFP-fusion was exclusively cytoplasmic in transfected HeLa cells ( Figure 4 ) , which is consistent with its experimentally determined localisation in Xenopus oocytes ( Supplementary file 1 ) . A CRM1-block by 10 nM leptomycin B , however , caused the fusion to equilibrate between nucleus and cytoplasm , suggesting a significant nuclear entry rate and rapid CRM1-dependent retrieval in undisturbed cells . 10 . 7554/eLife . 11466 . 007Figure 4 . Validation of Xenopus CRM1-cargo candidates and identification of NESs . HeLa cells were transfected to express GFP- or GFP-NLS-fused candidate proteins , then incubated with or without the CRM1 inhibitor leptomycin B ( LMB ) , fixed , and analysed by confocal laser scanning microscopy ( CLSM ) . The co-transfected RFP-NLS-NES was detected in a separate channel as a control for the LMB-effect . Tested candidates: eukaryotic peptide chain release factor eRF3a ( Q91855 ) , subunit 1b of the ARP2/3 complex ( Q6GNU1 ) , Septin-2 ( B7ZR20 ) , Ap1-gamma subunit of the clathrin-associated adapter complex ( Q6GPE1 ) , the cAMP-dependent kinase type II-alpha regulatory subunit pRKAr2a ( F7CZT8 ) , and the regulator of nonsense transcripts UPF2 ( Q498G1 ) . UniProt entry names are given in parentheses . Figure also shows sequences of identified NESs , and their validations as transfected GFP-NLS-fusions with an LMB-sensitive cytoplasmic localisation . DOI: http://dx . doi . org/10 . 7554/eLife . 11466 . 007 We also transfected a fusion that included an SV40 nuclear localisation signal ( GFP-NLS-eRF3a ) to enforce a faster nuclear entry , which made the fusion indeed exclusively nuclear following Leptomycin B-treatment . In undisturbed cells , however , we observed a still nearly exclusively cytoplasmic localisation , suggesting that the eRF3a NES confers a considerably faster export from nuclei than nuclear import mediated by the ( very strong ) SV40 NLS . We identified this NES within the N-terminal unstructured region of eRF3a and confirmed its nuclear export activity by transfection assays as well ( Figure 4 ) . eRF3a binds also purified CRM1 in a RanGTP-dependent manner ( Figure 5 ) , suggesting that the interaction is direct and not bridged by another factor . Due to its efficient binding from Xenopus oocyte extract to CRM1 , eRF3a was classified as a ‘category A’ cargo . Next , we also confirmed a far lower scoring ‘category B’ candidate , namely the 1b subunit of the Arp2/3 complex , as a directly-binding , bona fide CRM1 export cargo with an NES at its extreme C-terminus . In the light of the rather weak CRM1-binding of Arp2/3-1b ( Supplementary file 2 ) , its NES turned out to be surprisingly strong . When fused to GFP , it behaved like a supraphysiological NES ( Engelsma et al . , 2004 ) and produced pronounced transport intermediates at NPCs ( best visible in weakly expressing cells ) . This difference is , however , plausible in the context of the Arp2/3 complex structure ( pdb 1K8K; Robinson et al . , 2001 ) , which shows this NES packing against the rest of the chain . The rather loose packing and high local B-factor suggest , however , that this NES region is sufficiently mobile to get transiently exposed for subsequent CRM1-binding . 10 . 7554/eLife . 11466 . 008Figure 5 . Identification of cargo candidates as direct CRM1-binders . The H14-ZZ-Sumo tagged candidate proteins ARP2/3 1b ( Q6GNU1 ) , eRF3a ( Q91855 ) , Haus1 ( Q3B8L5 ) , pRKAr2a ( F7CZT8 ) , Septin-2 ( B7ZR20 ) were expressed in E . coli , purified , immobilised on anti-zz beads , and incubated with CRM1 in the absence or presence of RanGTP . Immobilised candidate proteins were released , after washing , by Sumo-protease cleavage and co-eluting materials were analysed by SDS-PAGE ( Note that Septin-2 elution was less efficient than the others ) . An unfused H14-zz-Sumo module served as a negative and a fusion with a PKI-NES as a positive control for CRM1-binding . DOI: http://dx . doi . org/10 . 7554/eLife . 11466 . 008 In total , we tested 29 candidates from Xenopus , human and yeast , and validated 23 of them positive ( Figures 4 , 6 , 7 and 8 ) , suggesting that the majority of hits represent indeed CRM1 cargoes . Negative cases where , for example , the exclusively nuclear replication factor C ( subunit 3 ) or xDDX6 . Explanations could be an issue with NES-modulating post-translational modifications , a transport-independent interaction with CRM1 or that another subunit in a larger complex accounts for CRM1-binding . 10 . 7554/eLife . 11466 . 009Figure 6 . Validations of additional CRM1 cargo candidates from Xenopus . Analysis was as in Figure 4 . ( A ) Tested candidates that behave like true CRM1 cargoes: Asn-tRNA ligase ( Q6DD18 ) , LSM14b ( L14BB ) , COP beta’ ( Q7ZTR0 ) , and Haus1 ( Q3B8L5 ) . ( B ) Tested candidates that are not CRM1 cargoes: the peroxisomal 2 , 4-dienoyl-CoA reductase DECR2 ( Q6GR01 ) , the RNA helicase DDX6 , dynactin 6 ( Q6IRC3 ) and the replication factor complex subunit RFC 3 ( Q4QQP4 ) . DDX6 had been in cargo category A , but probably requires Lsm14 ( see panel A and main text ) for CRM1 interaction . We assume an analogous scenario for DECR2 . Dynactin 6 is in category ‘ambiguous’ and was therefore not considered a CRM1 cargo in the first place . DOI: http://dx . doi . org/10 . 7554/eLife . 11466 . 00910 . 7554/eLife . 11466 . 010Figure 7 . Validation of human CRM1 cargo candidates . Analysis was as in Figure 4 . UniProt identifiers correspond either to abbreviated protein names or are given in parentheses . ( A ) Positively tested CRM1 cargoes: the co-translational methionine aminopeptidase MetAP2 ( MAP2 ) , PEX5 , SRP54 , PEX19 , the Ser-tRNA ligase ( SYSC ) , CDC37L ( CD37L ) , and ATG3 . ( B ) The Cys-tRNA ligase ( SYCC ) was classified as a ‘CRM1-non-binder’ and accordingly shows a CRM1-independent nuclear exclusion . DOI: http://dx . doi . org/10 . 7554/eLife . 11466 . 01010 . 7554/eLife . 11466 . 011Figure 7—figure supplement 1 . Validation of human CRM1 non-binders . ( A ) Re-tested CRM1 non-binders , which showed a nucleocytoplasmic equilibration: the F-box only protein 7 ( FBX7 ) , the 26S proteasome non-ATPase regulatory subunit 10 ( PSD10 ) , and the Adapter molecule CRK ( CRK ) . Note that leptomycin B shifted none of them to a more nuclear localization . Analysis was as in Figure 4 . ( B ) The following proteins were tested: the m7GpppX diphosphatase ( DCPS ) , the Cyclin-dependent kinase inhibitor 1 ( CDN1A ) , the mesoderm induction early response protein 1 ( MIER1 ) , and the Acidic leucine-rich nuclear phosphoprotein 32 family member B ( AN32B ) . Note that all these proteins showed an exclusively nuclear localization already without leptomycin B treatment , which is consistent with the assumption of negligible steady state export . DOI: http://dx . doi . org/10 . 7554/eLife . 11466 . 01110 . 7554/eLife . 11466 . 012Figure 8 . Validation of CRM1 cargo-candidates from S . cerevisiae . Analysis was as in Figure 4 and included: the translation initiation factor eIF3j , the ribosome biogenesis factors LSG1 , REI1 , and ENP1 , the α-subunit of the Phe-tRNA ligase ( SYFA ) , as well the peroxisome biogenesis factor PEX19 . This positive validation of yeast cargoes in HeLa cells also emphasises the extreme conservation of NES-recognition by CRM1 . DOI: http://dx . doi . org/10 . 7554/eLife . 11466 . 012 We reasoned that the latter scenario might apply to xDDX6 , which has been reported to interact with Lsm14 ( isoforms a and b; Tanaka et al . , 2006; Arthur et al . , 2009 ) . Indeed , the transfected Lsm14b-GFP fusion was exclusively cytoplasmic , but shifted upon leptomycin B treatment to a nuclear localisation . Given that Lsm14 had even better CRM1-binding scores than xDDX6 and that the two factors were recovered in ≈1:1 stoichiometry within the ‘CRM1+RanGTP’-bound fraction ( Supplementary file 2 ) , it is most likely that Lsm14 is the direct CRM1-binder , while xDDX6 piggybacks . Similar considerations will probably apply to numerous other cargo candidates that occur in complexes with other proteins . Supplementary files 2–4 provide the information for interpreting such cases , because they list not only the already mentioned binding-scores , but also estimate the molar ratios in which cargo candidates were recovered in the ‘CRM1+RanGTP’-bound fractions . The sheer number of CRM1 cargoes already suggests a very broad impact of this nuclear export pathway on cellular physiology . Yet , there is a remarkable bias towards or against individual functional categories ( Figure 9 ) . Highly abundant metabolic enzymes ( including glycolytic enzymes ) , for example , are greatly under-represented amongst the CRM1 cargoes . Here , it is remarkable that many of these CRM1 non-cargoes , including enzymes that are part of large complexes , show a rather even distribution between cytoplasm and nucleus of the Xenopus oocyte without having a detectable NLS ( Supplementary file 2 , sheets ‘glycolysis’ and ‘metabolic enzymes’ ) . This suggests that a large size alone cannot guarantee a cytoplasmic confinement . 10 . 7554/eLife . 11466 . 013Figure 9 . Correlation between functional groups , nucleocytoplasmic partitioning and CRM1-interaction . ( A ) Panel shows a ‘density plot’ to illustrate how many unique proteins ( see Figure 1D ) show a given N:C partition coefficient in Xenopus oocytes . The yellow area covers all proteins , the blue area only proteins that belong to CRM1 cargo categories ‘A’-‘C’ . ( B–L ) Density plots are analogous to ( A ) , but each panel represents just one functional group and the curves were re-scaled to account for the smaller number of proteins in a given group . Functional groups were initially defined by KEGG BRITE hierarchies and then manually refined ( see Supplementary file 2 for included proteins ) . DOI: http://dx . doi . org/10 . 7554/eLife . 11466 . 013 Translation initiation factors represent the other extreme ( Supplementary files 2–4 , sheets ‘Translation factors’; Figure 9B ) . eIF2 , eIF2B , eIF4B , eIF4G , eIF5 and eIF5B all behaved like high-scoring cargoes in either Xenopus , human or yeast , which coincides nicely with their complete exclusion from the nuclei of Xenopus oocytes . We assume that CRM1 serves here the purpose of suppressing intranuclear translation and possibly also avoiding an interference with ribosome biogenesis if translation factors bound pre-maturely to ribosomal assembly intermediates . Leakage from the cytoplasm into the nuclear compartment should be fast for small individual proteins , but slow for large entities . It is therefore remarkable that even very large translation factor complexes behaved like nuclear export substrates , examples being the 125 kDa eIF2αβγ complex ( 150 kDa with tRNA ) or the 270 kDa eIF2Bαβγδε complex ( Supplementary files 2–4 , sheets ‘Translation factors’ ) . This suggests that even the presumably slow leakage of large cytoplasmic assemblies into nuclei can be so deleterious that countermeasures are required . None of the translation elongation factors appeared to be a convincing CRM1 cargo , possibly because appending an NES to the elongation factors might be incompatible with efficient translation . Nevertheless , they are also subject to exportin-mediated nuclear export . Human EF1A was previously identified as the most prominent export cargo of Xpo5 , while the elongation factor for proline-rich regions , eIF5A , is exported by Xpo4 ( Lipowsky et al . , 2000; Bohnsack et al . , 2002; Calado et al . , 2002 ) . It is thus well possible that other translation factors that fail to interact with CRM1 also use a more specialized nuclear export pathway . The translation termination factor eRF3a is again a bona fide CRM1 cargo ( see Figure 4 ) , as is eRF1 ( Supplementary file 2 and Bohnsack et al . , 2002 ) . Furthermore , CRM1-dependent nuclear depletion also applies to several translation-associated factors , such as the signal recognition particle SRP , where SRP54 appears to be the NES-carrying component ( Figure 7 ) , or the human start-methionine aminopeptidase MetAP2 , which binds translating ribosomes and cleaves the start methionine from nascent polypeptides . The normally strictly cytoplasmic MetAP2 accumulates upon leptomycin B treatment in nucleoli ( Figure 7 ) , not only validating it as a true CRM1 cargo , but also suggesting that it can bind to assembling ribosomal subunits prematurely and possibly interfere with the maturation process . CRM1 is known to be essential for ribosome biogenesis . It exports pre-60S ribosomal subunits ( rSUs ) through the export adapter Nmd3 ( Ho et al . , 2000 ) , while Ltv1 and Rio2 behave like export adapters for yeast 40S rSUs ( Seiser et al . , 2006; Fischer et al . , 2015 ) . Yeast Arx1 and the Mex67•Mtr2 dimer also bind pre-60S rSU subunits and promote export through direct interactions with FG repeats ( Bradatsch et al . , 2007; Yao et al . , 2007 ) . Furthermore , Ecm1 , Bud20 , Alb1 , Tif6 , Rlp24 , Nog1 , and Mtr4 are known to escort 60S rSUs , while Nob1 and Enp1 have been shown to accompany 40S rSUs to the cytoplasm , without remaining a constituent of mature ribosomes ( reviewed in Panse and Johnson , 2010; Thomson et al . , 2013 ) . We now found that yeast Enp1 ( essential nuclear protein 1 ) behaves like an autonomous CRM1 substrate ( Figure 8 ) , suggesting that it might actually act as yet another adapter for CRM1-mediated export of 40S rSUs . The use of multiple export adapters , as in here , might not just be an issue of robustness and redundancies , but a more fundamental requirement for making such large cargoes sufficiently ‘soluble’ in the gel-like permeability barrier of NPCs ( Ribbeck and Görlich , 2002; Schmidt and Görlich , 2015 ) . Furthermore , our analysis revealed that additional predominantly nuclear , ribosomal biogenesis factors behave like CRM1 cargoes , namely: Brx1 , Pno1 , Tsr1 , Rpf1 , Rlp7 , Ytm1 , Tsr3 , Rix7 , Erb1 , Nsa1 , Ssf1 , Dim1 , Nol10 , Loc1 , and Rpf2 ( see Supplementary file 4 sheet ‘Ribosome biogenesis’ ) . Thus , they might also escort rSUs to the cytoplasm , and possibly facilitate the export process by providing additional binding sites for CRM1 . Newly exported 40S and 60S rSUs acquire translation competence only after a series of maturation steps in the cytoplasm . In the case of yeast 60S rSUs , this involves the activities of Drg1/ Afg2 , Rei1 , Reh1 , Jjj1 , Yvh1 , Lsg1 , Efl1/ Ria1 as well as of Sdo1 ( reviewed in Panse and Johnson , 2010; Thomson et al . , 2013 ) . While some of these might get loaded already inside the nucleus , it appears that Drg1 , Rei1 , Reh1 , Lsg1 , and Ria1 act exclusively in the cytoplasm . All these factors showed a strong and strictly RanGTP-dependent interaction with CRM1 ( Supplementary file 4 sheet ‘Ribosome biogenesis’ ) . This suggests that CRM1 keeps them cytoplasmic , possibly to avoid the occurrence of translation-active ribosomes inside the nucleus ( at least Lsg1 and Rei1 behave like true CRM1 cargoes , see Figure 8 ) . A cytoplasmic confinement of these factors appears crucial for yet another reason: They displace the export mediators Arx1 , Nmd3 , Mex67 , and Mtr2 from newly exported 60S species ( Loibl et al . , 2014 ) . A premature intra-nuclear displacement by mis-localized cytoplasmic maturation factors could thus abort ribosomal export . Cytoplasmic maturation also involves the incorporation of additional ribosomal proteins , a prominent example being rpS3 . rpS3 is highly abundant in the ‘CRM1+RanGTP’-bound fraction , and thus perhaps hindered ( directly or indirectly ) by CRM1 to bind 40S rSUs prematurely already inside nuclei . The same might apply to rpL24a . These enzymes are essential for translation , and a cytoplasmic confinement of tRNA-charging would also contribute to the exclusion of intranuclear translation . Nevertheless , nuclear aminoacylation has been proposed as a proof-reading step for correct pre-tRNA processing and maturation prior to export ( Lund and Dahlberg , 1998; Sarkar et al . , 1999 ) . This proposal was based on the observations that blocking aminoacylation in yeast results in nuclear accumulation of tRNA and that a similar treatment in Xenopus oocytes prevents tRNAs , which had been injected into the nucleus , from reaching an exclusively cytoplasmic steady state distribution . There are , however , also arguments against this scenario , foremost that a loss of cytoplasmic retention of tRNA by translation elongation factor eEF1A ( which binds only aminoacylated tRNA ) would explain the phenotypes as well . Furthermore , a structural and functional analysis of tRNA•Xpo-t•Ran export complexes revealed that this exportin proof-reads a correct 3'-CCA end ( Arts et al . , 1998; Lipowsky et al . , 1999 ) , but cannot sense aminoacylation ( Cook et al . , 2009 ) . The latter applies also to the alternative tRNA exporter Xpo5 ( Bohnsack et al . , 2002; Calado et al . , 2002 ) . To clarify this issue at least for Xenopus oocytes , we analysed the nucleocytoplasmic distribution of the aminoacyl tRNA ligases and found a strong cytoplasmic bias ( Figure 9C; Supplementary file 2 , sheet ‘Aminoacyl tRNA ligases’ ) . The cytoplasmic concentration exceeded the nuclear one by more than a factor of 100 in most cases , and even the least excluded ones had ≈10 times higher levels in the cytoplasm than in the nucleus . This makes is rather unlikely that nuclear aminoacylation is a general proof-reading criterion prior to tRNA export . At least the Ser , Thre , and Asn tRNA-ligases showed a strong RanGTP-dependent interaction with CRM1 . We tested the Xenopus Asn tRNA ligase by transfection assays and observed a perfect CRM1-dependent nuclear exclusion ( Figure 6 ) . Human Ser tRNA ligase and the α-subunit of the yeast Phe tRNA ligase behave the same way ( Figures 7 and 8 ) , suggesting that cells make a true effort to keep tRNA aminoacylation cytoplasmic . We also tested the human Cys tRNA ligase that failed to interact with CRM1 ( Supplementary file 3 ) . Nevertheless , the transfected GFP-fusion protein showed a perfect nuclear exclusion , which was also not impaired by leptomycin B-treatment ( Figure 7B ) . How this cytoplasmic confinement is maintained is still unclear , but the formation of larger complexes or alternative export pathways are plausible possibilities . In contrast to a CRM1-dependent confinement , the cytoplasmic localisation can , however , not be maintained against a fused SV40-type NLS . These transfection experiments also exemplify the validation of a ‘CRM1-non-binder’ ( for more such validations , see below ) . We identified several high-scoring CRM1 cargoes amongst cytoplasmic mRNA degradation factors , components of P-bodies and stress granules . These include Upf1 , Upf2 and Upf3 ( Supplementary files 2–4 and Figure 5 ) , which function in nonsense-mediated decay ( NMD ) of incorrectly spliced mRNA during a pioneering round of translation ( reviewed by Popp and Maquat , 2013 ) , as well as the de-capping enzymes ( DCP1 and DCP2 ) and enhancers of de-capping ( EDC proteins ) , which typically reside within P-granules and initiate mRNA degradation ( reviewed by Parker and Sheth , 2007 ) . Thus , CRM1 probably enables cells to control nuclear and cytoplasmic RNA turnover independently from each other . This CRM1 function appears very well conserved from yeast to human . So far , no connection has been made between CRM-mediated nuclear export and vesicle formation along the secretory pathway ( reviewed by Kirchhausen , 2000 ) . Yet , we observed that COPI and COPII coat proteins as well as , e . g . , the AP-1 and AP2 adapter complexes or the AP180 clathrin coat assembly protein behave like CRM1 cargoes ( Supplementary files 2–3 , sheets ‘Vesicle coat proteins’; as well as Figures 4 and 6 ) . They are also extremely well excluded from the oocyte nucleus ( Figure 9E ) . A formation of intranuclear vesicles was so far observed only in oocytes of rather exotic species , such as the ascidian Botryllus schlosseri ( Manni et al . , 1994 ) . The absence in other cell types can now be explained by an active depletion of the vesicular budding machineries from the nuclear interior . The data also suggest an unanticipated connection of CRM1 to autophagy ( reviewed in Reggiori and Klionsky , 2013 ) . Atg1 , Atg13 , Atg17 , and Vps30 are all required for autophagy in yeast and all of them are high-scoring CRM1 cargoes ( Supplementary file 4 , sheet ‘Autophagy’ ) . It thus appears as if CRM1 counteracted an initiation of autophagy from the nuclear interior . The situation is very similar in human cells , though the spectrum of CRM1-interacting autophagy components is slightly different ( Supplementary file 3 sheet ‘Autophagy’ ) . For example , here the ATG8-conjugating enzyme ATG3 is a major CRM1-cargo ( Figure 7 ) . Our data set also revealed high-scoring CRM1 cargoes that make an unexpected link between nuclear export and protein import into peroxisomes ( reviewed in Ma et al . , 2011 ) : The peroxisomal targeting ( PTS1 ) receptor Pex5 was not only identified as a high-scoring CRM1-binder , but also showed a strictly CRM1-dependent nuclear exclusion ( Figure 7 ) . This connection might point to a general challenge for post-translational transport from the cytosol , namely that diffusive transport will not necessarily lead to the destination organelle , but also to and possibly into nuclei . In the case of peroxisomal proteins , this poses a particular danger as many of them produce reactive oxygen species that might damage the genome . A first line of defence against such incidents is a trapping of peroxisomal proteins by dedicated targeting receptors . The resulting receptor•substrate complexes , however , still need to reach peroxisomes by diffusive transport . If this fails and the complex ends up inside nuclei , then CRM1-mediated export can rectify the problem and give the targeting complex another chance to reach its correct destination . PEX19 from yeast or human is another example , which behaves like a perfect CRM1 cargo and shows a CRM1-dependent nuclear exclusion ( Figures 7 and 8 ) . PEX19 targets membrane proteins to peroxisomes and allows pre-peroxisomes to bud from the ER ( reviewed in Ma et al . , 2011 ) . A mis-targeting of PEX19 to nuclei by a fused NLS causes interesting consequences , namely nuclear accumulation of newly synthesised peroxisomal membrane proteins ( Sacksteder et al . , 2000 ) . It now seems very likely that a loss of CRM1-mediated export would have the same effect . Apart from a cytoplasmic confinement of e . g . VASP or the Arp2/3 complex ( Supplementary file 3 , Figure 5 ) , it appears that CRM1 has only little direct impact on the actin cytoskeleton , which is consistent with the fact that metazoans possess a dedicated exportin ( Xpo6 ) to deplete actin from their nuclei ( Stüven et al . , 2003 ) . Xenopus and human septins , in contrast , are highly abundant amongst the CRM1-bound proteins . We validated Septin 2 and found that leptomycin B treatment of transfected cells caused indeed a shift from a cytoplasmic to a nuclear localisation ( Figure 5 ) . Septins interconnect actin and microtubule networks , function in cytokinesis , formation of cilia and defence against pathogens ( Mostowy and Cossart , 2012 ) , and it appears that their nuclear accumulation needs to be actively suppressed . The data also suggests some contribution of CRM1 to nuclear exclusion of α- and β-tubulin in metazoan cells ( Supplementary files 2 and 3 ) . We further found high-scoring CRM1 cargoes amongst the components of the centrosomes ( Figure 9F ) , which function as microtubule organizing centres ( reviewed by Bornens , 2012 ) . These strong CRM1 interactors include the HAUS augmin complex , γ-tubulin and γ-tubulin complex components as well as a number of additional centriolar proteins ( e . g . CEP41 , CEP55 , and CEP170 ) . This could point not only to an active suppression of microtubule nucleation in metazoan interphase nuclei , but also to additional mitotic function of CRM1 ( see Arnaoutov et al . , 2005 ) —namely at centrosomes . Almost every aspect of cellular physiology is under the control of protein kinases , phosphatases or the ubiquitin/ proteasome system . Our data now suggest that CRM1 is more heavily involved in such regulatory circuits than previously thought . Alone in HeLa cells , we found ≈ 70 kinases as high-scoring CRM1 cargoes ( Supplementary file 3 ) , only a fraction of which had been described as such before . The new cargoes include heavily studied kinases such as protein kinase A , the interleukin-1 receptor-associated kinase 1 , the pro-apoptotic serine/ threonine kinase 3 , the tank-binding kinase 1 , Raf-1 , and several isoforms of casein kinase I and II . In addition , we identified numerous kinase regulators as new CRM1 cargoes , examples being the already mentioned regulatory subunit of PKA ( PRKAr2a; Figures 4 and 5 ) or the TSC1•TSC2 complex ( Supplementary file 2 ) , which is a key component of the mTOR signalling pathway ( Laplante and Sabatini , 2009 ) . Such action of CRM1 will contribute to compartment-specific phosphorylation patterns , but in many cases also exert control by granting or denying those kinases access to their substrates . Likewise , numerous phosphatases and components of the ubiquitin or the ubiquitin-like modifier system showed up as potential new CRM1 cargoes , alone in human cells 22 and 57 , respectively ( Supplementary file 3 , Figure 9H and 9K ) . CRM1-dependent nuclear export signals are usually short ( 9–15 residues long ) peptides with 4–5 hydrophobic Φ residues that are spaced according to characteristic patterns ( Wen et al . , 1994; Fischer et al . , 1995; Güttler et al . , 2010; Xu et al . , 2012a ) . The most common one is a PKI-type NES . Several other spacings are also allowed , which can be explained either by such NESs binding in a different conformation to CRM1 ( snurportin-type or Rev-type NES ) or by skipping one Φ-residue in a 5Φ NES . A functional NES also needs to be solvent-exposed and not buried in a globular fold . We applied these criteria to identify NESs in a set of validated , new CRM1 cargoes . We were successful in six cases ( Figure 5 ) , but failed with some others ( e . g . Pex5 , rpS3 , Haus1 , Cop beta’ , MetAP2 , CDC37L , Enp1 , Phe tRNA ligase α ) . These negative cases illustrate how difficult a reliable NES prediction still is . These difficulties originate from at least two problems . First , many published NESs turned out to be incorrect ( see e . g . Xu et al . , 2012b and discussion therein ) . NES prediction will therefore remain unreliable and produce frequent false-positive hits as long as it is based on an unreliable list of positive cases . Second , we probably miss true NESs , because we do not yet know all NES patterns . Our study should now provide solutions to these limitations . First , we provide large test sets for benchmarking current and future NES predictions , first of all a total of ≈1300 ‘category A cargoes’ , which should contain an NES ( in the case of oligomers at least one per complex ) . The subset of such true CRM1 cargoes , which lack a so far recognizable export signal , provides an ideal starting point for identifying NESs that conform to new patterns . We also found ≈2200 clear non-CRM1-binders that can serve as a negative control group for NES prediction . The just mentioned non-binders include several cases ( 11 ) that have been listed as CRM1-cargoes in the NESdb ( Supplementary file 5; Xu et al . , 2012b ) . This could now point to the false-positives in the NESdb or to contaminating false-negatives in our dataset . To address this issue , we analysed seven of these conflicting cases and tested the behaviour of the corresponding GFP fusions in the HeLa cell transfection assay ( Figure 7—figure supplement 1 ) . All of them showed a considerable nuclear signal already in untreated cells and this nuclear signal did not further increase upon blocking CRM1 by leptomycin B . These seven re-tested candidates thus behaved as predicted from their classification as CRM1 non-binders , at least for the tested protein isoforms ( cloned from HeLa cell cDNA ) , for the tested cell type ( human HeLa cells ) and under standard cell culture conditions . An interesting twist to CRM1-mediated export is that it can be regulated by post-translational modifications . Cyclin B , for example , is kept cytoplasmic by CRM1 until prophase , when the NES gets inactivated by phosphorylation of adjacent serine residues and the protein suddenly accumulates inside nuclei ( Yang et al . , 1998 ) . PHAX , the export adapter for U snRNAs , on the other hand , requires phosphorylation for an efficient interaction with CRM1 ( Kitao et al . , 2008 ) . Given these examples , it will now be very interesting to obtain a global view on how phosphorylation impacts individual cargo-CRM1 interactions , for example , by testing in how far CRM1 selects for or against the corresponding phosphoforms . Surprising outcomes of our study not only were the sheer number of CRM1 cargoes ( ≈ 1/4 of all detectable cytoplasmic and nuclear proteins ) , but also that the majority of cargoes are actually exclusively cytoplasmic proteins ( Figure 9A; Supplementary files 2–4 ) . This poses the questions of why there is a need to actively maintain a cytoplasmic localization and why evolution has not come up with a better barrier system ? The perhaps best answer is that is impossible to preclude nuclear entry of unwanted proteins in the first place , because there are several leakage routes into the nucleus . First , the NPC permeability barrier is per se probably imperfect and allows leakage: It is based on the sieving effect of reversibly crosslinked FG repeat domains and probably represents compromise in the sense that a tighter barrier would also restrict facilitated transport ( see Schmidt and Görlich , 2015 and discussions therein ) . In addition , there are situations , where even the most perfect NPC barrier gets bypassed . The open mitosis in metazoans , for example , leads to a complete mixing of nuclear and cytoplasmic contents and requires an unmixing following reformation of the nuclear envelope . Nuclear proteins require re-import , while obligatory cytoplasmic proteins , such as translation factors , components of the vesicular transport machinery and many others , are obviously subject to active nuclear export . Another source of leakage might be NPC assembly , where inner and outer nuclear membrane might already fuse to a pore , before all FG Nups are in place to maintain a permeability barrier . Finally , cells should survive a temporary damage of their nuclear envelope , which is known to occur rather frequently in certain cultured cancers cells ( Hatch et al . , 2013 ) , but might also happen in normal cells , in particular when exposed to mechanical stress . Such incidents not only require a repair of the NE , but also a rapid unmixing of nuclear and cytoplasmic contents by active import as well as by active export . Thus , the barriers of the NE alone cannot maintain the compartment identities . Instead , a robust separation of nuclear and cytoplasmic contents requires active corrective mechanisms , whereby the exportin CRM1 appears to play a very central role .
Manual microdissection of the oocytes was performed in ‘5:1/HEPES buffer’ ( 10 mM HEPES/ KOH pH 7 . 5 , 83 mM KCl , 17 mM NaCl , supplemented with Roche complete protease inhibitor ( EDTA-free ) as previously described ( Liu and Liu , 2006 ) . Isolated nuclei were gently washed several times , and proteins were recovered by ethanol precipitation . The respective cytoplasmic fractions were diluted with ‘5:1/HEPES buffer’ and homogenized with a pestle . Pigments , yolk and membranes were removed from the extract by two rounds of centrifugation ( 17 , 000 g , 15 min , 4°C ) . Note that thereby also insoluble protein complexes , such as intermediate filaments were removed , and that our analysis explicitly aimed at the nucleocytoplasmic partitioning of soluble complexes and proteins . We therefore also excluded proteins that co-purified with nuclei through their association with the nuclear envelope ( NE ) . Such proteins were identified and ‘flagged’ by a manual dissection of nuclei into nuclear interiors and crude NEs ( kindly performed by Volker Cordes ) and by asking which ‘nuclear’ proteins did not fractionate with the nuclear interiors . A total of 775 unique proteins ( see below ) were flagged as possibly not being cytosolic or intranuclear , 748 of them were removed from the list of Supplementary file 1 , while 27 were kept in the final list because they were manually qualified as cytosolic or intranuclear proteins according to previous literature , or according to their subcellular localization at Human Proteome Atlas ( Uhlén et al . , 2015 ) . Quantitative mass spectrometric analysis of the obtained nuclear and cytosolic fractions involved three biological replicates with two technical replicates each . For quantification of nucleocytoplasmic partitioning , we considered that a nucleus is 10-fold smaller in volume than a yolk-free cytoplasm . Therefore , we compared for each analysis ≈60 nuclei with ≈6 cytoplasms . In these experiments , we used biotinylated CRM1 versions carrying an Avi-tag , which is an optimised biotin-acceptor sequence for enzymatic biotinylation by BirA ( Schatz , 1993 ) . HeLa S100 extract ( Abmayr et al . , 2006 ) and Saccharomyces cerevisiae whole cell extract ( Gottschalk et al . , 1999 ) were kindly supplied by the group of Reinhard Lührmann . Xenopus laevis oocyte extract was prepared as described ( Leno et al . , 1996 ) . Xenopus laevis oocyte , Saccharomyces cerevisiae and cytosolic HeLa extracts were diluted 1:5 in binding buffer ( 20 mM HEPES/ NaOH pH 7 . 5 , 90 mM KAc , 2 mM MgOAc , 250 mM Sucrose , 5 mM DTT ) and cleared by a 1 hr centrifugation step in a S55A rotor at 4°C and 37 , 000 rpm . Supernatants were incubated with Phenyl-Sepharose ( low substitution ) for the selective depletion of endogenous nuclear transport receptors as described previously ( Ribbeck and Görlich , 2002 ) . The flow-throughs were incubated with 0 . 1 μg/ml RNAse A for 20 min on ice ( to detach ‘indirect’ cargoes that interact through RNA with direct ones ) and then 100 units/ml RNasin ( Promega ) were added . 1 ml of a such treated extract was supplemented with 5 μM RanGTP ( hsRanQ69L5-180 ) , centrifuged in a S45A rotor for 1 hr at 4°C at 37 , 000 rpm . 500 pmoles mmCRM1 or scCRM1 were immobilized on 20 μl streptavidin agarose beads ( Sigma Aldrich ) . Free biotin-binding sites were quenched thereafter with 1 mM biotin , and the beads were rotated for 3 hr at 4°C with the RanGTP-supplemented extract . Beads were then washed three times with 500 μl binding buffer . Bound material was eluted with 60 μl SDS sample buffer at 45°C . Minus CRM1 and minus RanGTP controls were processed analogously . Input extracts as well as elutions were analysed by SDS-PAGE followed by Coomassie-staining and/or analysed by mass spectrometry as described below . An overview of the sample preparation for mass spectrometry and LC-MS/MS instrumentation: Analysis of CRM1 binders ( all species ) Analysis of nucleocytoplasmic partitioningReplicate 1Replicate 2Replicate 3Sample preparationSDS-PAGE , digestion with trypsinSDS-PAGE , digestion with trypsinSDS-PAGE , digestion with trypsinDigestion with Lys-C and trypsin , reverse phase HPLC at pH 10LC-MS instrumentationDionex Ultimate 3000 HPLC Q-Exactive HFEASY nLC-1000 Q-ExactiveDionex Ultimate 3000 HPLC Q-Exactive HFDionex Ultimate 3000 HPLC Orbitrap Fusion To estimate absolute protein concentrations , Universal Proteomics Standard-2 ( UPS2; Sigma-Aldrich ) was added to the analysed samples ( at a 1:10 ( w/w ) ratio between the standards and the total sample protein ) . Since UPS2 is based on human proteins , this standard was only employed for the non-human samples . In one workflow , proteins were first separated by SDS-PAGE ( 4-12% Bis/Tris gradient mini-gel , NuPAGE , Novex ) and visualized by colloidal Coomassie-staining . Proteins were then in-gel digested as described before ( Shevchenko et al . , 2006 ) with minor modifications . Briefly , proteins were reduced with 10 mM DTT for 30 min at 55°C , and then alkylated with 55 mM iodoacetamide ( IAA ) in 50 mM ammonium bicarbonate ( BC ) for 20 min at 26°C in the dark . Protein digestion was performed overnight at 37°C at a 1:50 ( w/w ) trypsin ( Promega #V5111 ) to protein ratio . Following digestion , peptides were extracted from the gel pieces , and concentrated by vacuum evaporation of the solvent in a SpeedVac to near dryness . Dried peptides were dissolved in 20 µl of 1% ( v/v ) formic acid , and 6 µl were analysed LC-MS/MS for each technical replicate . In the second workflow , ethanol-precipitated proteins were dissolved in 1% ( v/v ) RapiGest SF surfactant ( Waters # 186002122 ) at 70°C for 10 min . Proteins were then reduced with 5 mM DTT in 50 mM BC for 30 min at 50°C , and alkylated with 10 mM IAA in 50 mM BC for 20 min . Excess IAA was reacted with an additional 5 mM DTT at RT for 20 min . Proteins were first digested with Lys-C ( Roche , 1:100 enzyme to protein ratio ) for 4 hr at 37°C , then overnight with trypsin at a final RapiGest concentration to 0 . 1% . Following digestion , the samples were acidified with trifluoroacetic acid ( final concentration of 1% , v/v , 37°C , 1 h ) to break down the RapiGest surfactant , the resulting by-products were pelleted by centrifugation ( 13 , 000 rpm , 15 min , RT ) , and the supernatant containing the digested peptides was transferred to a new tube . Peptides were desalted on reversed phase-C18 solid-phase extraction cartridges ( SPE; SepPak , Waters ) and concentrated in a SpeedVac to near dryness . Then , peptides were resuspended in 10 mM ammonium hydroxide ( pH 10 ) and loaded onto a reverse phase HPLC column ( XBridge C18 , Waters , 3 . 5 µm , 1 . 0 mm x 150 mm ) and eluted in a 5-35% ( v/v ) acetonitrile gradient at a flow rate of 60 µl/min . 45 initial fractions were collected , which were combined into 17 peptide pools . Each pool was concentrated as described above and dissolved in 20 µl 1% FA ( v/v ) . 6 µl each were then analysed by LC-MS/MS for a technical replicate . First , extracted peptides were loaded onto an in-house packed C18 ‘trapping’ column ( 0 . 15 mm x 20 mm , Reprosil-Pur 120 C18-AQ 5 μm , Dr . Maisch GmbH , Germany ) . Then a second C18 column was connected in tandem ( an analytical C18 capillary column; 0 . 075 mm x 250 mm column self-packed with 3 μm Reprosil-Pur 120 C18-AQ ) . Peptides were then eluted using a 105 min linear gradient ( 5– 35% acetronitrile in 0 . 1% FA at 300 nl/min ) on an EASY nLC-1000 system in-line coupled to a Q Exactive hybrid quadrupole/orbitrap mass spectrometer ( Thermo Scientific , Dreieich ) . The instrument was operated in data-dependent acquisition mode with a survey scan resolution of 70 , 000 at m/z 200 and an AGC target value of 1 x 106 . Up to 15 of the most intense precursor ions with charge state 2 or higher were sequentially isolated at an isolation width of 2 . 0 m/z for higher collision dissociation ( HCD ) with a normalized collision energy of 25% . Dynamic exclusion was set to 30 s to avoid a repeating sequencing of the same precursor ion . The LC setup was as described above , but a Dionex Ultimate 3000 HPLC ( Thermo Scientific , Dreieich ) and a 350 mm capillary C18 column were used . Both mass spectrometers were operated in data-dependent acquisition mode with a survey scan resolution of 120 000 ( for Orbitrap Fusion ) or 60 000 ( for Q-Exactive HF ) at m/z 200 , with an AGC target of 1 x 10e6 . Up to 30 of the most intense precursor ions with charge state 2 or higher were sequentially isolated for HCD with normalized collision energy of 27% ( Q Exactive HF ) or 30% ( Orbitrap Fusion ) , respectively . MS/MS scans were recorded in the Orbitrap for Q Exactive HF , and in the LTQ ion trap for the Orbitrap Fusion . Dynamic exclusion was set to 50 s . MS raw files were processed with the MaxQuant software package ( version 1 . 5 . 0 . 30 ) and peak lists were searched with the in-built Andromeda search engine ( Cox and Mann , 2008; Cox et al . , 2011 ) . FASTA Sequence Databases: for X . laevis samples an mRNA derived X . laevis protein database with 79 , 214 entries ( Wühr et al . , 2014 ) was used , for human samples a human UniProt FASTA database ( download date: June 2014 , 20 , 258 entries ) , and for yeast samples a S . cerevisiae Uniprot FASTA database ( download date: June 2014 , 6743 entries ) . These databases were supplemented with common contaminants ( e . g . keratins , serum albumin ) and with the reverse sequences of all entries for false discovery rate estimations . The Andromeda search engine parameters were: carbamidomethylation of cysteine was set as a fixed modification , whereas oxidation of methionine and N-terminal protein acetylation were set as variable modifications; tryptic specificity was considered with proline restriction; up to two missed cleavages were allowed; and the minimum peptide length was set to seven amino acids . The MS survey scan mass tolerance was set to 6 ppm , and MS/MS mass tolerances to 20 ppm ( Orbitrap ) and 0 . 5 Da ( LTQ ion trap ) , respectively . The false discovery rate was set to 1% at both the peptide and the protein level . Several levels of criteria were applied for confident estimation of protein concentrations . First , peptides having posterior error probability ( PEP ) above 0 . 01 were excluded from the estimation of protein concentrations in the cytosolic and the nuclear fractions of X . laevis samples . Second , proteins identified with single ‘only identified by site’ peptides in only one compartment were excluded from the calculation of protein concentrations . The absolute protein concentration of proteins in the cytosolic and the nuclear fractions were estimated by correlation with the absolute concentrations of UPS2 standard proteins and their respective iBAQ intensities ( Schwanhäusser et al . , 2011 ) , assuming volumes of 50 nl for the nucleus and 500 nl for the yolk-free cytosol . A regression curve of the absolute amount of UPS2 standard proteins were plotted against their measured iBAQ intensity ( log 10 scale ) in each biological replicate to generate linear regression equations . These equations were then used to estimate protein concentrations in each biological replicate . Then , the nuclear-to-cytosolic ( N:C ) ratio was calculated from average protein concentration in the nucleus and the cytosol . Annotations were largely based on the UniProt database ( UniProt Consortium , 2015 ) . For each hit , relevant UniProt data were fetched . These data were simplified to ‘simplified localization’ ( Cytoplasm , Nucleus or Both ) and Flags ( transmembrane , mitochondrial , ER proteins ) . UniProt annotations of X . laevis proteins are still sparse . Therefore , human and X . tropicalis entries were used as additional references . Appropriate orthologues were mapped by blasting identified X . laevis contigs against human and X . tropicalis databases ( with an E value cut-off of 10-11 ) . Functional protein groups were acquired from the Kyoto Encyclopedia of Genes and Genomes ( KEGG ) database ( Kanehisa et al . , 2004 ) if not stated otherwise . For assessment of protein complexes in Xenopus laevis oocyte and HeLa cells , human data sets from Ruepp et al . , 2010 and Havugimana et al . , 2012 were used . Yeast protein complexes were assigned according to Gavin et al . , 2006 . For previous CRM1 interaction data , BioGRID ( Chatr-Aryamontri et al . , 2015 ) , the NESdb ( Xu et al . , 2012b ) , and data from Thakar et al . , 2013 were used . Proteins identified with ‘only identified by site’ ( MaxQuant version 15 . 0 . 30 ) within the ‘CRM1+RanGTP’ sample were excluded from further analysis . For X . laevis samples , these additional parameters were applied: peptides with PEP values higher than 0 . 01 were excluded , and a minimum of two unique peptides were required to consider a protein for quantification . For calculating the molar fraction of a given protein in the ‘CRM1+RanGTP’ sample , its iBAQ intensity was divided by the sum of iBAQ intensities of all proteins detected in this sample . ‘Enrichment from input’ for a given protein was obtained by dividing its molar fraction within the ‘CRM1+RanGTP’ sample by its molar fraction in the input extract . The ‘RanGTP-stimulation’ for a given protein was calculated by dividing its iBAQ intensities in the 'CRM1+RanGTP’ and the ‘CRM1 w/o Ran’ samples . When a protein was not identified in the ‘CRM1 w/o Ran’ or ‘Input’ sample , then its iBAQ intensity was replaced by a conservative estimate , namely a baseline intensity for detectability ( the median of iBAQ intensities of the least 30 abundant proteins in this sample ) . This was to avoid in the calculation of parameters divisions by zero and to avoid over-estimating the significance of low abundance cargo candidates . Based on these three parameters , distinct significance categories were constructed for three species as summarized in the ‘Category thresholds’ tables ( Supplementary files 2–4 ) . For proteins that are part of the nuclear transport machinery ( Importins , exportins , Nups and related factors ) separate categories ( NUPs , NTRs , NPC , CRM1 cofactor ) were assigned . mmCRM1 , scCRM1 and hsRanQ69L5-180 were expressed with an N-terminal His14-ZZ-bdSUMO tag and purified by ( i ) immobilisation via Ni2+ chelate chromatography , ( ii ) on column bdSENP1 protease elution ( Frey and Görlich , 2014 ) and ( iii ) gel filtration . Candidate cargoes and controls were expressed as His14-ZZ-bdSUMO fusions and purified via Ni2+ chelate chromatography and imidazole elution . 250 pmoles of His14-ZZ-bdSUMO tagged cargo candidates and controls were immobilized on 20 μl anti-ZZ beads , and incubated with 300 pmoles mmCRM1 either in the presence or absence of 1500 pmoles RanGTP ( hsRanQ69L5-180 ) in a volume of 500 µl . After 3 hr incubation at 4°C , bound material was eluted by adding bdSENP1 protease , and eluted fractions were analysed by SDS-PAGE and Coomassie-staining . Cargo and NES candidates were cloned behind either GFP or a GFP-NLS ( SV40 ) module in modified pEGFP-C1 ( Invitrogen ) vectors . Also a control vector coding for a RFP-NLS-NES fusion was prepared for cotransfection with the GFP/GFP-NLS constructs . HeLa Kyoto cells were grown on coverslips in 24-well plates , and transiently cotransfected with FuGENE6 ( Promega ) according to manufacturer’s instructions . After 24 hr , cells were incubated with either 10 nM Leptomycin B ( LMB , dissolved in DMSO ) or DMSO alone for 3 hr . They were then fixed for 30 min with 3% paraformaldehyde and 0 . 1% glutaraldehyde , and the aldehydes were quenched with 1 mg/ml NaBH4 . Fluorescence signals were recorded on an SP5 confocal laser scanning microscope ( Leica ) , using sequential scans with excitations at 488 and 561 nm .
The mass spectrometry proteomics data have been deposited to the ProteomeXchange Consortium ( Vizcaíno et al . , 2014 ) via the PRIDE partner repository with the dataset identifier PXD002899 . The dataset of high confidence CRM1 interactions ( cargo categories A , B , and ‘low abundance’ ) has been submitted to the IntAct databases ( Orchard et al . , 2014 ) with the identifier IM-24624 . | Animals , plants and other eukaryotic organisms subdivide their cells into compartments that carry out specific tasks . For example , the cell nucleus hosts the genome and handles the genetic information , whereas the surrounding cytoplasm is specialized in making proteins . These proteins are then either used in the cytoplasm or transported to the nucleus and other cell compartments . Since proteins are not made in the nucleus , all proteins in this compartment must be imported from the cytoplasm . Two layers of membrane separate the nucleus and cytoplasm from each other . Any exchange of material must therefore proceed through channels called nuclear pore complexes , or NPCs for short . The NPCs have filters that allow only small molecules a free transit , while larger ones are typically rejected . However , larger proteins may also rapidly pass through the nuclear pore complexes if loaded onto dedicated shuttle molecules; for example , “exportins” transport proteins out of the nucleus . Kırlı , Karaca et al . used an approach called proteomics to measure the levels of 5 , 000 different proteins within the nucleus and the cytoplasm . Such a census helps to predict where a given protein works and where it might cause problems . Further experiments also used proteomics to identify which proteins are carried by an exportin called CRM1 . This revealed that a remarkably large number of different proteins ( around 1 , 000 ) are exported by CRM1 from either yeast , human or frog cell nuclei . Most of these “cargo” proteins were previously thought to never leave the cytoplasm . It now seems , however , that these proteins can leak into the nucleus , but CRM1 recognizes them as cytoplasmic proteins and expels them from the nucleus . These findings suggest that the border control at NPCs is less perfect than was previously believed . If not remedied , this would pose a serious problem for the cell , because the accumulation of "wrong" proteins inside the nucleus would disturb the processes that occur there and could destabilize the genome . Kırlı , Karaca et al . propose that the export of such accidentally displaced proteins by CRM1 is a crucial measure to protect the nucleus . | [
"Abstract",
"Introduction",
"Results",
"and",
"discussion",
"Materials",
"and",
"methods",
"Database",
"depositions"
] | [
"biochemistry",
"and",
"chemical",
"biology",
"cell",
"biology",
"tools",
"and",
"resources"
] | 2015 | A deep proteomics perspective on CRM1-mediated nuclear export and nucleocytoplasmic partitioning |
Eukaryotic DNA is highly organized within nuclei and this organization is important for genome function . Fluorescent in situ hybridization ( FISH ) approaches allow 3D architectures of genomes to be visualized . Scalable FISH technologies , which can be applied to whole animals , are needed to help unravel how genomic architecture regulates , or is regulated by , gene expression during development , growth , reproduction , and aging . Here , we describe a multiplexed DNA FISH Oligopaint library that targets the entire Caenorhabditis elegans genome at chromosome , three megabase , and 500 kb scales . We describe a hybridization strategy that provides flexibility to DNA FISH experiments by coupling a single primary probe synthesis reaction to dye conjugated detection oligos via bridge oligos , eliminating the time and cost typically associated with labeling probe sets for individual experiments . The approach allows visualization of genome organization at varying scales in all/most cells across all stages of development in an intact animal model system .
Eukaryotic genomes are non-randomly organized within mitotic and interphase nuclei . The basic unit of genome organization is the nucleosome . Nucleosomes assemble into higher order structures whose biogenesis , maintenance , regulation , and purpose are poorly understood ( Bickmore , 2013; Bonev and Cavalli , 2016; Meaburn , 2016; Yu and Ren , 2017 ) . DNA fluorescent in situ hybridization ( FISH ) technologies and chromatin conformation capture techniques allow 3D architectures of genomes to be assessed ( Bauman et al . , 1980; Beliveau et al . , 2012; Bienko et al . , 2013; Dekker et al . , 2013 , Dekker et al . , 2002; Lieberman-Aiden et al . , 2009 ) . Studies using these technologies have begun to reveal how DNA is organized within nuclei . For instance , chromatin capture experiments show that many eukaryotic genomes are assembled into megabase-sized structures termed topologically associated domains ( TADs ) . TADs , and larger organizational units termed compartments , are thought to allow subregions of chromosomes to share and integrate long-range transcriptional regulatory signals ( Dekker and Heard , 2015; Dekker and Mirny , 2016; Dixon et al . , 2012; Lieberman-Aiden et al . , 2009; Vernimmen and Bickmore , 2015 ) . Additionally , DNA FISH and chromatin immunoprecipitation ( ChIP ) experiments have shown that the position of genes within nuclei is often not random: active genes tend to localize near nuclear pores and/or the nuclear interior while inactive genes tend to localize to the nuclear periphery , distant from nuclear pores ( Casolari et al . , 2004; Gonzalez-Sandoval and Gasser , 2016; Lemaître and Bickmore , 2015; Pickersgill et al . , 2006; van Steensel and Belmont , 2017 ) . Finally , DNA FISH experiments have shown that individual chromosomes tend to occupy distinct non-overlapping regions of subnuclear space , even in interphase nuclei ( termed chromosome territories ) ( Bolzer et al . , 2005; Cremer and Cremer , 2010 ) . Many questions concerning the large-scale architecture of genomes remain unanswered , including the following: how the various aspects of genome architecture , such as gene position , TADs , or territories differ in different cell types or across developmental time , and how such changes relate to gene expression during development . Technologies that enable rapid and flexible analysis of genome organization in an intact animal would allow such questions to be addressed . The nematode Caenorhabditis elegans is an excellent model for studying genome organization in an intact animal due to its size ( 1 mm ) , lifespan ( ~3 days to reproductive maturity ) , genome size ( 100 Mb across five autosomes and one sex chromosome ) , and transparent body . Whole-chromosome DNA FISH experiments have been instrumental for our current understanding of chromosome architecture and dynamics in model organisms such as C . elegans ( Lau et al . , 2014; Nabeshima et al . , 2011 ) . Such studies are limited , however , by the cost and time associated with generating DNA FISH probe sets . For example , a previous chromosome level DNA FISH experiment in C . elegans used 127 yeast artificial chromosome ( YAC ) clones split into 50 amplification reactions for three chromosome probe generation ( Nabeshima et al . , 2011 ) . Oligopaint technology has made DNA FISH probe production faster , cheaper , and more flexible ( Beliveau et al . , 2012 ) . Oligopaint takes advantage of massively parallel DNA synthesis technologies to create user defined libraries containing hundreds of thousands of individual DNA oligos each comprised of a short ( 42 bp ) DNA sequence that hybridizes to a genome , as well as additional ‘barcode’ sequences that serve two major functions . First , barcodes allow probes to be repeatedly amplified from an Oligopaint library , thus providing a virtually inexhaustible supply of oligos for DNA FISH ( Beliveau et al . , 2012; Chen et al . , 2015; Murgha et al . , 2014 ) . Second , barcodes allow pre-labeled detection oligos to be used to detect Oligopaint oligos , thus obviating the need for fluorescently labeling probes for each DNA FISH experiment ( Beliveau et al . , 2015; Beliveau et al . , 2012; Nir et al . , 2018; Rosin et al . , 2018 ) . Previously , the Oligopaint technology has been used to visualize chromosome territories in Drosophila cultured cells in normal and mutant contexts ( Rosin et al . , 2018 ) . Here , we describe a rapid , flexible and inexpensive Oligopaint strategy that enables visualization of chromosome territories and sub-chromosome regions in a whole intact organism . Specifically , we report methods for using this library to simultaneously visualize all six C . elegans chromosomes , as well as three megabase and 500 kilobase subregions of these chromosomes , in all/most cells of C . elegans across all stages of development .
An Oligopaints bioinformatics pipeline was used to identify 42 bp DNA sequences in the C . elegans genome ( genome reference Ce10 ) that ( 1 ) uniquely map to the genome , ( 2 ) exhibit similar melting temperatures and similar GC content , ( 3 ) lack repetitive stretches , and ( 4 ) lack predicted secondary structures ( Beliveau et al . , 2012 ) . [Note , an updated bioinformatics pipeline for identifying probes is now available ( Beliveau et al . , 2018 ) . ] The results of our bioinformatic search revealed approximately nine suitable probe sequences per kilobase of C . elegans genomic DNA ( Table 1 ) . We generated an Oligopaint library that contained , on average , ~2 probe sequences per kb of genomic DNA across each C . elegans chromosome ( Figure 1a and Table 1 ) . Figure 1a shows the distribution of all ( 25 , 174 ) chromosome I probes . Few gaps between probes exceeded 5 kb ( 33 out of 25 , 174 ) , with the largest gap spanning ~18 kb . A subtle decrease in probe density is observed on chromosome arms , perhaps due to an increase in repetitive sequences in these regions , which biased against probe selection ( C . elegans C . elegans Sequencing Consortium , 1998 ) . Similar probe distributions are observed for the other five C . elegans chromosomes ( Figure 1—figure supplement 1 ) . In total , the library consisted of 170 , 594 oligos ( termed primary Oligopaint oligos ) , which each contain 42 bp of unique genomic sequence flanked by barcode sequences that allow for DNA FISH targeting each of the six C . elegans chromosomes , as well as three megabase , or 500 kb subregions of these chromosomes ( Figure 1b ) . Bridge oligos ( also see Nir et al . , 2018 ) were designed to base pair with barcode sequences contained within primary probes as well as base pair with dye-conjugated detection oligos ( Figure 1c ) . Detection oligos were designed that base pair with bridge oligos and are conjugated to three fluorophores ( Alexa 488 , Cy3 , and Alexa 647 ) ( Figure 1c ) . Thus , bridge oligos are intermediate probes that hybridize to the primary probe and provide a docking site for labeled detection probes . Bridge oligos provide versatility ( and cost savings ) to DNA FISH experiments as these oligos allow any primary probe set to be coupled to any detection probe set with minimal additional cost . Bridge oligos also allow for more than one fluorophore to be targeted to primary probes , which expands the number of objects that can be visualized with a standard three channel microscopy system ( see six chromosome FISH experiments below ) . By using 1 ) detection oligos with fluorophores on both 3’ and 5’ termini , 2 ) two detection oligos per bridge oligo , and 3 ) bridge oligos that target the 5’ and 3’ barcode sequence of primary probes , it is possible to have each primary oligo recognized by eight fluorophores . To conduct DNA FISH , unlabeled primary probes are first PCR amplified from the Oligopaint library as described in Beliveau et al . ( 2017 ) ; Beliveau et al . , 2012; Chen et al . , 2015; Murgha et al . , 2014 ( also see Materials and methods ) . Second , primary Oligopaint oligos are hybridized to fixed samples of C . elegans overnight ( see below and Materials and methods ) . Third , samples are hybridized with a mixture of bridge oligos and dye-conjugated detection oligos for 3 hr the following day ( Figure 1c ) . Together , this three-step strategy allows many DNA FISH experiments to be conducted fairly cheaply after a single primary probe synthesis step . DNA FISH in C . elegans is typically done on dissected tissue . We developed a fixation and hybridization protocol that allowed for efficient DNA FISH on intact C . elegans . As part of this protocol , hybridization steps are conducted in microcentrifuge tubes , which allows large numbers of animals to be simultaneous assayed by FISH . A detailed description of this fixation and hybridization protocol can be found in Materials and methods . To test our C . elegans Oligopaint library , we amplified a primary probe set targeting chromosome II ( 27 , 360 unique probes ) and asked if this probe set was able to specifically label chromosome II . The behavior and morphology of chromosomes in the C . elegans germline are well-established ( Albertson et al . , 2011 ) . For instance , homologous chromosomes pair at the pachytene stage of Meiosis I at a defined region of the germline ( termed pachytene region ) . Oocytes are arrested in diakinesis of meiosis I and chromosomes are highly condensed with homologs connected via a single chiasmata ( termed bivalents ) ( Villeneuve , 1994 ) . Mature sperm harbor highly condensed chromosomes and are haploid . To address specificity , we imaged germ cells of animals subjected to chromosome II DNA FISH . This analysis detected the expected chromosomal structures in pachytene germ cells , oocytes , and sperm; fluorescent staining was observed on a single bivalent in oocytes and in one region of the nucleus in sperm and pachytene germ cells ( Figure 2 ) . HIM-8 is required for X chromosome homolog pairing during meiosis and , consequently , X chromosomes are present as two univalents ( and not a single bivalent ) in him-8 mutant animals ( Phillips et al . , 2005 ) . To further address specificity , we imaged oocytes of wild type and him-8 ( e1489 ) mutant animals subjected to X chromosome Oligopaint . As expected , X chromosome DNA FISH stained a single bivalent in wild-type oocytes ( 58/58 ) and two univalents in 84% of him-8 oocytes ( 51/61 ) ( Figure 2—figure supplement 1 ) . Together the data show that the C . elegans Oligopaint library is specific . To quantify the efficiency of our method , we first measured the percentage of whole animals stained by chromosome II DNA FISH . Staining presented in an all or nothing fashion with 1085/1303 ( 83% ) of larval stage animals , 317/326 ( 97% ) of adult animals , and 50% of the embryos housed within uteri of adult animals displaying FISH signal ( Figure 2—figure supplement 2 ) . [Note , an alternative protocol that allows for greater efficiency using isolated embryos ( 90% ) is described in Materials and methods . ] We next measured the % of nuclei within a given animal that were stained by chromosome II DNA FISH . We randomly chose DAPI-stained nuclei ( from animals that showed staining ) and asked if these nuclei were positive for chromosome II DNA FISH signals . Out of 50 randomly chosen somatic nuclei 50/50 had DNA FISH signal . Likewise , 50/50 germline nuclei were positive for DNA FISH signals . We conclude that our library and hybridization strategy allows for robust and specific labeling of a whole chromosome in many cell types and many developmental stages simultaneously in large numbers of animals . It is possible that DNA FISH signals in every cell and at every stage of development can be visualized with this approach . C . elegans possess six chromosomes . Most lab microscopy systems are equipped to image 3-4 fluorophores and , thus , are not capable of imaging all six C . elegans chromosomes simultaneously . To circumvent this issue , we took advantage of our bridge oligo strategy to target combinations of fluorophores to each C . elegans chromosome in order to image all six chromosomes using a three-channel microscopy system ( Figure 3 ) . For instance , Figure 3—figure supplement 1 shows an example of multi-probe labeling of the X chromosome: Detection oligos labeled with either Alexa 647 or Cy3 were targeted to the X chromosome ( Figure 3—figure supplement 1a ) . Overlapping Alexa 647 ( green ) or Cy3 ( red ) channels produces a yellow pseudocolor that can be differentiated from Alexa 647 ( green ) or Cy3 ( red ) alone ( Figure 3—figure supplement 1b ) . We next amplified primary probe sets targeting all six C . elegans chromosomes and hybridized primary probe sets to fixed adult C . elegans . We then used bridge oligos to couple these primary probe sets to detection probes labeled with Alexa 488 , Cy3 , or Alexa 647 , or combinations of these fluorophores , in order to simultaneously image all six C . elegans chromosomes ( Figure 3a ) . We imaged oocytes in these animals and detected six bivalents that were each labeled a distinct color ( Figure 3b ) . In pachytene stage adult germ cells , C . elegans chromosomes are paired , condensed , and localized near the nuclear periphery ( Albertson et al . , 2011 ) . DNA FISH illuminated six regions of distinct colors concentrated near the nuclear periphery in pachytene germ cells ( Figure 3c ) . Six chromosome DNA FISH staining was robust: 50/50 randomly chosen DAPI positive nuclei were stained successfully . Six chromosome FISH staining was also successful in somatic nuclei . Six distinct colors were often distinguishable in the nuclei of intestinal and hypodermal nuclei , as well as nuclei whose small size and positioning within the animal were indicative of ventral cord neurons ( Figure 3d–f and Figure 3—video 1 ) . These data show that our DNA FISH approach is capable of labeling all six C . elegans chromosomes simultaneously in many different cell types of an intact animal . The data also show that , like interphase chromosomes in other eukaryotes , C . elegans chromosomes occupy largely distinct territories within interphase nuclei and that these chromosome territories persist in post-mitotic cells . Note , because the six color strategy described above visualizes an overlap of two colors ( each of which is also used to define a separate chromosome ) , rigorously defining the subnuclear space occupied by individual chromosomes is not possible using this six color approach . Single channel probe sets are recommended for experiments in which knowing the precise space occupied by a chromosome is relevant ( see Figure 5 for examples ) . We designed our Oligopaint library to include 3 Mb and 500 Kb barcode sequences that should permit visualization of chromosomal subregions ( Figure 1c ) . To test this aspect of our library , we amplified Oligopaint oligos targeting chromosome I and hybridized these probes to adult C . elegans . We then used bridge oligos that recognized all Chromosome I Oligopaint oligos ( ~15 Mb ) , a 3 Mb subregion of chromosome I ( 0–3 Mb ) , or a 500 kb subregion of this 3 Mb region ( 1 . 0–1 . 5 Mb ) . Detection oligos coupled to Alexa 488 , Cy3 , and Alexa 647 were used to illuminate each genomic region , respectively . We imaged pachytene germ cells and , as expected , observed a single contiguous DNA FISH signal after chromosome I DNA FISH ( Figure 4a ) . 3 Mb DNA FISH illuminated a subregion of the chromosome I signal and 500 kb DNA FISH illuminated a subregion of this 3 Mb signal ( Figure 4a ) . Staining was robust , with 50/50 randomly chosen nuclei possessing all three fluorescent signals in successfully stained animals . Similar patterns were observed when chromosome IV , a 3 Mb subregion of chromosome IV ( 0–3 Mb ) , or a 500 kb subregion of this 3 Mb region ( 2 . 5–3 . 0 Mb ) were analyzed ( Figure 4b ) . We conclude that the Oligopaint library has the capability to visualize 3 Mb and 500 kb subregions of the C . elegans genome . Our C . elegans Oligopaint library and hybridization protocol should allow many questions relating to the biology of genome organization to be asked within the context of a whole animal . We started this process by using our library to ask two simple questions: 1 ) Whether genomic architecture changes during aging , and ( 2 ) what cellular factors are needed to establish and/or maintain chromosome territories in post-mitotic cells ? Recent studies suggest that higher order chromatin structures may break down during aging , and in aging-related diseases such as Alzheimer’s ( Winick-Ng and Rylett , 2018 ) . Age-related alterations in nuclear morphology have also been noted in C . elegans ( Haithcock et al . , 2005 ) . We used our Oligopaint library to simultaneously visualize all six chromosomes in 1- and 10-day-old animals ( C . elegans typically live about 2 weeks ) to ask if the aging process might affect the genomic architecture . For this analysis , we imaged intestinal nuclei as these cells are postmitotic , have large nuclei , and are easily identifiable due to their idiosyncratic size , shape , and location within the animal . As expected , all six chromosomes occupied largely distinct territories in intestinal cells of 1-day-old animals ( Figure 5a ) . Interestingly , in 10-day-old animals , chromosomes were no longer organized into discrete territories ( Figure 5a ) . Quantification ( see Materials and methods ) of the space occupied by chromosomes I , II and III in young and aged animals revealed an ~50% increase in the volume of all three chromosomes ( Figure 5b ) as well as an increase in the degree to which chromosomes I , II and III overlapped in subnuclear space ( Figure 5—figure supplement 1a ) . The data confirm that the discrete chromosome territories observed in young intestinal nuclei are lost as the worms ages . We asked if the loss of chromosome territories in older animals were a consequence of aging or a function of time . To do so , we conducted a similar analysis on young and old animals harboring a mutation ( e1370 ) in daf-2 . daf-2 encodes a insulin-like receptor and loss-of-function mutations in daf-2 and mutations in daf-2 such as e1370 cause animals to live twice as long as wild-type animals ( Kenyon et al . , 1993; Kimura et al . , 1997 ) . Chromosome territories did not become enlarged or disorganized in 10-day-old daf-2 mutant animals versus young animals , indicating that the loss of chromosome territories we see in older wild-type animals is linked to aging and not chronological time ( Figure 5a/b and Figure 5—figure supplement 1a ) . The data are consistent with a model in which higher order chromatin structures are lost during aging . Further studies will be needed to address if genome organization in other/all cell types is similarly affected by aging and , more importantly , if the loss of chromosomal territories that occur in aged animals is a cause or consequence of the aging process . Very little is known about how chromosome territories are established or maintained in animals . The Oligopaint DNA FISH library described above could be used to identify and characterize genes and pathways mediating and regulating these processes . As a first attempt to identify such factors , we conducted six chromosome DNA FISH on animals subjected to RNAi targeting seven candidate genes , which we suspected might be involved in establishing/maintaining chromosome territories in C . elegans ( Figure 5—source data 1 ) . mes-3 , which encodes a component of the Polycomb Repressive Complex 2 ( PRC2 ) , was included in this small-scale screen because PRC2 is a known regulator of chromatin architecture in many organisms ( Capowski et al . , 1991; Holdeman et al . , 1998; Ross and Zarkower , 2003; Xu et al . , 2001 ) . RNAi targeting mes-3 caused a loss of chromosome territories in adult hypodermal cells ( Figure 5c ) . Note: we chose to image hypodermal cells for this analysis as these cells are , like intestinal cells , easy to identify and because ( for unknown reasons ) the effects of mes-3 RNAi on genome architecture appeared to be most dramatic in this cell type ( Figure 5—figure supplement 2 , and see below ) . Quantification of chromosomes I , II , and III volumes in hypodermal nuclei revealed an ~45% increase in chromosome volume when animals were exposed to mes-3 RNAi , suggesting that MES-3 is important for maintaining chromosome territories of hypodermal cells during the normal course of growth and development ( Figure 5d ) . Quantification of the degree to which DNA FISH signals for chromosomes I , II , and III overlapped revealed an ~2 fold increase in overlap after mes-3 RNAi , suggesting that chromosome territories are not just growing , they are also mixing ( Figure 5—figure supplement 1b ) . In summary , the data suggest that MES-3 and , therefore , PRC2 is needed to establish and/or maintain chromosome territories in C . elegans . Additional studies will be needed to understand the source of cell type specificity of mes-3 knockdown on genome architecture and related studies using mutant alleles of mes-3 , as well as loss-of-function alleles in other components of the PRC2 , will be needed to confirm the link between PRC2 and the maintenance of chromosome territories .
The invariant cell lineage , transparency , and small genome ( 100 Mb ) of C . elegans make this animal an excellent system in which to explore how genome architecture relates to gene expression , development , growth , reproduction , and aging . DNA FISH experiments in C . elegans have historically relied on 1 ) labeling PCR products that cover a single small ( 5–10 kb ) region , or 2 ) YACs to generate probes targeting larger regions ( up to whole chromosomes ) . Such approaches are low throughput and rigid in the sense that new probe sets need to be produced for each new DNA FISH experiment . Such experiments have also been limited by the types of cells that can be queried , as most DNA FISH protocols rely on dissection of tissues , which is low throughput and limits the number of cell types that can be analyzed at one time . Here we describe an Oligopaint DNA FISH library and hybridization strategy that allow for visualization of all six C . elegans chromosomes at varying scales . The ability to rapidly and cheaply produce C . elegans DNA FISH probes , in conjunction with improvements to hybridization protocols , enables DNA FISH in all/most cells across all stages of development in an intact animal . These improvements should empower studies asking if/how higher-order chromatin structures regulate , and/or are regulated by , changes in gene expression that occur during growth and development . Given the invariant cell lineage of C . elegans , it should now also be possible to ask if chromosome- chromosome interactions or homolog pairing , or the size , morphology , or sub-nuclear positioning , of chromosomal territories ( or subregions of these territories ) vary predictably by cell type , age , or developmental trajectory .
We used a previously described pipeline to mine the C . elegans genome ( build ce10 ) for desirable ( see main text ) oligonucleotide nucleotide sequences 42 base pairs in length ( Supplementary file 1 ) ( Beliveau et al . , 2012 ) . 872 , 946 oligonucleotide sequences met this criteria , and 170 , 594 probes sequences were chosen for the library ( Supplementary file 2 ) . A series of barcode sequences were appended to each 42 bp hybridization sequence , which resulted in each primary probe being 150 bp . Barcode sequences can be found in Supplementary file 5 . The 170 , 594 sequences were ordered as two 90 k oligonucleotide chips from Custom Array ( Bothell , WA ) . Note that a step by step protocol for Oliogpaint probe synthesis can be found in Supplementary file 6 . To obtain primary probes for Oligopaint experiments , desired primary probes were first amplified using primers specific to the outermost barcode sequences , which correspond to the individual chromosome barcodes shown in Figure 1a ( Supplementary file 5 ) . Single stranded probe ( primary probe ) generation was conducted as previously described ( Chen et al . , 2015 ) . Briefly , PCR was used to append a T7 polymerase site to the 5’ end of chromosome specific barcode sequence , followed by T7 polymerase reactions to generate ssRNA . ssRNA was reverse transcribed into ssDNA . Unwanted ssRNA species were degraded using base hydrolysis . Finally , long ssDNA oligos were purified using the Zymo-100 DNA Clean and Concentrator Kit with oligo binding buffer . Probes were stored at 100 pmol/ul at −20C . An aliquot of the library is available to qualified labs upon request . Bridge oligos were ordered from IDT at 25 or 100 nmole scales using standard desalting procedures . Fluorescent detection oligos were ordered from IDT with 5’ and 3’ fluorescent modifications on the 250 nm or 1 um scale and subjected to HPLC purification . Bridge and detection probe sequences are listed in Supplementary file 5 . See Supplementary file 7 for detailed protocol on worm collection and Oligopaint FISH . 10 cm plates containing adult ( or mixed stage ) C . elegans were washed with M9 solution ( 11 mM KH2PO4 , 21 mM Na2HPO4 , 4 mM NaCl , 9 mM NH4Cl in H2O ) and collected in 15 ml conical tubes . Animals were pelleted ( 3 k rpm for 30 s ) , and washed two times with M9 solution . Animals were resuspended in 10 ml of M9 solution and rocked for ~30 min at room temperature . Animals were pelleted and aliquoted to 1 . 5 ml microcentrifuge tubes ( 30–50 ul of packed worms per tube ) . Samples were placed in liquid nitrogen for 1 min . Frozen worm pellets were resuspended in cold 95% ethanol and vortexed for 30 s . Samples were rocked for 10 min at room temperature . Samples were spun down ( 3 k rpm for 30 s ) , supernatant discarded , and washed twice in 1X PBST ( 10X Phosphate- Buffered Saline ( Thermo Fisher Scientific: 70011–044 ) diluted to 1X in H2O , 0 . 5% Triton X-100 ( Sigma: X100 ) . 1 ml of 4% paraformaldehyde solution ( 4% paraformaldehyde in 1X PBS ) was added and samples were rocked at room temperature for 5 min , washed twice with 1X PBST , and resuspended in 2X SSC ( 20X saline-sodium citrate ( SSC ) buffer ( Thermo Fischer Scientific: 15557–044 ) diluted in H2O ) for 5 min at room temperature . Samples were spun down and resuspended in a 50% formamide 2X SSC solution at room temperature for 5 min , 95°C for 3 min , and 60°C for 20 min . Samples were spun and resuspended in 60 ul of hybridization mixture ( 10% dextran sulfate , 2X SSC , 50% formamide , 100 pmol of primary probe per chromosome and 2 ul of RNAse A ( sigma 20 mg/ml ) ) . Hybridization reactions were incubated in a 100°C heat block for 5 min before overnight incubation at 37°C in a hybridization oven . The next day , samples were washed with prewarmed 2X SSCT ( 2X SSC with 0 . 5% Triton X-100 ) ( rotating at 60°C ) for 5 min , followed by a second 2X SSCT wash at 60°C for 20 min . Wash buffer was removed and samples were resuspended in 60 ul of bridge oligo hybridization mixture ( 2X SSC , 30% formamide , 100 pmol of bridge oligo per targeted region ( ie whole chromosome , three megabase , or 500 kb spots ) and 100 pmol of each detection oligo . Bridge/detection oligo hybridization reactions were incubated at room temperature for 3 hr . Samples were washed in prewarmed 2XSSC at 60°C for 20 min , followed by a 5-min wash with 2XSSCT at 60°C and a 20 min wash in 2XSSCT at 60°C . Samples were then washed at room temperature in 2XSSCT . Wash buffer was removed and samples were resuspended in mounting medium ( Vectashield with DAPI or Slowfade Gold with DAPI ) . Samples were mounted on microscope slides and sealed with nail polish . DNA FISH on in utero embryos was only 50% efficient . The following protocol improves this efficiency to 90% . This protocol is an adaptation of an existing C . elegans DNA FISH protocol ( Crane et al . , 2015 ) . Briefly , adults were dissected in 8 ul of 1X egg buffer on a coverslip ( 25 mM HEPEs , pH 7 . 3 , 118 mM NaCl2 , 48 mM KCl , 2 mM CaCl2 , 2 mM MgCl2 ) to release embryos . Coverslips were placed on a Superfrost Plus Gold slide ( Thermo Scientific ) and placed in liquid nitrogen for 1 min . Coverslips were popped off with a razor blade and slides were submerged in 95% cold ( −20C ) ethanol for 10 min . Slides were washed twice in 1XPBST before fixation in 4% paraformaldehyde solution ( described above ) for 5 min . Slides were washed twice in 1XPBST . 20 ul Primary hybridization mixture ( described above ) was added to each sample and a coverslip was placed on top . Slides were placed on a 90°C heat block for 10 min . Slides were placed in a humid chamber at 37°C overnight . Wash steps and bridge/detection oligo hybridization was carried out as described above . 15 ul of mounting medium was added to each sample , and coverslips were sealed with nail polish . Standard fluorescent microscopy was conducted on a widefield Zeiss Axio Observer . Z1 microscope using a Plan-Apochromat 63X/1 . 40 Oil DIC M27 objective and an ORCA-Flash 4 . 0 CMOS Camera . The Zeiss Apotome 2 . 0 was used for structured illumination microscopy using three phase images . All image processing were done using the Zen imaging software from Zeiss . Confocal microscopy was done using a Nikon Eclipse Ti microscope equipped with a W1 Yokogawa Spinning disk with 50 um pinhole disk and an Andor Zyla 4 . 2 Plus sCMOS monochrome camera . A 60X/1 . 4 Plan Apo Oil objective or a 100X/1 . 45 Plan Apo Oil objective was used . Ten adult animals were picked to 6 cm NGM plates seeded with OP50 , and 10 plates were used for each condition . Adult animals were picked off 24 hr later and sacrificed . Once the offspring reached the fourth larval stage , 50 animals were transferred to 6 cm NGM plates seeded with OP50 that were soaked in 1 ml of FUDR solution ( 3 mg of FUDR ( abcam ) per plate ) the previous day . 20 FUDR soaked plates were used per condition . After 24 hr , 10 plates per condition were collected and animals were frozen as pellets in liquid nitrogen before storage at −80C ( Day 1 adult samples ) . Ten days later , the same collection was repeated on the remaining 10 plates per condition ( Day 10 adult samples ) . Dead animals , as determined by animals that did not respond to a light touch , were removed prior to sample collection for each condition . Wild-type ( N2 ) embryos were collected via hypochlorite treatment ( see Supplementary file 7 for description of embryo isolation by hypochlorite treatment ) and placed on RNAi plates ( NGM plates with 2 . 5 mM KH2PO427 mM Carbenicillin , 1 mM IPTG ) seeded with either HT115 bacteria , or HT115 bacteria expressing mes-3 dsRNA for two generations: the embryos of adult animals grown on either treatment were placed back onto either treatment , grown to adulthood , and collected for FISH analysis ( see Supplementary file 7 for description of sample collection ) . The mes-3 RNAi clone was obtained from the Ahringer library and confirmed to target mes-3 by Sanger sequencing ( Kamath and Ahringer , 2003 ) . For a step-by-step protocol for the image analysis used in this study see Supplementary file 8 . All territory quantifications were done using standard tools in ImageJ along with the 3D objects counter plugin ( Bolte and Cordelières , 2006 ) . First , each individual nucleus was segmented from the original file to generate individual nuclei files . The four-channel stack was then split to create individual files for each chromosome ( each chromosome is represented by a single fluorophore/channel ) . To remove background noise and create a binary mask , each image was subjected to thresholding using the default ImageJ thresholding using ‘auto’ across every image . Once masks were obtained the 3D objects counter tool was utilized to select objects larger than 30 voxels ( eliminating further background signal ) . Object masks for each channel were loaded into the 3D Manager plugin for ImageJ , and all objects for a given chromosome were merged into a single object . The colocalization and measure 3D functions within 3D manager were used to determine the volume of each chromosome as well as the volume of overlap between each chromosome . | DNA contains the instructions needed to build and maintain a living organism . How DNA is physically arranged inside a cell is not random , and DNA organization is important because it can affect , for example , which genes are active , and which are not . Researchers often use a technique called “fluorescence in situ hybridization” ( or FISH for short ) to study how DNA is organized in cells . FISH tethers fluorescent molecules to defined sections of DNA , making those sections glow under the right wavelength of light . It is possible to collect images of the fluorescent DNA regions under a microscope to see where they are in relation to each other and to the rest of the cell . Fields , Nguyen et al . have now created a new library of FISH molecules that can be used to analyze the DNA of a microscopic worm known as Caenorhabditis elegans – a model organism that is widely used to study genetics , animal development , and cell biology . The library can be used to visualize the worm’s whole genome at different scales . The library enables accurate and reliable investigations of how DNA is organized inside C . elegans , including in intact worms , meaning it also offers the first chance to study DNA organization in a whole organism through all stages of its life cycle . This new resource could help to reveal the relationships between DNA organization , cell specialization and gene activity in different cells at different stages of development . This could help to clarify the relationships between physical DNA organization and biological change . This design strategy behind this whole genome library should also be adaptable for similar studies in other animal species . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"chromosomes",
"and",
"gene",
"expression",
"tools",
"and",
"resources"
] | 2019 | A multiplexed DNA FISH strategy for assessing genome architecture in Caenorhabditis elegans |
Hematopoiesis leads to the formation of blood and immune cells . Hematopoietic stem cells emerge during development , from vascular components , via a process called the endothelial-to-hematopoietic transition ( EHT ) . Here , we reveal essential biomechanical features of the EHT , using the zebrafish embryo imaged at unprecedented spatio-temporal resolution and an algorithm to unwrap the aorta into 2D-cartography . We show that the transition involves anisotropic contraction along the antero-posterior axis , with heterogenous organization of contractile circumferential actomyosin . The biomechanics of the contraction is oscillatory , with unusually long periods in comparison to other apical constriction mechanisms described so far in morphogenesis , and is supported by the anisotropic reinforcement of junctional contacts . Finally , we show that abrogation of blood flow impairs the actin cytoskeleton , the morphodynamics of EHT cells , and the orientation of the emergence . Overall , our results underline the peculiarities of the EHT biomechanics and the influence of the mechanical forces exerted by blood flow .
Hematopoiesis is an essential biological process leading to the genesis of all blood lineages throughout life . Currently , producing full-potential hematopoietic stem cells ( HSCs ) in vitro remains a challenge ( Yvernogeau et al . , 2016 ) , in particular for regenerative purposes . This points to the relevance of reinforcing the fundamental knowledge of the process as it occurs in vivo . HSCs with long-term replenishment potential are formed in the embryo according to a genetic program highly conserved between vertebrate species ( Ciau-Uitz et al . , 2014; Ciau-Uitz and Patient , 2016; Jaffredo and Yvernogeau , 2014 ) . HSCs emerge during the so-called definitive wave of hematopoiesis , from vascular components , in particular the ventral floor of the dorsal aorta , in the aorta-gonad mesonephros ( AGM ) region as well as from the umbilical and vitelline arteries in mammals . The zebrafish has proven to be an extremely valuable model for studying the developmental aspects of hematopoiesis ( Frame et al . , 2017 ) ; the combination of cell tracing and live-imaging have notably allowed to trace the migration of definitive hematopoietic stem and progenitor cells ( HSPCs ) to successive hematopoietic organs ( Jin et al . , 2007; Kissa et al . , 2008; Murayama et al . , 2006 ) . Real-time in vivo imaging also established unambiguously the aortic origin of definitive HSPCs , that were visualized emerging as single cells directly from the hemogenic endothelium in the floor of the dorsal aorta ( Bertrand et al . , 2010; Kissa and Herbomel , 2010 ) . This very unique type of cell emergence , called the endothelial-to-hematopoietic transition ( EHT ) , is characterized by an unusual bending of the cell that increases while the emergence is progressing , and so until the exit from the vascular wall ( Kissa and Herbomel , 2010 ) . The process was also described ex vivo in slices of mouse embryos , although at a lower resolution ( Boisset et al . , 2010 ) . An apparent difference between zebrafish and mouse is the direction of the emergence , with HSPCs extruding into the sub-aortic space in the former and into the aortic lumen in the latter ( where they initiate the so-called intra-aortic clusters ) . The reasons for such a difference are currently unknown and may relate to different biophysical constraints of the aortic and sub-aortic environments in either species . The vascular and aortic systems have very unique characteristics ( Aird , 2012 ) . They are constituted by extremely flat endothelial cells with luminal/apical and basal membranes having virtually equal surfaces ( a so-called squamous epithelium ) and organized around a lumen filled with blood cells and fluid , and exposed to high mechanical loading , including shear stress exerted on the surface of endothelial cells as well as circumferential wall stress ( Lu and Kassab , 2011 ) . Blood flow also appears to regulate endothelial cell fates in development ( García-Cardeña and Slegtenhorst , 2016 ) and , in regard to hematopoiesis , to have an influence on the expression of key transcription factors such as Runx1 , the transcription factor essential for hematopoiesis , both in mouse embryonic stem cells ( Adamo et al . , 2009 ) and in the zebrafish ( North et al . , 2009 ) . Hence , comprehending the fundamental mechanisms of HSC emergence requires taking into account the peculiarities of the vascular environment and the mechanical constraints . The EHT allows single live cells to leave a squamous epithelium without perturbing its integrity . Most other cell extrusion processes that have been described concern thick , columnar/cuboidal epithelia , and among them , only the epithelial mesenchymal transition ( EMT , [Lamouille et al . , 2014; Nieto et al . , 2016] ) gives rise to live wandering cells endowed with a new potential , as the EHT does . Hence , in order to specify further the peculiarities of the EHT and to be able to establish comparisons with the other cell extrusion processes described so far , we delved deeper into the process , taking the vascular environment into account . To do so , we developed a novel algorithm allowing the deployment of the aortic wall into high-resolution 2D-cartographies and 2D-time-lapse sequences . We also addressed the influence of blood flow on the EHT morphodynamics by abrogating heart beating . This led us to uncover some of the key and unique features of the EHT ( and to draw a comprehensive model of its main steps ) and identify the contribution of some of the essential molecular components involved in the morphological changes accompanying the emergence . Further , our analysis brought insight into the dynamic interplay between cells undergoing EHT and adjoining endothelial cells and pointed to the essential role of blood flow in HSCs emergence .
In order to unravel the critical biomechanical features of the EHT , we performed confocal live-imaging at high spatio-temporal resolution . In comparison to previous work ( Kissa and Herbomel , 2010 ) , we improved the spatial resolution of our time-lapse ( TL ) sequences by reducing the distance between the sequential optical planes ( 0 . 43 μm instead of 2 or 0 . 60 μm ) and improved the temporal resolution by reducing the 5 min intervals between Z-stacks acquisitions down to 2 min when performing spinning-disk confocal imaging . We imaged the dorsal aorta ( DA ) of transgenic zebrafish embryos ( Figure 1A ) , starting at 48 hpf ( corresponding to the timing at which the EHT process is culminating , see Kissa and Herbomel , 2010 ) , until 72 hpf . We noticed that the angles of emergence of EHT undergoing cells ( hereafter designated as 'EHT cells' ) varied between approximately 0–45° from the dorso-ventral axis of the DA and 0–45° from its antero-posterior ( A-P ) axis . In general , the most valuable information was obtained from cells for which the two most distant poles were parallel to the A-P axis ( Figure 1B ) . We initiated the study using Tg ( kdrl:Ras-mCherry; kdrl:eGFP ) fish so as to visualize cellular membranes as well as the cytoplasmic volume . As previously described ( Kissa and Herbomel , 2010 ) , the morphological criterion allowing unambiguous identification of cells having initiated the EHT is their cup-shaped morphology , with bending toward the sub-aortic space . Hence , many of our TL sequences were initiated at this stage , increasing chances to image completion of the process and minimizing the risk of phototoxicity ( see Figure 1C for a 3D-rendering view , and Figure 1—video 1 , Figure 1—video 2 ) . Ras-mCherry allowed visualizing the luminal and basal membranes ( Figure 1H ) , revealing that the latter underwent more or less extensive blebbing at the cup-shaped stage ( Figure 1D , I ) . This blebbing preceded the protrusion of large membrane extensions that were formed hours before the cell exit and were reminiscent of cell shape changes occuring during amoeboid migration ( Figure 1—video 1 ) . Finally , at the end of the process , Ras-mCherry delineated a transient narrow membrane foot that remained connected to the aorta floor and preceded release in the sub-aortic space ( Figure 1F , G and L and Figure 1—video 1 and Figure 1—video 2 ) . We then measured the time-to-exit , starting from the onset of imaging at the cup-shaped stage and found that it varied widely , with an average duration of 5 . 48 hr ( n = 66 , see Figure 1—figure supplement 1B , right panel ( note that we define the cup-shaped stage as the steps at which the angle between the axis tangential to the inner cell border and the plane of the aortic floor evolves from 150 to 90° , see A: middle top panel ) ) . We also imaged the initial phase of the cell bending , starting from a flat-shaped to the cup-shaped stage in order to determine the time required to reach the latter . We followed a limited amount of cells because , in addition to increasing the risk of phototoxicity owing to long hours of laser exposure , it is currently impossible to determine the precise starting point at which EHT committed cells , engulfed in the hemogenic endothelium , will initiate the emergence ( hence hampering the precise determination of EHT initiation and duration ) . The time intervals between the flat morphology to the cup-shaped stage were also very variable , with an average of 6 . 5 hr ( n = 7 , see Figure 1—figure supplement 1B , left panel and Figure 1—video 3 ) . Finally , to quantify the progression of the process , starting from the cup-shaped stage , we manually measured through time the distance between the anterior and posterior poles of the EHT cell ( within the plane of the aortic wall ) , which we will refer to as the A-P luminal distance ( Figure 1—figure supplement 1C ) . Firstly , we found little correlation between the A-P luminal distance at the beginning of our imaging and the time of exit of the EHT cell ( the longest A-P luminal distances did not imply the longest times-to-exit , see Figure 1—figure supplement 1D ) . Secondly , we noticed that the narrowing down of the A-P luminal distance did not occur at a uniform pace , but with phases of contraction and of relative stabilization of variable durations taking place in the limit of an A-P luminal distance of approximately 2 µm , followed by steep decrease until the release of the cell ( this fast drop may represent a collapse following the culmination of mechanical tensions ) . Hence , the heterogeneous duration of the EHT might be explained by these discontinuities , but the reason for this is unknown . A specific feature of the EHT is that EHT cells maintain contact with their endothelial neighbors until the very end of the process . This is most probably because of the need to maintain the tightness of the endothelium . In order to visualize the local density and spatial organization of junctional complexes established between EHT cells and contacting endothelial neighbors , we followed ZO1 , a protein that interacts preferentially with tetraspan membrane constituents of tight junctions , and also with α-catenin ( hence with cadherin complexes , [Maiers et al . , 2013] ) . We performed live-imaging of hemogenic regions of the aortic floor using Tg ( fli1:Gal4; UAS:RFP; UAS:eGFP-ZO1 ) ( Herwig et al . , 2011 ) embryos ( with cytosolic RFP facilitating the localization of cells in the ventral side of the aorta whose increase in thickness indicates hemogenic potential ) . We observed a heterogeneous density of eGFP-ZO1 at the interfaces between EHT cells and their endothelial neighbors , with an apparent increase at the anterior and posterior poles of the cell that persisted until their convergence to apparent coalescence , shortly before cell exit ( Figure 2A and Figure 2—video 1 ) . We then tracked semi-automatically the displacement of these eGFP-ZO1 enriched regions at the poles and used them as reference to follow the apical constriction as they converge concomitant to the closure of the circumferential junctional belt . This complemented , with a higher spatio-temporal resolution , our results presented in Figure 1—figure supplement 1D on the variability of the temporal course of the EHT process per se and led to two main conclusions . Firstly , the spots of eGFP-ZO1 high density at the two opposite poles of the EHT cell moved mainly along the A-P axis , often with an imbalance of migration distance ( Figure 2B and legend ) . Secondly and importantly , the distance between the two eGFP-ZO1 high-density regions decreased non-uniformly ( Figure 2C and Figure 2—figure supplement 1 ) , with phases of contraction and relative stabilization ( already visible in Figure 1—figure supplement 1D and better resolved spatio-temporally here owing to semi-automatic tracking , see Materials and methods ) . The successive phases of contraction and stabilization are characterized by oscillations of the closing speed of relatively homogenous periodicity ( Figure 2C , D ) , each oscillation resulting from acceleration and deceleration phases . The oscillations taking place during the contraction phases are the largest in amplitude and are interspersed by short periods of time during which the progression of the A-P distance reaches a minimum ( referred to as pseudo-plateaus , see Figure 2—figure supplement 1 ) . This oscillatory behavior of the apical constriction is reminiscent of the pulsatile/oscillatory behavior observed in several morphogenetic events during development ( Levayer and Lecuit , 2012 ) , although with significant differences in the biomechanical parameters of these oscillations ( see Discussion ) . We then thought of investigating more systematically the localization of eGFP-ZO1 in junctional regions across the entire aortic wall , as well as using the delineation of cell contours by this reporter to study in detail the EHT cellular landscape , notably the morphology of EHT cells and adjoining endothelial ones . Precise morphological information on the aortic cellular landscape is difficult to extract from Z projections or 3D views , and projections mask structures in the medial Z planes and distort distances . This prompted us to develop a specialized software tool able to unwrap the signal of the roughly cylindrical cell layer of the aorta onto a plane ( 2D plane images will be referred to as 2D-maps thereafter ) . This software allows the treatment of information through time , taking into account heterogeneities and variations in the shape of the aorta along its axis using semi-automatic adjustment of the surface ( see Figure 3A and Materials and methods for details ) . Using this software , applied mainly to eGFP-ZO1 TL sequences , we obtained a series of significant results . Firstly , at 48 hpf , as exemplified in Figure 3C–E and Figure 4 , the aortic circumference spans over three cells along most of its length . Secondly , its cellular landscape is composed of two major populations of cells , displaying similar A-P length but a different aspect ratio; one spanning the aortic floor and distinctly narrower in the medio-lateral axis , and the other one spreading across the entire aortic wall , including occasionally the floor ( Figure 4B , D and legend ) . In the TL sequence from which images presented in Figure 3C–D were extracted ( see Figure 3—video 1 ) , the cells of the former population ( cells e1 and e2 ) underwent shrinkage along the A-P axis , and performed an EHT . These data support the idea that the laterally narrower cells spanning the aortic floor constitute the hemogenic endothelium . Thirdly , hemogenic cells are capable of performing mitosis contemporarily to their EHT , as exemplified in Figure 3C , D in which cell e1 divided symmetrically ( based on their morphology ) along the A-P axis . After mitosis , the daughter cells continued their shrinkage along the A-P axis until their emergence . Mitosis during EHT took place in the aortic plane in 16% of our TL sequences , and in the sub-aortic space in 31% of them . Fourthly , the junctions between EHT cells and their neighbors are characterized by increased density of eGFP-ZO1 , particularly at the level of tricellular junctions and contacting surfaces that are oriented more or less aligned with the medio-lateral axis ( Figure 3C ( cells e1 and e2 ) , Figure 4 and Figure 3—video 1 , Figure 4—video 1 ) . During the EHT , these regions of high junctional density flanking the EHT cell converged while the cell apex constricted and merged ( Figure 4A and Figure 3—video 1 ) , confirming what we described in Figure 2A . At the end of the process , the junctional contacts newly established between adjoining endothelial cells merged into a single line of high eGFP-ZO1 density oriented perpendicular to blood flow ( Video 1 , t = 07:00 , yellow arrowheads ) . Fifthly , the bijunctional interfaces between endothelial cells that are not in contact with EHT cells and are oriented perpendicular to the A-P axis also often exhibit an apparent increase in eGFP-ZO1 density ( Figure 4A ) . Hence , we are tempted to deduce that the increase in tight junction components at bi- and tricellular contacts is aimed at reinforcing membrane interfaces that are under high mechanical tension . This would be consistent with the proposed function of tight junctions in the cross-talk with VE-cadherin dependent control of cell-cell tension ( Tornavaca et al . , 2015 ) . Finally , the number of EHT surrounding cells does not vary much during the EHT , with an average of five contacting cells at the beginning , getting down to three at the end of the process ( Figure 4C ) . In the case of EHT cells undergoing mitosis , the daughter cells were spread apart from each other by junctions being established between two adjoining endothelial cells ( Figure 3D , blue arrow ) . Most cases of morphogenic movements of tissues during development involve changes in cell geometry that rely on the longitudinal , medial-apical and/or circumferential contraction of actomyosin complexes composed of non-muscular myosin II ( Heisenberg and Bellaïche , 2013; Munjal and Lecuit , 2014 ) . In order to investigate actomyosin function in the morphological changes taking place during the EHT , particularly in the contraction of the circumferential junctional belt , we first looked at the localization of filamentous actin ( F-actin ) in the vascular system , using Lifeact-eGFP as a reporter ( Lifeact is a small non interfering peptide of 17 amino acids that allows visualization of actin dynamics and stains filamentous actin ) . We observed the recruitment of Lifeact-eGFP mainly at the cortico-apical side of the EHT cell , assembled as a belt at the interface with its endothelial neighbors with an apparent enrichment at its A and P poles ( Figure 5A , upper two rows ) . The 2D-map algorithm revealed F-actin enrichment at contacting surfaces that are more or less perpendicular to the blood flow ( Figure 5A , bottom panels ) . This localization is very similar to what we observed with eGFP-ZO1 , suggesting that F-actin contributes to the organization of sub-junctional regions and perhaps the stabilization of junctional complexes that are , as proposed before , the most exposed to mechanical tension . In addition , the data also support the idea that the apical constriction of EHT cells requires the contractile activity of a circumferential actomyosin belt . TL sequences performed with embryos expressing Lifeact-eGFP allowed us to follow the dynamics of cortico-apical circumferential actin ( Figure 5—video 1 , with its most informative images composing Figure 5B ) . This confirmed and extended our previous results , showing ( i ) an apical contractile actin belt that ultimately shaped into a contractile ring upon the narrowing down of the A-P luminal distance and ( ii ) the subsequent formation of an F-actin enriched foot that preceded the full release of the cell . We speculate that this latter pool of F-actin may result from the dissolution of the belt , and might serve as a propeller to trigger mechanical force required to complete the exit and/or the mechanical uncoupling of cell-cell junctions . The anisotropic organization of F-actin and ZO1 in hemogenic and EHT cells , with their enrichment at the antero-posterior poles more or less aligned with the blood flow axis , suggested that hemodynamic forces may have an influence on the cytoskeleton architecture in these cells as well as on the cytoskeletal/junctional interface . The involvement of blood flow in hematopoiesis was initially shown in seminal work in the zebrafish ( North et al . , 2009 ) in which the abrogation of heart beating and blood circulation impaired arterial identity and HSC formation ( using the silent heart ( sih ) mutant of cardiac troponin T [Sehnert et al . , 2002] ) . Hence , to question the involvement of hemodynamic forces on the cytoskeleton of hemogenic cells and , more broadly , the dynamics of HSC emergence , we interfered with heart beating using a sih morpholino ( injected in Tg ( kdrl:Ras-mCherry; kdrl:Lifeact-eGFP ) embryos to visualize F-actin and cellular membranes ) . We observed , in addition to major alterations in the morphology of the dorsal aorta ( Figure 6A ) , the recurrent absence of EHT cells with cup-shaped morphology and emerging toward the sub-aortic space ( 14 embryos injected with the sih morpholino were carefully analysed , all along the dorsal aortic wall ) . Rather , the aortic floor was decorated with ovoid cells , with their luminal/apical membrane protruding into the lumen of the aorta ( Figure 6B ) . The lumen also contained cells labelled with Lifeact-eGFP and Ras-mCherry ( hence from vascular origin and supposedly having performed their extrusion from the surrounding aortic wall since circulation is abrogated and cells in the aortic lumen do not migrate extensively in comparison to cells in the sub-aortic space , see Figure 6—video 1 for Z-sectioning of the aorta and visualization of the content of the aortic lumen ) . TL sequences performed in the trunk region revealed that the ovoid cells in the aortic floor do undergo mitosis and are able to emerge either toward the sub-aortic space or , more unexpectedly , toward the aortic lumen ( Figure 6—video 2 , right ) , which we never saw in the control condition ( this intra-aortic emergence was observed in 3/3 movies obtained from three independent injection experiments ) . Interestingly , in one of our TL sequences ( 1/3 , the longest TL sequence that lasted 13 . 30 hr ) , two of the released cells ended up bursting into pieces ( see Figure 6—video 2 , left ) , suggesting that cells released in the absence of mechanical constraints exerted by blood flow are prone to cell death . This is consistent with previous studies showing a reduction in the number of hematopoietic stem cells and progenitors upon sih morpholino treatment ( in particular in the caudal hematopoietic tissue , see Murayama et al . , 2006; Murayama et al . ( 2006 ) ; Bertrand et al . , 2008 ) . Altogether , these results support the idea that hemodynamic mechanical forces contribute to the inward bending of the apical/luminal membrane of EHT cells ( the cup-shaped cells ) as well as the direction of the emergence ( toward the sub-aortic space ) . They also suggest that the biomechanics of the emergence may impact downstream cellular functions and/or commitments that , if not fulfilled , impede cell survival . Interestingly , when heart beating and blood flow were temporarily arrested using tricaine methanesulfonate ( at 1 mg/ml ) , the inward bending of the apical/luminal membrane was not observed anymore and cells performing the emergence toward the sub-aortic space maintained their path ( 10 embryos were observed , all along the dorsal aortic wall , between a time window of 35–135 min after interruption of blood flow , data not shown ) . This result strongly supports the idea that components of hemodynamic forces ( among which the mechanical strain , the force perpendicular to the flow axis acting on the vascular wall ) are exerting a direct action on the apical/luminal membrane bending . Finally , we also investigated the effect of obliterating blood flow on F-actin and ZO1 organization . We managed to obtain reasonably good images after 2D-maping with the Tg ( kdrl:Ras-mCherry; kdrl:Lifeact-eGFP ) fish line , although with difficulties because of the strong distortion of the aortic shape in the sih embryos . Images clearly revealed a perturbation of F-actin organization in EHT cells , with little reinforcement of that staining at the antero-posterior poles of the cells in comparison to the control ( Figure 6B , bottom panels ) . Unfortunately , in the case of eGFP-ZO1 , fluorescence signals in the sub-cortical regions of hemogenic cells were too low and aorta too distorted to perform 2D-maping . However , 3D-projections and 3D-analysis with the Imaris software also revealed changes in eGFP-ZO1 localization in cells of the hemogenic endothelium , with altered recruitment at the sub-plasmalemmal level and apparent increase of the eGFP-ZO1 cytoplasmic pool ( Figure 6C and see Figure 6—video 3 for 3D-Imaris rendering ) . Altogether , these results support the idea that hemodynamic forces exert an action on the cytoskeletal architecture of hemogenic cells and impact on junctional organization . Time-lapse imaging of Lifeact-eGFP in embryos in which the EHT cells were best aligned with the A-P axis also suggested an active role of the adjoining endothelial cells in the process . The A and P neighbors moved toward each other over the actin-rich poles and appeared to extend actin-rich filopodia to establish contacts . Subsequently , filopodia anastomosis should contribute to the sealing of the aorta floor ( Figure 7—video 1 , with the main sequential steps composing Figure 7 , left panels ) . The live-imaging data presented so far led us to define the main , sequential steps of the EHT ( see the cartoons illustrating steps 1–5 , Figure 7 , middle and right panels , and details in the legend ) . We propose that the key features of the process are: ( i ) the initiation of the bending of an hemogenic cell toward the sub-aortic space ( with its antero-posterior axis more or less aligned in the direction of blood flow and the inward bending of its luminal/apical membrane depending on hemodynamic forces ) ; ( ii ) the anisotropic organization of circumferential F-actin and of junctional contacts influenced by the blood flow ( with enrichment at trijunctions and membrane interfaces perpendicular to blood flow/aorta A-P axis ) ; ( iii ) the anisotropic , stepwise and oscillatory constriction of the cell apex , accompanied by concomitant contraction of a circumferential F-actin belt ( essential molecular components of this contractile activity are myosin and associated light chains , see below ) ; ( iv ) the apparent deployment of pushing forces by adjoining endothelial cells possibly contributing to cellular bending and apical constriction; ( v ) the extension of filopodia by adjoining endothelial cells; ( vi ) the anastomosis of filopodia to seal the aortic floor; ( vii ) the release of the EHT cell after remodelling of the luminal membrane ( see Discussion ) and down-regulation of remaining junctional complexes . We then addressed the mechanism of contractility of the circumferential actin belt by investigating the role of myosin and , more specifically , the regulation of its activity by myosin regulatory light chains ( RLCs ) . We focused on the regulatory light chain 9 ( Myl9 ) because the myl9 gene was shown to be a target of Runx1 in a genome-wide analysis of the transcriptional profile and of Runx1 binding in a model of the mouse hemogenic endothelium in cell culture ( Lie-A-Ling et al . , 2014 ) . Owing to gene duplication in teleosts , the myl9 gene was duplicated in the zebrafish , leading to the expression of two paralogs bearing more than 93% identity and referred to as Myl9a and Myl9b ( for sequence comparison , see Figure 8—figure supplement 1A ) . By WISH , we confirmed that the myl9a mRNA is expressed in the dorsal aorta of 48 hpf embryos ( Figure 8A ) and we found that so is the case for myl9b . More precisely , we detected the myl9b mRNA in the brain , somites , and dorsal aorta ( owing to strong signals in the ventral part of the somites , transverse sectioning was necessary to visualize unambiguously the localization of myl9b mRNA in aortic cells , Figure 8A , right panel ) . We next investigated the incidence of decreasing the expression level of Myl9a or Myl9b on definitive hematopoiesis . We designed two pairs of splice-blocking morpholinos targeting each mRNA ( morpholino efficiencies were validated by qPCR , see Figure 8—figure supplement 1B , C ) , and injected them in Tg ( CD41:eGFP ) embryos . The overall development and blood circulation of injected embryos appeared normal . GFP-positive cells were then scored at 48 hpf , in the sub-aortic space ( AGM ) ( to be most proximal to the site of HSPCs emergence ) , and in the caudal hematopoietic tissue ( CHT ) , the first niche where aorta-derived HSPCs home . In both knock-downs , we observed a significant decrease in HSPC number in the AGM and the CHT ( Figure 8B–E ) . This phenotype was phenocopied by two additional morpholinos targeting myl9b and myl9a mRNAs respectively and was rescued for the spe3i3 myl9b morpholino ( Figure 8—figure supplement 2 ) . These data support the involvement of both Myl9a and Myl9b in definitive hematopoiesis , and possibly in the EHT process . Finally , in order to address the intracellular localization and dynamics of the two Myl9 isoforms during the EHT , we generated Tg ( kdrl:Myl9a-eGFP ) and Tg ( kdrl:Myl9b-eGFP ) fish lines . Since the fluorescence intensities of Myl9a and Myl9b fused to eGFP were much lower than in the case of Lifeact-eGFP , we were unable to treat our movies with the 2D algorithm . However , as seen in Video 2 , regular TL sequences show convincingly that Myl9a-eGFP and Myl9b-eGFP assemble as cortico-apical belts with enrichment at spots between EHT cells and neighbors that move , upon apical constriction , along the A-P axis , and coalesce . This is reminiscent of what we observed with Lifeact-eGFP and eGFP-ZO1 ( see Video 3 for a comparative ‘en face’ visualization of the dynamics of actin and myosin belts/rings ) . We also noticed that Myl9a and b were recruited at the contracting cytokinesis ring in EHT cells that underwent mitosis ( see Video 2 and its legend ) . Myl9b and Myl9a are very similar ( 93% identity calculated by ClustalO ) , and the most significant difference resides in their amino-terminus , where the two serines following the starting methionine ( which are strictly conserved in mammals and in most zebrafish RLCs , see Figure 8—figure supplement 1A ) are present in Myl9b , but not in Myl9a . These two residues are the substrates of protein kinase C ( PKC ) activity ( Ikebe et al . , 1987 ) . It was shown that their phosphorylation impairs the phosphorylation of the downstream Rho-kinase/MLCK/Citron-kinase target site ( Figure 8—figure supplement 1A ) that is essential for the stimulatory activity of RLCs on the ATPase activity of the myosin heavy chain ( Turbedsky et al . , 1997 ) . Hence , phosphorylation of this amino-terminal site inhibits myosin activity and actomyosin contraction . We mutagenized the Myl9b cDNA to substitute the two amino-terminal serine residues by alanines , and expressed the wt or mutant form ( Myl9bA2A3 ) fused to eGFP in zebrafish embryos by transient transgenesis . Expression of the mutant form using the kdrl promoter caused mild morphological alterations of the dorsal vascular system ( with maintenance of blood flow , no obvious change in the shape of the dorsal aorta and a few distortions at the base of intersegmental vessels , data not shown ) . To restrict the expression of Myl9bA2A3 or the wt form to aorta-derived HSPCs , we used the Runx’1 + 23’ enhancer ( Tamplin et al . , 2015 ) , which we confirmed to drive transient expression specifically in these cells ( Figure 9—figure supplement 1 ) . Using this promoter , we transiently expressed Myl9b-eGFP and Myl9bA2A3-eGFP in Tg ( Kdrl:Ras-mCherry ) embryos to visualize Myl9b-expressing HSPCs together with the vascular system ( Figure 9 ) . At 40 and 50 hpf , and in comparison with the wt form , we found a much lower number of HSPCs expressing the mutant form , both in the sub-aortic space and in the CHT ( Figure 9A , B ) . These results suggest that preventing N-terminal phosphorylation impairs definitive hematopoiesis and may interfere with the emergence from the aortic floor . Finally , we performed live imaging in this context , and made a series of observations . Firstly , while we occasionally found hemogenic cells in the aortic floor that expressed the wt Myl9b-eGFP and underwent EHT , we were unable to find such cells upon expressing Myl9bA2A3-eGFP , suggesting that the latter cells emerged at a significantly lesser frequency . Secondly , cellular fragments were often seen in the sub-aortic area , usually indicative of cell bursting . This cell fragmentation may have occurred during the EHT as described upon runx1 knock-down ( Kissa and Herbomel , 2010 ) , or in EHT-derived HSPCs as upon scl-α knock-down ( Zhen et al . , 2013 ) . An example of cell fragmentation probably taking place short after the emergence is shown in Figure 9C and Figure 9—video 2 . Thirdly , EHT-derived HSPCs expressing Myl9bA2A3-eGFP exhibited abnormal shapes and migration , characterized by anarchic membrane extensions in many directions at once ( Figure 9D and Figure 9—video 1 ) . Interestingly , they were still able to undergo mitosis , with Myl9bA2A3 relocating to the cytokinesis ring as wt Myl9b would do ( data not shown ) .
The inward bending of the luminal membrane of EHT cells is one of the hallmarks of the EHT that , to our knowledge , has never been described so far for any other type of cell extrusion . In the context of the EHT , the direction of extrusion is opposite depending on the species , i . e towards the sub-aortic space in the zebrafish and the aortic lumen in the mouse ( Klaus and Robin , 2017 ) . Although the two emergences proceed in opposite directions , it appears that the EHT , in these two species , may have something in common . Indeed , electron microscopic images of cells of the embryonic mouse aorta floor showed the presence of large intracellular vacuole-like structures ( Marshall and Thrasher , 2001 ) reminiscent of what is observed in the zebrafish embryo , raising the idea that the inward bending of the apical/luminal membrane may be followed by its intracellular release after the emergence . If the reason for the discrepancy regarding the direction of the emergence between the mouse and the zebrafish is currently unknown , this study contributes shedding light on this issue . When abrogating the blood flow by obliterating heart beating from the very beginning , we show that cells from the aortic floor can emerge in the aortic lumen ( presumably hemogenic cells endowed with the potential to become hematopoietic ) . Unless , in the zebrafish , hemogenic cells that have never been exposed to hemodynamic forces do not express the relevant cytosolic machineries essential for performing a sub-aortic emergence , this raises the possibility that the variation between the two species is a direct consequence of differences in the strength of components of hemodynamic forces , such as , for example , the mechanical strain ( the force perpendicular to the direction of the flow and that relates to the rhythm of heart beating , Hahn and Schwartz , 2009 ) . However , the differences in heart beating frequencies between the two species are not so far from each other at that developmental stage ( approximately 2–3 beats/s , [Xing et al . , 2018; Hahurij et al . , 2014] ) . Hence , it is not so clear if the sub-aortic orientation of the emergence of HSCs in the zebrafish may be a direct consequence of the mechanical strain; what might count as well regarding the difference between the two species are the physical properties of the respective aortic walls that should impact the way they respond to the force because of their diameters , much wider in the mouse than in the zebrafish ( approximately 300 μm in the mouse and 25 μm in the zebrafish embryos at the time window of the EHT ) . The period of the aortic development during which the EHT takes place is the one at which aortic cells are highly exposed to mechanical tensions . Indeed , it has been shown by using a VE-cadherin mechanical sensor , that young aortic cells ( at 48 hpf ) , are more exposed to tension than 2 days later , when the aorta has matured and the aortic cells align with the direction of the flow ( Lagendijk et al . , 2017 ) . Hence , elongated hemogenic and EHT cells , on the aortic floor , may also orientate and contract along that same axis ( as we show in our work with the 2D-maping results ) , to minimize environmental constraints and membrane tension . In this context , the membrane interfaces that remain the most exposed to mechanical tensions are most probably the ones that are enriched in ZO1 and actin and are oriented perpendicular to blood flow ( consistently with ZO1 enrichment at the interface between endothelial cells that spread over a large surface of the aorta and whose orientation is also perpendicular to blood flow ) . According to our results obtained upon abrogation of heart beating and showing alteration of actin anisotropy ( i . e the reduction in the recruitment of sub-cortical actin at the two poles of the cells more or less aligned with the blood flow axis ) , tensions exerted by the hemodynamics control sub-plasmalemmal cytoskeletal architecture in EHT cells and , consequently , the actin/junctional interface ( as suggested by the delocalization of ZO1 that we observe in these conditions ) . Altogether , these results support the conclusion that the EHT biomechanics are specifically adapted to - and influenced by - the mechanical tensions exerted by blood flow . It should be noted that this control by the flow might be direct ( ex: via cell surface mechanosensors ) , and/or indirect ( ex: via the induction of Runx1 , the transcription factor essential for hematopoiesis [Adamo et al . , 2009] ) . Indeed , hemodynamics and in particular shear stress was shown to control endothelial cell morphology and influence the expression of many transcription factors ( for example , for a link between flow , mechanotransduction , actin , transcription see Nakajima et al . , 2017 ) . The constriction of the apical/luminal side of EHT cells , like many apical cell constriction events described so far , should depend on the activity of actomyosin recruited at the junctional circumferential belt . The membrane interfaces that are the less enriched in actomyosin ( oriented along the antero-posterior axis ) , are the ones that shrink the most , supporting the idea that the constriction is also coupled with their consumption , possibly by endocytosis . EHT cell apical constriction exhibits peculiar biomechanical characteristics in comparison to other apical constriction mechanisms described so far . It is pulsatile , with successive phases of contraction and of stabilization , each one with a time-length varying stochastically ( see Figure 1—figure supplement 1D and Figure 2—figure supplement 1 ) . The pulsatility relies on variations of the apical closing speed that oscillates ( owing to acceleration and deceleration cycles ) , with more or less large amplitudes ( with larger amplitudes during the constriction than the stabilization phase ) and with periods of approximately 40 min on average . This average period value is longer than in the most extensively studied and quantified processes taking place during development , such as , for example , the apical-medial contraction of cells during ventral mesoderm invagination in Drosophila melanogaster embryos ( see the seminal article [Martin et al . , 2009] ) , or the apical-medial and apical-junctional mediated contractions during neuroblasts ingression ( see [An et al . , 2017; Simões et al . , 2017] ) ; the latter being more comparable to the EHT process since a single cell extrudes from an epithelium , the ectoderm . In these two processes , for which the duration of the apical constriction takes approximatey 6 min and 30 min , respectively , the period of each pulsatile cycle is in the minute range ( with , for example , an average of 3 . 2 constriction pulses over 6 min for ventral cells ( interspersed with inter-periodic intervals of similar values ) and of 28 . 8 constriction pulses over 30 min for neuroblasts ( value extrapolated from Simões et al . , 2017 ) ) . Finally , for the EHT , the inter-periodic intervals of the contraction phase take approximately 21 min on average ( which we referred to as pseudo-plateaus , see Figure 2—figure supplement 1 ) . Currently , the reason for such a discrepancy in the periods of pulsatile cycles of the apical constriction is not known . However , in comparison to the time scale of apical contraction in the neuroblast ingression process that also leads to cell extrusion ( lasting for approximately 30 min , as mentioned already ) , the timescale of the EHT process per se is relatively long ( average duration of approximately 10 hr ) . This suggests that the specific parameters of the oscillations of the apical closing speed for EHT cells that impact on contraction efficiencies ( in particular for periods and amplitudes ) correlate with the time scale of the whole process , suggesting that they may be controlled at the mesoscopic level , perhaps involving gene regulatory networks controlling cyclic expression of genes ( such as in the case of ultradian oscillations that have periods in the time-range of the hour , see Isomura and Kageyama , 2014 ) . Indeed , EHT cells are integrated into a developing lumenized tissue that is submitted to a series of mechanical tensions owing to aortic morphological maturation including ( i ) the adaptation of the endothelial wall to EHT cell emergence and the significant reduction of the aortic diameter with time ( Isogai et al . , 2001; Kissa and Herbomel , 2010; Sugden et al . , 2017 ) , ( ii ) the extensive modification of the structure of the sub-aortic space ( with the modification of the basal lamina contacting EHT cells and transducing forces via integrins; see the incidence of basal lamina on periods of oscillations [Gorfinkiel , 2016; Koride et al . , 2014] ) and , last but not least , ( iii ) complex hemodynamic forces . In such environment , the high mechanical tensions might tune the downstream molecular events controlling actin-driven directional forces and actomyosin recruitment and contractility ( for the relevance of high-order information , at the tissue level , on the biomechanics and role of contractile pulses and the theoretical description of pulsatile/oscillatory behaviour , see [Levayer and Lecuit , 2012] ) . In this study , we do show that the contraction of circumferential actomyosin accompanies the constriction of EHT cells; however , we do not show a correlation between apical pulses and oscillatory recruitment of actomyosin components in that specific region as performed in other studies , essentially for technical reasons ( the weakness of signals of myosin RLCs fused to fluorescent reporters in all the transgenic lines established is too close to background to allow extraction of oscillatory signals from fluctuating noise ) . Hence , we can only speculate that the peculiarities of the biomechanical characteristics of apical constriction discussed above rely on the regulation of actomyosin recruitment and contraction . However , interfering with Myl9b function by mutagenesis of its PKC phosphorylation site impairs hematopoiesis , which is consistent with the requirement for actomyosin activity during the EHT and points to the importance of a particular regulatory pathway . Indeed , the N-terminal phosphorylation site in Myl9b , shared with most other myosin RLCs , was shown to be the substrate of PKC ( Ikebe et al . , 1987 ) and , upon phosphorylation , to prevent the activating phosphorylation of a downstream site ( the MLCK site , the most characterized phosphorylation site of RLCs , see [Turbedsky et al . , 1997] ) . Hence , this reinforces the possibility that a major regulatory axis of the EHT process relies on controlling actomyosin activity , by regulating a complex interplay between activating and inhibitory kinases . In this scenario , there exists the possibility that Myl9b , via the phosphorylation of its N-terminus , contributes to the tuning of the mechanical parameters of the contraction . Finally , there may be additional complexity to the system , for another RLC , Myl9a , is expressed in the aorta and may be co-expressed with Myl9b in EHT cells . Myl9a is a paralog that appears to be more restricted to the zebrafish and lacks the N-terminal , PKC phosphorylation site . Since the knock-down of myl9a , as is the case for myl9b , impairs definitive hematopoiesis , both isoforms may be involved in the EHT , with a fine tuning of their molecular ratio . In conclusion , our study reveals key biochemical and biomechanical features of hematopoietic stem cell emergence in the zebrafish embryo and strengthens the significance of the mechanical constraints imposed by the environment and more specifically the blood flow . This work paves the way for future studies on this very unique process , including studying the contribution of forces exerted by the adjoining endothelial cells . Overall , it invites to biophysical modelling to point to the most relevant biomechanical features of the process .
Further information and requests for resources and reagents should be directed to and will be fulfilled by the corresponding Authors , Anne A . Schmidt ( anne . schmidt@pasteur . fr ) and Philippe Herbomel ( philippe . herbomel@pasteur . fr ) . Zebrafish ( Danio rerio ) of the AB background and transgenic fish carrying the following transgenes Tg ( kdrl:HsHRAS-mCherry ) s916 ( Chi et al . , 2008 ) also referred to as Tg ( kdrl:Ras-mCherry ) , Tg ( fli1a:Lifeact-eGFP ) zf495 ( Phng et al . , 2013 ) , Tg ( fli1ep:Gal4ff ) ubs3 ( Zygmunt et al . , 2011 ) , Tg ( 5xUAS:RFP ) nkuasrfp1a ( Asakawa et al . , 2008 ) , Tg ( UAS:eGFP-ZO1 ) ubs5 ( Herwig et al . , 2011 ) , Tg ( kdrl:eGFP ) s843 ( Jin et al . , 2005 ) , Tg ( −6 . 0itga2b:eGFP ) la2 ( Lin et al . , 2005 ) also referred to as Tg ( CD41:eGFP ) , Tg ( kdrl:Gal4; UAS:RFP ) , Tg ( kdrl:Myl9a-eGFP ) ip5 and Tg ( kdrl:Myl9b-eGFP ) ip6 ( this study ) were raised and staged as previously described ( Kimmel et al . , 1995 ) . Fish were maintained on a 14 hr light/10 hr dark cycle , and embryos were collected and raised at 28 . 5 or 24°C in N-Phenylthiourea ( PTU , Sigma Aldrich , Cat#: P7629 ) /Volvic source water ( 0 . 003% , final ) to prevent pigmentation implemented with 280 μg/L methylene blue ( Sigma Aldrich , Cat#: M4159 ) . The embryos used were of early developmental stages ( 26–50 hpf ) , precluding sex determination of the animals . The general fish maintenance at the Institute follows the regulations of the 2010/63 UE European directives and is supervised by the veterinarian office of Myriam Mattei . All constructs were generated using the Gibson Assembly method ( NEB , Cat#: E2611S ) . The myl9a and b sequences were amplified from a pool of cDNAs made using 24 to 48 hpf whole zebrafish embryos ( see Key Resources Table for the primers used , Super Script III kit , Cat#: 8080093 and Expand High Fidelity polymerase , Sigma Aldrich , Cat#:11732641001 ) . The kdrl promoter ( Flk ) ( Jin et al . , 2005 ) was used to drive endothelial expression of Myl9a and Myl9b fused at their carboxy-terminus , with eGFP ( Clontech ) or mKate2 ( Evrogen ) . The murine runx’1 + 23’ enhancer ( Tamplin et al . , 2015 ) coupled with the murine betaglobin was used to drive hemogenic expression . Final constructs were purified using the NucleoBond Xtra Midi endotoxin free kit ( Macherey Nagel , Cat#: 740420 . 10 ) and 25 ng/μl were co-injected into one-cell stage zebrafish embryos along with 25 ng/μl tol2 transposase mRNA transcribed from linearized pCS-zT2TP ( Suster et al . , 2011 ) plasmid using the mMESSAGE mMACHINE SP6 kit ( Ambion , Cat#: AM1340 ) ( Kawakami , 2007; Kotani et al . , 2006 ) . Embryos were screened for fluorescence between 24 hpf and 48 hpf , selected for imaging or raised to adulthood for stable transgenesis . Founders were isolated by screening for fluorescence . For stable transgenesis , the following plasmids have been injected: pG1-flk-MCS-tol2-Myl9a-eGFP or pG1-flk-MCS-tol2-Myl9b-eGFP . For transient transgenesis , the following plasmids have been used: pTol2-Runx1+23-1-betaglobin-myl9b-eGFP , pTol2-Runx1 + 23-1-betaglobin-myl9bA2A3-eGFP , pTol2-Runx1 + 23-1-betaglobin-Myl9a-mKate2 , pUAS:eGFP-hs-ZO1 . The Myl9b phosphorylation mutant form was generated by site directed mutagenesis using a sense 5-prime primer containing three nucleotide substitutions and PCR amplification using the wild-type myl9b cDNA as template . The sense primer also contained a Kozak sequence . Sense and anti-sense primers contained extensions for cloning the PCR product into the pTol2-Runx1+23-1-beta-globin:eGFP using the Gibson Assembly method . The double mutation was verified by sequencing . Sense mutagenesis primer ( with the substituted nucleotides in underlined bold ) : CAGACATCCCGCGGTGGAGCTCCAGGTCGCCACCATGGCCGCCAAAAGAGCAAAGGGGAAGACCACCAAG Reverse primer: CGCCCTTGCTCACCATGGTGGCGACGAATTCCATGTCGTCTTTGTCTTTGGCTCCGTG Splice blocking morpholinos ( MOs , see Key Resources Table for sequences ) , myl9aspe2i2 , myl9aspi3e4 , myl9bspe2i2 , myl9bspe3i3 , and as well as the sih Tnnt2 translation start codon and flanking 5-prime sequence MO ( Sehnert et al . , 2002 ) were prepared in stock solutions at 2 mM in ddH2O . 1 . 5 ( sih MO ) or 4–12 ng ( myl9a and myl9b MOs ) were injected into one-cell stage zebrafish embryos . The same amount of MO targeting a mutated form of the human beta-globin pre-mRNA was used as control ( Standard Control from GeneTools ) . Splice blocking efficiency was confirmed for each myl9 MO using qRT-PCR assays ( see Key Resources Table for sequences ) and the resulting phenotypes were assessed by manual counting of CD41:eGFP-positive cells in the AGM and the CHT of 40 and 50 hpf embryos . Results obtained using the MOs myl9aspe2i2 ( MO myl9a , 12 ng ) and the MO myl9bspe3i3 ( MO myl9b , 12 ng ) are presented in Figure 8 . For sih morphants , embryos that were imaged were checked for absence of heart beating 24 hr after injection as well as before and after experimental analysis . Capped and poly-adenylated full-length myl9b mRNAs were synthesized using the mMESSAGE mMACHINE kit ( Ambion , Cat#: AM1340 ) and the Poly ( A ) Tailling kit ( Ambion , Cat#: AM1350 ) from myl9b cDNA amplified by PCR from our pG1-flk-MCS-tol2-Myl9b-eGFP expression vector . 100 or 200 pg of myl9b mRNA were microinjected into one-cell stage Tg ( CD41:eGFP ) embryos either alone or in combination with 6 ng or 12 ng of myl9bspe3i3 morpholino . The resulting phenotypes were assessed by manual counting of CD41:eGFP-positive cells in the AGM and the CHT of 48 and 50 hpf embryos . Whole-mount chromogenic in situ hybridization was performed as described previously ( Thisse and Thisse , 2014 ) . Briefly , embryos were hand dechorionated , fixed in 4% FA ( Polysciences , Cat#: 040181 ) overnight at 4°C and stored in 100% MeOH at −20°C . Embryos were permeabilized using 10 μg/ml Proteinase K ( Ambion , Cat#: AM2546 ) in PBT ( 1x PBS + 0 . 1% Tween ) for 5 min and post-fixed in 4% FA for 20 min . Hybridization was performed using 200 ng Digoxigenin-labeled myl9a ( RefSeq: NM_001006027 ) and myl9b ( RefSeq: NM_213212 ) probes synthetized using the manufacturer recommendation ( T7 RNA polymerase , Promega , Cat#: P2075 , DIG-nucleotides , Jena Bioscence , Cat#: NU-803-DIGX ) . Detection of the probes was performed by incubating the embryos with an anti-Digoxigenin antibody coupled to alkaline phosphase ( Roche , Cat#: 11207733910 ) diluted 1:5000 in blocking solution 1X WBR ( Roche , Cat#: 11921673001 ) . The staining reaction was performed using NBT/BCIP ( Sigma , Cat#: N6639 , Cat#: B-8503 ) in NTMT buffer ( 100 mM Tris HCl pH 9 . 5 , 50 mM MgCl2 , 100 mM NaCl , 0 . 1% Tween ) . Subsequently to the revelation , embryos were washed in PBT and incubated in 30% sucrose for 2 days at 4°C and with agitation before embedding in tissue freezing medium blue ( Electron Microscopy Sciences , Cat#: 72592-B ) and frozen in liquid nitrogen . Transverse sectioning was performed at 5 μm using a CM 3050 cryostat ( Leica ) and slices were dried for 2 hr before mounted with AquaPoly Mount medium ( Biovalley , Cat#: 18606–5 ) . Images were captured on an Olympus vs120 slide scanner using an ORCA Flash4 . 0 V2 CMOS camera ( Hamamatsu ) and a UPLSAPO 20X objective ( Olympus ) . To determine the splice-blocking efficiency of myl9a and myl9b MOs , cDNA was extracted from 20 whole 48 hpf embryos and qRT-PCR oligos were designed to hybridize on the joined sequential exons initially surrounding the splicing sites targeted by the MOs ( see Key Resources Table for sequences ) . One-cell-stage embryos were injected with 4–8 ng of myl9a or myl9b or Standard control MOs ( see Key Resources Table for sequences ) and raised until 48 hpf . Total mRNA was extracted using the RNeasy Plus Mini Kit ( Qiagen , Cat#: 74134 ) as indicated by the manufacturer and the concentration was obtained by nanodrop quantification ( Nanodrop1000 , ThermoFisher ) . Equivalent amounts of RNA ( 300 ng ) were used as template for cDNA synthesis , which was performed using the M-MLV Reverse Transcriptase kit ( Invitrogen , Cat#: 28025013 ) . qRT-PCR was performed using Takyon Rox SYBR Master mix blue dTTP kit ( Eurogentec , Cat#: UF-RSMT-B0701 ) , 15 ng of cDNA and 10pM of each primer . qRT-PCR was carried out for three biological replicates with measurements taken from three technical replicates on an Applied Biosystems 7300 Real Time PCR system ( Thermofisher ) . The relative expressions of myl9a and 9b were determined after normalization to zebrafish elongation factor 1a ( eif1α ) . The conditions from the samples injected with the myl9 MOs were compared to the Control ones using the delta-delta-Ct method . To determine the potential off target effect of myl9a MOs on myl9b expression and of myl9b MOs on myl9a expression ( see Key Resources Table for oligo sequences ) , RNA extraction was performed as described above; cDNA synthesis was performed using the SuperScript III kit ( ThermoFisher , Cat#: 8080093 ) , equivalent amount of RNA ( 200ng ) and qRT-PCR were performed as described above . The graph presented in Figure 8—figure supplement 1D corresponds to two independent experiments carried out in triplicates . Due to the low n number ( n = 2 ) , this graph indicates a tendency . Embryos were hand dechorionated and anesthesized using tricaine ( Sigma Aldrich , Cat#: A5040 ) . Then , they were mounted in lateral position and embedded in 1% low melting agarose ( Promega , Cat#: V2111 ) in a glass-bottom μ-Dish ( Ibidi , Cat#: 81158 ) . To avoid movements and pigmentation during image acquisitions , 1x tricaine/1 x PTU Volvic water was incorporated to the low melting agarose and , after solidification , added on top of it . Embryos were imaged on an inverted microscope ( UltraVIEW VOX , Perkin Elmer ) equipped with a Yokogawa CSU-X1 spinning-disk , an ImagEM-X2 camera ( Hamamatsu ) , 488 nm and 561 nm laser lines for excitation , a Zeiss LD C-Apochromat 40x corrM27 water objective ( NA 1 . 1 , WD 0 . 62 mm at cover glass 0 . 17 μm ) and the Volocity ( Perkin Elmer , http://www . perkinelmer . com/ ) acquisition software . Optical Z planes were spaced by 0 . 6 μm , the power of the lasers ( around 150 mW ) and exposure time ( between 50 ms and 200 ms ) were adjusted depending on the fluorophore and the transgenic line used . In all cases , acquisitions of a Z stacks containing the whole dorsal aorta was reduced to 1 min or less and spaced every 2 min . For TL sequences longer than 1 hr , samples were maintained at 28 . 5°C using an Okolab cage incubator and the objective was immersed in Immersol W 2010 oil ( Zeiss , Cat#: 444969-0000-000 ) . Embryos were imaged using an inverted microscope ( Leica TCS SP8 , CTR 6500 ) , equipped with 488 nm and 561 nm diodes for excitation , a Hybrid Detector , a Leica HC PL APO CS2 40x water objective ( NA 1 . 1 , WD 0 . 65 mm at cover glass 0 . 17 μm ) and the LAS X acquisition software ( Leica , http://www . Leica-microsystems . com ) . Optical planes were spaced by 0 . 43 μm and Z stacks containing the whole dorsal aorta were acquired every 5-10 min . For long time-lapse imaging , samples were maintained at 28°C using a INU stage Top Incubator ( Tokai Hit ) and the objective was immersed in 1:1 AquaSound Clear ( Free-med , Cat#: FU00071 ) ultrasound gel with water . For cell counting in live embryos , an HC PL APO CS2 20x IMM objective ( NA 0 . 75 , WD 0 . 67 mm ) and a resonant scanner were used for high-speed imaging of the whole AGM or CHT . Subsequent maximum projection and manual counting was performed . Some embryos were imaged with an upright microscope ( Leica SPE , CTR 6000 ) , equipped with 488 nm and 532 nm diodes for excitation , an immersion HCX APO LUV-I 40x water objective ( NA 0 . 8 , WD 3 . 3 mm ) and the LAS AF version 2 . 6 . 0 . 7266 acquisition software ( Leica ) . Optical planes were spaced by 0 . 6 μm and Z stacks containing the whole dorsal aorta were acquired every 5 min . The algorithm developed for the projection of the aortic wall on a 2D plane , using fluorescent signals , was called AortaTracker . It is decomposed in two sections aimed at ( 1 ) the segmentation of the aortic tube , ( 2 ) the unwrapping . To perform the segmentation of the tube , the algorithm first needs to be initialized by the user . To circumvent the variations in the aortic shape , movements , and signals coming from neighboring structures , a strong constrained surface is imposed to the aortic wall ( this part is available as an Icy plugin ( http://icy . bioimageanalysis . org ) called TubeSkinner; [de Chaumont et al . , 2012] ) . The aorta tube is rotated so that its axis becomes almost aligned with the Z-axis . On the first Z-section , the user then manually adjusts a circle fitting the inner part of the aorta . In such a section , the aorta is represented as a few numbers of bright spots representing the intersection of the plane with cellular membranes . The circle is then adjusted to maximize the sum of the fluorescent signals in a crown around its diameter , and to minimize the sum of intensity inside the inner circle of the crown . This procedure is repeated along the tube , effectively locating its central axis . Then for each circle , the intensity is collected along its perimeter as the maximal signal in the crown for a given angle . Finally , the unwrapper algorithm creates the image by transforming the point in the z-stack to the ( theta , r ) coordinate system considering the circle fit , making sure the final calibration is isotropic . No interpolation is performed on pixel intensity . The X-axis of the final image corresponds to the perimeter of the circles , and its Y-axis corresponds to the tube central axis . The obtained 2D-maps were then duplicated to merge the rims of unwrapped aorta and visualize the continuity of all cell contours . 3D-rendering of live-imaging have been processed using the Imaris software ( Bitplane , http://www . bitplane . com/ ) . We considered the Antero-Posterior luminal distance ( A-P distance ) , as the length between the most anterior and posterior luminal poles of the EHT undergoing cell , aligned with the A-P axis . In the case of EHT cells emerging parallel to the A-P axis , we used the ‘Straight Line’ tool on Fiji software ( NIH , https://imagej . net/Fiji ) and measured the A-P distance ( for each time points in Figure 1—figure supplement 1 ) on 2 or three consecutive Z-planes localized in the central part of the EHT cell ( where the apparent A-P distance was the longest ) and averaged . For cells emerging at a small angle relative to the A-P axis , we measured the apparent A-P distance on the maximum projection along the Z-axis; when the angle was larger , we first performed a 3D-rotation on Imaris software so as to align the cell in the X-Y plane before doing the measurement . In order to measure A-P luminal distances and dynamics ( Figure 2 ) , dense eGFP-ZO1 extremities were independently tracked using the Fiji plugin TrackMate ( Tinevez et al . , 2017 , https://imagej . net/TrackMate ) after maximum projection along the Z-axis and X , Y translation using the Fiji plugin ‘Descriptor based series registration’ to correct the X , Y drift ( Preibisch et al . , 2010 ) . Tracks display and subsequent distance analysis were performed using the MATLAB software ( MATLAB 2017a , , The MathWorks , https://fr . mathworks . com ) . To focus on long scale activity rather than short displacements , the distance between the two extremities was filtered twice using an unweighted sliding-average over 20 min . Following this process , we characterized the speed at which the A-P luminal distance decreases by automatically extracting the closing speed local minima and filtering them based on their prominence ( >0 . 01 µm/min ) . For additional details see https://fr . mathworks . com/help/matlab/ref/islocalmin . html . Distance between successive minima was used to measure the acceleration-deceleration cycle duration distribution ( for the algorithm and codes in MATLAB ( version 2017b ) , see the link: https://github . com/sebherbert/EHT-analysis ( Herbert , 2018; copy archived at https://github . com/elifesciences-publications/EHT-analysis ) . Distances in x were measured between the most anterior and posterior positions of the tracks . Cells were manually delimited using the ‘Polygon Selections’ tool on Fiji software and areas directly extracted from the measurement tool in Fiji . The length ( parallel of the blood flow axis ) and width ( along the circumference of the aortic wall ) were extracted from the rectangle that best fits the cell . To classify the aortic cells in one category or another ( endothelial , potentially hemogenic or EHT cell , see Figure 4 ) , we defined a series of criteria that take into account their respective length ( along the antero-posterior axis ) , width ( along the circumference of the aortic wall ) and positioning in regard to the medio-lateral axis . By looking through the Z-planes and selecting the one for which the aorta is of the largest diameter in the dorso-ventral axis , we first determined the ventral floor axis and placed it on the 2D-map ( see Floor on the maps ) . The roof ( see Roof on the maps ) was subsequently positioned at a distance corresponding to half the length of the unwrapped aorta ( allowing to determine the surfaces of the frontside and backside ) . We then applied a relatively stringent cut-off and considered that the pseudo-hemogenic cell category did not extend beyond the medio-lateral axis of the aorta while spanning the ventral floor . Cells were considered as neighbors when at least a part of their lateral junctions were in direct contact . Statistical analysis was performed using Student unpaired t test with Welch’s correction ( does not assume similar SD ) in GraphPad Prism six software ( https://www . graphpad . com/ ) . Graphs depict Mean ± SEM or SD or median alone and differences with p value ≤ 0 . 05 ( * ) were considered statistically significant . The AortaTracker algorithm for 2D projections is available as an Icy plugin ( http://icy . bioimageanalysis . org ) . | As humans , we have two major types of blood cell: our red blood cells transport oxygen around the body , while our white blood cells fight disease . Both types of cell come from the same stem cells , which first appear early in embryonic development . These stem cells emerge from the walls of major blood vessels , including the aorta – which carries blood from the heart . Stem cells have not yet decided which adult cell to become . Given the right signals , blood stem cells can form red blood cells or any of the different types of white blood cell . Understanding this process could allow scientists to recreate it in the laboratory , making blood stem cells that can give rise to all blood cells found in the body . But , this is not yet possible because we do not know all the conditions needed to make the cells and ensure they survive . One crucial gap in our understanding concerns the importance of blood flow . As the main blood vessel leaving the heart , the aorta experiences mechanical stress every time the heart beats . Lancino et al . wanted to find out whether this influences the development of the blood stem cells . Zebrafish embryos are transparent , making it easy to see their bodies developing under a microscope . Like humans , they also produce both red blood cells and white blood cells meaning Lancino et al . could watch the birth of blood stem cells in these embryos from a part of the aorta called the aortic floor . A new software tool unwrapped pictures of the tube-shaped blood vessel into flat , two-dimensional maps , making it possible to see how the aorta changed over time . This revealed that , as blood stem cells leave the aortic floor , they bend and contract with the direction of the blood flow . Rings of actin and myosin proteins that formed around the stem cells as they are born helped the process along , while stopping the heartbeat changed the way the blood cells emerged . Without any blood flow , the actin proteins did not assemble properly; the stem cells also emerged in the wrong direction and some of them even died . These findings show that physical forces play a role in the formation of blood stem cells . Understanding this process brings scientists a step closer to recreating the conditions for making different kinds of blood cells outside of the body . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"developmental",
"biology",
"cell",
"biology"
] | 2018 | Anisotropic organization of circumferential actomyosin characterizes hematopoietic stem cells emergence in the zebrafish |
Kinesin super family protein 2A ( KIF2A ) , an ATP-dependent microtubule ( MT ) destabilizer , regulates cell migration , axon elongation , and pruning in the developing nervous system . KIF2A mutations have recently been identified in patients with malformed cortical development . However , postnatal KIF2A is continuously expressed in the hippocampus , in which new neurons are generated throughout an individual's life in established neuronal circuits . In this study , we investigated KIF2A function in the postnatal hippocampus by using tamoxifen-inducible Kif2a conditional knockout ( Kif2a-cKO ) mice . Despite exhibiting no significant defects in neuronal proliferation or migration , Kif2a-cKO mice showed signs of an epileptic hippocampus . In addition to mossy fiber sprouting , the Kif2a-cKO dentate granule cells ( DGCs ) showed dendro-axonal conversion , leading to the growth of many aberrant overextended dendrites that eventually developed axonal properties . These results suggested that postnatal KIF2A is a key length regulator of DGC developing neurites and is involved in the establishment of precise postnatal hippocampal wiring .
In the mammalian nervous system , kinesin super family proteins ( KIFs ) play a crucial role in intracellular transport , microtubule ( MT ) dynamics , and signal transduction and , hence , are key players in brain function , development , and disease ( Hirokawa et al . , 2010; Hirokawa , 1998 ) . KIF2A belongs to the kinesin-13 family ( M-kinesin: internal motor domain family ) ( Aizawa et al . , 1992; Noda et al . , 1995; Miki et al . , 2001 ) , which destabilizes MTs in an ATP-dependent manner ( Desai et al . , 1999; Howard and Hyman , 2007; Moores and Milligan , 2008 ) . In the early stages of the developing murine nervous system , KIF2A controls neurite elongation by regulating MT dynamics at neuronal growth cones and plays a crucial role in neuronal migration , axonal elongation and axon pruning in vivo ( Homma et al . , 2003; Noda et al . , 2012; Ogawa and Hirokawa , 2015; Maor-Nof et al . , 2013 ) . Recently , studies of KIF2A function in humans have primarily focused on cortical development because KIF2A mutations in residues Ser317 and His321 have been identified in patients with malformed cortical development ( MCD ) ( Cavallin et al . , 2017; Poirier et al . , 2013 ) . Both mutants are expected to lose MT destabilizing activity , due to a disruption in ATP binding or hydrolysis , thus resulting in a classic form of lissencephaly . After reaching its peak in the early postnatal weeks , however , the expression of postnatal KIF2A throughout the brain is gradually restricted to specific brain regions , including the hippocampus ( Lein et al . , 2007 ) , in which new neurons are generated throughout an individual’s life in established neuronal circuits ( Gonçalves et al . , 2016; Eriksson et al . , 1998; Kaplan and Hinds , 1977 ) . This postnatal expression pattern suggests that KIF2A might be involved in adult neurogenesis , neuronal migration , and the establishment of refined neuronal circuits in these brain regions , but this role has not yet been confirmed because conventional knockout mice die within 1 day of birth ( Homma et al . , 2003 ) . In this study , we generated tamoxifen-inducible Kif2a conditional knockout ( Kif2a-cKO ) mice to demonstrate the postnatal role of KIF2A in the postnatal brain . We began tamoxifen injections in postnatal week 3 ( 3w ) , when the cortical neurons or major cranial nuclei had nearly completed their migration . The 3w-Kif2a-cKO mice showed successful neuronal migration , but all mice died by postnatal week 6 and showed signs of hyperactivity , weight loss , and temporal lobe epilepsy ( TLE ) . In the postnatal hippocampus , KIF2A expression was histologically restricted to the dentate mossy fibers ( MFs ) , and the loss of KIF2A-induced MF sprouting ( MFS ) and aberrant recurrent excitatory circuits . In our 3w-Kif2a-cKO mouse model , unlike the typical TLE mouse model , the dentate granule cells ( DGCs ) extended aberrant axons through both the inner and outer molecular layers ( IML and OML , respectively ) . Intriguingly , primary cultured P3-Kif2a-cKO DGCs did not regulate axonal or dendritic length , and consequently , the characteristics of the overextended dendrites changed , resulting in axonal conversion . These results suggested that postnatal KIF2A is a key length regulator of DGC developing neurites and is crucial for establishing postnatal hippocampal wiring .
To determine the role of KIF2A in the postnatal brain , we generated tamoxifen-induced Kif2a conditional knockout mice ( Kif2a-cKO , Figure 1A ) . Before tamoxifen injection , these mice were normal in appearance and did not exhibit any abnormal phenotypes . We injected both wild-type ( WT ) and Kif2a-cKO siblings with tamoxifen for 7 days during the third postnatal week , after the peak expression of KIF2A in the hippocampus ( Figure 1B and C ) . In addition , by the end of the second postnatal week , cortical neurons have almost finished migration and the injection timing was chosen to minimize the neuronal migratory defects in the developing cortex , which are severe in conventional knockout mice ( Homma et al . , 2003 ) . As a result , KIF2A expression was lost in the cKO brain within 1 week of tamoxifen injection ( Figure 1D ) . These cKO mice were designated 3w-Kif2a-cKO mice because the tamoxifen injections began at postnatal week 3 . As a control for the Kif2a-cKO mice , WT siblings were used in all experiments after confirmation that the phenotypes of all siblings except for tamoxifen-injected Kif2a-cKO mice were not significantly different . During the postnatal week 4 , the 3w-Kif2a-cKO mice became smaller than the WT siblings ( Figure 1E ) and showed weight loss ( Figure 1F ) . They also gradually developed hyperactivity ( Figure 1G ) , twitching , and seizures . An open-field test showed that almost half of the 3w-Kif2a-cKO mice experienced an epileptic seizure within 30 min ( Figure 1H ) . Eventually , all 3w-Kif2a-cKO mice died by postnatal day 42 ( P42 ) , the end of the postnatal week 6 ( Figure 1I ) . Among them , some 3w-Kif2a-cKO mice died immediately after experiencing severe epileptic convulsions , which may have been one of the causes of death . The source data of body weight and activity of 3w-Kif2a-cKO mice and all siblings were shown in Figure 1—source data 1 . To determine the focal point of the seizures , we simultaneously recorded electroencephalograms ( EEGs ) and behavior during postnatal week 5 . The electrodes were inserted into the hippocampus and the cortex of WT and 3w-Kif2a-cKO siblings . In the hippocampus , 3w-Kif2a-cKO mice showed aberrant spikes in the EEG ( arrowheads in Figure 2A ) that coincided with twitching in the resting or locomotive state ( Figure 2B and C ) . Moreover , during the epileptic seizure ( Figure 2—video 1 ) , the paroxysmal EEG events were clearly detected in the hippocampus but not in the cortex ( Figure 2D ) . The source data of those EEG was shown in Figure 2—source data 1 . These results suggested that the loss of postnatal KIF2A resulted in an epileptic hippocampus in 3w-Kif2a-cKO mice . Supporting evidence for the epileptic hippocampus was provided by three experiments: a histological analysis of the hippocampus of 3w-Kif2a-cKO siblings using Bodian’s silver staining method , an immunohistological analysis of frozen hippocampal sections of 3w-Kif2a-cKO siblings using an anti-glial antibody to detect gliosis , and a physiological analysis of the cultured hippocampal slices from P3-Kif2a-cKO siblings collected on P5 . The third experiment required P3-Kif2a-cKO mice because P4-P6 mice should be used for hippocampal slice cultures ( Ikegaya , 1999 ) . The first analysis demonstrated that , although there were no apparent laminar defects ( Figure 2E ) , the DGCs of 3w-Kif2a-cKO mice developed many kinks and defasciculated axons in the CA3 region ( Figure 2F ) , and were hypertrophic , scattered ( Figure 2G ) and swollen ( Figure 2H ) at the end of the fifth postnatal week , all possible features of hippocampal sclerosis ( Shibley and Smith , 2002 ) , a subsequent complication of hippocampal epilepsy . Moreover , when we began the tamoxifen injections 1 week later , at postnatal week 4 , 4w-Kif2a-cKO mice clearly showed features of hippocampal sclerosis ( Figure 2—figure supplement 1A ) with CA1 loss ( Figure 2—figure supplement 1B ) , mossy fiber sprouting ( Figure 2—figure supplement 1C ) , and hypertrophic scattered DGCs ( Figure 2—figure supplement 1D ) . In the second analysis , frozen sections were stained with an anti-glial fibrillary acidic protein ( GFAP ) antibody because gliosis is also a sign of an epileptic hippocampus ( Pollen and Trachtenberg , 1970; Gupta et al . , 1999; Loewen et al . , 2016 ) . Importantly , the 3w-Kif2a-cKO hippocampus contained more GFAP-positive astroglia ( Figure 2—figure supplement 1F and Figure 2—figure supplement 1G , Figure 2—figure supplement 1—source data 1 ) than the WT ( Figure 2—figure supplement 1E ) . In the third analysis , we attempted to demonstrate the endogenous development of excitatory recurrent circuits in the P3-Kif2a-cKO hippocampus . We dissected the hippocampus at postnatal day 5 ( P5 ) , sliced it for culturing , and performed an electrophysiological analysis at 10 days in vitro ( DIV10 ) . We then placed a stimulating electrode into the hilus and a detecting electrode into the granule cell layer ( GCL ) in the dentate gyrus ( Figure 2—figure supplement 1H ) to record the presence of excitatory signals , which would indicate the development of excitatory recurrent circuits in the P3-Kif2a-cKO slice . As shown in Figure 2J , an apparent paroxysmal depolarization shift ( PDS ) was observed in the P3-Kif2a-cKO slice ( Figure 2K , Figure 2—source data 2 ) but not in the WT hippocampus ( Figure 2I ) , suggesting that the P3-Kif2a-cKO hippocampal slices had endogenously developed excitatory recurrent circuits without application of any excitatory drugs , such as picrotoxin . Together , the results suggested that recurrent excitatory circuits are endogenously induced by the loss of KIF2A without extrinsic excitation . To demonstrate how the loss of postnatal KIF2A contributes to the development of an epileptic hippocampus , we first analyzed neurogenesis and cell migration in the dentate gyrus because abnormally generated or migrated DGCs often affect epileptogenesis , seizure frequency , and seizure severity ( Korn et al . , 2016; Hester and Danzer , 2013; Koyama et al . , 2012; Houser , 1990 ) . Two types of thymidine analogs , 5-chloro-2’-deoxyuridine ( CldU ) and 5-iodo-2’-deoxyuridine ( IdU ) , were injected for 7 days before and after tamoxifen injection to detect the newly synthesized DNA of replicating cells before and after the loss of KIF2A ( Figure 3A ) . The brains of injected mice were fixed at P35 , and sliced sections were stained with anti-CldU and anti-IdU antibodies . Importantly , the numbers of CldU- and IdU-positive cells were not significantly different between WT and 3w-Kif2a-cKO slices ( Figure 3B , Figure 3—source data 1 ) . Then , the vertical distance from the baseline of the GCL ( white broken lines in Figure 3C–3F ) to the dU-positive cells was calculated . When cells migrated into the GCL or hilus ( white and blue arrowheads in Figure 3E and F , respectively ) , the distance was given a plus ( + ) or minus ( - ) value , respectively . Migration histograms show that CldU-positive cells migrated farther than IdU-positive cells , but the difference in the cellular distribution between the WT and 3w-Kif2a-cKO mice was not significant ( Figure 3H , compared with 3G , see also Figure 3I , Figure 3—source data 1 ) . Before further experiments were conducted to elucidate the contribution of KIF2A to the development of an epileptic hippocampus , we analyzed the detailed distribution of KIF2A in the hippocampus by using an anti-KIF2A antibody at P21 . As shown in Figure 4A , KIF2A expression was highly localized in the hilus and stratum lucidum of WT mice , where MFs were found ( white arrowhead ) , whereas this effect was absent in 3w-Kif2a-cKO mice ( Figure 4B ) . MFs are the excitatory axons of DGCs ( Watson et al . , 2012 ) and create synapses with their targets , which are pyramidal cells in the CA3 , mossy cells in the hilus , and basket cells in the dentate gyrus ( Amaral and Dent , 1981 ) . In addition , MFs are closely related to TLE as MFS is often observed in the hippocampus of patients and animal models of TLE . Thus , we hypothesized that KIF2A specifically regulates MF elongation and that the loss of KIF2A induces MFS , thus resulting in aberrant excitatory circuits and an epileptic hippocampus . Early reports have also suggested that sprouted MFs contribute to TLE pathogenesis ( Kwak et al . , 2008 ) . However , MFS is known to be intimately involved in the deterioration and chronicity of TLE ( Koyama et al . , 2004 ) . When TLE occurs , excitation results in MFS , after which these collaterals recurrently elongate into the IML of the dentate gyrus where they form excitatory recurrent circuits . In short , MFS is thought to be a result of TLE . To verify our hypothesis , we first determined the final destinations of the MFs by using Timm’s staining methods . Timm sulfide silver staining is a histochemical technique used to visualize the spatial distribution of MF terminals , which specifically express high levels of Zn2+ ( Danscher and Zimmer , 1978 ) . For controls , we prepared two different samples: a pilocarpine-induced TLE mouse model ( Shibley and Smith , 2002 ) as a positive control for hippocampal epilepsy and a carbamazepine ( CBZ ) -injected 3w-Kif2a-cKO mouse model as a negative control in which CBZ blocks voltage-gated sodium channels and suppresses epileptic seizures during continuous use . We aimed to distinguish the effects of KIF2A deficiency from the effects of epileptic excitation on MFS . In WT mice , Zn-positive axons of DGCs were observed in the hilus and stratum lucidum but not in the stratum oriens ( so ) or molecular layer ( ML ) ( Figure 4C ) . In 3w-Kif2a-cKO mice ( Figure 4D ) , however , the Zn-positive axons were aberrantly elongated in the stratum oriens ( yellow arrowheads in Figure 4F compared with Figure 4E ) and throughout the entire ML ( yellow bar in Figure 4H compared with Figure 4G ) . Importantly , the same phenotypes were observed in the CBZ-injected 3w-Kif2a-cKO mice ( Figure 4—figure supplement 1B and Figure 4—figure supplement 1D , compared with Figure 4—figure supplement 1A and Figure 4—figure supplement 1C , Figure 4—figure supplement 1—source data 1 ) . Furthermore , the Timm grain intensities in the ML of both 3w-Kif2a-cKO and CBZ-injected 3w-Kif2a-cKO mice were >2-fold higher than those in the respective controls ( Figure 4I and Figure 4—figure supplement 1E , Figure 4—source data 1 , Figure 4—figure supplement 1—source data 1 ) . Intriguingly , the signal patterns in the ML of both 3w-Kif2a-cKO and CBZ-injected 3w-Kif2a-cKO mice were different from those in the TLE mouse model in which the signal was restricted to only the IML ( Figure 4J ) . These results suggested that the aberrant DGC axons of 3w-Kif2a-cKO mice extended throughout the ML , regardless of the presence of epileptic seizures . Supporting those results , a different axon marker ( neurofilament M , NFM ) , which specifically detects axons but not dendrites of DGCs in the hippocampus ( Kron et al . , 2010; Parent et al . , 1997 ) , and a DGC-axonal synaptic marker ( synaptoporin ) both exhibited wider distributions in 3w-Kif2a-cKO mice than in WT mice ( Figure 4—figure supplement 2B and E compared with Figure 4—figure supplement 2A and D . See also Figure 4—figure supplement 2C and F . Figure 4—figure supplement 2—source datas 1 and 2 ) . To investigate the identity of the aberrant DGC axons in the entire ML , we attempted to visualize the morphology of a single DGC in the hippocampus . To this end , Kif2a-cKO mice were crossed with thy1-YFP transgenic mice ( M-line ) in which yellow fluorescent protein ( YFP ) is genetically encoded downstream of the Thy1 promoter ( Feng et al . , 2000 ) and selectively expressed in a specific neuronal subset . YFP allowed for full visualization of the hippocampal neurons , including their axons , nerve terminals , dendrites , and dendritic spines . The offspring of the cross , thy1; YFP; Kif2a-cKO mice , were injected with tamoxifen beginning at postnatal week 3 , and their tissues were fixed 3 weeks later . We treated 300-μm-thick sliced sections with ScaleView , an optically transparent reagent ( Hama et al . , 2011 ) , to clarify the structure of granule cells without decreasing their fluorescence signal . We focused on three morphological queries ( Figure 5A ) . The first was whether the origin of aberrant axons in the ML of Kif2a-KO mice originated predominately from MFs in the hilus or directly from the cell bodies . The second was whether there were phenotypic differences between immature and mature DGCs . As shown in Figure 5A , after neurogenesis in the subgranular zone ( SGZ ) , DGCs migrate into the GCL and develop a primary axon and an apical dendritic tree ( Kempermann et al . , 2004 ) . Therefore , at postnatal week 3 , immature DGCs in the inner GCL lost KIF2A in the early developmental stage , but DGCs near the ML in the outer GCL lost KIF2A after maturation . The third query related to the effects of KIF2A loss in the dendrites of mature DGCs , because alteration in MT-dynamics often affect spine morphology and function ( Jaworski et al . , 2009; Hoogenraad and Bradke , 2009; Penazzi et al . , 2016 ) . As shown in Figure 5B , WT DGCs in both the inner and outer GCL ( orange and red asterisks , respectively ) projected a single primary axon into the hilus and extended several apical dendrites into the ML ( white arrows ) . However , 3w-Kif2a-cKO mice ( Figure 5C ) , which were more than 10% of outer mature DGCs , extended an aberrant axon recurrently to the ML ( similarly to the cell with a white asterisk ) , and more than 5% of outer mature DGCs extended aberrant axons directly from the cell body ( similarly to the cell with a red asterisk ) ( Figure 5C and F ) . In contrast , among inner immature DGCs , almost 30% of the cells had some aberrant protrusions on the cell bodies ( orange arrowheads of the cell with orange asterisk in Figure 5C and G ) . Moreover , there were more ectopic DGCs in the 3w-Kif2a-cKO inner ML than in the ML of WT mice ( Figure 5H ) . The source data of those processes were shown in Figure 5—source data 1 . In addition , the spine density of the dendrites of outer mature cells was higher in 3w-Kif2a-cKO mice ( Figure 5E ) than in WT mice ( Figure 5D ) . Morphologically , the number of thin spines , not mushroom or stubby spines , was specifically higher in both the IML and OML of 3w-Kif2a-cKO dendrites than in those of WT dendrites ( Figure 5I , Figure 5—source data 2 ) . The results suggested that the loss of KIF2A results in more unstable or immature spines . To analyze how DGCs develop and differentiate their axons and dendrites , we prepared a primary culture of dissociated DGCs from P3-Kif2a-cKO mice at P5 , and characterized their processes with axonal markers ( Tau1 or NFM ) and dendritic markers ( MAP2 ) at different stages . Before this analysis , we confirmed the DGC characteristics of the cultured cells and the loss of KIF2A from the DGCs by immunostaining the cells with anti-Prox1 ( a DGC marker ) and anti-KIF2A antibodies . More than 80% of cultured cells were Prox1-positive ( Figure 6—figure supplement 1A and B ) , and almost all cells had lost KIF2A expression ( Figure 6—figure supplement 1C ) . At DIV1 , both WT and P3-Kif2a-cKO DGCs generated a short Tau1-dominant axon ( Figure 6B and F ) and a MAP2-dominant dendrite ( Figure 6C and G ) . At DIV3 , however , P3-Kif2a-cKO DGCs gradually generated more aberrant axons than they did dendrites . At that time , in WT DGCs , ankyrin G , the marker of the axon hillock , was detected at the neck of one axonal neurite ( an arrow in Figure 6—figure supplement 1E ) . In the P3-Kif2a-cKO DGCs , however , ankyrin G was detected in more than one neurite ( arrows in Figure 6—figure supplement 1H ) and the population of DGCs with multiple axonal nurites was significantly larger in P3-Kif2-cKO DGCs than in WT DGCS ( Figure 6—figure supplement 1J ) . Eventually , at DIV5 , the WT DGCs developed a single long axon ( a white arrowhead in Figure 6I ) and several dendrites ( a white arrow in Figure 6K ) , whereas P3-Kif2a-cKO DGCs developed long , defasciculated , NFM-positive axons ( white arrowheads in Figure 6J ) and generated some dendrites with multiple additional axons around their cell bodies ( Figure 6L ) . A statistical analysis also demonstrated significant neogenesis of aberrant axons in P3-Kif2a-cKO DGCs ( Figure 6M green bars , Figure 6—source data 1 ) . These phenotypes were rescued by the KIF2A transfection ( Figure 6—figure supplement 2A–G ) . The observations suggested that the neogenesis of aberrant axons in P3-Kif2a-cKO DGCs could be the result of a cell autonomous process rather than an altered response to the external environment , such as alterations in chemo-attraction/repulsion . Finally , we recorded living DGCs for 24 hr at DIV2 . In a WT DGC ( Figure 6—video 1 ) , both an axon with a single branch and a dendrite are shown to gently elongate and contract . In contrast , P3-Kif2a-cKO DGCs did not exhibit length control for axons or dendrites ( Figure 6—video 2 ) . In the video , an axon with a single branch dramatically sprouted and extended multiple branches . Even dendrites actively generated many thin spinous processes . Moreover , some protrusions instantly appeared from the cell body , and then elongated , seemingly contacting one another . After the recording , the DGCs were fixed and stained with anti-Tau1 and anti-MAP2 antibodies ( Figure 6—figure supplement 2H and I ) , revealing that the overextended dendrites had axonal ( Tau1-positive , arrows in Figure 6—figure supplement 2I ) , rather than dendritic characteristics ( MAP2-positive , arrowheads in Figure 6—figure supplement 2I ) . These results suggested that the loss of postnatal KIF2A might disrupt axon/dendrite determination and induce the development of multiple short axons in the hippocampus , thus resulting in complex networks in the dentate gyrus .
KIF2A was predicted to play an important role in postnatal proliferation or migration , due to its effect as an MT destabilizer ( Desai et al . , 1999; Wordeman and Mitchison , 1995; Manning et al . , 2007 ) and its critical role in proliferation ( Ems-McClung and Walczak , 2010; Chen et al . , 2016 ) and neuronal migration in the prenatal hippocampus ( Homma et al . , 2003 ) . In this study , however , neither abnormal neurogenesis nor significant migratory defects were detected in 3w-Kif2a-cKO mice within 3 weeks after tamoxifen injection ( Figure 3 ) . Thus , from our present results , whether KIF2A is crucial for postnatal neurogenesis and migration is difficult to confirm . However , 3 weeks may be too short a period to allow for the detection of migratory defects in postnatal neuronal migration as the variation in the migratory distance was greater than the average DGC migration distance . The use of adult-Kif2a-cKO such as 8w-Kif2a-cKO , which can survive long enough for newborn DGCs to migrate through the entire DGC layer , may reveal the function of KIF2A in neuronal migration in the future . Previously , KIF2A has been shown to be a key axonal collateral suppressor of prenatal hippocampal pyramidal neurons . The loss of KIF2A , a MT destabilizer , resulted in the activation of MT polymerization at the growth cone and the overextension of neuronal processes ( Homma et al . , 2003 ) . In agreement with its prenatal functions , the expression of postnatal KIF2A was strongly distributed in the MF tract of DGCs in the hippocampus ( Figure 4A ) , and the loss of KIF2A induced MFS both in vivo and in vitro ( Figures 2C , 4F , 6J and Figure 6—figure supplement 1B ) , whereas KIF2A transfection rescued the aberrant collaterals ( Figure 6—figure supplement 2D ) . In addition to KIF2A expression in axons , cultured DGCs expressed KIF2A in the cell bodies and dendrites at DIV1 ( Figure 6A ) . The loss of KIF2A from DGCs induced the generation of multiple protrusions ( Figure 5C and Figure 6—video 2 ) and aberrant axons from the cell bodies both in vivo and in vitro ( Figures 4H , 5C and 6L ) , and resulted in a change in the appearance and characteristics of dendrites ( Figures 5E , I and 6L ) . These effects of KIF2A loss are interesting and may indicate a new function of KIF2A in axon/dendrite determination . The activation of MT polymerization by collapsin response mediator protein ( CRMP ) −2 is known to be crucial for axon/dendrite determination and morphogenesis in hippocampal pyramidal neurons ( Bretin et al . , 2005; Quach et al . , 2015; Zhang et al . , 2016 ) . Aberrant MT polymerization due to KIF2A loss might abnormally determine the fate of developing neuronal processes as axons . In other words , KIF2A may suppress the elongation of future dendrites by destabilizing MTs during the early stages of neural development . More intriguingly , similar phenotypes of aberrant neurite growth have been associated with knockout mutations of phosphatase and tensin homologs on chromosome 10 ( PTEN ) in DGCs , which extended multiple long axons both in vitro and in vivo . The loss of PTEN constitutively activated Akt kinase activity ( Gary and Mattson , 2002 ) and the mTORC1 pathway ( Zhou et al . , 2009 ) . In addition , another phenotype of 3w-Kif2a-KO mice resembles the phenotype of Pten-KO mice ( Kwon et al . , 2006 ) . Pten-KO mice showed signs of gliomas , microcephaly , and an epileptic hippocampus ( Luikart et al . , 2011; Althaus and Parent , 2012 ) . Moreover , PTEN expression in neurons starts postnatally , and Pten-KO neurons developed neuronal hypertrophy and a loss of neuronal polarity . Therefore , the association of KIF2A with a PTEN-related cascade should be further elucidated . We schematically present KIF2A function in DGC development in Figure 7A . The function of KIF2A in the development of DGCs might affect postnatal hippocampal wiring because DGCs continue to proliferate in the subgranular layer ( SGL ) , migrate through the GCL , and mature , incorporating their new processes into preliminary existing neuronal networks in the hippocampus throughout their life ( Figure 7B , center panel ) ( Gonçalves et al . , 2016; Kaplan and Hinds , 1977 ) . In Kif2a-cKO mice ( Figure 7B , right panel ) , inner immature DGCs might develop aberrant protrusions , migrating DGCs might develop aberrant axons , and outer maturing DGCs might incorporate their aberrant processes into the dentate ML and hippocampal circuits . In addition , the spine morphology is altered in dendrites lacking KIF2A . These phenotypes might mainly occur because of a lack of control of MT dynamics and might result in aberrant hippocampal wiring and epileptogenesis . In addition , some reports have suggested that the integration of aberrantly migrated hilar ectopic granule cells into the dentate gyrus circuitry is responsible for TLE in the pilocarpine mouse model ( Cameron et al . , 2011 ) and in TLE patients with early-life status epileptics ( Muramatsu et al . , 2008 ) and febrile seizures ( Koyama et al . , 2012 ) . In the IML in 3w-Kif2A-cKO mice , ectopic granule cells into the dentate morphologically integrated into the dentate ML , potentially affecting gyrus circuitry and resulting in epileptogenesis in the 3w-Kif2A-cKO hippocampus . However , the influence of TLE on the Kif2a-cKO hippocampal phenotypes is still arguable . In ordinary TLE ( left panel of Figure 7B ) , epileptic excitation induced MFS , recurrent extension of sprouted MF into the dentate IML , and aberrant excitatory recurrent circuits in the hippocampus ( McNamara , 1999; Gu et al . , 2015; Liu et al . , 2013 ) . Although we succeeded in decreasing the influence of TLE through anticonvulsant CBZ treatment in vivo ( Figure 4—figure supplement 1D ) , and by reproducing the aberrant neurites in dissociated DGC cultured cells ( Figure 6 , Figures 6—figure supplement 1 and 2 ) , this approach must still be carefully evaluated when examining the contribution of the loss of KIF2A to the epileptogenesis and phenotypes of 3w-Kif2a-cKO mice . For example , regarding the expansion of the aberrant axons throughout the ML in 3w-Kif2a-cKO mice , in TLE , many other types of axons in the hippocampus in addition to MFs exhibit sptouting , but those axons are not detected by Timm's stain . In 3w-Kif2a-KO mice , MFS may extend into the OML simply because other types of axons did not sprout in our model , thus leaving more ‘space’ for MF sprouting . However , we believe that the ‘space’ might not be greatly different between 3w-Kif2a-cKO mice ( Figure 4H ) and the 3w-induced TLE animal models ( Figure 4J ) because both mice showed TLE beginning at postnatal week 4 , and the sprouting of many types of axons , including MFs in the hippocampus , might be induced by TLE in the ML at the same level . Regarding the regulation of KIF2A function , the phenotypes of tamoxifen-injected Kif2a-KO DGCs suggested that KIF2A may possibly contribute to MFS in ordinary TLE . After epileptic fits in the hippocampus , hyper-excitation locally up-regulates brain-derived neurotrophic factor ( BDNF ) in the target site of the MF projection , the stratum lucidum ( SL ) . The local activity of BDNF in the hilus initiates MFS , thus eventually resulting in hippocampal hyper-excitability ( Koyama et al . , 2004; Tamura et al . , 2009 ) . As shown in Figure 4A , KIF2A is highly expressed in the SL , and previous research has shown that BDNF-derived kinases p21-activated kinase 1 ( PAK1 ) and cyclin-dependent kinase 5 ( CDK5 ) block the function of KIF2A by phosphorylation in cortical neurons ( Ogawa et al . , 2012 ) ; thus , the hyper-excitation-induced local activation of BDNF might potentially block KIF2A function and therefore induce MFS only in the hilus . Detailed analysis of tamoxifen-injected Kif2a-KO DGCs would demonstrate the contribution of KIF2A to MFS in ordinary TLE . The 3w-Kif2a-cKO mice exhibited weight loss ( Figure 1F ) and hyperactivity ( Figure 1G ) , first becoming hyperactive , then gradually losing interest in food , becoming weak , and eventually dying . The link between these phenotypes and epilepsy remains unclear , but some reports have suggested that these phenotypes are associated with hippocampal dysfunction ( Fallet-Bianco et al . , 2008 ) . Doublecortin ( DCX ) -KO mice , which exhibit both weight loss and hyperactivity , harbor a neuronal lamination defect in the hippocampus ( Håvik et al . , 2012 ) . Tuba1a-KO mice , which are hyperactive , exhibit prominent hippocampal lamination defects ( Fallet-Bianco et al . , 2008 ) . Although 3w-Kif2a-cKO mice did not show an apparent hippocampal laminar defect at P35 ( Figure 2E ) , ectopic cells in the dentate gyrus or pyramidal cells displaced by sprouted MFs would result in weight loss and hyperactivity . Moreover , both phenotypes have also often been reported in MCD patients and animal models ( Stottmann et al . , 2013; Stottmann et al . , 2017 ) . Four recently reported KIF2A-mutated human pediatric patients with MCD display band heterotopia , posterior predominant pachygyria , a thin corpus callosum , severe congenital microcephaly , and neonatal-onset seizures ( Poirier et al . , 2013 ) . Because the patients were young at the time of the report ( 3 and 5 months old ) , further research is necessary to confirm the hippocampal phenotypes of Kif2a-mutated MCD patients . We herein demonstrated that postnatal KIF2A regulates the development of DGCs and the wiring of neuronal circuits in the hippocampus . However , the relevance of the hippocampal phenotypes of Kif2a-KO mice to human patients with mutations in Kif2a is not certain . In the future , the molecular mechanisms of KIF2A regulation in DGC development and hippocampal wiring should be explored in both KO mice and in human patients . The progress of this line of research will allow for analysis of the pathogenesis of Kif2a-related diseases , including schizophrenia ( Li et al . , 2006 ) , juvenile myoclonic epilepsy ( Kapoor et al . , 2007 ) , mental retardation , ocular defects ( Jaillard et al . , 2011 ) , and MCD ( Poirier et al . , 2013 ) . We hope that the collection of data on KIF2A-deficient mice will clarify the pathogenesis of these diseases and lead to a more accurate diagnosis in humans .
A 3loxP-type targeting vector was constructed by using a genomic clone obtained from an EMBL3 genomic library , and genomic fragments were amplified from the 129/Sv-derived ES cell ( ESC ) line CMT1-1 ( Chemicon/Millipore , Billerica , MA ) by using an LA-PCR kit ( Takara , Japan ) . The CMT1-1 ESCs were transfected with the targeting vector and screened for homologous recombinants using PCR . The 3loxP/+ESCs were electroporated using a pCre-Pac plasmid to remove the selection cassette flanked by loxP sequences . The 2loxP/+ESCs were injected into blastocysts , and chimeric male mice were obtained and bred with C57BL/6J female mice . Germline transmission was confirmed by PCR using tail DNA samples . Kif2afl/fl mice were produced by an intercross with Kif2afl/+ mice . To conditionally delete exons flanked by loxP and driven by the chicken beta-actin promoter/enhancer coupled with the cytomegalovirus ( CMV ) enhancer ( CBA ) , tamoxifen-inducible Cre transgenic mice ( CreERt; CAG-Cre/Esr1; Jax #004453 , JAX MICE Laboratories , Bar Harbor , ME ) were used . The CBA-CreERt mice were characterized by using lacZ expression with ROSA26 reporter mice ( R26R ) , which have a loxP-flanked STOP sequence followed by the lacZ gene inserted into the gene trap ROSA26 locus ( By courtesy of Prof . Soriano [Soriano , 1999] ) . Conventional knockout mice ( Homma et al . , 2003 ) were crossed with CBA-CreERt+/− mice to obtain Kif2a+/−;CBA-CreERt+/− mice . Male Kif2a+/−; CBA-CreERt+/− mice were mated with female Kif2afl/fl mice to produce offspring that contained the Kif2afl/+ , Kif2afl/+; CBA-CreERt+/− , Kif2afl/− , and Kif2afl/−; CBA-CreERt+/− alleles . Tamoxifen ( Sigma , St . Louis , MO , 10 mg/ml ) was dissolved in sunflower oil and administered at a dose of 30 mg per kg body weight daily for 7 consecutive days . The Kif2a deletion occurred when the tamoxifen-induced Cre recombinase deleted the floxed DNA domain , which was followed by a frameshift during the Kif2a RNA translation . Deletion was confirmed by a western blot analysis of the crude extracts of whole brain tissues at P21 by using a monoclonal antibody against the N-terminal region of KIF2A ( Noda et al . , 2012 ) . For control mice , we generally used wild-type mice after ensuring that the Kif2afl/−; CBA-CreERt+/− mice and WT mice were not significantly different . The genotypes were determined by PCR of tail DNA or DNA from ES cells with the following primers ( see Figure 1A ) : F1 , 5’-CGCTCATGTGTTTTAAGCTG-3’; R1 , 5’- CACCCCACTATAACCCAGCATTCG-3’; F2 , 5’-GCTGCCAGTGACATAGACTAC-3’ , and the Neo and Cre transgenes . The mice were maintained by repeated backcrossing with C57BL/6J mice ( >12 times ) in a pathogen-free environment . The mice received an intraperitoneal ( i . p . ) injection of scopolamine methyl bromide ( Sigma , St . Louis , MO , 1 mg/kg ) in a sterile saline vehicle ( 0 . 9% NaCl , 0 . 1 ml total volume ) 30 min prior to an injection of pilocarpine to decrease the peripheral cholinergic effects of pilocarpine . The experimental animals were then i . p . injected with a single dose of pilocarpine ( Sigma , 290 mg/kg ) , as previously described ( Shibley and Smith , 2002 ) . The WT mice were age-matched with treated mice and received a comparable volume of vehicle . WT male mice and 3w-Kif2a-cKO ( P25 littermates ) were used in all behavioral tests in a blinded manner . The home cage activity tests were conducted using a MicroMax Monitor ( AccuScan Instruments , Columbus , OH ) and quantified using a computer-operated MicroMax 1 . 3 ( AccuScan Instrument ) . The monitor displayed 16 invisible infrared light beams per axis with synchronous filtering , double modulation and digital hysteresis . These beams provide information that describes the movement of an animal in its home cage , thus allowing an animal’s behavior to be monitored . Mice that were housed singly in their home cages were placed in the beam boxes for 5 min , and their activity was continually recorded . The measurements used to assess home cage activity included active time . The average amount of active time was analyzed using Student’s t-tests . For epilepsy , five mice were isolated in a cage and observed for 30 min . The epileptic mice were genotyped after the observation . WT and 3w-Kif2a-cKO siblings were anesthetized in the postnatal week 4 by using ketamine/xylazine and were surgically implanted with a set of electrodes . Two 0 . 1 mm diameter silver wires were bonded , including a 1 . 2-mm-long reference electrode and a 2 . 0-mm-long working electrode with a hard epoxy resin coat ( except for its 0 . 2-mm-long exposed tip ) , which served to electrically insulate the probe from the reference electrode . Dental cement ( GC Dental , Tokyo , Japan ) was used to fix the electrode set to the skull . The electrode positions in the left hemisphere and the CA1 of the left hippocampus were stereotaxically determined as 1 . 3/1 . 3 mm or 2 . 0/1 . 8 mm anterior to the bregma and 1 . 2/1 . 2 mm or 1 . 5/1 . 5 mm lateral to the midline at a depth of 1 . 3/1 . 2 mm or 1 . 5/1 . 3 mm for the WT or 3w-Kif2a-cKO mice , respectively . These differences were due to the differences in the average brain sizes between the two genotypes . EEG recordings were obtained from mice after complete recovery . The electrodes , measurement system , and software were all purchased from Unique Medical ( Tokyo , Japan ) . EEG recordings were obtained from five mice for each genotype . After EEG recordings , we confirmed the electrode position using a histological examination . The patch-clamp recordings of DGCs were obtained at room temperature using an Axopatch 1D amplifier ( Axon Instruments , Union City , CA ) . Patch pipettes ( 3–5 MΩ ) were filled with 122 . 5 mM Cs gluconate , 17 . 5 mM CsCl , 10 mM HEPES , 0 . 2 mM EGTA , 8 mM NaCl , 2 mM Mg-ATP , and 0 . 3 Na-GTP ( pH 7 . 2 , 290–300 mM mOsm ) . A slice was transferred to a recording chamber and continuously perfused with cold oxygenated ACSF containing 119 mM NaCl , 2 . 6 mM KCl , 1 . 3 mM MgSO4 , 1 mM NaH2PO4 , 26 mM NaHCO3 , 2 . 5 mM CaCl2 , and 11 mM D-glucose . Single pulse stimuli were delivered by bipolar tungsten electrodes , which were positioned on the hilus far from the recorded cells , to avoid antidromic activation . Absence of antidromic activation contamination was concluded if CNQX-AP-5 completely eliminated any responses to the stimulus . The signals were filtered at 2 kHz , digitized at 10 kHz , and analyzed using Clampex 9 . 2 software ( Axon Instruments , Union City , CA ) . The nervous elements were stained using the standard Bodian method ( Bodian , 1936; Bodian , 1937 ) . Briefly , the brains were fixed in FEA ( formalin-ethanol-acetic acid: 90 ml of 80% ethanol with 5 ml of formaldehyde and 5 ml of glacial acetic acid ) , dehydrated with ethanol , and embedded in paraffin . The tissue was sectioned at a thickness of 7 μm . The paraffin sections were hydrated in distilled water ( DW ) , and the slides were then incubated in 2% Protargol solution for 48 hr at 37°C in the dark with 5 g of polished copper shot . The samples were then rinsed three times with DW; reduced in 1% hydroquinone for 10 min; rinsed three times with DW; immersed in 1% aqueous gold chloride for 10 min; rinsed three times with DW; developed in 2% oxalic acid for 20 min; rinsed twice with DW; fixed in 5% sodium thiosulfate for 5 min; rinsed five times with DW; dehydrated; and mounted with cover slips . For brain tissue immunohistochemistry , the mice were perfused with a solution of 4% paraformaldehyde ( PFA ) and 0 . 1% glutaraldehyde ( GA ) in 0 . 1 M sodium phosphate buffer ( PB , pH 7 . 4 ) . Subsequently , 30-μm-thick frozen slices were rinsed in PBS , fixed for 10 min , permeabilized with 0 . 1% Triton X-100 in PBS for 10 min , incubated in blocking solution ( 5% normal goat serum in PBS ) for 30 min , and incubated with primary antibodies at 4°C overnight . After the tissues were washed with PBS , secondary antibodies were applied at 4°C overnight . To stain thick sections , 0 . 1% Triton X-100 was added to the blocking solution . All antibodies were as described in the previous section . To visualize YFP-expressing GCs in mouse brains , the mice were perfused with a fixation solution , and 300-μm-thick sections were immersed in ScaleView ( Olympus , Japan ) , an optically transparent reagent , at 4°C for 24 hr . For all experiments , we used littermates for controls and selected slices at comparable positions determined with a brain atlas . The samples were observed under an LSM710 or LSM 780 confocal microscope ( Zeiss ) . Immunofluorescence detection of two thymidine analogues ( CldU and IdU ) was performed along with Tuttle’s methods ( Tuttle et al . , 2010 ) . Briefly , 5-iodo-2’-deoxyuridine and 5-chloro-2’-deoxyuridene ( IdU and CldU; Sigma , St . Louis , MO ) were dissolved in saline at 10 mg/ml as stock solution . Proliferating cells in the hippocampus were labeled by sequential intraperitoneal injection at 50 mg/kg for 1 week before or after tamoxifen injection . All mice were perfused with 4% PFA/PBS , dehydrated with ethanol , and embedded in paraffin . The 7-μm-thick sections were rehydrated with ethanol , washed for 5 min in PBS , and permeabilized with 0 . 1% Triton-X 100 for 5 min; then , antigen retrieval was performed in boiling 0 . 01 M pH 6 . 0 sodium citrate buffer for 20 min by using a microwave oven . Slides were immersed in 1 . 5 N HCL for 40 min at RT , washed twice in PBS for 5 min , and immersed in blocking solution ( 5% goat serum in PBS ) . Then an anti-IdU antibody diluted in blocking solution was applied and incubated overnight at 4°C . After being agitated in PBS for 20 min in a shaking jar at 37°C , the slices were washed four times in PBS , and anti-CldU antibodies diluted in blocking solution were applied and incubated overnight at 4°C . The slides were washed twice for 5 min per wash in PBS , and then , the appropriate secondary antibody solution ( 1:300 ) was applied for 2 hr at RT . The slices were washed 5 times for 5 min per wash in PBS , and a cover glass was applied with PBS . The distance between the bottom edge of the GCL and the dentate granule cells was measured using IMARIS software ( Zeiss ) . To visualize zinc and other metals in the hippocampus , 30-μm-thick frozen brain sections were stained using the neo-Timm method with some modifications ( Danscher and Zimmer , 1978 ) . The pixel intensities were measured as previously described ( Koyama et al . , 2004 ) . Briefly , in an image acquired using a 20 × objective , at least five 20 μm × 20 μm cursor points at 20 μm intervals were positioned in each hilus , granular and IML , OML , and subicular area located just outside the hippocampal sulcus . The mean signal intensity ( I ) within these cursor points was measured at an 8-bit resolution using ImageJ software ( NIH , USA ) . The Timm grain intensity was determined by dividing the values of these subregions by the value of the subiculum ( background ) . The same method was used to measure the NFM intensity . As an internal control , we used the intensity of NFM staining in the hilus , because the intensity in the subicular or other areas in the dentate gyrus was not stable . P3-Kif2a-cKO and WT mice were euthanized at P5 , and their hippocampal dentate gyri were dissected ( Hagihara et al . , 2009 ) . Each dentate gyrus was trypsinized and gently triturated to isolate cells ( 3 . 5 × 104 cells/cm2 ) , which were placed in a four-well glass chamber ( Nunc , 155411 ) . Chambers were coated with poly-L-lysine overnight at room temperature ( Sigma , St . Louis , MO ) , washed with DW for 2 hr twice , and then coated with laminin overnight at 4°C ( inquiry ) to clearly visualize any morphological differences in the rapid growth of neurites . Dispersed cells were cultured in MEM ( Gibco Thermo Fisher , MA ) /Neuro Brew-21 ( MACS , Bergisch Gladbach , Germany ) at 37°C in a humidified atmosphere containing 5% CO2 . To confirm the characteristics of the cultured neurons , the cells were stained with anti-NFM , anti-MAP2 , anti-Prox1 , and anti-AnkyrinG antibodies . To confirm excitatory recurrent circuits in the tamoxifen-injected Kif2a-cKO hippocampus , organotypic hippocampal slice cultures were prepared as previously described ( Koyama et al . , 2004 ) . Briefly , P4 mice were deeply anesthetized , and their brains were removed and cut into 300-μm-thick transverse slices using a VR-1200S ( Leica Biosystems , Wetzlar , Germany ) in a cold oxygenated Gey’s balanced salts solution supplemented with 25 mM glucose . Entorhino-hippocampi were dissected and cultured using a membrane interface technique ( Stoppini et al . , 1991 ) . Briefly , the slices were placed on sterile 30-mm-diameter membranes ( Millicell-CM; Millipore , Bedford , MA ) and transferred to six-well tissue culture trays . The cultures were fed with 1 ml of 50% minimal essential medium ( Invitrogen , Gaithersburg , MD ) , 25% horse serum ( Cell Culture Lab , Cleveland , OH ) , and 25% HBSS and the cells were maintained in a humidified incubator at 37°C containing 5% CO2 . The medium was changed every 3 . 5 days . | The brain contains billions of neurons that connect together to form ‘circuits’ that control behavior and process information . By the time we are born , most of the neurons in our brain have already formed and connected into these circuits . But there are some brain areas that continue to make new neurons throughout our lives . One such area is the hippocampus , a region of the brain involved in learning and memory . There , neurons called dentate granule cells keep their ability to divide , migrate to new locations , and develop new connections . Like most neurons , at the heart of each dentate granule cell is a cell body that contains the cell's nucleus and protein-making machinery . Attached to this are a set of small branch-like structures called dendrites that receive signals from surrounding neurons . Extending away from the cell body is another , longer branch called an axon , which transmits signals to other neurons . A protein called KIF2A plays several roles in the developing brain of mammals , including helping neurons to migrate to the right place and controlling how their axons form . Before birth , neurons across the brain make KIF2A . After birth this gradually changes until only the dentate granule cells in the hippocampus produce KIF2A . In humans , mutations that prevent KIF2A from working are thought to cause brain malformations . They may also lead to disorders such as schizophrenia , epilepsy and eye defects . To investigate the role of KIF2A in more detail , Homma et al . genetically engineered mice so that giving them a drug called tamoxifen would inactivate the gene that produces KIF2A . Mice that had this gene switched off three weeks after birth – when KIF2A levels in the hippocampus are normally at their highest – lost weight and became hyperactive . They also developed severe temporal lobe epilepsy . To find out why these problems ocurred , Homma et al . used a microscope to study sections of the brains of the mice . The neurons had divided and migrated to the correct location of the brain with no significant problems . However , dentate granule cells that lacked KIF2A looked unusual . They had too many dendrites , the dendrites were longer than they should be and they showed markers usually only found on axons . This suggests that KIF2A helps to control the length of axons and dendrites and the wiring of the hippocampus . At the moment , it's not known whether the same defects also occur in humans . If the results are reproducible in people , future work could help to diagnose and understand conditions linked to KIF2A , like schizophrenia and epilepsy . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"cell",
"biology",
"neuroscience"
] | 2018 | KIF2A regulates the development of dentate granule cells and postnatal hippocampal wiring |
Three-dimensional ( 3D ) culture systems have fueled hopes to bring about the next generation of more physiologically relevant high-throughput screens ( HTS ) . However , current protocols yield either complex but highly heterogeneous aggregates ( ‘organoids’ ) or 3D structures with less physiological relevance ( ‘spheroids’ ) . Here , we present a scalable , HTS-compatible workflow for the automated generation , maintenance , and optical analysis of human midbrain organoids in standard 96-well-plates . The resulting organoids possess a highly homogeneous morphology , size , global gene expression , cellular composition , and structure . They present significant features of the human midbrain and display spontaneous aggregate-wide synchronized neural activity . By automating the entire workflow from generation to analysis , we enhance the intra- and inter-batch reproducibility as demonstrated via RNA sequencing and quantitative whole mount high-content imaging . This allows assessing drug effects at the single-cell level within a complex 3D cell environment in a fully automated HTS workflow .
A number of uniquely human diseases , including Parkinson’s disease , would greatly benefit from a comprehensive human cellular in vitro model that recapitulates key characteristics of midbrain tissues in a high-throughput-compatible format . Three-dimensional ( 3D ) cell culture in general and the ability to generate organ-like aggregates ( ‘organoids’ ) in particular have found a rapid following over the past few years ( Sato et al . , 2009; Eiraku et al . , 2011; Nakano et al . , 2012; Lancaster et al . , 2013; Quadrato et al . , 2017; Paşca et al . , 2015; Iefremova et al . , 2017; Takasato et al . , 2015; Dye et al . , 2015; Takebe et al . , 2013; McCracken et al . , 2014; Pașca , 2018 ) due to their potential to mimic cellular niches more closely than 2D cell cultures . These approaches promise to develop next-generation high-throughput screens ( HTS ) that can provide more relevant predictions of drug efficacy and toxicity ( Fatehullah et al . , 2016; Ranga et al . , 2014; Fang and Eglen , 2017; Dutta et al . , 2017; Ho et al . , 2018; Chen et al . , 2018; Friese et al . , 2019 ) as they may allow better modeling of pathologies with complex interactions of several cell types in specific cellular niches ( Qian et al . , 2016; Mariani et al . , 2015; Ogawa et al . , 2015; Verissimo et al . , 2016; Vlachogiannis et al . , 2018; Czerniecki et al . , 2018 ) . 3D culture in the form of spheroids has long been established , especially in the field of cancer biology ( Sutherland et al . , 1971 ) , and used in various HTS applications ( Kelm et al . , 2003; Senkowski et al . , 2015; Wenzel et al . , 2014; Kenny et al . , 2015 ) . While these model systems are already more complex and potentially more physiologically relevant than 2D culture ( Pickl and Ries , 2009 ) , they display a much simpler , less organ-like 3D cell organization ( Fang and Eglen , 2017 ) and do not mimic functional features of the organ as broadly and as closely as organoid tissues . Moreover , many of the 3D-based screens performed so far depend on whole-aggregate-based readouts such as size , morphology , and cell viability ( Vlachogiannis et al . , 2018; Ivanov et al . , 2014; Friedrich et al . , 2009; Ivanov et al . , 2015; Hou et al . , 2018; Kang et al . , 2015 ) , which make it challenging to gain mechanistic insights into cells or sub-populations of cells in the context of their niches . In contrast , complex organoids have emerged as a promising research tool due to their unique resemblance to human tissues , defined by an organ-like architecture composed of different tissue-specific cell types and the capability to mimic organ functions ( Lancaster and Knoblich , 2014 ) . One structure of particular interest in the context of disease modeling and drug development is the midbrain due to its role in the highly prevalent Parkinson’s disease ( affecting 2–3% of the population aged 65 and above Poewe et al . , 2017 ) and many other developmental disorders ( Barkovich , 2012; Doherty et al . , 2013 ) . While the current state of the art allows the generation of different types of midbrain organoids ( Qian et al . , 2016; Jo et al . , 2016; Monzel et al . , 2017 ) , the rigorous standardized organoid production and quantification methods needed for high-throughput screening have been elusive . The established protocols tend not to focus on scalable , homogeneous organoids with quantitatively predictable morphology , cellular composition , and local cell organization . Obligatory extensive manual handling often including cumbersome matrix embedding steps render them challenging for scale-up ( Tong et al . , 2018 ) . Furthermore , common analysis methods ( e . g . sectioning and immunostaining , RNA sequencing ) do not scale well for HTS applications . Here , we present a fully automated workflow in a standard 96-well format that combines generation , maintenance , whole mount immunostaining , tissue clearing , and high-content imaging of automated midbrain organoids ( AMOs ) ( see Figure 1a ) . The resulting AMOs are similar to published midbrain organoids with regard to their expression of midbrain-specific markers and cell populations , yet maintain a reproducible homogeneous phenotype . They mimic relevant organ function in the form of spontaneous , highly synchronized neural activity indicating functional cellular coupling across the entire AMO . Their high homogeneity , reproducibility , culture format , and fast development of approximately one month render them ideal for high-throughput screening applications . Moreover , our combined whole mount immunostaining and clearing workflow abolishes the need for labor-intensive tissue sections and allows for quantitative whole mount high-content analysis of entire organoids with single-cell resolution . Our automation of the entire workflow from seeding to analysis in standard plates allows for easy scale-up and implementation into existing screening facilities .
Screening applications require biological systems that operate within predictable physiological parameters . In order to limit cellular heterogeneity during differentiation , we produced human AMOs starting from small molecule neural precursor cells ( smNPCs ) ( Reinhardt et al . , 2013a ) , which in turn originate from pluripotent stem cells ( PSCs ) . The neural-restricted developmental potential of smNPCs still allows the self-organization required for the formation of a 3D architecture ( Monzel et al . , 2017; Di Lullo and Kriegstein , 2017 ) during differentiation toward a midbrain fate . To further reduce batch-to-batch variability , we also omitted matrigel embedding and standardized mechanical stresses by using an automated liquid handling system ( ALHS ) . Starting from seeding the organoids , all following steps including maintenance , fixation , whole mount staining , and clearing are performed in a fully scalable automated fashion using a 96-channel-pipetting head in a robotic ALHS ( see Figure 1a ) . The resulting AMOs show little intra- and inter-batch variability in size distribution ( see Figure 1b , average coefficient of variation ( CV ) within one batch 3 . 56%; min 2 . 2% , max 5 . 6% ) , morphology ( see Figure 1c ) , and cellular composition and organization ( see Figure 2 ) , making them ideal for HTS-approaches . Furthermore , our workflow generates one aggregate per well , maintained independently from others , thus minimizing batch effects due to paracrine signaling observed in bioreactor-based strategies ( Quadrato et al . , 2017 ) . If paracrine signaling is desired , our workflow can easily accommodate several aggregates per well . The fully automated workflow operates with very high efficiency , retaining 99 . 7% ( standard deviation 0 . 7% ) of samples for automated seeding , aggregation , and maturation steps over 30 days and 96 . 5% ( standard deviation 3 . 1% ) of samples for fixation , whole mount staining , clearing , and transfer to flat bottom imaging plates over 12 days . Lastly , 6 . 1% ( standard deviation 1 . 3% ) of these samples are rejected during high-content imaging for presence of dust , damage , or fibers ( see Figure 1d and Supplementary file 1 for source data ) . In order to characterize protein localization in our AMOs ( >600 μm diameter ) and assess the efficiency of their neural/midbrain differentiation at a cellular resolution and in a HTS-compatible manner , we adapted an extended 3D-staining protocol ( Lee et al . , 2016 ) and combined it with benzyl alcohol and benzyl benzoate ( BABB ) -based tissue clearing ( Dent et al . , 1989 ) . BABB-based clearing proved to be both the fastest and most efficient method in a comparison of different clearing protocols ( see Figure 1—figure supplement 1 ) . The combination of whole mount staining and clearing allows the 3D reconstruction of entire AMOs via confocal imaging and enables further detailed 3D quantification and analysis , for example tracing of neurites throughout the whole aggregate , which cannot be performed using typical tissue sectioning procedures ( see Video 1 ) . The immunostaining results are depicted as either single confocal optical slices ( see Figure 2a–d , f/g ) or maximum intensity projections ( MIP , see Figure 2e ) . Already at day 25 , the AMOs contained large numbers of neurons as indicated by the expression of Map2 ( Shafit-Zagardo and Kalcheva , 1998 ) ( also see Video 1 ) , β-tubulin III ( TUBB3 ) ( Leandro-García et al . , 2010 ) ( see Figure 2e ) , and doublecortin ( Gleeson et al . , 1999 ) ( DCX , see Figure 2c/d ) . Presence of high levels of tyrosine hydroxylase ( TH , Figure 2a/b , also see Figure 2—figure supplements 1 and 2 ) , the rate-limiting enzyme in dopamine synthesis ( Nagatsu , 1995 ) , as well as the expression of the transcription factors Foxa2 , Lmx1a , Nurr1 , and Pitx3 ( Hegarty et al . , 2013 ) ( see Figure 2—figure supplements 1 and 2 ) , are consistent with differentiation toward a dopaminergic midbrain fate . While other neuronal subtypes , specifically GABAergic ( vGAT ) and glutamatergic ( vGLUT1 ) neurons , are present in AMOs ( see Figure 2—figure supplement 1 ) , their abundance is low compared to dopaminergic neurons . As commonly seen in all 3D neural cultures , AMOs retain a population of neural precursors identified by the expression of Sox2 ( Ellis et al . , 2004 ) ( see Video 1 , Figure 2a/b , and Figure 2—figure supplement 2 ) , Brn2 ( Dominguez et al . , 2013; Figure 2c/d and Figure 2—figure supplement 2 ) , and the more general neural marker nestin ( Hendrickson et al . , 2011 ) ( see Figure 2a/b and Figure 2—figure supplement 2 ) . Over time , AMOs matured further . Expression of synapsin ( Thiel , 1993 ) ( see Figure 2—figure supplements 1 and 2 ) as well as the presynaptic marker synaptophysin and postsynaptic marker homer ( Tadokoro et al . , 1999 ) frequently colocalized with each other on Map2-positive neurites ( see Figure 2f ) and indicated the presence of synapses . Since gliogenesis follows neurogenesis in vivo ( Miller and Gauthier , 2007 ) , we expected the emergence of astrocytes after the initial formation of neurons . Consistently , AMOs contained GFAP and S100b double-positive astrocytes ( Götz et al . , 2015 ) at later stages ( see Figure 2g ) . In the cortex , neurons form cortical layers with distinct markers . Cortical 3D models recapitulate this layer organization to a degree ( Lancaster et al . , 2013; Paşca et al . , 2015; Qian et al . , 2016; Mariani et al . , 2015; Bhaduri et al . , 2020 ) . In contrast , the midbrain does not possess the typical layer organization of the cortex , hence published midbrain organoids are devoid of cortex-like layers ( Qian et al . , 2016; Monzel et al . , 2017 ) . Due to their self-organizing nature , typical published midbrain organoids initially form random local subdomains of organized tissue within the bulk of the organoid , often in the form of rosettes ( Jo et al . , 2016 ) , a hallmark of the very early stages of neural development ( Perrier et al . , 2004; Elkabetz et al . , 2008 ) . This makes their morphology harder to predict within batches and leads to heterogeneity rendering screening strategies more challenging . We have optimized our AMOs to not form distinct random local subdomains; rather , the different cell types within the AMOs ( i . e . neurons , astrocytes , and neural progenitors ) self-organize into different concentric zones with distinct cellular orientations spanning the entire organoid ( see Figure 2a–c ) . The outermost layer of the AMOs contains few nuclei with a dense , circumferentially oriented layer of TH+/nestin+/DCX+ cell processes . Cellular orientation changes in the underlying zone closer to the core , with TH+ dopaminergic and DCX+ neurons showing a clear radial alignment ( see Figure 2b/c ) . The next zone , separating this region of radially organized neurons and the core , contains circumferentially oriented DCX+ neurons and few Brn2+ neural precursors ( see Figure 2b/c ) . The core itself includes mostly neural precursors and few neurons . Within each zone , the different cell types are homogeneously distributed . In the context of HTS-compatibility , this homogeneity is advantageous , as this trait allows us to extrapolate data even from single medial confocal slices , drastically shortening acquisition times ( see Figure 8—figure supplement 1 ) . This inter-sample similarity can provide a uniform baseline for compound testing and render self-organized 3D human neural cell niches amenable to HTS strategies . Ultrastructural analysis of AMOs ( see Figure 2—figure supplement 3 ) supported the immunofluorescence data , revealing a dense 3D cell architecture consistent with neuronal cell bodies surrounded by nerve fibers . Analyzing the nerve fibers at a higher magnification revealed regular-spaced neurofilaments and microtubules . Moreover , vesicles with the characteristic size and localization of synaptic vesicles are frequently found within these nerve fibers . Further quantitative real-time PCR ( qPCR ) analysis demonstrated increasing expression levels of various neural ( DCX , Map2 , NEFL , NeuN , TBR2 , TUBB3 , Syt1 ) , midbrain ( TH , NURR1 , NKX6-1 , EN1 , GIRK2 , AADC ) , and glia-specific ( GLAST , MBP , S100b ) markers at different developmental stages with concomitant decreases in neural precursor markers ( Brn2 , nestin , Pax6 , Sox1 , Sox2 ) , confirming neural maturation toward a midbrain identity over time ( Figure 3 ) . Finally , we replicated these immunostaining and gene expression results with a second independent cell line ( ‘AMO line 2’ , see Figure 2—figure supplement 2 and Figure 3—figure supplement 1 ) , demonstrating the applicability of our workflow to cells with different genetic backgrounds and origin . To assess functional coupling of individual cells within the AMOs we first performed Fluo-4 acetoxymethyl ester ( AM ) -based calcium imaging , which can be used as a readout for spiking activity of neurons ( Grienberger and Konnerth , 2012 ) . In addition to frequent spontaneous activity of individual cells , we observed aggregate-wide synchronous and periodic calcium spikes ( see Video 2 ) in all analyzed AMOs . To characterize this behavior further , we defined different regions of interest ( ROIs ) and assessed the change in fluorescence intensity over time in each region ( see Figure 4 ) . Measuring the entire AMO reveals two consecutive spikes in Fluo-4 brightness , with a period of approximately 30 s ( see Figure 4a ) . When we subdivided the measured area into four quadrants , we observed synchronized spiking activity in all four resulting ROIs ( see Figure 4b ) . This parallel activity pattern could be found at many structural levels of the AMO , even for single cells ( see Figure 4c/e ) . Changing the time scale revealed additional levels of synchronicity between selected single cells , in addition to aggregate-wide spikes ( see Figure 4d/e ) . Considering the calcium-imaging analyses along with the existence of synaptic vesicles on the ultrastructural level ( see Figure 2—figure supplement 3 ) , the verification of synapses via immunostaining ( see Figure 2f and Figure 2—figure supplements 1 and 2 ) as well as synaptotagmin 1 ( Syt1 ) via qPCR ( see Figure 3 and Figure 3—figure supplement 1 ) , our results support the presence of functionally coupled and spontaneously active neurons within the AMOs . The synchronous spiking patterns suggest that not only a small number of neurons but , in fact , the entire aggregate may be functionally connected . Large-scale synchronous bursting behavior can also be observed in several developing brain regions in vivo ( Ben-Ari , 2001 ) and in brain slices in vitro ( Silva et al . , 1991 ) . Given the high reproducibility of synchronized calcium activity across all tested samples , we decided to evaluate the feasibility of using it as a functional readout in screening settings . Thus , we performed Fluo-4 AM-based calcium imaging on younger ( day 35 ) AMOs and measured the resulting fluorescence signals on a standard plate reader . While they showed a shorter periodicity than the older samples , all tested AMOs displayed distinct peaks in fluorescence intensity resembling the synchronous activity patterns seen during spinning disk microscopy analysis ( see Figure 4f ) , also in a second independent AMO line ( see Figure 2—figure supplement 2 ) . Treatment with the known calcium channel blocker cobalt ( II ) chloride completely abolished these peaks ( see Figure 4g ) . Since the synchronous calcium activity of the AMOs and its modulation by inhibitors can be measured easily via HTS-friendly standard plate readers or specialized FLIPR Ca imagers ( Sirenko et al . , 2019 ) , AMOs may be a promising 3D model of human neural activity that allows directly assessing midbrain related organ functions in HTS . Multielectrode array ( MEA ) measurements revealed spontaneous electrical activity in 35-day-old AMOs ( see Figure 4h/i ) . The field potentials of several electrodes in proximity to a 35-day-old AMO ( see Figure 4i ) oscillated in synchrony over time with two concurrent main frequencies at 1 Hz and 14 Hz ( see Figure 4j , upper and lower panels ) . Figure 4k shows a Fast Fourier Transformation ( FFT ) based on the data shown in Figure 4j . Uncoordinated single-cell activity can hardly account for such robust and spatially long-ranging electric field oscillations . Rather , the concurrent and covariant signals at disparate electrodes support a widespread , synchronized electrical activity encompassing the entire AMO . Taken together , this data further supports the functional coupling of entire AMOs indicated by the calcium imaging experiments . Finally , we characterized the electrophysiological properties of single cells from the AMOs using voltage patch-clamping . A stepwise increase of the membrane holding potential from −70 to +60 mV with 10 mV increments elicited transmembrane currents that consisted of a fast-activating , fast-inactivating inward current followed by a slower activating , slowly deactivating outward current ranging from a few hundred pA to several nA ( see Figure 4l ) . The I–V curves of both currents are typical for sodium inward and potassium outward currents through voltage-gated channels ( see Figure 4m; Reinhardt et al . , 2013a; Simard et al . , 1993; Cummins et al . , 1994; Reinhardt et al . , 2013b ) . Furthermore , the current-clamp recordings demonstrated that these cells generated action potentials ( APs ) in response to current injections ( see Figure 4n ) . The average membrane potential of the recorded cells was −41 . 9 ± 15 . 2 mV ( n = 29 ) . These typical excitable , neuron-like electrophysiological properties could be detected as early as day 25 ( see Figure 4—figure supplement 1 ) and in 29 of the 62 recorded cells . The rest of the cells possessed only outward currents of a few hundred pA by stepping to +60 mV ( see Figure 4—figure supplement 1 ) and were unable to generate APs in response to current injections . These may represent other cell types present in AMOs like astrocytes and neural precursors . To characterize AMOs on the level of global gene expression , we performed RNA sequencing of single organoids from three independent batches of AMOs ( i . e . cells were separately thawed , seeded , and cultured ) and compared the results with published RNA sequencing data sets of primary human tissues ( Roost et al . , 2015 ) and established midbrain organoids ( Jo et al . , 2016 ) . Consistent with successful neural differentiation , AMOs were most similar to the brain and spinal cord in a panel with data from 21 human fetal tissues ( see Figure 5a ) . Moreover , AMOs also correlated well with published data sets from different midbrain ( -like ) samples including primary human tissue ( see Figure 5b ) . On a global gene expression level , AMOs more closely resembled the primary human midbrain samples than published midbrain organoids ( AMOs = correlation 0 . 78 , published midbrain organoids = 0 . 72 , see Figure 5b ) . We also included publicly available RNAseq data from three prenatal human cortex samples ( Jaffe et al . , 2015 ) in the comparison as non-midbrain controls . As expected , the cortical samples showed high correlation with each other but less with the midbrain samples ( see Figure 5b; for further comparisons between AMOs and cerebral organoids via the protocol by Lancaster et al . , 2013 . see Figure 5—figure supplement 1 ) . Taken together , AMOs resemble published midbrain organoids as well as primary human midbrain tissue at the level of global gene expression . Since homogeneity and reproducibility are crucial for screening applications , we next examined the variance of AMOs on the gene expression level and compared it to that of published midbrain organoids ( Jo et al . , 2016 ) . This revealed that AMOs were consistently more reproducible within and between different batches than current midbrain protocols , as illustrated by the principal component analysis ( PCA ) plotted in Figure 5c . The AMOs from three independently thawed and cultured batches ( n = 64 separately processed single organoids in total ) clustered much more closely together than the published midbrain organoids ( n = 6 ) . This further underlines the utility of AMOs as a 3D cellular platform for HTS strategies . In screening settings , the wells at the edges of plates often display different readouts than those located toward the center of the plate ( ‘edge-effects’ ) ( Malo et al . , 2006 ) . Therefore , we sequenced half of a 96-well plate for one AMO batch and tested for differences resulting from well location within the plate ( group ‘one inside’ = center of the plate vs . ‘one outside’ = edge in Figure 5a–c ) . Importantly , in the PCA plot the AMOs clustered independently of their position on the plate ( Figure 5c ) and the groups also showed no apparent differences in any of the other analyses ( Figure 5a/b ) , indicating that AMOs exhibit no measurable edge effects at the global gene expression level and further substantiating the high reproducibility of our protocol . To further investigate the differences between AMOs and established midbrain organoid protocols , we performed gene ontology ( GO ) ( Ashburner et al . , 2000; Supek et al . , 2011 ) analysis of the genes significantly upregulated ( padj . < 0 . 05 ) in AMOs compared to the previously used published midbrain organoids ( Jo et al . , 2016 ) . This analysis yielded almost exclusively GO terms connected to neuronal and synaptic activity ( Figure 5d; for a complete list of GO terms see Supplementary file 2 ) . Consistent with the previously described synchronous activity patterns ( see Figure 4 ) , this further illustrates the physiological relevance and efficient neural differentiation of AMOs . While immunofluorescence-based screening-compatible techniques of whole 3D aggregates have been reported , they can only detect cells in the outer layers of large organoids ( Vergara et al . , 2017 ) , or they use small aggregates of approximately 100 μm diameter ( Verissimo et al . , 2016 ) or cystic organoids ( Czerniecki et al . , 2018 ) , both of which can be penetrated by antibodies and fluorescence illumination more easily . In contrast , our workflow is custom-tailored for automation and allows the quantification of entire dense , large-scale aggregates ( >800 μm diameter ) with single-cell resolution and high sensitivity , as highlighted by a dose-response assay for 3D cellular detection ( see Figure 6a ) . We mixed cells labeled with CellTracker deep red dye with unlabeled cells at known proportions , aggregated them to sizes similar or exceeding that of AMOs ( 750 μm and 950 μm , see Figure 6a ) , cleared them , and then analyzed them on a confocal high-content imaging system . The resulting relationship between the amount of tracked cells and measured brightness was highly linear ( R2 >0 . 99 ) , illustrating the quantitative nature of our optical HTS 3D whole mount analysis workflow . Next , we demonstrated the homogeneity of AMOs at the protein level . A fully automated 96-well based whole mount optical analysis ( see Figure 6b left ) illustrated the ability to detect both abundant filamentous structures ( neural marker Map2 ) and nuclear markers ( Sox2 ) in a HTS-compatible manner ( see Figure 6b right , single slice from one aggregate ) . Using nuclear markers like Sox2 , our technique allowed quantification at single-cell resolution by identifying , counting , and summing the brightness of Sox2+ nuclei for each imaged confocal plane ( see Figure 6c/d/f/h ) . Filamentous , abundant signals like Map2 could be quantified throughout 3D aggregates by summing the overall mean brightness for each confocal plane ( see Figure 6e/g ) . The comparison of three 96-well plates from independent batches revealed the uniform cellular composition of AMOs within and between independently thawed and cultured batches ( see Figure 6d–g ) ( Average CVSox2 = 5% , CVMap2 = 9% ) . Positional analysis detected effects of plate position ( edge effects ) for Map2 levels but not Sox2 levels with about 10% reduced Map2 brightness of samples in the center of the plate ( Figure 6—figure supplement 1 ) compared with the wells at the edge . Considered together with the absence of edge effects in the RNA sequencing results , this may indicate that only a specific subset of proteins is altered by edge conditions , while the vast majority of cellular processes is uniform throughout the plate . Since differentiation outcomes and kinetics are known to vary considerably between cell lines , we decided to benchmark our AMOs against a protocol that can be implemented using the same starting cell line as our AMOs and that is adaptable to the same automation and analysis workflow that we established for our midbrain model . In this comparison , smNPC-derived AMOs ( line 2 ) and the hiPSC organoids share the same cell line of origin . We compared our smNPC-derived cultures to hiPSC-derived 3D neural organoids based on a core protocol by Paşca et al . , 2015 ( also described in more detail by Sloan et al . , 2018 ) with modifications , as they share a number of key traits with our AMOs ( for an overview of our automated protocol and the differences to the published original , see Figure 7—figure supplement 1 ) . They are self-aggregated and self-organized , and they do not require the addition of an external matrix for proper development . Furthermore , to eliminate any potential bias due to manual handling , we adapted the cortical protocol to our automation pipeline . As a result , any remaining variability did not originate from handling but from stochastic biological processes . For detailed characterization , hiPSC cortical organoids underwent our established automated whole mount staining and clearing procedure ( see Figure 7—figure supplement 2 ) as well as qPCR ( see Figure 3—figure supplement 1 ) , confirming differentiation toward their correct cortical fate in our workflow . Out of twelve full 96-well plates we were , with our workflow , able to generate and maintain cortical organoids in all but one well , where the organoid got lost during the 30 days of automated culture . This further underlines the adaptability and efficiency of our midbrain protocol for other organoid types . Compared to smNPC-derived AMOs , morphology and zonal arrangement of neural subpopulations in hiPSC organoids varied to a larger degree ( see Figure 7a–d ) . Three independently cultured batches of automated hiPSC organoids showed up to a 5-fold higher coefficient of variation in cellular viability and up to a 10-fold higher coefficient of variation for organoid size than AMOs ( see Figure 7g/h , for individual organoid size and viability data see Figure 7e/f ) . We performed high-content analyses at the protein level analogous to the data in Figure 6 and found that the variation of Sox2 and Map2 content of the automated hiPSC-organoids was larger than for AMOs , even when we normalized for strongly variable sizes ( see Figure 7i/j ) . The acquisition of high-content data at the same hardware settings for both types of organoids ( from the same cell line of origin ) also allowed for a direct comparison of cell-type-specific signals . hiPSC-organoids contained a distinctly lower amount of Map2 per area , indicating less efficient/delayed neuronal maturation . In conclusion , AMOs are more homogeneous with regard to their morphology , cellular structure , size , viability , and protein expression than the automated hiPSC organoids generated from the same cell line and under strictly standardized conditions . A clinically relevant midbrain model requires dopaminergic activity . To further confirm midbrain-specific function of AMOs , we generated electrophysiological data from multielectrode arrays together with specific molecular agonists and antagonists for dopaminergic , GABAergic , and glutamatergic pathways . Functional responses were consistent with a midbrain identity of AMOs ( see Figure 8e/f/g/k ) . Electric field oscillations of AMOs responded strongly to the addition of modulators of the dopamine pathway , but yielded more limited responses when targeting the GABA pathway ( see Figure 8e/f/k ) . This correlates well with the lower abundance of GABAergic neurons in AMOs as detected by immunostaining ( see Figure 2—figure supplement 1 for a vGAT staining ) . The electrical activity increases following the stimulation with glutamate agonists and subsequently decreases after addition of the antagonist ( Figure 8g/k ) , despite the low abundance of glutamatergic cells ( see vGLUT1 staining in Figure 2—figure supplement 1 ) . However , this is consistent with previous studies reporting an increase in firing rate and burst-firing for dopaminergic neurons treated with glutamate agonists which can be counteracted by NMDA receptor antagonists ( Meltzer et al . , 1997; Wang and French , 1993 ) . Automated hiPSC-derived cortical organoids served as negative controls to validate midbrain-specific MEA data . As expected following cortical differentiation , control hiPSC organoids had a weak response to dopaminergic modulation and reacted most strongly to glutamatergic modulation ( see Figure 8h–j , k ) . The overall higher electric field amplitude in AMOs compared to automated hiPSC-organoids further illustrated the accelerated differentiation and faster maturation achieved by our workflow over hiPSC-based organoid protocols of the same age . To further evaluate cellular maturity as one key factor in physiologically relevant screens , we determined the percentage of Sox2-negative mature cell types within a large number of AMOs via high-content imaging . Testing AMOs from two separate lines with three independently cultured batches each , we detected between 70% and 80% mature cell types after only 30 days of differentiation with very little variation within and between batches and cell lines ( see Figure 8d ) . The maturity and dopaminergic midbrain identity of AMOs was further confirmed by the spontaneous and unstimulated secretion of dopamine into the culture medium at a similar level as found in adult human cerebrospinal fluid ( CSF , see Figure 8c; Goldstein et al . , 2012 ) . Sequestered dopamine may constitute an additional readout that is functionally relevant in models of the human midbrain . To further assess the physiological relevance of our workflow as well as its ability to quantify drug effects in 3D cell cultures , we treated AMOs from two different cell lines with increasing concentrations of toxins specific for dopaminergic neurons , namely 6-Hydroxy Dopamine ( 6OHD ) and 1-Methyl-4-phenylpyridinium ( MPP+ ) ( Reinhardt et al . , 2013b; Blum et al . , 2001; Simola et al . , 2007; Meredith and Rademacher , 2011 ) . High-content imaging allowed us to discern distinct effects of these nigral toxins on TH-positive neurons compared to general MAP2-positive neurons in AMOs . While both compounds reduced the number of all neurons in a dose-dependent manner , levels of TH-positive dopaminergic neurons decreased more strongly and at lower concentrations ( see Figure 8a/b ) , with 6OHD having the more specific effect . Overall , our HTS workflow detected subpopulation-specific effects within dense , whole-mount-stained organoids with very little variation between replicates and between AMOs from separate cell lines . We next evaluated the use of cleaved caspase 3 ( cCasp3 ) as a general marker for cell death in AMOs for HTS toxicity studies ( see Figure 8—figure supplement 1 ) . If used judiciously , cCasp3 staining can yield quantitative dose-response curves , for example for the addition of known cell toxicants such as G418 ( see Figure 8—figure supplement 1a , b , d ) . When used in co-staining scenarios , this makes cCasp3 potentially attractive to quantify the viability of a wide range of cellular subpopulations , such as Sox2+ precursor cells ( see Figure 8—figure supplement 1c , e ) . However , cCasp3 only labels cells that are currently undergoing apoptosis , and this signal disappears when dead cells are cleared from the tissue . Consequently , we observed that the cCasp3 signal dropped at high concentrations of toxicants ( see Figure 8—figure supplement 1g ) , necessitating a careful optimization of assay timing and assay windows . This makes cCasp3 challenging to use for primary screens , where compounds with a wide range of toxicities and kinetics may need to be evaluated in parallel . Other viability assays , such as the ubiquitous CellTiter-Glo 3D assay , provide a much broader dynamic range and are less sensitive to timing , as they quantify living cells instead of a transient signal of dying cells ( see Figure 8—figure supplement 1i ) . Unfortunately , CellTiter-Glo cannot distinguish survival of cellular subpopulations , since it relies on a non-discriminatory lysis protocol . Thus , CellTiter-Glo may be most suitable for primary screens and cCasp3 staining as a follow up to probe effects on specific cellular subpopulations . Another common readout in 3D cell culture screens is aggregate size ( Yao et al . , 2020; Mittler et al . , 2017; Kim et al . , 2020 ) . Interestingly , the largest cross-sectional area of organoids did not correlate with G418 doses or CellTiter-Glo survival data in the conditions and time scales tested here , but remained largely constant despite escalating toxin levels ( see Figure 8—figure supplement 1h ) . Our automated workflow is compatible with all three types of assay and the results demonstrate that AMOs show lower variance in cell death/survival in toxicity studies than iPSC-organoids derived from the same cell line ( see Figure 8—figure supplement 1g , i ) . Overall , this data highlights the suitability of AMOs as an ideal tool for midbrain-specific drug and toxicity screening with various high-throughput-compatible and functional readouts that can be directly applied to large-scale screening campaigns as well as more detailed secondary analyses to gain mechanistic insight into the behaviors of distinct cellular subpopulations in the context of their niche .
In this report , we present a fully scalable , automated workflow for the generation , maintenance , fixation , immunostaining , clearing , and optical high-content analysis of human midbrain organoids . In designing our workflow , we used a standard 96-well format along with optimized protocols providing full compatibility with standard liquid handlers in most screening facilities . By omitting cumbersome matrix embedding steps , we eradicate variation originating from positional differences of cells within extracellular matrix ( ECM ) droplets during embedding . Importantly , our workflow is also compatible with manual pipetting , and thus can benefit labs without access to liquid handlers . By starting with neural precursor cells rather than PSCs , we can accelerate and streamline the neural differentiation process resulting in highly reproducible , homogeneous , functionally coupled , and electrophysiologically active human 3D microtissues . In contrast to the highly organized layers of the cortex , which can be partially recapitulated in cerebral/cortical organoids ( Lancaster et al . , 2013; Qian et al . , 2016; Mariani et al . , 2015; Bhaduri et al . , 2020 ) , different tissues of the midbrain have a much less distinctive morphological organization at the cellular scale . While our AMOs do not represent the complexity of the actual human midbrain , we demonstrate the presence of key cell populations with a degree of structure and organization similar to that of other published midbrain organoids ( Qian et al . , 2016; Jo et al . , 2016; Monzel et al . , 2017 ) . AMOs’ uniformity in key parameters such as size , cellular organization , gene expression , and protein levels enables large-scale 3D-based HTS strategies . Moreover , the presence of aggregate-wide , reproducible , synchronous calcium and electrical activity in human neural 3D structures may enable neural activity to be used as a HTS-compatible , simple readout for phenotypic screens including toxicology studies and drug screening for disorders with altered brain activity . While other groups have previously reported the use of human neural precursor cell ( hNPC ) -derived neurospheres for toxicity testing ( Moors et al . , 2009 ) , as well as a hiPSC-derived high-throughput-compatible spheroid model including the use of calcium oscillations as a readout to evaluate neurotoxicity ( Sirenko et al . , 2019 ) , these aggregates do not display structural tissue self-organization to the same extent as our organoids . Importantly , none of the available high-throughput-compatible protocols generates midbrain-specific organoids , but rather they focus on a cortical fate ( Sirenko et al . , 2019 ) . Our model of the human midbrain opens up the potential to perform 3D organoid-based HTS of midbrain-specific disorders including the highly prevalent Parkinson’s disease . One of the most widely used models for Parkinson’s disease is the ablation of dopaminergic neurons with toxins such as 6-hydroxy dopamine or MPP+ ( Reinhardt et al . , 2013b; Blum et al . , 2001; Simola et al . , 2007; Meredith and Rademacher , 2011 ) . Using these compounds as a dose-response model , our workflow demonstrates the unbiased , quantitative , cell-type-specific assessment of dopaminergic neurons in a physiologically relevant and complex human 3D model system with full scalability and high-throughput compatibility . Our automated whole mount optical analysis can also be directly applied to other established 3D protocols and may help uncover phenotypes that manifest themselves primarily in 3D or in distinct subpopulations of cells . The neural precursor-based proof-of-principle demonstrated here may be suitable , for example , in screens for diseases affecting primarily neural precursors including Zika virus infection . While this condition has been intensively studied using organoids , platform technologies for 3D high-throughput screens are still lacking and in high demand ( Qian et al . , 2017 ) . Furthermore , other non-midbrain organoid protocols may benefit from our workflow; for example , we have adapted cortical organoids by Paşca et al . , 2015/Sloan et al . , 2018 to be scaled-up and analyzed in our system as automated hiPSC-based control organoids for our midbrain model . Automation of this existing manual protocol necessitated some modifications ( summarized in Figure 7—figure supplement 1 ) . For example , in order to create a workflow that is as scalable and as repeatable as possible , we decided to use accutase-digested single cell suspensions originating from feeder-free iPSC-cultures rather than dispase-treated colonies lifted off partially or whole from feeder-based iPSC-cultures . The use of colonies as a starting point makes colony/aggregate sizes difficult to control , especially in an automated setting , resulting in a variable starting point for each organoid generated in this fashion . We also aggregated and maintained the resulting organoids in SBS-compatible U-bottom plates that facilitate automated handling of all organoids as separate biological replicates , each in its own well , rather than in 100 mm dishes as in the original publication . These modifications to the original protocol may have changed the outcome compared to the published original; however , we demonstrate proper forebrain differentiation at the RNA and protein levels , and provide detailed single organoid-based high-content data from three independent batches for select parameters . We also believe that these modifications provide the most rigorous comparison between iPSC- and precursor-based organoid approaches by starting both protocols with well-defined and easily controllable cell numbers as single cell suspensions . By utilizing colonies as starting point , we would have biased reproducibility against the automated iPSC-protocol , thus weakening our comparison . Despite our efforts to standardize iPSC-based organoids in a modified automated workflow , AMOs displayed a distinctly lower variance in a broad set of parameters , including size , cellular subpopulations , and survival in toxicity studies . This may indicate that mechanical standardization alone cannot compensate for the innate variability of PSCs during their self-organization . Although the use of precursors cells as a starting population provides benefits , the comparison to hiPSC-based organoids also highlights certain limitations of this approach . Our choice of generating AMOs by seeding more committed neural precursor cells instead of hiPSCs improves the predictability of the differentiation outcome while maintaining the ability to form self-organized tissues . However , it also sacrifices some of the hiPSCs’ broader cell fate potential and complexity resulting in simpler structures compared to hiPSC-based organoids . In addition , smNPCs can only give rise to cells of the ectoderm , which excludes , for example , the formation of microglia . Although smNPCs are capable of efficiently generating oligodendrocytes ( Reinhardt et al . , 2013a; Ehrlich et al . , 2017 ) , they generally arise very late in neural development ( Goldman and Kuypers , 2015 ) , and other organoid protocols also demonstrate the presence of oligodendrocytes after 100 days of differentiation or more ( Marton et al . , 2019 ) . These long time scales are not very attractive for the design of high-throughput screening strategies . Instead , we favor approaches that ectopically add a known number of oligodendrocytes or microglia to our AMOs and thus reach a screenable state much earlier . Similar strategies have recently been described for the introduction of microglia into hiPSC-based neural organoids ( Lin et al . , 2018; Abud et al . , 2017; Muffat et al . , 2018 ) , albeit without the ability to perform HTS . AMOs fill the gap between heterogeneous PSC-derived organoids and established but less physiological spheroid- and 2D-based screening formats . The emerging science of 3D cell culture promises to probe the effects of drugs on single cell types ( here: dopaminergic neurons ) in an intact niche as part of a complex 3D in vitro culture . Understanding the effects in context will be an essential factor to better understand biology at the tissue level . Currently , 3D cell culture science is in a phase of transition , where traditional manual low throughput protocols need to be adapted to unbiased higher throughput workflows . This is a prerequisite to mine the information necessary to merge organoid science with the promises of big data and machine learning to better tackle the complexities of understanding 3D biology . Therefore , we see our workflow as a contribution to help others translate successful strategies such as cell painting ( Bray et al . , 2016 ) and other big-data-generating high-content strategies ( Friese et al . , 2019; Scheeder et al . , 2018 ) into biologically relevant 3D models . If successful , this will help to open up single-cell-based phenotypic discovery to the third dimension by taking 3D-based HTS approaches beyond bulk techniques such as cell survival or gross morphology . Taken together , we hope that our automation approach can contribute to establishing a next generation of cellular 3D in vitro disease models that allow unbiased , quantitative , high-throughput access to human tissue-surrogates in a dish .
All cell lines used in this study tested negative for mycoplasma contamination in PCR- and sequencing-based analyses . Unless otherwise noted , all cells and 3D aggregates were maintained at 37°C and 5% CO2 . The human small molecule precursor cells ( smNPCs ) were generated and characterized during a previous study ( Reinhardt et al . , 2013a ) and cultured as described there , with minor modifications . We grew smNPCs in 1% ( v/v ) Matrigel ( BD ) -coated 6-well plates ( Sarstedt ) in N2B27 medium supplemented with the small molecules smoothened agonist ( SAG ) ( 0 . 5 μM , Cayman Chemical ) and CHIR 99021 ( 3 μM , Axon MedChem ) . N2B27 consisted of DMEM-F12 ( Thermo Fisher ) and Neurobasal Medium ( Thermo Fisher ) at a 1:1 ratio , enriched with 1:400 diluted N2 supplement ( Thermo Fisher ) , and 1:200 diluted B27 supplement without vitamin A ( Thermo Fisher ) , 1% penicillin/streptomycin/glutamine ( Thermo Fisher ) , and 200 μM ascorbic acid ( Sigma-Aldrich ) . Typically , we exchanged medium every other day . The cells were split every 5–7 days at a splitting ratio of 1:10 to 1:20 via accutase ( Sigma-Aldrich ) treatment for approximately 15 min at 37°C , yielding a single-cell solution . To stop the digestion , the cells were diluted in DMEM-F12 with 0 . 1% bovine serum albumin ( BSA ) ( Thermo Fisher ) and centrifuged at 220 g for 2 min . The cell pellet was resuspended in fresh smNPC medium ( N2B27 with 0 . 5 μM SAG and3 μM CHIR ) and plated on Matrigel-coated 6-well plates . Each batch consisted of independently frozen and thawed cells of the same passage and cell line , passaged separately and seeded , maintained , fixed , stained , and cleared separately from the other batches . All liquid handling steps ( seeding , maintenance , and fixation of organoids ) were performed by a Beckman Coulter Biomek FXP liquid handling station equipped with a 96-channel-pipetting head and an attached Cytomat incubator ( Thermo Fisher ) . After digestion by accutase , we seeded 9000 smNPCs in each well of a conical 96-well plate ( Thermo Fisher ) in smNPC medium and allowed them to aggregate for 2 days . To increase inter-cell adhesion , we added 0 . 4% ( w/v ) polyvinyl alcohol ( PVA , Sigma-Aldrich ) . Starting at day 2 , cells undergo ventral patterning over 4 days in two feedings by removal of CHIR99021 in the continued presence of 1 μM SAG and the addition of 1 ng/mL brain derived neurotrophic factor ( BDNF , PeproTech ) and 1 ng/mL glial cell line-derived neurotrophic factor ( GDNF , PeproTech ) . After ventralization , we removed SAG on day 6 , further supported midbrain differentiation and maturation by the addition of 1 ng/mL transforming growth factor beta 3 ( TGFβ−3 , PeproTech ) , and 100 μM dibutyryl cyclic adenosine monophosphate ( dbcAMP , Sigma-Aldrich ) . To boost maturation and cell survival during the rest of the neural maturation , we increased the concentration of BDNF and GDNF to 2 ng/mL each starting at day 6 . A single dose of 5 ng/mL Activin A ( eBioscience ) was added on day 6 only . Depending on the desired degree of maturity , the duration of the maturation phase can be prolonged to 100 days and longer . Organoids were fed every other day for the entire duration of culture . For size measurements , we took brightfield images of randomly selected AMOs or hiPSC organoids using a stereo microscope ( Leica MZ10 F , camera: Leica DFC425 C ) . Images were processed with ImageJ/Fiji ( Schindelin et al . , 2012 ) using a custom-tailored standardized workflow . The auto threshold function was used to discriminate samples from the background followed by a measurement of their area with the analyze particles function . The measured area corresponds to the largest cross-section of the organoid . Data were outputted to Microsoft Excel and GraphPad Prism v8 . 4 . 2 ( Graphpad Software , Inc ) for further analysis . Coefficients of variation ( CVs ) were calculated via CV = standard deviation/mean . In order to analyze protein expression in 3D in a HTS-compatible manner , we adapted a whole mount staining protocol based on Lee et al . , 2016 . for large-scale 3D aggregates and optimized it for use in an automated liquid handling system ( Beckman Coulter Biomek FXP , 96-channel-pipetting head ) . After fixation with 4% PFA ( VWR ) for 10–15 min , we stained the aggregates with primary and secondary antibodies ( Alexa Fluor secondary antibodies , Thermo Fisher ) for 6 days each . A list of all primary antibodies and concentrations can be found in Supplementary file 3 . We diluted the antibodies in a blocking and permeabilization solution ( 6% BSA , 0 . 5% Triton-X 100 ( Roth ) , 0 . 1% ( w/v ) sodium azide ( Sigma-Aldrich ) in PBS ( Sigma-Aldrich ) ) and renewed it every 2 days . Between primary and secondary antibody incubation as well as after the staining procedure we washed the samples 5 times for 1 hr with 0 . 1% Triton X-100 in PBS . This extremely long staining procedure allows the antibodies to fully penetrate the aggregates despite their large size and high density . To enable full penetration by microscope illumination , the whole mount staining procedure is followed by BABB-based tissue clearing ( Dent et al . , 1989 ) . First , the samples were dehydrated stepwise through a methanol ( Roth ) series ( 25% , 50% , 70% , 90% , 100% , 15 min each ) . Next , they were transferred to an organic solvent-resistant cyclo-olefin 96-well plate ( ‘Screenstar’ , Greiner Bio-One ) . The samples were incubated for 30 min in 1:1 methanol/BABB ( benzyl benzoate ( Sigma-Aldrich ) and benzyl alcohol ( Sigma-Aldrich ) 1:1 ) and subsequently kept in BABB for imaging . We used Imaris v9 . 1 . 2 ( Bitplane , Oxford Instruments ) for 3D rendering of confocal slices to produce Video 1 . We performed RNA isolation for quantitative real-time PCR ( qPCR ) analysis using the NucleoSpin RNA XS kit ( Macherey-Nagel ) according to the manufacturer’s instructions . Depending on the age and protocol used , we pooled organoids from one batch in order to yield enough RNA for downstream analysis; AMOs: 32 ( d6 ) , 24 ( d16 ) , or 18 ( d22 and d30 ) , hiPSC organoids: 32 ( d6 ) , 14 ( d22 ) , or 10 ( d30 ) . We determined RNA concentration and purity using a NanoDrop 8000 spectrophotometer ( Thermo Fisher ) and performed reverse transcription according to standard protocols using 1000 ng RNA per reaction . qPCR was done in triplicates on a Quantstudio 5 Real-Time PCR System ( Applied Biosystems ) with iTaq Universal SYBR Green Supermix ( Bio-Rad ) and 3 . 2 ng RNA equivalents per 10 μL reaction . Cycling conditions were 2 min at 95°C followed by 40 cycles of 15 s at 95°C and 60 s at 60°C . We calculated the relative expression using the ΔΔct method and normalized to undifferentiated smNPCs ( AMO line 1 ) collected before aggregation as well as using GAPDH as a housekeeping gene . Alternatively , gene expression was quantified with the Biomark 48 . 48 integrated fluidic circuit ( IFC ) Delta Gene assay ( Fluidigm ) according to the manufacturer’s instructions . Briefly , following 14 cycles of preamplification , the samples were subjected to an exonuclease I ( New England Biolabs ) treatment ( 37°C for 30 min and 80°C for 15 min ) and diluted twentyfold with DNA Suspension buffer ( TEKnova ) . The samples ( in duplicates ) and assay mixtures were loaded onto a 48 . 48 microfluidic IFC chip and run on the BioMark real-time PCR reader ( Fluidigm ) where they were amplified and measured according to manufacturer’s instructions . Here , data analysis was performed using the BioMark real-time PCR analysis software 4 . 3 . 1 ( Fluidigm ) with standard settings . Again , undifferentiated smNPCs ( AMO line 1 ) were used as a reference and GAPDH served as housekeeping gene . All data was transferred to Microsoft Excel for further processing and GraphPad Prism v8 . 4 . 2 for plotting . A list of all used primers can be found in Supplementary file 4 . For calcium imaging , we added 10 μM cell-permeant Fluo-4 AM ( Thermo Fisher ) diluted in AMO medium to the aggregates and incubated for 60 min at 37°C . For inhibitor studies cobalt ( II ) chloride ( Sigma-Aldrich ) was added to the medium at a concentration of 2 mM together with the Fluo-4 AM . Imaging was performed using a Dragonfly spinning disc confocal microscope ( Andor , Oxford Instruments ) at a frequency of 10 Hz for 4 min . Data analysis was performed using ImageJ/Fiji ( Schindelin et al . , 2012 ) . First , different ROIs were defined as depicted in Figure 4 . Then , the mean fluorescence intensity in those ROIs was measured over time and plotted using GraphPad Prism v8 . 0 . 2 . The video was assembled via ImageJ/Fiji ( Schindelin et al . , 2012 ) and the frame rate accelerated to compress 4 min real time at 10 Hz into 20 s running time . Alternatively , we measured fluorescence intensity on a Synergy Mx plate reader ( BioTek ) , acquired data with the Gen5 software ( BioTek ) and outputted it to Microsoft Excel and GraphPad Prism v8 . 0 . 2 for further analysis and plotting . Electrophysiological measurements on microelectrode arrays ( USB-MEA256system , Multichannel Systems ) were performed on electrode areas of 9-well MEAs as previously described ( Piccini et al . , 2017 ) . The MEAs were plasma-cleaned and coated with 1:75 diluted Matrigel ( Corning ) in KO-DMEM ( Invitrogen ) overnight and additionally for 2 hr before seeding with a 0 . 1% gelatin ( Sigma-Aldrich ) in PBS solution at room temperature . The MEAs were pre-warmed to 37°C , organoids were transferred to the electrode area of the MEAs , and allowed to attach for approximately 28 hr . Subsequently , MEA chambers with attached samples were used for electrophysiological recordings at 37°C . To study the effects of different pharmacological modulators , organoids were first measured under basal conditions ( i . e . no addition of compounds ) to record a reference signal . We then added specific pathway activators to each sample chamber of a MEA and recorded the electric field potential after a brief period of equilibration . After recording the signal for the activators , we next added the inhibitors for the respective pathways to the same well and repeated the measurement procedure as before . This guaranteed that we measured the change of electric field potential of each sample and the basal activity of each sample could serve as an internal control . The compounds used were dopamine hydrochloride ( 10 μm , Sigma-Aldrich ) , risperidone ( 10 nM , Sigma-Aldrich ) , GABA ( 10 μm , Sigma-Aldrich ) , bicuculline ( 1 μm , Sigma-Aldrich ) , glutamate/glycine ( 100 μm each , Sigma-Aldrich ) , and ketamine ( 10 μm , Sigma-Aldrich ) . It is possible to culture organoids on MEAs for longer and to achieve a more widespread contact with MEA electrodes . However , organoids flatten and change morphology and possibly cellular composition upon prolonged attachment , and we sought to measure the organoids’ electrical activity as close to their usual spherical state as possible . 28 hr allowed sufficient time for attachment of the aggregates to record electric fields from still spherical samples while providing enough mechanical cell-substrate connection to transport MEA substrates from the incubator to the recording rig . After the recordings , it was possible to gently remove the organoids without causing observable damage . Although not optimized ( as we only performed measurements at one time point per organoid ) , this opens up the possibility to transfer the organoids back to standard culture conditions after MEA measurements and thus perform multiple measurements over time . Datasets were recorded with Cardio2D software ( Multichannel Systems ) . Analyses were performed using the software Cadio2D+ ( Multichannel Systems ) and Origin v9 . 0 ( OriginLab ) on exported data . Discrete fast Fourier analyses in Origin ( Blackman window ) was used to assess frequencies of autonomous activity of the organoids . To compare neural sample activity via electric field oscillations ( Figure 8k ) , we pre-processed the raw MEA data with a Savitzky-Golay filter ( curves were smoothed with a window size of 50 to preserve peak data while removing noise ) in Origin v9 . 0 and formed the total sum of the absolute values of each oscillation over 15 s . The results were outputted to Microsoft Excel , reformatted and then transferred to GraphPad Prism v8 . 4 . 2 for plotting . Due to the morphology of AMOs ( high optical density and the fact that most cell bodies are located in a depth of at least 10–20 µm ) , it was technically impossible to perform the patch-clamp measurements on intact aggregates . Therefore , the organoids were treated with 1 mg/ml trypsin and then mechanically dispersed to obtain single cells . These were seeded on PDL-coated coverslips and cultured for 1–3 days in AMO medium ( we stated the age of AMOs at the time of dissociation ) . The transmembrane currents were recorded from isolated cells using the whole-cell configuration of the patch-clamp technique ( Hamill et al . , 1981 ) . The patch pipettes were fabricated from borosilicate glass on a Sutter P1000 ( Sutter Instrument company ) pipette puller . When filled with pipette solution , they had a tip resistance of 4–6 MΩ . Recordings were done using an EPC-10 amplifier ( HEKA Elektronik ) and Patchmaster acquisition software ( HEKA Elektronik ) . Series resistance , liquid junction potential , pipette and whole-cell capacitance were cancelled electronically . Bath solution contained ( mM ) : NaCl 140 , KCl 2 . 4 , MgCl2 1 . 2 , CaCl2 2 . 5 , HEPES 10 , D-glucose 10 , pH 7 . 4 and the pipette solution contained ( mM ) : K-aspartate 125 , NaCl 10 , EGTA 1 , MgATP 4 , HEPES 10 , D-glucose 10 , pH 7 . 4 ( KOH ) . We performed all experiments at room temperature . Recordings of current-voltage relationship ( I-V curves ) were done in voltage-clamp mode at a holding potential of −70 mV . Recordings of evoked action potentials were performed in current-clamp mode . Data were analyzed using Patcher’s Power Tool routine for IgorPro ( WaveMetrics ) , SciDAVis ( http://scidavis . sourceforge . net/ ) and Origin Pro 2019 ( Origin Lab ) . To reveal the shape of I-V curves , single traces were normalized to the peak amplitude and then averaged . To isolate RNA of single AMOs and organoids we used the Direct-zol-96 RNA kit ( Zymo Research ) according to the manufacturer’s instructions . We assessed RNA concentration and purity using a NanoDrop 8000 spectrophotometer and RNA integrity with a Bioanalyzer ( Agilent Technologies ) per standard protocols . Next , mRNA was enriched using the NEBNext Poly ( A ) Magnetic Isolation Module ( NEB ) followed by strand-specific cDNA NGS library preparation ( NEBNext Ultra II Directional RNA Library Prep Kit for Illumina , NEB ) . The size of the resulting library was controlled by use of a D1000 ScreenTape ( Agilent 2200 TapeStation ) und quantified using the NEBNext Library Quant Kit for Illumina ( NEB ) . Equimolar pooled libraries were sequenced in a single read mode ( 75 cycles ) on the NextSeq 500 System ( Illumina ) using v2 chemistry yielding in an average QScore distribution of 95% >= Q30 score and subsequent demultiplexed and converted to FASTQ files by means of bcl2fastq v2 . 20 Conversion software ( Illumina ) . We aligned the RNA sequencing reads to the human genome hg19 with TopHat2 aligner ( v2 . 1 . 1 ) ( Kim et al . , 2013 ) , using default input parameters . Gene annotation from Ensembl ( version GRCh37 . 87 ) were used in the mapping process . The number of reads that were mapped to each gene was counted using the Python package HTSeq ( v0 . 7 . 2 ) ( Anders et al . , 2015 ) with ‘htseq-count – mode union – stranded no’ . For the correlation analysis , sequencing data of 21 different human fetal organs and midbrain ( -like ) samples were obtained from GSE66302 and E-MTAB-4868 , respectively . Human prenatal cortex ( at 24 weeks post-conception ) RNA-seq datasets were obtained from www . nature . com/neuro/journal/v18/n1/extref/nn . 3898-S9 . zip . Reads were mapped to the human genome as described above . RPKM values ( Reads Per Kilobase of transcript per Million mapped reads ) for each gene were computed by Cufflinks ( v2 . 2 . 1 ) . We selected genes with high expression ( log RPKM > 1 ) for further analysis . Based on the expression of the selected genes , Pearson correlations were calculated . We averaged the correlation coefficients for biological replicates . PCA and differential expression analysis were performed with raw counts using the R package DESeq2 ( v1 . 18 . 1 ) . Genes were considered as deregulated if |log2FC| > 2 and FDR < 0 . 05 using Benjamini-Hochberge multiple test adjustment ( Benjamini and Hochberg , 1995 ) . Gene Ontology ( GO ) term enrichment was analyzed with the bioinformatics web server Gorilla ( Eden et al . , 2009 ) and visualized with REViGO ( Supek et al . , 2011 ) . For the comparison with cerebral organoids ( Figure 5—figure supplement 1 ) , the dispersion within groups ( Figure 5—figure supplement 1b ) was calculated using the average distance between data points and centroids on the PCA plot ( Figure 5—figure supplement 1a ) . All RNA sequencing data generated by us was deposited to the NCBI GEO database ( GSE119060 ) and can be accessed at https://www . ncbi . nlm . nih . gov/geo/query/acc . cgi ? acc=GSE119060 . To assess how quantitative our imaging workflow is , we performed a dilution experiment . We mixed unlabeled smNPCs with different percentages ( 1 . 25% , 2 . 5% , 5% , 10% , 20% , 40% ) of CellTracker deep red dye ( Life Technologies ) -labeled cells ( labeling according to standard protocols , dye concentration 1:20 , 000 ) and aggregated them in smNPC maintenance medium with 0 . 4% PVA . To explore the effects of overall aggregate size on quantitation , we generated aggregates with 100 , 000 as well as 200 , 000 cells in total . After 1 day of aggregation , the aggregates were fixed with 4% PFA , subjected to BABB-based tissue clearing , imaged , and analyzed as described below . After staining and clearing , we achieved uniform aggregate positioning within the wells by tilting the plates off the horizontal plane at 60 degrees for 1 min . Image acquisition was carried out in an Operetta high content imager ( Perkin Elmer ) and images were analyzed in Harmony 4 . 1 software and/or Columbus version 2 . 6 . 0 . Each high-content imaging experiment requires a customized set of parameters adjusted for the size of the aggregates , marker-wavelengths , their morphology and distribution within the sample , marker combinations , and signal and background intensities . As the details of each workflow are particular to our hardware and software setup , we first provide a general description of our workflow that focuses on explaining the principal steps in a manner that allows platform-independent reproduction . We then provide the detailed analysis pipeline for the data shown in Figure 6 ( quantification of Sox2 and Map2 in AMO line 1 ) with all its unique parameters and steps in the Materials and methods section titled ‘Detailed high-content imaging analysis example for the data shown in Figure 6' . After staining and clearing , we achieved uniform aggregate positioning within the wells by tilting the plates off the horizontal plane at 60° for 1 min . Image acquisition was carried out in an Operetta high content imager ( Perkin Elmer ) and images were analyzed in Harmony 4 . 1 software . We acquired a total of 16 confocal planes in three channels ( DAPI , Sox2 488 nm , and MAP2 647 nm ) with an inter-plane spacing of 36 . 6 μm for a total stack of 549 μm , covering the entire aggregate height ( clearing and dehydration steps tend to shrink and flatten aggregates slightly ) . To define the aggregate region on each image plane , all three channels were summed , filtered with a median filter to remove small localized features , and bright areas were identified via the 'find image region' function . After cleaning the edge of the aggregate region by dilation and erosion steps of 10 and 3 pixels , respectively , we identified bona fide AMOs by selecting for regions with a minimum of 300 arbitrary brightness units ( abu ) and 4000 μm2 size . In order to better isolate Sox2+ nuclei from the general background , we ran a sliding parabola algorithm with a curvature setting of 2 across each image plane in the 488 nm channel . Nuclei were then identified within each aggregate region via the ‘find nuclei’ function , algorithm 'M' and further selected to be Sox2+ if they were larger than 10 µm2 and brighter than 1200 abu . We excluded image artifacts , small dust particles , and overlapping nuclei by omitting nuclei brighter than 6000 abu and larger than 70 μm2 from further quantification . For final output , the number and total brightness of nuclei in 488 nm and of aggregate regions in 647 nm were summed for all planes and all fields of view in each well and transferred to Microsoft Excel and TIBCO Spotfire for further annotation , analysis , and plotting . We omitted data from wells that contained dust particles within the same FOV as organoids , incompletely imaged aggregates due to improper positioning , or AMOs that have been damaged or lost during culture or downstream processing . Plate 1 , 2 , and 3 represent independent batches of separately thawed and cultured cells of the same frozen batch . For cortical organoid generation ( see method section below ) , human induced Pluripotent Stem Cells ( hiPSCs ) ( parental line of AMO line 2 smNPCs ) were generated and characterized during a previous study ( Reinhardt et al . , 2013b ) . Cells were cultured in Vitronectin ( Thermo Fisher ) -coated 6-well plates in mTeSR Plus medium ( Stemcell technologies ) supplemented with 1% penicillin/streptomycin . The medium was changed every other day and cells were passaged at a ratio of 1:10 to 1:15 , using accutase when they reached 80–90% confluency . After splitting , the medium was supplemented with the ROCK inhibitor Y-27632 ( 10 μM , tebu-bio ) until the first media exchange . For the generation of iPSC-based cerebral organoids ( see method section below and Figure 5—figure supplement 1 ) hiPSC culture was performed feeder-free using modified FTDA medium ( Frank et al . , 2012 ) in 1% ( v/v ) Matrigel-coated 6-well plates with iPSCs from the same line as AMO line 2 . FTDA medium consisted of DMEM-F12 supplemented with 1% human serum albumin ( Biological Industries ) , 1% Chemically Defined Lipid Concentrate ( Life Technologies ) , 0 . 1% Insulin-Transferrin-Selenium ( BD ) , 1% penicillin/streptomycin/glutamine . We fed the iPSCs daily and added 10 ng/mL FGF2 ( PeproTech GmbH ) , 0 . 2 ng/mL TGFβ3 ( PeproTech GmbH ) , 50 nM Dorsomorphin ( Enzo Life Sciences ) , 5 ng/mL Activin A ( eBioscience ) , 20 nM C59 ( Tocris ) before each media exchange . We split the iPSCs as single cells every 3–5 days using accutase for approximately 10 min at 37°C . We transferred 600 , 000 cells per well of a 6-well plate to be seeded to DMEM-F12 with 0 . 1% BSA and centrifuged at 220 g for 2 min . We resuspended the cell pellet in fresh FTDA medium supplemented with 1:2000 ROCK inhibitor Y-27632 ( tebu-bio ) and plated the iPSCs on Matrigel-coated 6-well plates . As a control for our AMOs , we generated cortical hiPSC organoids from the same hiPSC line used to derive smNPCs for AMO line 2 ( Reinhardt et al . , 2013b ) . After manual 2D culture of hiPCs , all steps were fully automated using our liquid handling system ( Beckman Coulter ) with attached incubator ( Thermo Fisher ) . Generally , we followed the protocol previously published by Paşca et al . , 2015 . ( and also described in more detail by Sloan et al . , 2018 ) , with adaptations for our automation pipeline ( see Figure 7—figure supplement 1 ) . Starting with 90–100% confluent cultures , we detached hiPSCs with accutase and seeded 10 , 000 cells per well in ultra-low attachment U-bottom plates ( Corning ) . Cortical organoid medium consisted of DMEM F-12 , 20% Knock-out Serum replacement ( GIBCO ) , 1% penicillin/streptomycin/glutamine , 1% Non-essential amino acids ( Sigma-Aldrich ) , and 0 . 2% 2-Mercaptoethanol ( Thermo Fisher ) . For the first 6 days , we supplemented the cortical organoid medium with 5 μM dorsomorphin ( Enzo Life Sciences ) and 10 μM SB-431542 ( Biomol ) . During seeding only , we also added 10 μM ROCK inhibitor Y-27632 . Aggregates were fed every 3 days using an automated liquid handling system . From day 6 to 24 , culture medium supplements were exchanged to EGF and FGF2 ( both 20 ng/ml , PeproTech ) and afterwards BDNF and NT3 ( metabion ) ( both 20 ng/ml ) . To measure the viability of individual organoids we used the CellTiter-Glo 3D Cell Viability Assay ( Promega ) according to the manufacturer’s instructions . The entire procedure was performed using an automated liquid handling system ( Beckman Coulter ) and is thus fully scalable and HTS-compatible . In short , the reagent and the AMOs were brought to room temperature in their 96-well culture plates for 30 min and the media volume of each 96-well was adjusted to 55 µl . We added an equal volume ( 55 µl ) of the CellTiter Glo 3D reagent and let it shake on a Thermomixer ( Eppendorf ) at 900 rpm for 5 min before incubating the samples protected from light at room temperature for 25 min . To prevent cross-talk between wells when measuring the luminescence , we next transferred the contents from the clear 96-well culture plates to opaque white 384-well Lumitrac plates ( Greiner ) with two technical replicates per sample . Luminescence was recorded immediately after transfer with a Synergy Mx plate reader ( BioTek ) . The results were outputted to Microsoft Excel , reformatted and then transferred to GraphPad Prism v8 . 4 . 2 for plotting . Coefficients of variation ( CVs ) were calculated via CV = standard deviation/mean . On day 47 of differentiation , AMOs received Tox medium ( TM ) overnight to remove antioxidants present in the B27 media supplement . TM consisted of DMEM-F12 supplemented with 1% N2 and 1% penicillin/streptomycin/glutamine . The next day , AMOs were treated with different concentrations ( 0 , 50 , 100 , 250 , 500 μm ) of either 6-Hydroxydopamine hydrochloride ( Sigma-Aldrich ) or 1-Methyl-4-phenylpyridinium iodide ( MPP+ , Sigma-Aldrich ) in TM . To refresh compounds , AMOs received TM including toxins at the same concentrations as before 24 hr after the first application . After a total incubation of 48 hr , the medium was changed back to standard AMO medium , and the samples were cultured for 6 more days to allow cell death to occur and the dead cells to be cleared from the organoids . After 6 days , the samples were fixed with 4% PFA , whole mount stained for Map2 and TH , BABB-cleared , and subjected to high-content confocal imaging as described above . We performed image analysis following the steps as outlined in the high-content analysis section with slight modifications to accommodate the individual brightness , morphology , and background characteristics of the staining . Briefly , after identifying AMOs , we ran a Gaussian smoothing algorithm across the TH 647 nm ( 10 px width ) and Map2 488 nm ( five px width ) channels and subtracted the smoothed images from the raw images to better isolate TH+ and Map2+ cells from the background . In the TH 647 nm channel , cells were then identified with the ‘find cells’ function , algorithm ‘C’ and further selected to be TH+ if they were brighter than 200 abu and larger than 25 μm2 . For Map2 , an additional sliding parabola algorithm with a curvature of 10 was run across the subtracted image to further reduce background noise . Map2+ cells were then also identified using the ‘find cells’ function algorithm ‘C’ . As final output , the total intensity of the identified TH+ cells in the raw 647 nm channel and the Map2+ cells in the raw 488 nm channel were summed for all fields of view and confocal planes per well . The data was then transferred to Microsoft Excel for further analysis , including normalization to the organoid area , and plotted using GraphPad Prism v8 . 4 . 2 . We collected the cell culture supernatant from 35 days old organoids 40 hr after feeding and measured its dopamine content using a Dopamine ELISA Kit ( Abnova ) according to the manufacturer’s instructions . Measurements were performed in duplicates and the sample concentrations were calculated based on a standard curve . Since there is no single marker that identifies all different mature cell populations within the AMOs , but all immature smNPCs express the precursor marker Sox2 , which is downregulated upon maturation , we defined Sox2-negative cells as mature for the purpose of this analysis . Using confocal high-content imaging analysis as described above , we quantified the number of Sox2-positive cells as well as the total cell number ( based on DAPI-stained nuclei ) within the entire organoids . We then calculated the ratio of Sox2-negative cells as 1 - ( number of Sox2-positive cells ) / ( total cell number ) . Calculations were performed in Microsoft Excel and data was transferred to GraphPad Prism v8 . 4 . 2 for plotting . ScalesSQ tissue clearing was performed as previously described by Hama et al . , 2015 . Briefly , we incubated organoids in ScaleSQ ( 22 . 5% ( w/v ) ) D-sorbitol ( Sigma ) and 9 . 1 M urea ( Sigma ) for 2 hr at 37°C . Next , we exchanged the solution to ScaleS4 ( 40% ( w/v ) ) D-sorbitol , 10% ( w/v ) glycerine ( Roth ) , and 15–25% ( v/v ) DMSO ( Sigma ) also at 37°C . ScalseS4 was renewed after 2 hr and the organoids were maintained at 37°C until analysis . This is critical , as the clearing effect decreases visibly after 1 hr at room temperature . We performed X-Clarity tissue clearing using an X-CLARITY Polymerization System ( Logos biosystems ) per manufacturer’s instructions . First , organoids were fixed with 4% PFA overnight at 4°C . The next day , the organoids were embedded in a hydrogel monomer solution and incubated 12–24 hr , followed by a 3-hr polymerization step at −90 bar and 37°C in the X-CLARITY Polymerization System . The embedded organoids were transferred to a tissue container and cleared in the electrophoretic tissue chamber for 2 hr at 1 . 2 A and 37°C . For ClearT tissue clearing we followed a protocol by Kuwajima et al . , 2013 . In short , fixed organoids were incubated for 30 min at room temperature in increasing formamide ( AppliChem ) concentrations: 20% , 40% , 80% , and 95% ( AppliChem ) . Samples were directly used for further applications or stored at 4°C . CUBIC tissue clearing was modified after Susaki et al . , 2015 . Briefly , we incubated the organoids for 6 hr at room temperature in CUBIC-1 ½ ( one part water and one part CUBIC-1: 25% ( w/w ) quadrol ( Sigma ) , 25% ( w/w ) urea , and 15% ( w/w ) triton X-100 in water ) . Afterwards , samples were incubated for 24 hr in CUBIC-1 at 4°C and 24 hr in CUBIC-2 ( 25% ( w/w ) urea , 50% ( w/w ) sucrose , and 10% ( w/w ) triethanolamine ( Sigma ) ) . We stained the organoids with 0 . 5 μg/mL DAPI ( Sigma ) in PBS for 24 hr and subsequently subjected them to the different tissue clearing protocols . Afterwards , a LSM 700 scanning confocal microscope ( Zeiss ) was used to acquire z-stacks of the stained and cleared organoids . Three XZ and three YZ cross-sections per aggregate ( n = 10 aggregates per clearing method ) were used to quantify the maximum depth at which the DAPI signal could still be detected at a given brightness threshold . The depth of each cross section was measured manually at n = 5 different positions for each slice and n = 10 organoids per clearing protocol using ImageJ/Fiji . The data was exported to Microsoft Excel for further processing and GraphPad Prism v7 . 0 for plotting and statistical analysis . Following clearing , we took brightfield images of the organoids ( n = 10 per protocol , n = 6 for X-Clarity ) using an upright stereomicroscope ( Leica MZ10 F , camera: Leica DFC425 C ) under standardized brightfield transmission conditions . The mean brightness of the organoid area , measured by ImageJ/Fiji , served as measure for the amount of light transmitted through the aggregate . The data was exported to Microsoft Excel for further processing and GraphPad Prism v7 . 0 for plotting and statistical analysis . For statistical analysis of the clearing protocol comparisons , we used GraphPad Prism v7 . 0 and performed unpaired , two-tailed t-tests with α = 0 . 05 as normal distribution and equal variances can be assumed for the analyzed data . For electron microscopy analysis , we used n = 3 AMOs from one batch at day 32 . Samples were initially fixed for 3 hr with 2% glutaraldehyde , 2% paraformaldehyde in 0 . 2 M cacodylate buffer , pH 7 . 2 . Afterwards , the specimen was dissected into smaller pieces and post-fixed with 1% osmium tetroxide containing 1 . 5% potassium hexacyanoferrate . Samples were dehydrated stepwise , including an over-night uranyl-block staining step in 70% ethanol . The specimen was orientated and flat embedded in epon . In total , three samples sectioned under different angles were analyzed at the electron microscope to visualize different aspects of the AMOs . For iPSC-derived cerebral organoid generation , we followed the protocol by Lancaster et al . , 2013 . with minor modifications . Briefly , we dissociated iPSCs to single cells by accutase treatment and plated 9000 cells per well in a conical 96-well plate in low FGF stem cell medium ( DMEM-F12 with knockout serum replacement ( KOSR , Thermo Fisher ) 1:5 , fetal bovine serum ( Biochrom ) 1:33 . 3 , 1% penicillin/streptomycin/glutamine , 1% non-essential amino acids ( NEAA , Sigma-Aldrich ) , 2-mercaptoethanol ( Thermo Fisher ) 1:143 , 4 ng/μL FGF2 , 50 μm ROCK inhibitor Y-27632 , and 0 . 4% PVA on seeding day only to facilitate aggregation ) . We exchanged the medium every other day , FGF2 and Y-27632 were withdrawn on day 6 . Neural induction was started on day 8 ( neural induction medium: DMEM-F12 with KOSR 1:5 , 1%penicillin/streptomycin/glutamine , 1% non-essential amino acids , N2 supplement 1:100 , and Heparin ( Sigma-Aldrich ) 1 μg/mL ) and continued for 6 days with media changes every other day . On day 13 , we embedded the aggregates into 30 μL matrigel droplets and transferred them to 6 cm2 suspension tissue culture dishes ( Sarstedt ) in cerebral organoid differentiation medium ( DMEM-F12 and Neurobasal 1:1 with 1% penicillin/streptomycin/glutamine , 1% NEAA , N2 upplement 1:200 , B27 supplement without vitamin A 1:100 , Insulin ( Sigma-Aldrich ) 1:4000 , and 2-mercaptoethanol 1:285714 ) . We placed the culture dishes on a shaker at 37°C and 5% CO2 and fed the organoids every other day . On day 20 the B27 supplement was replaced by B27 with Vitamin A ( Thermo Fisher ) and organoids were cultured until day 30 or 45 . At day 50 , we treated AMOs with increasing concentrations ( 0 , 5 , 50 , 100 , 250 , 500 , 1000 μg/mL ) of G418 ( Sigma-Aldrich ) added directly to the culture medium . After 2 days , we renewed the medium ( including identical toxin concentrations ) and fixed the aggregates after a total of 4 days of treatment . Fixation , whole mount immunostaining for cCasp3 and Sox2 as well as BABB-based clearing was performed as outlined above . Image analysis followed the steps as outlined in the high-content analysis section with slight modifications to accommodate the individual brightness , morphology , and background characteristics of the cCasp3 staining . Briefly , after identifying AMOs and Sox2+ cells as described previously , the cCasp3 channel was background corrected by running a sliding parabola algorithm with a curvature setting of 10 across each confocal slice of the AMO . We identified apoptotic cells via the ‘find nuclei’ function in the 647 nm channel , algorithm 'M' and further selected them to be cCasp3+ if they were larger than 11 μm2 , smaller than 100 μm2 , and brighter than 2700 abu . We considered cells to be Sox2/cCasp3 double-positive if they fulfilled the criteria for both filters at the same time . The results were outputted to Microsoft Excel , reformatted and then transferred to GraphPad Prism v8 . 0 . 2 for plotting , data analysis , and curve fitting . On day 30 of differentiation , AMOs received Tox medium ( TM ) overnight to remove antioxidants present in the B27 media supplement . TM consisted of DMEM-F12 supplemented with 1% N2 and 1% penicillin/streptomycin/glutamine . On day 31 , we added different concentrations of G418 ( ranging from 0 to 10 , 000 μg/mL ) dissolved in TM and renewed the medium ( including the different compound concentrations ) once two days later . After a total treatment time of 96 hr , the samples were analyzed by their size , the CellTiter-Glo 3D Cell Viability Assay , and cCasp3 staining as described before . In the case of cCasp3 staining , the image analysis parameters were used with minor modifications to account for the individual characteristics of the staining , for example signal intensity , aggregate size , background . The calculation of the organoid size/area plotted in Figure 8—figure supplement 1h was performed on the same fluorescence images as the cCasp3 quantification in subfigure g ) according to the methods outlined in the description of the high content analysis . Data was transferred to Microsoft Excel for further analysis and GraphPad Prism v8 . 4 . 2 for plotting and curve fitting . | In 1907 , the American zoologist Ross Granville Harrison developed the first technique to artificially grow animal cells outside the body in a liquid medium . Cells are still grown in much the same way in modern laboratories: a single layer of cells is placed in a warm incubator with nutrient-rich broth . These cell layers are often used to test new drugs , but they cannot recapitulate the complexity of a real organ made from multiple cell types within a living , breathing human body . Growing three-dimensional miniature organs or 'organoids' that behave in a similar way to real organs is the next step towards creating better platforms for drug screening , but there are several difficulties inherent to this process . For one thing , it is hard to recreate the multitude of cell types that make up an organ . For another , the cells that do grow often fail to connect and communicate with each other in biologically realistic ways . It is also tough to grow a large number of organoids that all behave in the same way , making it hard to know whether a particular drug works or whether it is just being tested on a 'good' organoid . Renner et al . have been able to overcome these issues by using robotic technology to create thousands of identical , mid-brain organoids from human cells in the lab . The robots perform a series of precisely controlled tasks – including dispensing the initial cells into wells , feeding organoids as they grow and testing them at different stages of development . These mini-brains , which are the size of the head of a pin , mimic the part of the brain where Parkinson's disease first manifests . They can be used to test new drugs for Parkinson's , and to better understand the biology of the brain . Perhaps more importantly , other types of organoids can be created using the same technique to model diseases that affect other areas of the brain , or other organs altogether . For example , Renner et al . also generated forebrain organoids using an automated approach for both generation and analysis . This research , which shows that organoids can be grown and tested in a fully automated , reproducible and scalable way , creates a platform to quickly , cheaply and easily test thousands of drugs for Parkinson's and other difficult-to-treat diseases in a human setting . This approach has the potential to reduce research waste by increasing the chances that a drug that works in the lab will also ultimately work in a patient; and reduce animal experiments , as drugs that do not work in human tissues will not proceed to animal testing . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"stem",
"cells",
"and",
"regenerative",
"medicine",
"tools",
"and",
"resources"
] | 2020 | A fully automated high-throughput workflow for 3D-based chemical screening in human midbrain organoids |
The skull of ‘Ligulalepis’ from the Early Devonian of Australia ( AM-F101607 ) has significantly expanded our knowledge of early osteichthyan anatomy , but its phylogenetic position has remained uncertain . We herein describe a second skull of ‘Ligulalepis’ and present micro-CT data on both specimens to reveal novel anatomical features , including cranial endocasts . Several features previously considered to link ‘Ligulalepis’ with actinopterygians are now considered generalized osteichthyan characters or of uncertain polarity . The presence of a lateral cranial canal is shown to be variable in its development between specimens . Other notable new features include the presence of a pineal foramen , the some detail of skull roof sutures , the shape of the nasal capsules , a placoderm-like hypophysial vein , and a chondrichthyan-like labyrinth system . New phylogenetic analyses place ‘Ligulalepis’ as a stem osteichthyan , specifically as the sister taxon to ‘psarolepids’ plus crown osteichthyans . The precise position of ‘psarolepids’ differs between parsimony and Bayesian analyses .
Scans of AM-F101607 reveal for the first time some of the sutures between the skull roofing bones ( Figure 1 ) showing a pattern different in important respects to the previous interpretation ( Basden and Young , 2001 ) . Viewing the scan in Drishti reveals a set of parallel bands tracing what we assume to be bone sutures in the posterior part of the skull ( Figure 1B ) . Closer inspection of the scan data reveals these bands to be high-density thickenings in the basal layer of the dermal skull roof bones ( Figure 1—figure supplement 1 ) . These show the outline of the postparietals ( of sarcopterygians; parietals of actinopterygians ) and the posterior edges of the parietals ( of sarcopterygians; frontals of actinopterygians ) ( Figure 1B , C ) . No midline suture is evident between the postparietals , but a very faint suture between the parietals is suggested . The lateral margin of the postparietal is scalloped in such a way as to provide contact faces for a series of three bones , with faint lines visible demarcating them . The most posterior bone presumably corresponds to the tabular ( of sarcopterygians; supratemporal of actinopterygians ) . Anterior to this is a supratemporal ( of sarcopterygians; intertemporal of actinopterygians ) , and a broad and elongate intertemporal ( of sarcopterygians; dermosphenotic of actinopterygians ) borders the orbit . Unfortunately , sutures cannot be visualized in the same way in ANU V3628 , despite the higher scan resolution , because no high density growth bands are evident . Instead the basal layer of the skull roof dermal bone is of uniform density and thickness . This suggests that they may vary between individuals or growth phases . Further specimens of ‘Ligulalepis’ are required to unambiguously determine the pattern of skull roof bones in this taxon . The presence of middle and posterior pitlines , and the supraorbital canals extending to the posterior edge of the postparietals , is confirmed in ANU V3628 . ANU V3628 preserves the previously unknown anterior portion of the skull roof ( Figure 2 ) . A pineal foramen is preserved , but due to a crack in the specimen it is unclear if a separate pineal plate was present . Sutures in the anterior part of the skull are unclear . The pattern of ornamentation anterior to the pineal opening is suggestive of a median ossification ( i . e . a median rostral ) , but this is not evident from the CT data . The profile of the snout has a sharply downturned anterior face ( Figure 3 ) , as is general for gnathostomes ( Gardiner , 1984; Long , 1988; Zhu et al . , 2013; Zhu et al . , 2009 ) . There is an abrupt change in ornamentation on the snout , from short anteriorly directed ridges to elongate transverse ridges ( Figure 2C ) . A similar pattern is known in Dialipina ( Schultze and Cumbaa , 2001 ) . The incurrent nostrils are large and widely separated from the excurrent nostrils , which appear to lie entirely within the orbits ( Figure 3 ) . Basden and Young ( 2001 ) also assumed communication of the posterior nostril with the orbit , including a notch for the nostril on the anterior margin of the orbital fenestra . Neither specimen of ‘Ligulalepis’ show evidence for such a notch , although the ventral part of the nostril and orbital margin are unknown . A nostril confluent with the orbit is typically considered an actinopterygian character , but without preservation of the premaxilla and cheek bones in ‘Ligulalepis’ we cannot rule out the possibility that dermal bone separated the external opening of the nostril from the orbit – for example a postero-dorsal process of the premaxilla as in Psarolepis ( Yu , 1998 ) , and perhaps Cheirolepis ( Gardiner , 1984 , Fig . 49 ) . However , in ‘Ligulalepis’ the opening for the posterior nostril in the endocranium lies directly within the orbit ( Figure 3A ) . This is in contrast to the situation in both actinopterygians and sarcopterygians , where an endoskeletal lamina ( the postnasal wall ) separates the nostril and the orbit ( e . g . Gardiner , 1984 , Fig . 13 ) . ‘Ligulalepis’ lacks such a lamina , and in this respect more closely resembles some placoderms such as Parabuchanosteus ( Young , 1979 ) and Dicksonosteus ( Goujet , 1984 ) . The supraorbital canal ( soc ) extends nearly the full preserved length of the cranium , terminating a little way posterior to the downturned margin of the snout , and appears to be formed from two separate sections ( Figures 1C and 3C ) . The sections overlap slightly anterior to the level of the postorbital process , the posterior section pinching out and sitting on top of the anterior section ( Figure 3C , indicated by arrow ) . Tubuli connecting the supraorbital canal to the surface are small and few in number . Tubuli connecting the infraorbital ( ioc ) and otic ( otc ) canals to the surface are larger ( Figure 3D , arrows ) . The tubuli do not appear to be branched ( although they may have branched in the skin above the bone ) , in contrast to the highly branched tubuli of some early sarcopterygians ( Bjerring , 1972; Clément and Ahlberg , 2010; Jarvik , 1972 ) . It is not clear whether the pores for the sensory canals figured for Mimipiscis and Moythomasia originate from branched or individual tubuli ( Gardiner , 1984 ) . Anterior to the level of the pineal foramen , the supraorbital sensory canals open to the dorsal surface of the cranium ( Figure 2 ) , although the canal itself is housed in a ridge on the visceral surface of the skull roof . This is similar to the condition in Achoania ( Zhu et al . , 2001 ) , Guiyu ( Zhu et al . , 2009 ) , and Psarolepis ( Yu , 1998 ) , and may be equivalent to the ‘nasal pitlines’ described for Mimipiscis ( Gardiner , 1984 , fig . 41 , 102 ) , although in Mimipiscis the supraorbital canals continue anterior to the pitline . ANU V3628 is ventrally incomplete , so it is not clear if an ethmoid commissure was present . If an ethmoid commissure was present , the supraorbital canals did not communicate with it . Basden and Young ( 2001 ) described a lateral notch for a preopercular sensory line , however , scans show no evidence for a preopercular canal in either specimen . A short anterior canal at the intersection of the otic and infraorbital canals is present in ANU V3628 ( Figure 3D: ‘P’ ) , but less developed in AM-F101607 . This is the ‘P’ canal of Northcutt , 1989 . It is present in some acanthodians , for example Acanthodes ( Watson , 1937 ) and some actinopterygians , namely Mimipiscis and Moythomasia ( Gardiner , 1984 ) . The wider distribution of the ‘P’ canal is difficult to assess in other taxa in the absence of exceptionally preserved material or CT data . The preservation of the braincase is similar in both AM-F101607 and ANU V3628 . It is mostly well ossified , and comprises the basisphenoid , orbitotemporal and otic regions . The ethmoid region is preserved on the left side of AM-F101607 and is more complete , but less well ossified , in ANU V3628 . The posterior and ventral portion of the braincase , comprising the occiput and basioccipital , is absent in both specimens . The loss of this region , which is rarely preserved in early osteichthyans ( e . g . Yu , 1998; Zhu et al . , 2001; Lu et al . , 2012a2012 ) , presumably corresponds to the presence of well-developed otoccipital and ventral otic fissures , possibly in conjunction with a vestibular fontanelle . The orbital region is large , comprising nearly half of total braincase length , and the ethmoid region short . A comparison of the two cranial endocasts is shown in Figure 7 . Differences in appearance largely relate to the presence of extensive rock matrix surrounding ANU V3628 , in contrast to the acid-prepared cranium of AM-F101607 . The external walls of the endocranial cavity are largely complete in both specimens , although as the parachordal plate of the braincase is not preserved the ventral extent is uncertain . Overall , the endocast of ‘Ligulalepis’ is short and broad , particularly the otic region ( Figure 7A , B ) . The proportions occupied by different regions are similar to early chondrichthyans , with the forebrain section comprising less than 20% of the total length , the midbrain section around 15% , and the hindbrain section some 65% . Description of the endocast allows the identity of features within the cranial cavity to be revised . A distinct depression in the roof of the cranial cavity , medial to the otic capsule , was considered by ( Zhu et al . ( [2010] , fig . 4c ) to be evidence of a lateral cranial canal . This embayment is in fact the crus commune of the anterior and posterior semicircular canal ( Figure 4B ) . The groove anterior to this is somewhat shallower in the braincase and indicates where the roof of the utricular region joins the rest of the cranial cavity ( the groove for anterior and posterior semicircular canals of Basden and Young 2001: fig . 3 ) . AM-F101607 and ANU V3628 were coded into an updated phylogenetic analysis modified from Lu et al . , 2017 . As well as changes to anatomical scores for ‘Ligulalepis’ , codes for several taxa were updated and some characters were deleted or reformulated to give a total of 282 characters coded for 94 taxa ( for full details see the ‘phylogenetic methods’ section ) . This dataset was analysed using both parsimony and Bayesian inference . The parsimony analysis retrieves Dialipina , ‘Ligulalepis’ , and ‘psarolepids’ as successively branching sister taxa to the osteichthyan crown node ( Figure 10A ) . However , support for the clade that comprises crown osteichthyans ( as retrieved from this analysis ) is low , with Bremer support of 1 and a bootstrap of just 4 . This is very weak support , although we note that bootstrap values obtained from TNT are likely to be much more conservative than those produced by PAUP*: bootstrap values in TNT are calculated from the strict consensus trees found in each replicate ( Goloboff et al . , 2008 ) , whereas PAUP* uses all the shortest trees from each replicate , weighted by the reciprocal of the number of trees found in that replicate ( Swofford , 2003 ) . There are six unambiguous character state changes on the branch leading to crown osteichthyans . These are #78 ( enameloid on teeth gained ) , #110 ( shape of parashenoid splint shaped ) , #116 ( olfactory tracts long ) , #130 ( eyestalk absent ) , #184 ( median dorsal plate absent ) , #211 ( dorsal fin spines absent ) . Of these , only the olfactory tracts and eyestalk are known in 'Ligulalepis' . Alternative phylogenetic placements under parsimony were tested using two constrained searches , one with ‘Ligulalepis’ constrained within actinopterygians and another with ‘psarolepids’ constrained within sarcopterygians . A stem actinopterygian position for ‘Ligulalepis’ requires a single additional step , and the grouping of ‘Ligulalepis’ and actinopterygians was found in 18% of the bootstrap replicates . Enforcing this topology also resulted in ‘psarolepids’ being resolved as stem sarcopterygians ( Figure 10B ) . A single additional step is required to place ‘psarolepids’ on the sarcopterygian stem , and this grouping is found in 16% of bootstrap replicates . When this grouping is enforced it leads to ‘Ligulalepis’ falling into a polytomy with actinopterygians and sarcopterygians ( Figure 10B ) . The Bayesian analysis retrieves ‘psarolepids’ on the sarcopterygian stem with moderately strong support ( pp = 0 . 94 , Figure 11 ) . ‘Ligulalepis’ is resolved as a stem osteichthyan in the 50% majority rule tree ( Figure 11 ) , although the crown osteichthyan clade has weak support ( 0 . 61 ) . However , an actinopterygian position for ‘Ligulalepis’ has a posterior probability of 0 . 22 .
‘Ligulalepis’ is recovered as a stem osteichthyan in the phylogenetic analysis , specifically as the sister lineage to ‘psarolepids’ ( Guiyu , Sparalepis , Psarolepis , Achoania ) plus crown Osteichthyes ( Figures 9 and 10 ) . Dialipina is resolved as the sister taxon to all other osteichthyans . However , the placement of ‘Ligulalepis’ as the earliest diverging stem actinopterygian requires only a single additional step , and evidence for an actinopterygian affinity must be considered . Cranial features previously suggested as linking ‘Ligulalepis’ with actinopterygians ( Basden and Young , 2001 ) are now better considered to be general osteichthyan characters ( e . g . dermal ornament ) or of uncertain polarity ( skull roof pattern and overall structure ) . Of the three characters proposed by Lu et al . , 2017 as uniting ray-finned fishes inclusive of Meemannia , ‘Ligulalepis’ lacks two: posteriorly expanded tabulars ( supratemporals of actinopterygians ) and a spiracular canal . The remaining character , presence of a lateral cranial canal ( Coates , 1999; Gardiner , 1984 ) , is harder to assess . Primitively , the lateral cranial canal connects with the endocavity through the loop of the posterior semicircular canal , but in neopterygians it may connect with the cranial cavity anteriorly ( e . g . ‘Caturus’ , Rayner , 1948 , Giles et al . , 2018 ) , communicate with the fossa bridgei ( e . g Pteronisculus , Boreosomus: Nielsen , 1942; Polyodon: Bridge , 1878 ) or form an independent pocket ( e . g . Acipenser , Gardiner 1984 ) . Patterson ( 1975 ) claimed that the symmetry and even the presence of this character can vary between individuals of the same species – although investigation of several of Patterson’s specimens via CT scanning has identified only symmetrical lateral cranial canals ( Giles et al . , 2018 ) . In Mimipiscis the lateral cranial canal in some specimens can occupy the whole area between the posterior and anterior semicircular canals , while in others be ‘little more than a pocket in front of the posterior semicircular canal’ ( Gardiner , 1984 , pg . 242 ) . Gardiner ( 1984 ) suggested that the lateral cranial canal can be expressed simply in terms of the degree of ossification of the dorsal otic region . The two specimens of ‘Ligulalepis’ seem to confirm this idea , with the development of a lateral cranial canal variable between specimens , and the extent of the canal and perichondral lining also variable within a specimen . The endocranium of Meemannia is known only from a single skull specimen , so variability in development of the lateral cranial canal cannot be studied in this taxon . Mechanical preparation of Meemannia may also have obscured aspects of lateral cranial canal anatomy . Moreover , an actinopterygian identification for ‘Ligulalepis’ is also at odds with the lack of pore canal network . Topology tests reveal that the relationships of these early osteichthyans are somewhat interdependent , as constraining ‘Ligulalepis’ to the actinopterygian stem also leads to ‘psarolepids’ branching from the sarcopterygian stem , necessitating independant origins of a number of characters in 'psarolepids' and non-osteichthyan gnathostomes ( cf . Lu et al . , 2017 ) , and of tooth enamel in actinopterygians and sarcopterygians . This is because characters that support a stem osteichthyan position for ‘Ligulalepis’ ( i . e . the presence of an eyestalk and short olfactory tracts ) are also found in Psarolepis and Achoania ( Zhu et al . , 2001; Zhu et al . , 2013 ) and only support a stem osteichthyan position if all these taxa are recovered on the stem . Evidence for a stem osteichthyan position for ‘psarolepids’ is now accumulating , with characters such as dorsal fin spines , a median dorsal plate and absence of tooth enamel supporting this relationship ( Qu et al . , 2015; Zhu et al . , 2009; Lu et al . , 2017 ) . This in turn provides additional support for a stem osteichthyan position for ‘Ligulalepis’ . In summary , our current phylogenetic hypothesis is that ‘Ligulalepis’ is a stem osteichthyan . While an actinopterygian affinity requires only one extra step , this position seems to be at odds with the distribution of anatomical features amongst early osteichthyans . The skulls investigated herein are not necessarily disqualified from belonging to the same animal as the scales described for Ligulalepis ( Schultze , 1968 ) . However , we follow Giles et al . , 2015c in maintaining the position that disassociated material cannot be unequivocally attributed to the same taxon . Scale material of Ligulalepis was described as actinopterygian on the basis of an anterodorsal process on the scale , ‘ganoine’ ridges , and a narrow scale peg ( Schultze , 1968; Schultze , 2016 ) . However , the distribution of these characters amongst osteichthyans has subsequently been comprehensively addressed by Friedman and Brazeau ( 2010 ) . An anterodorsal process is primitive for osteichthyans , and as the presence ( and therefore relative width ) of a peg cannot be assessed in outgroups the polarity of this character is ambiguous . While ‘ganoine’ encompasses multiple character states , some of which are general for osteichthyans ( e . g . the presence of enamel , multiple layers of enamel ) the presence of superimposed layers of enamel applied directly to each other is known only in actinopterygians . This indicates that Ligulalepis—that is the scale-based taxon—is an actinopterygian , at odds with the osteichthyan identification of ‘Ligulalepis’—that is the cranium-based taxon . A scale-based Ligulalepis is still problematic , however , as constituent species are erected on the basis of widespread ( and often plesiomorphic characters ) and span from the Ludlow of China ( Wang and Dong , 1989 ) to the Emsian of Australia ( Schultze , 1968 ) . The tooth and jaw fragment attributed to Ligulalepis recently figured by ( Schultze , 2016 , fig . 13 ) presents an additional problem . A vertical thin section through the tooth clearly shows an acrodin tip . Acrodin is a highly mineralized capping tissue restricted to actinopterygians crownward of Cheirolepis ( Friedman and Brazeau , 2010 ) . It is unclear which characters were used to identify this specimen as Ligulalepis , but it most likely does not belong to the same taxon as the skulls investigated herein . Furthermore , this tooth comes from a different fossil site ( Troffs Formation , Trundle Group , Mid-Pragian-Lower Emsian of New South Wales ) than the skulls described in this study . As both the scales and jaw possess actinopterygian characters , it is possible that they belonged to the same taxon . However , in keeping with our protocol of not referring unassociated specimens ( at least in the absence of clear apomorphic characters ) , we hesitate to support a Ligulalepis identity for the jaw specimen .
This study involves the incomplete skull of ‘Ligulalepis’ AM-F101607 , which was previously described ( Basden and Young , 2001; Basden et al . , 2000 ) , and a new specimen , ANU V3628 , discovered by Ben King in late 2015 . Both specimens came from the limestone outcrops on private land ( Cathles' ‘Cooradigbee’ property ) at the southern end of Goodradigbee Inlet , Wee Jasper , New South Wales , Australia . ANU V3628 was found in the Bloomfield Limestone Member of the Taemas Formation near Rocky Flat , and AM-F101607 was probably from a similar horizon , possibly at Caravan Point about 300 m to the north , although precise locality and horizon were not recorded for this specimen ( although most likely from the Emsian pireneae-serotinus condont zone ) . ANU V3628 was found in a large limestone block which was trimmed with an angle grinder . The specimen was then bathed for approximately 2 hr in 5% acetic acid . The exposed bone was embedded in resin , and the block was trimmed further with an angle grinder . The specimen was then given a number of acid baths in 5% acetic acid whilst suspended upside down from a retort stand . After the skull roof became visible , further baths at progressively lower acid concentration were performed with the specimen fully immersed . Later acid baths were buffered using spent acid . Exposed bone was hardened with paraloid at intervals . AM-F101607 was scanned at the Australian National University ( ANU ) High Resolution Micro X-ray Computed Tomography facility ( Sakellariou et al . , 2004 ) with a resultant scan resolution of 30 . 4 microns ( SI:1 ) . ANUV3628 ( SI:2 ) was scanned at Adelaide Microscopy on a Skyscan 1076 . Specimen to source distance was 121 mm , camera to source distance was 161 mm . Source voltage was 100kV , and current 100 µA . 393 projections were taken on a Hamamatsu Orca-HRF camera . The resultant voxel size was 8 . 5 microns . Three-dimensional modeling and segmentation was completed using the software VGStudio Max , version 2 . 2 ( Volume Graphics Inc . , Germany ) , and Mimics 18 . 0 ( Materialise Medical Co , Belgium ) . Drishti version 2 . 6 ( Limaye , 2012 ) and Blender ( blender . org; Stitching Blender Foundation , Amsterdam , the Netherlands ) were also used for presentation purposes . Both CT datasets are available as Supplementary Information . 0 , canal for terminal nerve 0; I , canal for olfactory nerve I; II , canal for optic nerve II; III , canal for oculomotor nerve III; IV , canal for trochlear nerve IV; V , canal for trigeminal nerve V; acv , anterior cerebral vein; ant . amp , ampulla on anterior semicircular canal; ap . f , foramen in anterior pocket; asc , anterior semicircular canal; bpt , basipterygoid process; br . buc . 1 , lateralis nerve branches for the dorsal part of the infraorbital canal; br . prof , canal for branches of the profundus nerve V; br . max , canals for branches of the maxillary nerve in the postnasal wall; bsp , basisphenoid; cc , crus commune; cer , space for cerebellar auricles; com . V . jug , communication between the trigeminal nerve and the jugular canal; It ( Dsph ) , intertemporal bone ( dermosphenotic of actinopterygians ) ; die , space for the diencephalon; epsb , canal for the efferent pseudobranchial artery; esc , external semicircular canal; ext . amp , ampulla on external semicircular canal; eys , area for attachment of eyestalk; f . ica , foramen for entry of internal carotid artery; frla , foramina for ramus lateralis accessorius; g . dend , possible groove for endolymphatic duct; hmf , hyomandibular facet; hyp , space for hypophysis; hyp . v , hypophysial vein; ica , groove for internal carotid artery; ioc , postorbital branch of the infraorbital sensory line; jug . c , canal for jugular vein; lcc ? , possible lateral cranial canal; mcv , canal for middle cerebral vein; mpl , middle pit line; my . IV , myodome for superior oblique eye muscle/dorsal myodome; my . III , myodome for oculomotor-innervated eye muscle; my . VI , myodome for abducens-innervated eye muscle; n . cap , nasal capsule; olf . b , space for olfactory bulb; opha , ophthalmic artery; opt . l , space for optic lobes otc otic section of the infraorbital canal; ot . lat , otic lateralis nerve branches; otc , otic canal; "P" , extension of the main sensory canal beyond infraorbital canal; Par ( Fr ) , parietal ( frontal ) ; pcv , posterior cerebral vein; pdf , posterodorsal fontanelle; pin , pineal canal; pit , pituitary vein; por , postorbital process; PP ( par ) , postparietal ( parietal ) ; ppl , posterior pit line; prof , canal for profundus nerve; psc , posterior semicircular canal; pv , pituitary vein; r . lat , root of the anterior lateralis nerves; s . su , sinus superior; sac , sacculus; soc , supraorbital sensory canal; soph , canal for the superficial ophthalmic nerve; sp . n , spiracular notch; St ( It ) , supratemporal bone ( intertemporal of actinopterygians ) ; Tab ( St ) , tabular bone ( supratemporal of actinopterygians ) ; tel , space for telencephalon; vam , ventral anterior myodome; VIIhm , canal for hyomandibular branch of the facial nerve VII; vm , ventral myodome; Vmd , canal for mandibular trunk of trigeminal nerve V; vom , area for attachment of vomer . The character matrix used was based upon the dataset of Lu et al . for their recent work on Ptyctolepis , which contained 278 characters and 94 taxa ( Lu et al . , 2016a ) . ‘Ligulalepis’ was coded from the two skulls only; scale characters were not included . Based on new information from the scans , the coding for character #31 ( Sensory canals/grooves ) was updated from state 0 ( within thickness of skull bones ) to state 1 ( prominent ridges on visceral surface of skull bones ) . Seven other characters previously unknown in ‘Ligulalepis’ were coded for the first time: #41 , Pineal opening in dermal skull roof ( present ) ; #47 , Number of bones of skull roof lateral to postparietals ( two ) ; #132 , Canal for jugular in postorbital process ( present ) ; #152 , External/horizontal semicircular canal ( joins the vestibular region dorsal to posterior ampulla ) ; #259 , Position of anterior nostril ( facial ) ; #261 , Three large pores associated with each side of ethmoid ( absent ) ; #263 , Size of profundus canal in postnasal wall ( small ) . We clarified the definition of character #115 to refer only to presence or absence of dermal bone separating the nostril and orbit . Previously , the definition of this character simply referred to ‘association’ or ‘confluence’ of the nostril and the orbit , but this is not entirely satisfactory in the case of ‘Ligulalepis’ where the nostril directly enters the orbit , but the dermal bones around the external opening are not completely known . A new character was introduced to reflect the different conditions of the endoskeleton around the posterior nostril . This was character #281 endoskeletal lamina ( postnasal wall ) separating posterior nostril and orbit: 0 ( absent ) ; 1 ( present ) . Another new character was introduced concerning the pituitary vein , following Castiello and Brazeau , 2018 . This was character #282 pituitary vein canal: 0 ( discontinuous , enters endocranial cavity ) ; 1 ( discontinuous , enters hypophysial chamber ) ; 2 ( continuous transverse canal ) . Other minor changes were #240 from one to inapplicable for Cladoselache , Climatius and Cobelodus . State 1 of character #267 ( endoskeletal spiracular canal: partial enclosure or spiracular bar ) was changed to ( spiracular bar ) , to avoid grey areas as to what constitutes ‘partial enclosure’ . Raynerius was recoded as state 0 ( open ) , and Cheirolepis as 0/1 ( open/spiracular bar ) due to uncertainty interpreting the crushed specimen ( Giles et al . , 2015a ) . One character ( trigemino-facial recess present/absent ) was deleted following King et al . , 2017 . One skull roof character ( Lu et al . , 2017 ) character 43: Series of paired median skull roofing bones that meet at the dorsal midline of the skull ) was reformulated into four: #277 , Postparietals/centrals ( 0 absent/1 present ) ; #278 , Condition of postparietals/centrals ( 0 meet in midline/1 do not meet in midline/2 single median bone ) ; #279 , Parietals ( 0 absent/1 present ) , and #280 , Condition of parietals ( 0 meet in midline/1 do not meet in midline ) . The final matrix comprises 282 characters ( see SI 3 ) , scored for the same 94 taxa as Lu et al . , 2017 . Multistate characters were treated as unordered except for numbers 63 , 125 , 164 , 260 , 262 and 266 . Parsimony analysis was performed in TNT v1 . 5 ( Goloboff and Catalano , 2016 ) . Analyses initially used new technology search for 1000 replications , using ratchet , tree fusing , sectorial search and drift search algorithms with default settings . TBR branch swapping was then performed on the resulting trees to explore the tree islands more thoroughly . A total of 1936 trees ( using collapsing rule 1 ) of length 818 were found , and the strict consensus tree was saved . Gnathostomes ( i . e . all taxa except Galeaspida and Osteostraci ) were constrained to be monophyletic , and trees were rooted on Galeaspida . Bremer support values were calculated through a series of tree searches each with a negative constraint on a node in the strict consensus tree . Each of these constrained searches used the same new technology search settings as for the main analysis , for 200 replications . Bootstrap values were calculated using 1000 bootstrap replications . Within each bootstrap replication , the same new technology search settings as above were used , for 100 random addition sequence replications . A list of apomorphies was produced using ACCTRAN for one of the shortest trees using PAUP* ( Swofford , 2003 ) . All scripts for all analyses are included in the supplementary information ( see SI 3 ) . Bayesian analysis was performed in MrBayes 3 . 2 . 6 ( Ronquist et al . , 2012 ) . The same set of characters was ordered . The MkV model ( Lewis , 2001 ) was applied , with a gamma parameter to account for rate variation across characters . Four independent analyses were run ( each with four chains ) for 10 million generations . Convergence of the four runs was confirmed by standard deviation of split frequencies less than 0 . 01 and effective sample size greater than 1000 for all parameters . The following files are available for download from DRYAD ( https://doi . org/10 . 5061/dryad . 41dh5 ) , when using this data please cite the data package in addition to the original publication . Supplementary Information 1: Reconstructed TIFF slices of AM-F101607 . Supplementary Information 2: Reconstructed BMP slices of ANUv3628 . Supplementary Information 3: Folder with all files for phylogenetic analysis . Supplementary Information 4: Folder with Mimics files . The project was conceived by AMC and JAL . AMC , SG , BK and JAL generated the CT renderings . AMC , SG , BK , JAL and BC produced figures . GCY and BK conducted fieldwork . BK prepared and scanned one of the specimens . SG , BK , AMC and JAL conducted the phylogenetic analyses . PEA and JAL both contributed materials to the project . All authors participated in the interpretation of the specimen and writing of the manuscript . | All animals can be classified as either vertebrate ( those that have a spine ) or invertebrate ( those that do not ) . About 98% of all living vertebrate species belong to a group called Osteichthyes , otherwise known as bony fish . Despite the name , this group also includes all four-limbed vertebrates – amphibians , reptiles , birds and mammals – since they evolved from prehistoric bony fish millions of years ago . The oldest known bony fish can be traced back to around 425 million years . These ancient bony fish are all part of a sub-group called lobe-finned fish . Most modern bony fish , however , are part of a different sub-group called ray-finned fish , which can only be confidently traced back about 390 million years . A species called Ligulalepis was once thought to represent the oldest ray-finned fish . Scientists worked this out by examining a single Ligulalepis skull fossil from around 400 million years ago . However , subsequent studies have disputed its position in the evolutionary tree . So , the early evolution of bony fish remains poorly understood . To address this , Clement , King , Giles et al . re-examined the original Ligulalepis skull fossil , alongside a newly discovered second skull fossil of the same species . Modern x-ray scanning techniques were used to produce detailed 3D models of both skulls and compare them to other prehistoric bony fish . This allowed Clement , King , Giles et al . to find Ligulalepis’s exact place in the evolutionary family tree . The experiments identified many previously unknown features of the Ligulalepis skull . These features suggest that this species was not a ray-finned fish; rather , it existed just before bony fish split into two sub-groups ( lobe-finned and ray-finned ) . The analysis also suggests that Ligulalepis was the species most closely related to another group of fish called psarolepids . Overall , these findings clarify our understanding of the evolutionary tree of all vertebrates , including humans . Future research should continue using modern scanning techniques to uncover new information from old fossils and give further insights into the early evolution of vertebrates . | [
"Abstract",
"Introduction",
"Discussion",
"Materials",
"and",
"methods"
] | [
"evolutionary",
"biology"
] | 2018 | Neurocranial anatomy of an enigmatic Early Devonian fish sheds light on early osteichthyan evolution |
Microbial ecosystem modeling is complicated by the large number of unknown parameters and the lack of appropriate calibration tools . Here we present a novel computational framework for modeling microbial ecosystems , which combines genome-based model construction with statistical analysis and calibration to experimental data . Using this framework , we examined the dynamics of a community of Escherichia coli strains that emerged in laboratory evolution experiments , during which an ancestral strain diversified into two coexisting ecotypes . We constructed a microbial community model comprising the ancestral and the evolved strains , which we calibrated using separate monoculture experiments . Simulations reproduced the successional dynamics in the evolution experiments , and pathway activation patterns observed in microarray transcript profiles . Our approach yielded detailed insights into the metabolic processes that drove bacterial diversification , involving acetate cross-feeding and competition for organic carbon and oxygen . Our framework provides a missing link towards a data-driven mechanistic microbial ecology .
In MCM , a microbial community model is a set of differential equations for the population densities of the cell species comprising the community and of the ambient concentrations of utilized nutrients ( metabolites ) , coupled to optimization problems for the cell-specific rates of reactions involving these metabolites . Each cell is characterized by its metabolic potential , that is , the genetically determined subset of reactions it can catalyze , as well as any available metabolite transport mechanisms . The reaction rates and metabolite exchange rates ( i . e . the metabolism ) of each cell are assumed to depend on its metabolic potential as well as on the current environmental conditions , such as metabolite concentrations . Through their metabolism , in turn , cells act as sinks and sources of metabolites in the environment . Additional metabolite fluxes , such as oxygen diffusion from the atmosphere into the growth medium of a modeled bacterial culture , can be included in the model . At any point in time , individual cell metabolism is determined using flux balance analysis ( FBA ) ( Orth et al . , 2010 ) , a widely used framework in cell-metabolic modeling ( Varma and Palsson , 1994; Duarte et al . , 2004; Klitgord and Segrè , 2010; Freilich et al . , 2011; Chiu et al . , 2014 ) . In FBA , cell metabolism is assumed to be regulated in such a way that the rate of biosynthesis is maximized ( Varma and Palsson , 1994; Feist and Palsson , 2010 ) . The chemical state of cells is assumed to be steady , leading to stoichiometric constraints that need to be satisfied for any particular combination of intracellular reaction rates . Reaction rates , on the other hand , are limited due to finite enzyme capacities . Metabolite uptake/export rates can also be limited due to finite diffusion rates or limited transmembrane transporter efficiency . For example , uptake rates can be Monod-like functions of substrate concentrations ( Mahadevan et al . , 2002; Harcombe et al . , 2014 ) . Taken together , cell-metabolic potential , stoichiometric consistency , reaction rate limits and transport rate limits define the constraints of a linear optimization problem for each cell species at each point in time . The optimized biosynthesis rate is translated into a cell production rate by dividing by the cell's mass , thus defining the species' population growth ( Figure 1 ) . 10 . 7554/eLife . 08208 . 003Figure 1 . Framework used by MCM . ( A ) Conceptual framework used by MCM . Cells ( colored shapes ) optimize their metabolism for maximal growth and influence their environment via metabolite exchange ( small colored arrows ) . Additional external fluxes can also affect the environment ( large grey arrows ) . The environment , in turn , influences each cell's metabolism . ( B ) Computational framework used by MCM . Each iteration consists of four steps: flux balance analysis ( FBA ) is used to translate cell-metabolic potentials and environmental conditions ( 1 ) into a linear optimization problem for the growth rate of each cell species ( 2 ) . The set of possible reaction rates corresponds to a polytope in high-dimensional space . Solving the optimization problems ( 3 ) yields predictions on microbial metabolite exchange rates ( 4 ) . Metabolic fluxes and cell growth rates are used to predict metabolite and cell concentrations in the next iteration ( 1 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 08208 . 003 The central assumption of individual cells maximizing biosynthesis , subject to environmental and physiological constraints , is rooted in the idea that evolution has shaped regulatory mechanisms of unicellular organisms in such a way that they strive for maximum growth whenever possible . Biosynthesis has been experimentally verified as an objective for Saccharomyces cerevisiae and E . coli ( Burgard and Maranas , 2003; Gianchandani et al . , 2008; Harcombe et al . , 2013 ) . The assumption of maximized biosynthesis is less valid for genetically engineered organisms or those exposed to environments that are radically different from the environments that shaped their evolution ( Segrè et al . , 2002 ) . Despite its limitations , FBA has greatly contributed to the understanding of several genome-scale metabolic networks and metabolic interactions between cells ( Stolyar et al . , 2007; Klitgord and Segrè , 2010; Orth et al . , 2010; Freilich et al . , 2011; Chiu et al . , 2014; Harcombe et al . , 2014 ) . One advantage of FBA models over full biochemical cell models is their independence of intracellular kinetics and gene regulation , which limits the number of required parameters to stoichiometric coefficients and uptake kinetics . The combination of FBA with a varying environmental metabolite pool , as implemented by MCM , is known as dynamic flux balance analysis ( DFBA ) ( Mahadevan et al . , 2002; Chiu et al . , 2014; Harcombe et al . , 2014 ) . In contrast to conventional FBA , DFBA models are dynamical because cell densities and environmental metabolite concentrations both change with time , and the rate of change of each cell density and metabolite concentration depends on the current cell densities and metabolite concentrations ( Mahadevan et al . , 2002; Harcombe et al . , 2014 ) . Because metabolites can be depleted or produced by several cell species , the environmental metabolite pool mediates the metabolic interactions between cells ( Schink and Stams , 2006 ) . For example , oxygen uptake rates might depend on environmental oxygen concentrations , which in turn are reduced by cellular respiration . Similarly , cells might excrete acetate as a byproduct of glucose catabolism , which then becomes available to other cells . The metabolic optimization of individual cells striving for maximal growth , while modifying their environment , leads to non-trivial community dynamics that can include competition , cooperation and exploitation . The cell-centric nature of DFBA differs fundamentally from other flux balance analyses of microbial communities that assume an optimization of a community-wide objective such as total biomass synthesis ( Stolyar et al . , 2007; Klitgord and Segrè , 2011; Zomorrodi and Maranas , 2012 ) . Such an assumption is at least questionable from an evolutionary perspective and likely not appropriate for communities comprising several species ( Mitri and Foster , 2013 ) . Recent work suggests that DFBA is a promising approach to microbial ecological modeling ( Meadows et al . , 2010; Chiu et al . , 2014; Harcombe et al . , 2014 ) . For example , Harcombe et al . ( 2014 ) designed a computational tool ( COMETS ) based on DFBA , which was able to accurately predict equilibrium compositions of mixed bacterial cultures grown on petri dishes . However , COMETS offers limited model versatility in terms of uptake and reaction kinetics and only has few environmental feedback mechanisms ( namely , varying extracellular metabolite concentrations ) . Furthermore , it assumes complete knowledge of all required model parameters and provides no generic statistical model analysis . Hence , while COMETS sets an important precedent , considerable work is still needed to make DFBA a practical approach in microbial ecosystem modeling . MCM extends Harcombe et al . 's framework to more versatile microbial ecological models that include arbitrary reaction kinetics ( e . g . , subject to product-inhibition ) as well as dynamical environmental variables ( e . g . , pH ) that influence , and are influenced by , microbial metabolism . In addition , MCM supports cell models in which internal molecules act as dynamical constraints that further restrict the FBA solution space , for example to account for post-transcriptional regulation or delays in enzyme synthesis ( Blazier and Papin , 2012 ) . These so called regulatory FBA models have been shown to improve the fidelity of conventional FBA models for E . coli and S . cerevisiae ( Covert et al . , 2001; Covert and Palsson , 2002; Covert et al . , 2004; Herrgård et al . , 2006 ) , however their application to microbial communities remains untested . MCM can statistically evaluate models against data , analyze their sensitivity to varying parameters ( Cariboni et al . , 2007 ) , and estimate the uncertainty of model predictions in the face of stochasticity ( Hammersley and Handscomb , 1964 ) . Perhaps most importantly , MCM can automatically calibrate unknown model parameters to data , for example obtained from monoculture experiments ( as demonstrated below ) , from bioreactor experiments involving multiple species ( Louca and Doebeli , 2015 ) or from environmental samples of unculturable communities ( Figure 2; see the ‘Materials and methods’ and the Supplement for details ) . MCM can thus be used to understand the dynamics of realistic microbial ecosystems , ranging from the soil or groundwater to mixed laboratory cultures and bioreactors . 10 . 7554/eLife . 08208 . 004Figure 2 . Overview of MCM's working principle and functionalities: A microbial community model is specified using human-readable configuration files in terms of metabolites , reactions , the metabolic potential of cell species and any additional environmental variables . Models with multiple ecosystem compartments are also possible . A script with MCM commands controls the analysis of the model and , if needed , its calibration using experimental data . The calibrated model can also be used to create new , more complex models ( as exemplified in this article ) . DOI: http://dx . doi . org/10 . 7554/eLife . 08208 . 004
In a series of laboratory evolution experiments with E . coli ( strain B REL606; Yoon et al . , 2012 ) in glucose-acetate supplemented medium , two metabolically distinct strains consistently evolved from the ancestral ( A ) strain ( Le Gac et al . , 2008; Spencer et al . , 2008; Herron and Doebeli , 2013 ) . When grown in monoculture with the same medium composition , all three strains exhibit diauxic growth curves with a fast glucose-driven growth phase followed by slower growth on acetate . However , the three strains differ in their efficiencies to catabolize glucose and acetate: Strain SS ( slow switcher ) is a better glucose utilizer when compared to strain A , and the depletion of glucose only leads to a slow switch to acetate consumption . On the other hand , the FS ( fast switcher ) strain has evolved to be a better acetate utilizer , initiating acetate consumption at higher remnant glucose concentrations than strains A and SS . This acetate specialization is based on a tradeoff in the citric acid cycle and comes at the cost of being a less competitive glucose consumer . Replicated serial dilution experiments starting with strain A monocultures have shown a consistent phenotypic diversification , involving an initial invasion of the SS phenotype and a subsequent invasion of the FS phenotype , leading to the eventual extinction or near-extinction of the ancestor and the stable coexistence of the SS and FS phenotypes ( Figure 3 ) ( Le Gac et al . , 2008; Spencer et al . , 2008; Tyerman et al . , 2008; Herron and Doebeli , 2013 ) . Genome sequencing revealed that this metabolic diversification can be attributed to point-mutations in genes linked to glucose and acetate uptake kinetics and metabolism ( Herron and Doebeli , 2013 ) . The successional dynamics of the three phenotypes are thus likely driven by adaptations to a changing metabolic niche space , defined by fluctuating glucose , acetate and , potentially , oxygen availabilities ( Le Gac et al . , 2008; Tyerman et al . , 2008; Herron and Doebeli , 2013 ) . An understanding of the underlying ecological processes would shed light on the ecology and evolution of natural microbial communities with shared catabolic pathways . 10 . 7554/eLife . 08208 . 005Figure 3 . Estimated relative cell densities of the A , SS and FS types during three replicated evolution experiments by Herron and Doebeli ( 2013 , Figure 2 ) , starting with the same ancestral E . coli strain . Within each of the three experiments ( A–C ) , the illustrated SS or FS lineage comprises several strains with varyingly pronounced SS or FS phenotypes , respectively . Cell generations were translated to days by assuming an average of 6 . 7 generations per day ( Herron and Doebeli , 2013 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 08208 . 005 To mechanistically explain the observed community dynamics , we used MCM to construct a model comprising the ancestral and the two evolved E . coli types . By keeping track of pathway activation , cell densities , metabolic fluxes and nutrient concentrations , we gained detailed insight into the processes driving the successional dynamics of metabolic diversification . Based on a published cell-metabolic template for the ancestral E . coli strain comprising over 2000 reactions ( Yoon et al . , 2012 ) , we first constructed three separate cell models for the phenotypes A , SS and FS , respectively . In these preliminary models , cells grew on a substrate pool that resembled previous batch-fed monoculture experiments with glucose-acetate supplemented minimal medium ( Le Gac et al . , 2008 ) . Cell-specific oxygen , acetate and glucose uptake rate limits were Monod-like functions of substrate concentrations ( Emerson and Hedges , 2008; Millero , 2013 ) . We calibrated several physiological parameters for each cell type to measured chemical concentration and cell density profiles , using least squares fitting ( Figure 4 ) . MCM automatically calibrates free parameters to data through an optimization algorithm that involves step-wise exploration of parameter space and repeated simulations ( see ‘Materials and methods’ and Supplementary Material ) . 10 . 7554/eLife . 08208 . 006Figure 4 . Calibration of E . coli cell models . Continuous curves: Time course of cell densities , glucose concentration , acetate concentration and oxygen concentration ( columns 1–4 , respectively ) predicted by MCM for monocultures of strain A , SS and FS ( rows 1–3 , respectively ) grown on glucose-acetate medium . Points are data used for model calibration , and were obtained from analogous monoculture growth experiments ( Le Gac et al . , 2008 ) . Oxygen data were not available for strain A . DOI: http://dx . doi . org/10 . 7554/eLife . 08208 . 006 We then constructed the microbial community ( MC ) model by combining the three calibrated cell models into a community growing in a common substrate pool . The environmental context resembles Herron & Doebeli's evolution experiments ( Herron and Doebeli , 2013 ) . In particular , the model includes realistic oxygen depletion-repletion dynamics ( Gupta and Rao , 2003 ) , glucose and acetate depletion by microbial consumption , as well as daily dilutions into fresh glucose-acetate supplemented medium at a factor 1:100 . The microbial community initially consists mostly of type A ( 1010 cells/l ) , while both SS as well as FS cells are assumed to be rare ( 1 cell/l ) . Because the model is deterministic , the invasion or extinction of each type only depends on its growth rate in a possibly changing environment , but not on random mutation events , nor on demographic stochastic fluctuations . Simulations of the MC model reproduced the successional dynamics observed in Herron & Doebeli's experiments: An initial replacement of the ancestor by the SS type is followed by an invasion of the FS type , leading to the eventual coexistence of the SS and FS types and extinction of the ancestral strain ( Figure 5A ) . Interestingly , FS can also invade in the absence of SS , however invasion occurs much slower and FS reaches lower densities than in the presence of SS ( Figure 5—figure supplement 1 ) . This is consistent with an early presence of the FS lineage at low densities in the evolution experiments ( Figure 3 ) , indicating that some of the first FS mutations already confer a slight advantage over the ancestor when FS is rare ( Herron and Doebeli , 2013 ) . 10 . 7554/eLife . 08208 . 007Figure 5 . Dynamics of the E . coli microbial community model . ( A ) Relative cell densities of the A , SS and FS types over time . ( B ) Acetate concentration over time . ( C ) , ( D ) and ( E ) : SS and FS cell densities , relative cell densities and growth rates over time , respectively , during stable coexistence . ( F ) , ( G ) and ( H ) : Cell-specific glucose , acetate and oxygen uptake rates over time , respectively . Negative values correspond to export . ( I ) , ( J ) and ( K ) : Glucose , acetate and oxygen concentrations over time , respectively . Diurnal fluctuations in all figures are due to daily dilutions into fresh medium . Tics on the time axes in ( C–K ) mark points of dilution . DOI: http://dx . doi . org/10 . 7554/eLife . 08208 . 00710 . 7554/eLife . 08208 . 008Figure 5—figure supplement 1 . Predicted relative cell densities of the A and FS types in coculture , in the absence of SS . Initial cell densities were 1010 cells/L for type A and 1 cell/L for type FS . All other model parameters are identical to the microbial community model ( comprising the A , SS and FS types ) described in the main article . DOI: http://dx . doi . org/10 . 7554/eLife . 08208 . 00810 . 7554/eLife . 08208 . 009Figure 5—figure supplement 2 . Robustness of the predicted stable coexistence of the SS and FS types in coculture . Shown are the probability distributions of the relative SS ( A ) and FS ( B ) cell densities over time , when calibrated model parameters ( Table 3 ) are randomly chosen within an interval spanning 10% above and 10% below their fitted values . Initial cell densities were 1010 cells/L for both types , all other parameters were as described in the main article . Probability distributions were estimated using 50 Monte Carlo simulations . In all cases both the SS and FS type persisted . The analysis was performed using the MCM command UAMCM . DOI: http://dx . doi . org/10 . 7554/eLife . 08208 . 009 Time series of acetate concentrations ( Figure 5B ) link the observed successional dynamics of the three types to a gradually changing metabolic niche space: The replacement of type A by the more efficient glucose specialist SS leads to an accumulation of acetate and facilitates the invasion of the FS type . The specialization of the SS and FS types on glucose and acetate , respectively ( Figure 6A ) , enables their long-term coexistence on glucose-acetate enriched medium through frequency dependent competition ( Friesen et al . , 2004; Le Gac et al . , 2008; Herron and Doebeli , 2013 ) . In fact , cell-specific acetate exchange rates reveal that the SS type temporarily excretes acetate during short intervals , which is concurrently and subsequently consumed by the FS type ( Figure 5G ) . This periodic acetate cross-feeding is an evolutionarily emergent property of the microbial community ( Treves et al . , 1998 ) . The temporary production of acetate by the SS type is consistent with previous SS-FS coculture experiments , which revealed slightly increased acetate concentrations towards the end of the SS exponential growth phase ( Spencer et al . , 2007 ) . An evolved increase of acetate excretion by E . coli in glucose minimal medium has also been reported by Harcombe et al . ( 2013 ) . 10 . 7554/eLife . 08208 . 010Figure 6 . Metabolic differentiation of the A , SS and FS types . ( A ) Predicted cell-specific net metabolite uptake rates in coculture . ( B ) Predicted cell-specific reaction rates in coculture , for acs ( acetyl-CoA synthesis ) , ack ( acetate synthesis ) , pta ( acetyl phosphate synthesis ) , ppc ( oxaloacetate synthesis from phosphoenolpyruvate ) , pdh ( decarboxylation of pyruvate to acetyl-CoA ) and pyk ( pyruvate synthesis from phosphoenolpyruvate ) . Rates in ( A ) and ( B ) are averaged over all time points within the first 100 days of evolution , with reversed reactions or net metabolite export represented by negative rates . ( C ) and ( D ) : Simplified model subset of E . coli acetate and glucose metabolism , showing pathway activations in type SS ( C ) and FS ( D ) relative to type A during exponential growth in monoculture . Non-bracketed numeric values are ratios of predicted fluxes in the evolved types over fluxes in type A . Bracketed values are ratios of mRNA harvested from monoculture experiments by Le Gac et al . ( 2008 ) , for comparison . A ratio of 0/0 indicates zero flux in both the evolved and ancestral type , a ratio of 1 corresponds to an unchanged flux or mRNA , a ratio of 0 corresponds to complete deactivation in the evolved type . Darker arrows indicate increased predicted fluxes in the evolved type . Flux predictions correspond to the time points of mRNA measurements , that is , 3 . 5 hr after dilution for SS and 4 hr after dilution for A and SS ( Le Gac et al . , 2008 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 08208 . 010 It should be noted that cell metabolism depends on substrate concentrations and is subject to strong temporal variation . In particular , acetate excretion by SS cells correlates strongly with oxygen limitation ( Figure 5G , K ) . The excretion of acetate by E . coli as a byproduct of oxygen-limited glucose catabolism has been observed experimentally and explained using flux balance analysis ( Mahadevan et al . , 2002 ) . In the absence of oxygen limitation , complete aerobic glucose catabolism to carbon dioxide is preferred over incomplete glucose catabolism with acetate excretion . On the other hand , oxygen limitation leads to an energetic tradeoff between complete glucose catabolism and efficient oxygen utilization , resulting in the excretion of acetate . Furthermore , the depletion of oxygen during cell growth makes oxygen a temporary limiting resource for all cells ( Figure 5K ) . Shortly after dilution into fresh medium , the exponential growth of the SS type on glucose leads to a rapid drop of oxygen to nanomolar concentrations . Despite oxygen diffusion into the medium , oxygen remains at sub-saturation levels for several more hours because the slow-growing acetate-consuming FS cells still consume oxygen after the growth of SS cells has halted . Differences in SS and FS growth rates ( Figure 5C , E ) thus mitigate competition for oxygen through temporal niche separation . Hence , oxygen likely plays an important role in the metabolic diversification , as previously hypothesized by Le Gac et al . ( 2008 ) . This shows that the splitting of metabolic pathways across specialists can be caused by the composite effects of competition for electron donors and electron acceptors . Consistent with differential substrate usage , average cell-specific reaction rates ( Figure 6B ) reveal differences in pathway activation: The transformation of acetate into acetyl-CoA by acetyl-CoA synthetase ( acs ) is decreased in type SS and increased in type FS , when compared to the ancestral type . Furthermore , the conversion of phosphoenolpyruvate to oxaloacetate ( ppc ) , the conversion of phosphoenolpyruvate to pyruvate ( pyk ) and the decarboxylation of pyruvate to acetyl-CoA ( pdh ) , linking the glycolysis pathway to the citric acid cycle , are all upregulated in the SS type when compared to the FS type . Similar differences in pathway activation also exist during early exponential growth in monoculture ( Figure 6C , D ) , because FS grows partly on acetate and SS excretes acetate ( Figure 4F , J ) . Previous microarray profiles of mRNA concentrations during exponential growth in monocultures ( Le Gac et al . , 2008 ) found an upregulation of acetate consumption genes in FS and acetate excretion genes in SS compared to A , qualitatively confirming our predictions ( Figure 6C , D ) . Interestingly , our simulations suggest a significant downregulation of glucose catabolism ( pyk , pdh and ppc ) in FS compared to A , which contradicts the transcript profiles ( Figure 6D ) . However , mRNA was harvested from well-aerated flasks , while the monoculture experiments ( Figure 4 ) and evolution experiments ( Figure 3 ) were performed in test tubes where oxygen can become limiting ( Andersen and von Meyenburg , 1980 ) . Oxygen becomes particularly scarce in the FS tubes ( Figure 4K ) and temporarily limits glucose catabolism , which would explain the strong downregulation not reflected in the transcript profiles ( Le Gac et al . , 2008 ) . Furthermore , while broad pathway activation patterns could be qualitatively reproduced in our system , this might be harder in other cases due to post-transcriptional regulation or post-translational modifications ( Blazier and Papin , 2012 ) . The periodic ( seasonal ) changes in glucose and acetate concentrations in batch culture have previously been shown to promote coexistence of the SS and FS types , in analogy to the maintenance of phytoplankton diversity via fluctuations of resource availability ( Sommer , 1984; Spencer et al . , 2007 ) . Experiments with SS-FS batch cocultures revealed that the SS type quickly dominates over the FS type , when restricted to the first glucose-rich season through frequent dilution into fresh growth medium . Reciprocally , when SS and FS are grown in solution resembling the second glucose-depleted acetate-rich season , the FS type quickly dominates over the SS type ( Spencer et al . , 2007 ) . Accordingly , in a full batch cycle the relative SS cell density has been shown to culminate within 4–8 hr and to gradually decrease afterwards ( Friesen et al . , 2004 , Figure 6B ) , in consistence with our simulations ( Figure 5D ) . Simulations of the SS and FS batch coculture restricted to the first or second season , analogous to Spencer et al . 's experiments , reproduce these observations and verify the role of periodic variation of glucose and acetate concentrations in maintaining the coexistence of both types ( Figure 7 , see the ‘Materials and methods’ for details ) . 10 . 7554/eLife . 08208 . 011Figure 7 . Predicted relative cell densities of the SS and FS types in batch coculture when restricted to either the first glucose-rich ( A ) or second glucose-depleted ( B ) season . In ( A ) , restriction to the first season was achieved by shorter dilution periods which prevented the complete depletion of glucose . In ( B ) , restriction to the second season was achieved by using the glucose-depleted acetate-rich solution , produced by the full-batch coculture , as growth medium ( see the Methods for details ) . See Figure 7—figure supplement 1 for results from analogous experiments by Spencer et al . ( 2007 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 08208 . 01110 . 7554/eLife . 08208 . 012Figure 7—figure supplement 1 . Measured relative cell densities of the SS and FS types in batch coculture , when restricted to either the first glucose-rich ( left column ) or second glucose-depleted ( right column ) season for three independently evolved communities ( rows 1–3 ) , as reported by Spencer et al . ( 2007 , Figures 2A , B therein ) . Restriction to the first season was achieved by shorter dilution periods which prevented the complete depletion of glucose . Restriction to the second season was achieved by using the glucose-depleted acetate-rich solution , produced by the full-batch coculture , as growth medium . Initial population sizes differed between experiments . Strains used by Spencer et al . ( 2007 ) evolved in slightly different growth medium than in this article . Cell generations were translated to days by assuming an average of 6 . 7 generations per day ( Herron and Doebeli , 2013 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 08208 . 012 The models presented here make detailed predictions about the microbial dynamics in the considered experiments . First , after calibration the cell models largely explain the data from the monoculture experiments ( Figure 4 ) . Second , the predictions for pathway activation in the three strains ( Figure 6 ) are roughly consistent with transcription profiles . Third , simulations of the microbial community consisting of all three strains ( Figure 5 ) reproduce the successional dynamics of diversification observed in the evolution experiments ( Figure 3 ) . Fourth , simulations of the SS-FS cocultures restricted to either the glucose-rich or glucose-depleted season reproduce the dominance of the SS or FS type ( Figure 7 ) , respectively , in consistence with previous co-culture experiments . It is important to note that only data from monoculture experiments were used to calibrate the cell models for the three strains ( A , SS and FS ) . In particular , no information from co-culture experiments was used in the setup of the microbial community model , and thus there was no a priori knowledge about what the emergent community dynamics would be . Hence , our work conceptually produced non-trivial predictions that could be compared to experimental observations , although all experiments had already been performed . Our work sheds light on the fundamental problem of metabolic diversification and the emergence of shared catabolic pathways . In particular , our model allowed quantitative predictions for the metabolic fluxes for each strain in coculture , revealing temporary cross-feeding as an emergent property of the evolved community ( Treves et al . , 1998 ) . Cross-feeding , conventionally seen as a beneficial interaction ( Morris et al . , 2013 ) , thus emerged as a form of niche segregation driven by competition for organic carbon and oxygen . Because both evolved types prefer glucose whenever available at high concentrations , but exchange acetate under oxygen limitation , the community constantly switches between competitive and beneficial interactions . Natural microbial populations might thus also oscillate between negative and positive interactions , for example depending on oxygen levels . Our findings also support previous suggestions that microbial evolution can be driven by deterministic ecological processes ( Wood et al . , 2005; Oxman et al . , 2008; Herron and Doebeli , 2013 ) . In this case , the observed diversification is due to competition for limiting resources whose use is constrained by basic metabolic tradeoffs . Other instances of ecological diversification in microbial evolution experiments , for example as reported by Plucain et al . ( 2014 ) , might be explained using a similar approach . More generally , we have demonstrated how MCM can be used to experimentally calibrate and combine genome-based cell models to predict the emergent dynamics of microbial communities . Our framework thus provides a starting point for designing microbial communities with particular metabolic properties , such as optimized catabolic performance . While MCM is designed for genome-based metabolic models , it can also accommodate conventional functional group models . In these models , different ecological functions such as photosynthesis , heterotrophy or nitrification are performed by distinct populations whose metabolic activity is determined , for example , by Michaelis–Menten kinetics and whose growth is described by simple substrate-biomass yield factors ( Hood et al . , 2006; Reed et al . , 2014 ) . Hence , natural microbial communities could be modeled even if annotated genomes are not available for each member species . While functional group models general require fewer parameters , their calibration remains a challenge ( Panikov and Sizova , 1996 ) . In MCM , model calibration becomes analogous to coefficient estimation in conventional multivariate regression , and can be used to estimate poorly known parameters such as stoichiometric coefficients , growth kinetics or extracellular transport coefficients ( MCM user manual , Supplementary file 1 , section 12 ) . To our knowledge , no existing comparable framework offers the flexibility combined with the statistical functionality of MCM . In view of the increasing availability of genome-scale metabolic models ( Feist et al . , 2008 ) , our work provides a missing link to a predictive and synthetic microbial ecology .
MCM is a mathematical and computational framework for the construction , simulation , statistical analysis and calibration of microbial community models ( Figure 2 ) . Models are specified in special files that define all metabolites , reactions , cell species and environmental variables . MCM is controlled through custom scripts , that is , text files containing a sequence of special commands , such as for running simulations or fitting parameters . MCM includes tools for the conversion of conventional genome-scale FBA models , such as generated by the Model SEED pipeline ( Henry et al . , 2010 ) based on sequenced genomes , into a draft MC model . MCM can accommodate microbial communities comprising genome-based cell models with arbitrary environmental variables , metabolite exchange kinetics and regulatory mechanisms . For example , environmental variables may be stochastic processes ( e . g . , representing climate ) , or specified using measured data ( e . g . , redox potential in bioreactor experiments ) , or depend on metabolite concentrations ( e . g . , pH determined by acetate concentration ) or even be dynamical ( e . g . , temperature increasing at a rate proportional to biomass production rates ) . This versatility allows for the incorporation of complex environmental feedbacks , such as host immune responses in gut microbiota ( Karlsson et al . , 2011 ) . Metabolite uptake and export rate limits can be arbitrary functions of metabolite concentrations or environmental variables . Similar interdependencies are possible for reaction rate limits , thus allowing the inclusion of inhibitory or regulatory mechanisms ( Covert et al . , 2008 ) . Metabolite concentrations can be explicitly specified , for example , using measured time series , or depend dynamically on microbial export and other external fluxes . Effects of phage predation ( Jensen et al . , 2006 ) , reaction energetics ( Reed et al . , 2014 ) or stochastic environments can also be incorporated . MCM keeps track of a multitude of output variables such as cell densities , reaction rates , metabolite concentrations and metabolite exchange rates . Because each reaction can be formally associated with a particular enzyme , in turn encoded by a particular gene , MCM also makes predictions about gene densities as a product of cell densities and gene copy numbers per cell . Metabolic activity statistics ( e . g . , Figure 6A , B ) facilitate the identification of metabolic interactions such as cross-feeding ( Morris et al . , 2013 ) . The predicted time courses of output variables can be statistically evaluated against time series ranging from chemical concentrations , rate measurements to cell densities and metagenomics . MC models can include arbitrary abstract ( symbolic ) numeric parameters with a predefined value range or probability distribution . Symbolic parameters can represent , for example , stoichiometric coefficients , gene copy numbers , cell life expectancies , half-saturation constants or environmental variables . The inclusion of symbolic parameters enables a high-level analysis of microbial communities: For example , MCM can automatically calibrate ( fit ) unknown symbolic parameters to time series using maximum–likelihood parameter estimation ( Eliason , 1993 ) . The likelihood of the data , given a particular parameter choice , is calculated by assuming a mixed deterministic-stochastic model in which the deterministic part is given by the model predictions , and the stochastic part is given by normally distributed errors . The likelihood is minimized using an iterative optimization algorithm involving step-wise parameter adjustments and repeated simulations . Other fitting algorithms are also available , such as maximization of the average coefficient of determination ( R2 ) , which is equivalent to weighted least-squares fitting . Because MCM can calibrate unknown measurement units , raw uncalibrated data ( e . g . , optical cell densities with no calibration to colony forming units , Figure 4A ) can also be used . In this paper single-cell models were calibrated to monoculture experiments , however models can also be calibrated using data from experimental or natural communities that include unculturable species ( MCM user manual , Supplementary file 1 , sections 7 and 12; Louca and Doebeli , 2015 ) . In general , fitted parameters need not be directly connected to the data used for calibration , as long as a change in the parameters influences the predictions that are being compared to the data . While this is a general principle of parameter estimation ( Tarantola , 2005 ) , in practice the uncertainty of calibrated parameters ( e . g . , in terms of confidence intervals ) increases when their influence on the ‘goodness of fit’ is weaker . Moreover , alternative parameter combinations can sometimes yield a comparable match to the data , especially if multiple parameters influence the same variables ( inverse problem degeneracy ) . Local fitting optima can be detected through repeated randomly seeded calibrations ( see next section ) , and overfitting can be partially avoided by keeping the number of free parameters at a bare minimum . Nevertheless , in certain cases good knowledge of the system or previous literature may be required to identify the most plausible calibrations . Finally , we emphasize that MCM is , after all , merely a framework enabling the construction , calibration and analysis of microbial community models . MCM models are thus limited by the same caveats and assumptions as other constraint-based metabolic models ( Blazier and Papin , 2012; Antoniewicz , 2013 ) and any predictions made by MCM should be subject to similar scrutiny . E . coli strains were obtained from an evolution experiment performed in a batch culture environment with daily dilutions into glucose-acetate supplemented Davis minimal medium ( Spencer et al . , 2008; Tyerman et al . , 2008 ) . For each phenotype , three clones were isolated from population 20 after 150 days and used for three independent monoculture growth experiments . Optical densities , as well as glucose , acetate and oxygen concentration data from these experiments were used to calibrate the individual cell-metabolic models for the A , SS and FS phenotypes . Oxygen measurements were not available for type A . Experimental details and results are described by Le Gac et al . ( 2008 ) . In the models , the limiting nutrients are assumed to be oxygen , glucose and acetate; all other nutrients can be taken up at an arbitrary rate . Oxygen , glucose and acetate uptake rate limits were described by Monod-like kinetics . The maximum cell-specific oxygen uptake rate was set to 1 . 008×10−13 mol/ ( d⋅cell ) , according to Varma and Palsson ( 1994 ) . The oxygen half-saturation constant was set to 1 . 21×10−7 M according to Stolper et al . ( 2010 ) . Oxygen was assumed to be initially at atmospheric saturation levels ( 0 . 217 mM at 37∘ C ) and repleted at a rate proportional to its deviation from saturation ( Gupta and Rao , 2003 ) . The fitted parameters for each cell type were the maximum cell-specific uptake rates and half-saturation constants for glucose and acetate , as well as initial cell densities and non-growth associated ATP maintenance energy requirements . The initial glucose and acetate concentrations were set to the average value measured at the earliest sampling point ( 1 hr after incubation ) for each type . The oxygen mass transfer coefficient ( M/day per M deviation ) was initially fitted individually for each type together with all other parameters , and then fixed to the average of all three initial fits . All other parameters were then again fitted individually for each type . Parameter fitting was done by maximizing the average coefficient of determination ( R2 ) using the MCM command fitMCM . A total of 237 data points were used to fit 19 parameters ( Table in Supplementary file 2 ) . To reduce the possibility of only reaching a local maximum , fitting was repeated 100 times for each strain starting at random initial parameter values and the best fit among all 100 runs was used . While some fitting runs reached alternative local maxima , the best overall fit was reached in most cases . Cell densities were directly compared to optical density ( OD ) measurements . The appropriate calibrations were estimated by MCM and ranged within 8 . 2×1011 −1 . 3×1012 cells/ ( L⋅OD ) . These estimates are consistent with previous experimental calibrations ( Lawrence and Maier , 1977 ) yielding 0 . 26 g dry weight/ ( L⋅OD ) , which corresponds to 1 . 4×1012 cells/ ( L⋅OD ) ( assuming a cell dry weight of 1 . 8×10−13 g in the stationary phase; Fagerbakke et al . , 1996 ) . The microbial community model was simulated using the MCM command runMCM . Initial glucose and acetate concentrations were set to the average of all values measured at the earliest sampling point of the monoculture incubations . Cell death was not explicitly included , because of lack of appropriate data for calibration and because daily dilutions by far exceeded cell death as a factor of cell population reduction . To verify the robustness of the stable SS-FS coexistence in coculture , we randomly varied each fitted model parameter uniformly within an interval spanning 10% above and 10% below its calibrated value . Both types coexisted in 50 out of 50 random simulations ( Figure 5—figure supplement 2 ) . Simulations of the SS-FS cocultures restricted to the first glucose-rich or second glucose-depleted season , as opposed to the full batch cycle , were performed in analogy to the experiments by Spencer et al . ( 2007 ) . More precisely , to model the first season experiment we changed the dilution rate to 1/32 every 5 hr , so that at the end of each batch cycle glucose was not yet completely depleted . Similarly , for the second season experiment we changed the dilution rate to 1/32 every 19 hr , and adjusted the growth medium to resemble the glucose-depleted acetate-rich solution reported by Spencer et al . ( no glucose , 3 . 59 mM acetate ) . Initial cell densities were set to 1010 cells/l for both types . All other model parameters were kept unchanged . The original experiments by Spencer et al . ( 2007 ) were performed at higher dilution rates ( 4 and 15 hr for the first and second season experiment , respectively ) , however in our simulations neither the FS nor SS type could persist at these high dilution rates . We note that the strains used in our work ( Le Gac et al . , 2008 ) had evolved in separate evolution experiments using a different growth medium than those by Spencer et al . ( 2007 ) . MCM is open source and available at http://www . zoology . ubc . ca/MCM . | Microbes like bacteria and yeast play important roles in the environment , human health and even some industrial processes . However , it is difficult to understand the roles of microbes in these situations because many different types of microbes often live together in complex communities . Some of the microbes may compete with each other for resources like oxygen or sugar . Others may rely on one another for survival . For example , one microbe may feed on molecules that are released as waste from another microbe . To better understand these microbial communities , we first need to understand the processes by which each microbe uses nutrients and releases waste molecules that influence other microbes . Researchers have used a technique called ‘genome sequencing’ to reconstruct the networks of genes and chemical reactions that are involved in these processes , and to build computer models of microbial communities in different environments . However , the existing models can be labor intensive and do not allow researchers to easily use statistics to analyse them . To address this problem , Louca and Doebeli created a new computer model with built-in statistical tools that accurately predicts the interactions in communities that contain multiple strains of a bacterium called Escherichia coli . First , Louca and Doebeli grew a single strain of E . coli in the laboratory for many generations , which led to the evolution of the bacteria so that two new strains emerged . One of the new strains was more efficient at using sugar as a food source than the other and sometimes released a molecule called acetate . The other new strain became more efficient at using this acetate . Next , Louca and Doebeli used data that had been collected for each individual strain , to test whether the model could recreate the way that the new strains had evolved together . The model accurately predicted that the two new strains would gradually replace the original strain . The strain that was more efficient at using sugar emerged first , which led to extra acetate being available for the other new strain that became more efficient at using acetate . Louca and Doebeli's findings demonstrate for the first time that data collected for individual microbes can be used to explain the dynamics and evolution of small communities of microbes using mathematical models . The next step is to test this approach on larger communities in the environment . | [
"Abstract",
"Introduction",
"Results",
"and",
"discussion",
"Materials",
"and",
"methods"
] | [
"computational",
"and",
"systems",
"biology",
"tools",
"and",
"resources"
] | 2015 | Calibration and analysis of genome-based models for microbial ecology |
We present a novel mass spectrometry-based high-throughput workflow and an open-source computational and data resource to reproducibly identify and quantify HLA-associated peptides . Collectively , the resources support the generation of HLA allele-specific peptide assay libraries consisting of consensus fragment ion spectra , and the analysis of quantitative digital maps of HLA peptidomes generated from a range of biological sources by SWATH mass spectrometry ( MS ) . This study represents the first community-based effort to develop a robust platform for the reproducible and quantitative measurement of the entire repertoire of peptides presented by HLA molecules , an essential step towards the design of efficient immunotherapies .
Next-generation immune-based therapies are expected to facilitate the eradication of intractable pathogens , cancer and autoimmune diseases ( Koff et al . , 2013 ) . T cells play a critical role in such therapies by their ability to detect the presence of disease-specific antigens/peptides presented by major histocompatibility complex ( MHC ) molecules ( human leukocyte antigen [HLA] molecules in humans ) . Under steady-state or pathological conditions , thousands of HLA class I-associated peptides of 8–12 amino acids in length are displayed on the surface of virtually all nucleated cells for scrutiny by CD8+ T cells . HLA class II-associated peptides are 10–25 amino acids in length and are normally found on the surface of specialized antigen-presenting cells including macrophages and dendritic cells for presentation to CD4+ T cells . Collectively , HLA class I and class II peptides are referred to as the immunopeptidome , also known as HLA ligandome/peptidome ( Caron et al . , 2011; Kowalewski et al . , 2014 ) . The composition of the immunopeptidome in the human population is complicated by the presence of more than 3000 HLA alleles , resulting in a high diversity of peptide repertoires characterized by the presence of HLA allele-specific binding motifs ( Falk et al . , 1991 ) . To be successful in designing efficient immunotherapies against autoimmunity , cancer and infectious diseases , it is becoming increasingly important to comprehensively map the complexity of the human immunopeptidome and to gain a more quantitative understanding of its dynamics in various disease states . Mass spectrometry ( MS ) has evolved as the method of choice for the exploration of the human immunopeptidome ( Hunt et al . , 1992; Admon and Bassani-Sternberg , 2011; Granados et al . , 2015 ) . The largest HLA peptidomes reported to date using MS contain more than 10 , 000 class I or class II peptides ( Hassan et al . , 2013; Bergseng et al . , 2014; Bassani-Sternberg et al . , 2015 ) . Estimates from various analytical and cell-based techniques also indicate that individual peptides are expressed on average at 50 copies per cell with extremes ranging from 1 to 10 , 000 copies per cell ( Granados et al . , 2015 ) . Until recently , the most common strategy for the analysis of immunopeptidomes by MS has focused on the isolation of HLA-bound peptides by immunoaffinity chromatography and the collection of fragment ion spectra of selected peptides through automated MS operated in data-dependent acquisition ( DDA ) mode . Although DDA is a powerful strategy for exploring the peptidomic content of various cell and tissue types , it is not a reliable platform for solving problems that require the comparison of comprehensive , quantitative , and reproducible data sets across many samples or conditions . In fact , analyses of complex/unfractionated digests of cell lysate using DDA have shown that as many as 84% of peptides may remain unselected for fragmentation even though they are clearly detectable by the mass spectrometer ( Michalski et al . , 2011 ) . Although the complexity of isolated HLA peptides is hardly comparable with that of cell lysate digests , as many as 20% of the selected HLA peptides can vary between replicate analyses of the same sample ( Granados et al . , 2014 ) ( Figure 1—figure supplement 1A ) . A second strategy , referred to as selected/multiple reaction monitoring ( S/MRM ) , is a targeting MS technique capable of generating highly reproducible , quantitatively accurate and sensitive datasets ( Picotti and Aebersold , 2012 ) . S/MRM is , however , limited by its capacity to detect only tens to hundreds of peptides per sample injection and thus is not ideally suited to comprehensively quantify HLA peptidomes . To overcome this limitation , we recently introduced SWATH-MS , a new mass spectrometric technique that combines data-independent acquisition ( DIA ) with a targeted data extraction strategy ( Gillet et al . , 2012; Röst et al . , 2014 ) . In DIA mode , all peptides in a sample are fragmented and the corresponding fragment ion spectra are acquired , resulting in a digital recording of the peptide sample . DIA is an unbiased MS technique and therefore represents a suitable strategy for efficiently generating consistent , reproducible and quantitatively accurate measurements of peptides across multiple samples ( Gillet et al . , 2012; Collins et al . , 2013; Rosenberger et al . , 2014; Röst et al . , 2014; Guo et al . , 2015; Liu et al . , 2015; Schubert et al . , 2015a ) . To extract quantitative information from digital SWATH-MS data , high-quality assay libraries are required . Such libraries contain retention-time and fragmentation information of the peptides to be targeted . Assay libraries are generated from native and/or synthetic peptides using a SWATH compatible mass spectrometer operated in DDA mode . To date , several generic SWATH assay libraries were generated for the analysis of proteomes in various species . These include Mycobacterium tuberculosis ( Schubert et al . , 2015a ) , Saccharomyces cerevisiae ( Selevsek et al . , 2015 ) , and Homo sapiens ( Rosenberger et al . , 2014 ) . Assay libraries were successfully employed to measure a limited number of MHC class I peptides by S/MRM in various contexts—that is , viral infection ( Croft et al . , 2013 ) , autoimmunity ( Schittenhelm et al . , 2014a ) and cancer ( Gubin et al . , 2014 ) —but have never been created for robust quantitative and high-throughput measurement of HLA-associated peptides by SWATH-MS . For the SWATH-MS technology to meet its potential to support rapid advances in the design of next-generation vaccines and immunotherapies , comprehensive HLA peptide assay libraries have to be created and made readily available to basic and translational scientists . Generating such assay libraries could ultimately enable the fast and reproducible quantification of the entire repertoire of HLA peptides across many samples . Towards this end , we developed a workflow to ( 1 ) generate a pilot repository of HLA allele-specific peptide spectral and assay libraries , and to ( 2 ) analyze SWATH-MS HLA peptidomic data acquired from multiple international laboratories ( Figure 1 ) . In this study , libraries were created from natural and/or synthetic HLA class I and II peptides whereas analysis of SWATH-MS HLA peptidomic data focused mainly on naturally presented class I peptides . 10 . 7554/eLife . 07661 . 003Figure 1 . General workflow for building HLA allele-specific peptide assay libraries and for analyzing SWATH-MS HLA peptidomic data . ( Left panel ) A community-based repository of HLA class I allele-specific peptide spectral and assay libraries was created and stored in the SWATHAtlas database . HLA typed-biological samples and synthetic HLA peptides were used to build the repository . Our workflow integrates ( 1 ) data-dependent acquisition ( DDA ) of HLA peptidomic data , ( 2 ) multiple open-source database search engines and statistical validation tools , ( 3 ) HLA allele annotation of the identified peptides , and ( 4 ) spectral and assay library generation tools . ( Right panel ) HLA peptidomic data from HLA-typed biological samples were acquired in data-independent acquisition ( DIA ) mode . The matching HLA class I allele-specific peptide assay libraries were combined and DIA data were analyzed using the OpenSWATH and the Skyline software . DOI: http://dx . doi . org/10 . 7554/eLife . 07661 . 00310 . 7554/eLife . 07661 . 004Figure 1—source data 1 . Comparative analysis of DDA and SWATH-MS for the identification of HLA class I peptides . DOI: http://dx . doi . org/10 . 7554/eLife . 07661 . 00410 . 7554/eLife . 07661 . 005Figure 1—figure supplement 1 . Reproducibility of DDA and SWATH-MS for the identification of HLA class I peptides . HLA class I peptides were isolated from JYEBV+ cells . Six technical replicates were consecutively injected in a TripleTOF 5600 MS . The Venn diagrams indicate the number of peptides identified in each analysis and the number of peptides shared between the runs . ( A ) Three datasets were acquired in DDA mode and the peptides were identified using the open source database search engines ( 1% peptide-level FDR ) . ( B ) Three datasets were acquired in SWATH mode and the peptides were identified using OpenSWATH and a combined HLA-A and -B peptide assay library ( 1% peptide-level FDR ) . DOI: http://dx . doi . org/10 . 7554/eLife . 07661 . 00510 . 7554/eLife . 07661 . 006Figure 1—figure supplement 2 . Combining results of three open-source database search engines in immunopeptidomics using iProphet . ( A ) The HLA peptidome of fourteen PBMC samples were analyzed . Venn diagrams show the search results obtained from three database search engines ( i . e . , Comet , MS-GF+ and X ! Tandem ) at 5% peptide-level FDR . The search identifications were combined and statistically scored using PeptideProphet and iProphet within the Trans-Proteomic Pipeline ( TPP ) . Following annotation of all identified peptides to their respective HLA allele , all nonannotated peptides were removed from the iProphet combined search result and a corrected false discovery rate ( cFDR ) was manually calculated based on the target-decoy approach . cFDR is indicated for each PBMC sample . At peptide-level FDR 1% , the cFDR was estimated on average at 0 . 5% . At peptide-level FDR 5% , the cFDR was estimated at 2 . 5% . ( B ) The table shows the number of HLA class I peptides identified from the iProphet combined search results that were used to build the spectral libraries . The sum of peptides identified by the three search engines ( Union ) as well as the number of overlapping peptides ( Intersection ) for each venn diagram/sample is also indicated . DOI: http://dx . doi . org/10 . 7554/eLife . 07661 . 00610 . 7554/eLife . 07661 . 007Figure 1—figure supplement 3 . Combining both open-source and commercial database search engines in immunopeptidomics . Analysis of PBMC#2 is shown here as an example . ( A ) Comparison of search results obtained from multiple search engines and for different class I HLA alleles at 1% and 5% peptide-level FDR . Performance of two commercial search engines ( Mascot+Percolator or Mascot alone , and PEAKS ) is also shown here for comparison . ( B ) Venn diagram showing the performance of the search engines at 5% pep-level FDR ( 2 . 5% cFDR ) . DOI: http://dx . doi . org/10 . 7554/eLife . 07661 . 007
Large-scale DDA-based identification of immunoaffinity purified HLA class I peptides is supported by several software tools ( e . g . , MaxQuant , Perseus or X-PRESIDENT ) and results in thousands of unclassified peptides of various lengths . Since large HLA peptidomic datasets are generated at an increasing pace , additional computational frameworks facilitating the HLA annotation and storage of such datasets need to be developed . Here , we first created a computational workflow to support the identification , classification/annotation , visualization and storage of HLA peptidomic data in an allele-dependent manner . The software tools described in the section below enable ( 1 ) systematic annotation of peptides to their respective HLA allele , ( 2 ) visualization of HLA peptidomic datasets , and ( 3 ) generation of HLA class I allele-specific peptide spectral libraries , which can be converted into high quality assay libraries for the processing of SWATH-data ( Figure 2 , Figure 2—figure supplement 1 , Figure 2—source data 2 and Supplementary file 1 ) . 10 . 7554/eLife . 07661 . 008Figure 2 . Content and analysis of the pilot repository . ( A ) HLA peptides were isolated by immunoaffinity chromatography and were annotated to their respective HLA alleles following DDA mass spectrometry . ( B ) Heat map visualization of HLA class I peptides identified from 20 HLA-typed biological samples . HLA-A and -B alleles are indicated for each sample . ( C ) 35 , 812 distinct class I and class II HLA peptides were identified , annotated , and used to build 32 and 11 HLA allele-specific peptide spectral and SWATH assay libraries , respectively . ( D ) The distribution curve shows that 95% of the HLA-B07-annotated peptides were predicted to bind the HLA molecule with an IC50 below 531 nM . Inner pie chart: we assessed the predicted HLA binding affinity of all peptides contained in individual source proteins . The pie chart shows that 92% of naturally presented HLA-B07 peptides were ranked in the top 1% ( blue ) of predicted peptides ( see also Figure 2—figure supplement 6 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 07661 . 00810 . 7554/eLife . 07661 . 009Figure 2—source data 1 . Sources of HLA peptides used in this study . DOI: http://dx . doi . org/10 . 7554/eLife . 07661 . 00910 . 7554/eLife . 07661 . 010Figure 2—source data 2 . Annotation of HLA peptides . DOI: http://dx . doi . org/10 . 7554/eLife . 07661 . 01010 . 7554/eLife . 07661 . 011Figure 2—source data 3 . List of eluted HLA class I peptides that were identified at 1% and 5% peptide-level FDR . DOI: http://dx . doi . org/10 . 7554/eLife . 07661 . 01110 . 7554/eLife . 07661 . 012Figure 2—source data 4 . HLA class I allele-specific peptide spectral libraries stored in PeptideAtlas . DOI: http://dx . doi . org/10 . 7554/eLife . 07661 . 01210 . 7554/eLife . 07661 . 013Figure 2—source data 5 . HLA class I and II allele-specific peptide assay libraries stored in the SWATHAtlas database . DOI: http://dx . doi . org/10 . 7554/eLife . 07661 . 01310 . 7554/eLife . 07661 . 014Figure 2—figure supplement 1 . Automated NetMHC-based method for annotating and visualizing HLA allele-specific peptides . PBMC#2 was typed positive for HLA-A02 , -A03 , -B35 , -B39 , and is shown here as a representative sample . ( A ) The stand-alone software package of the HLA binding prediction algorithm NetMHC 3 . 4 was used to predict the binding affinity of all identified peptides to HLA-A02 , -A03 , -B35 and -B39 ( four peptides are shown for simplicity ) . For each peptide , an annotation score was calculated by dividing the second lowest IC50 value ( second best predicted allele ) by the lowest IC50 value ( best predicted allele ) . Peptides with a score ≥3 were annotated to the HLA allele predicted to bind best . Peptides with a score below 3 were considered as non-annotated . Non-annotated peptides were curated in the output files in Figure 2—source data 2 and correspond to 1 ) non-HLA peptides/contaminants , 2 ) peptides predicted to strongly bind more than one HLA allele ( supertype peptides ) , 3 ) peptides predicted to bind HLA-C alleles , 4 ) exceptional HLA peptides with no known binding motifs . Annotation scores of all eluted peptides are shown in Figure 2—source data 2 . Additional information is provided in Supplementary file 1 . ( B ) Curves showing the distribution of the predicted HLA binding affinities for all HLA-A03-annotated peptides with a score ≥3 . Overall , 91% of all HLA-A03-annotated peptides are predicted to have a binding affinity below 500 nM for the HLA-A03 molecule ( see also Figure 2—figure supplement 4 and Figure 2—figure supplement 5 ) . The same peptides are predicted to be non-binders for the other alleles – i . e . , HLA-A02 , -B35 and -B39 . ( C ) Heat map visualization following clustering of predicted HLA binding affinity values . The white box highlights HLA-A03-annotated peptides . The four peptides in the table in ( a ) are indicated by arrows and their respective predicted binding affinity for the HLA-A03 molecule is indicated in parenthesis . DOI: http://dx . doi . org/10 . 7554/eLife . 07661 . 01410 . 7554/eLife . 07661 . 015Figure 2—figure supplement 2 . Identification of HLA class I allele-specific peptides by DDA . ( A ) Number of distinct HLA class I allele-specific peptides identified using an Orbitrap-XL and a 5600 TripleTOF at peptide-level FDR 5% . ( B ) Logo showing the profile motif for peptides presented by different HLA-A and -B alleles . Profile motifs were created by using all annotated HLA class I peptides in this study and the sequence logo generator WebLogo . DOI: http://dx . doi . org/10 . 7554/eLife . 07661 . 01510 . 7554/eLife . 07661 . 016Figure 2—figure supplement 3 . Generation of assay libraries from a large collection of synthetic HLA class II peptides . ( A ) Workflow to generate an assay library from synthetic peptides . A total of 20 , 176 predicted peptides ( with a range of 2 to 10 per ORF , and an average of 5 ) , were synthesized and arranged into 23 peptide pools of ~900 peptides ( Lindestam Arlehamn et al . , PLoS Pathog , 2013 ) . Spiked-in reference iRT peptides were used and the pools of synthetic peptides were analyzed in DDA mode using a 5600 Triple-TOF and an Orbitrap ELITE ( CID and HCD fragmentation ) . The identified peptides were then processed through our computational pipeline to generate the assay library . ( B ) Venn diagram showing the overlap between peptides identified by the 5600 Triple-TOF and by the ELITE ( CID and HCD fragmentation methods ) . Number of peptides identified is indicated in parenthesis . ( C ) Histogram showing the distribution of the precursor charge state . DOI: http://dx . doi . org/10 . 7554/eLife . 07661 . 01610 . 7554/eLife . 07661 . 017Figure 2—figure supplement 4 . Distribution curves of peptide binding affinities for different HLA-A and -B alleles ( 1% peptide-level FDR; 0 . 5% cFDR ) . The predicted IC50 values of the annotated peptides in Figure 2—source data 3 were used to generate the distribution curves ( blue line ) . The proportion of peptides with a predicted affinity lower than the established 500nM threshold ( grey ) is indicated for individual HLA alleles . The plots also indicate that 95% of the annotated peptides ( green ) are predicted to bind their respective HLA molecules with an IC50 ranging from 72 nM ( HLA-A01 ) to 5682 nM ( HLA-B51 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 07661 . 01710 . 7554/eLife . 07661 . 018Figure 2—figure supplement 5 . Distribution curves of peptide binding affinities for different HLA-A and -B alleles ( 5% peptide-level FDR; 2 . 5% cFDR ) . The predicted IC50 values of the annotated peptides in Figure 2—source data 3 were used to generate the distribution curves ( blue line ) . The proportion of peptides with a predicted affinity lower than the established 500nM threshold ( grey ) is indicated for individual HLA alleles . The plots also indicate that 95% of the annotated peptides ( green ) are predicted to bind their respective HLA molecules with an IC50 ranging from 388 nM ( HLA-A01 ) to 5761 nM ( HLA-B51 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 07661 . 01810 . 7554/eLife . 07661 . 019Figure 2—figure supplement 6 . Binding scores of naturally presented HLA-A and -B peptides contained in individual source proteins . We assessed the predicted HLA binding affinity of all peptides contained in individual source proteins . The pie chart shows the proportion of naturally presented peptides isolated by immunoaffinity chromatography that ranked in the top 1% ( blue ) , top 5% ( red ) , top 10% ( yellow ) , or below the 90th percentile of peptides ( pale blue ) . DOI: http://dx . doi . org/10 . 7554/eLife . 07661 . 019 To test our workflow , the generated data and computational resources , we first assessed the feasibility of generating HLA class I allele-specific peptide spectral libraries from a panel of fourteen PBMC samples ( PBMC #1–14 ) expressing different combinations of HLA class I alleles . HLA class I-bound peptides were isolated from HLA-typed PBMC's by immunoaffinity chromatography and analyzed by DDA on an Orbitrap-XL mass spectrometer ( Figure 2 and Figure 2—source data 1 ) . Peptides were identified using multiple open-source database search engines . The search identifications were combined and statistically scored using PeptideProphet and iProphet within the Trans-Proteomic Pipeline ( TPP ) as shown previously ( Figure 1 ) ( Shteynberg et al . , 2011 , 2013 ) . We next annotated the identified peptides to their respective HLA allele . Previously , HLA binding prediction algorithms such as SYFPETHI , NetMHC and SMM were used for manual or semi-automated annotation of HLA peptides ( Fortier et al . , 2008; Berlin et al . , 2014; Granados et al . , 2014 ) . Here , we designed a fully automated annotation strategy integrating the stand-alone software package of the HLA binding prediction algorithm NetMHC 3 . 4 with a set of in-house software tools ( Figure 2—figure supplement 1 ) . The in-house software tools enable an automated , consistent and effective annotation of the majority of the identified peptides to their respective HLA allele ( Supplementary file 1 ) . Briefly , each identified peptide was given a predicted HLA binding affinity ( IC50 ) for each of the HLA alleles expressed in the corresponding healthy donor . An HLA annotation score was then computed for each individual peptide by dividing its second best IC50 value ( i . e . , the second best predicted allele ) by its best IC50 value ( i . e . , the best predicted allele ) . The higher this annotation score was , the higher the probability was for the peptide to be correctly annotated to a specific HLA allele . As an example , in PMBC#2 , an annotation score of 77 was computed for the KLEEQARAK peptide by dividing 21 , 400 nM ( second best IC50 value predicted for HLA-B39 ) by 278 nM ( best IC50 value predicted for HLA-A03 ) ( Figure 2—figure supplement 1A ) . Peptides with an HLA annotation score ≥3 ( selected cutoff value; see ‘Materials and methods’ and Supplementary file 1 ) were systematically annotated to the allele predicted to bind best ( e . g . , HLA-A03 for the KLEEQARAK peptide ) . Using this scoring strategy , ∼80% of all identified 8–12-mers were annotated to a specific HLA-A or -B allele ( Figure 2—source data 2 ) . HLA-A and -B alleles were prioritized due to the high reliability of the NetMHC 3 . 4 predictor for a broad diversity of HLA-A and -B alleles as well as for their high expression levels ( Kim et al . , 2014; Bassani-Sternberg et al . , 2015; Trolle et al . , 2015 ) . Peptides with an annotation score below 3 were considered as non-annotated in this study and were discarded for the process of building the HLA allele-specific peptide spectral libraries . Tables including scored peptides were then used to generate heat maps and visualize HLA-A and -B peptidomes of PBMC's as described ( Figure 2B and Supplementary file 1 ) . Of note , allele-supertype peptides ( i . e . , peptides predicted to strongly bind more than one allele with an IC50 below 500 nM ) were curated in the output files but were not visualized on the heat maps in this study . A corrected false discovery rate ( cFDR ) was estimated for each PBMC sample following removal of all non-annotated contaminant peptides ( Figure 1—figure supplement 2 and Figure 1—figure supplement 3 ) , resulting in a total of 4153 ( peptide-level FDR 1%; average cFDR 0 . 5% ) or 7921 ( peptide-level FDR 5%; average cFDR 2 . 5% ) distinct annotated peptides distributed across eighteen HLA class I alleles ( Figure 2—figure supplement 2A and Figure 2—source data 3 ) . All annotated peptides identified from the 14 PBMC samples were then used in SpectraST ( Lam et al . , 2008 ) to build the HLA class I allele-specific peptide spectral libraries ( ‘Materials and methods’ ) . The same procedure was applied to peptides identified from JYEBV+ and C1R cells . Notably , endogenous HLA-C04 peptides were recently shown to be significantly expressed on the surface of C1R cells ( Schittenhelm et al . , 2014b ) and were therefore considered in this study . In total , 3528 HLA-A peptides , 4208 HLA-B peptides and 205 HLA-C04 peptides were recorded in the spectral libraries , which were then stored in the public PeptideAtlas database ( Figure 2—source data 4 ) . In summary , we generated a computational workflow to effectively annotate and visualize HLA peptidomic data , which were finally converted and stored into HLA allele-specific peptide spectral libraries consisting of consensus fragment ion spectra . This strategy could be further refined to collect , store and share HLA peptidomic information obtained from various cell lines and from larger cohorts of donors . Importantly , this computational approach can be broadly applied to generate SWATH-compatible assay libraries as described below . Libraries of consensus fragment ion spectra can be converted into high quality assays for high-throughput targeted analysis of SWATH-MS data , an emerging approach for reproducible , consistent and accurate quantitative measurements of peptides ( Gillet et al . , 2012; Collins et al . , 2013; Rosenberger et al . , 2014; Röst et al . , 2014; Guo et al . , 2015; Liu et al . , 2015; Selevsek et al . , 2015; Schubert et al . , 2015a ) . Here , we aimed at initiating a worldwide community-based effort to generate pilot HLA allele-specific peptide assay libraries that could be further used for the analysis of SWATH-MS HLA peptidomic data . Naturally presented and/or synthetic HLA class I and class II peptides were provided from six independent laboratories and were analyzed using four distinct TripleTOF 5600 MS instruments operated in DDA acquisition mode in four different international institutions . Naturally presented HLA class I peptides from JYEBV+ ( HLA-A02 and -B07 ) , PBMC ( HLA-A03 , -A26 , -B51 and -B57 ) , and Jurkat ( HLA-A03 , -B07 and -B35 ) cells were isolated by immunoaffinity chromatography ( Figure 2—source data 1 ) . Natural class I peptides from three C1R cell lines—stably expressing HLA-C04 as well as HLA-B27 , -B39 or -B40 molecules—were also isolated using the same procedure . Synthetic EBV-derived peptides known to bind HLA-A02 or -B07 were also used to build the libraries ( Figure 2—source data 2 ) . All laboratories used the spiked-in landmark iRT peptides for retention time normalization ( Escher et al . , 2012 ) . The DDA data generated by the different groups were shared and pipelined through the computational workflow described above , resulting in the identification of 7668 ( peptide-level FDR 1%; average cFDR 0 . 5% ) or 11 , 275 ( peptide-level FDR 5%; average cFDR 2 . 5% ) distinct HLA class I peptides distributed across eleven different HLA class I alleles ( Figure 2—figure supplement 2B and Figure 2—source data 3 ) . To properly assess the efficiency of generating HLA peptide assay libraries from synthetic peptides , a large collection of 20 , 176 synthetic HLA class II peptides was analyzed by DDA using different mass spectrometers and fragmentation methods ( Figure 2—figure supplement 3 and Figure 2—source data 2 ) . Our results show that a total of 15 , 875 peptides ( ∼79% ) were identified ( Figure 2—source data 2 ) . A large collection of synthetic HLA class I peptides was not available but could be used in the future to extend the contents of the present class I libraries derived from native peptides . All identified peptides were used to build the HLA allele-specific peptide assay libraries ( ‘Materials and methods’ ) . To date , the pilot libraries contain a total of 223 , 735 transitions for 26 , 857 unique peptides and were stored by class and allele in the SWATHAtlas database ( Figure 2—source data 5 and http://www . swathatlas . org ) . By using the automated HLA peptide annotation method described above , we observed that similar binding affinities were predicted for HLA class I peptides identified at peptide-level FDR 1% and peptide-level FDR 5% ( Figure 2—figure supplement 4 and Figure 2—figure supplement 5 ) , suggesting that a large fraction of true positives were excluded at peptide-level FDR 1% . Our data also show that 95% of the annotated class I peptides in this study were predicted to bind their respective HLA molecules with an IC50 ranging from 72 nM ( for HLA-A01 ) to 5682 nM ( for HLA-B51 ) at peptide-level FDR 1% ( Figure 2—figure supplement 4 ) . Similar results were obtained at peptide-level FDR 5% ( Figure 2—figure supplement 5 ) . This result supports a recent study indicating that HLA class I alleles are associated with peptide-binding repertoires of different affinity ( Paul et al . , 2013 ) . Altogether , we demonstrated the feasibility of collecting DDA data from multiple international laboratories to generate standardized HLA allele-specific peptide assay libraries . We anticipate this global effort as a first step towards the development of a standardized Pan-human HLA peptide assay library , which could be used to rapidly and reproducibly quantify the entire repertoire of peptides presented by HLA molecules using SWATH-MS . SWATH-MS is emerging as a robust next-generation proteomics technique for efficiently generating reproducible , consistent and quantitatively accurate measurements of peptides across multiple samples ( Gillet et al . , 2012; Collins et al . , 2013; Rosenberger et al . , 2014; Röst et al . , 2014; Guo et al . , 2015; Liu et al . , 2015; Selevsek et al . , 2015; Schubert et al . , 2015a ) . To promote the worldwide development of SWATH-based MS platforms towards robust quantitative measurements of HLA peptidomes , we assessed whether the HLA allele-specific assay libraries described above could be used to extract quantitative information from digital SWATH maps acquired by different laboratories . Importantly , four independent laboratories generated their own digital SWATH maps using TripleTOF 5600 MS operated in DIA acquisition mode . Naturally presented HLA class I peptides were isolated from the cell types mentioned above ( i . e . , JYEBV+ , Jurkat , PBMC and C1R ) . Precursors in the range of 400–1200 Th were divided into 32 SWATH windows of 25 Da ( Gillet et al . , 2012 ) . All ionized peptide precursors in this mass range were fragmented , generating comprehensive and quantitative digital fragment ion maps . The HLA peptidome of JYEBV+ cells was analyzed using the OpenSWATH ( Röst et al . , 2014 ) software tool and a combined assay library containing 22 , 206 transitions for 1507 HLA-A02 and 2194 HLA-B07 peptides—the two dominant HLA alleles expressed on these cells . At an estimated peptide-level FDR of 1% ( m-score < 0 . 01 ) , a total of 3150 unique HLA class I peptides were identified from the digital SWATH map ( Figure 3A , B , C , Figure 3—figure supplement 1A , B , Figure 3—figure supplement 7 and Figure 3—source data 1 ) . Notably , assays generated from the synthetic EBV-related class I peptides enabled the identification of one EBV-derived HLA-A02 peptide ( Figure 3C ) , thereby demonstrating that building high-quality assay libraries from synthetic class I peptides of pathogen origin could be useful for the identification of non-self HLA-bound peptides by SWATH-MS . To analyze self-HLA peptides isolated from PBMC ( HLA-A03 , -A26 , -B51 and -B57 ) , Jurkat ( HLA-A03 , B07 and -B35 ) , C1R-B27 ( HLA-B27 ) and C1R-B40 ( HLA-B40 ) cells , the matching HLA class I allele-specific peptide assay libraries were combined accordingly using SpectraST and then processed in the OpenSWATH software . High-throughput targeted analysis from these four additional peptidomic datasets indicated that ∼81% of HLA class I peptides present in an assay library could be extracted from a quantitative digital SWATH map in a cell type-independent manner ( peptide-level FDR 1% ) ( Figure 3—figure supplement 1C , Figure 3—figure supplements 2–6 and Figure 3—source data 1 ) . We next optimized the SWATH acquisition conditions according to the size distribution of HLA class I peptides . Most class I peptide precursors ( ∼98% ) fall within the range of 400–700 Th and were divided in 30 SWATH windows of 10 Da width each . Using SWATH data generated from JYEBV+ cells , we found that narrowing the size of the windows by 2 . 5-fold resulted in a ∼13% fold-increase in the identification of class I peptides ( Figure 3—figure supplement 1A ) . The R2 value for SWATH-MS quantification was 0 . 979 from two technical replicates ( Figure 3D ) . In accordance with previous studies , we also observed that the dynamic range of peptides quantified in different cell types using SWATH-MS , based on their signal intensity , was about 3-4 orders of magnitude ( Figure 3E ) ( Hassan et al . , 2013; Bassani-Sternberg et al . , 2015 ) . Altogether , we demonstrate the feasibility of an international effort to build standardized HLA allele-specific peptide assay libraries , which were used to extract quantitative information from digital SWATH maps acquired in different sites . We therefore provide a proof of concept that acquisition of SWATH-MS HLA peptidomic data may enable robust analysis of the human immunopeptidome on a global scale . 10 . 7554/eLife . 07661 . 020Figure 3 . High-throughput targeted analysis of HLA peptidomic data by SWATH-MS . ( A ) SWATH-MS coordinates of two HLA class I allele-specific assay libraries ( HLA-A02 and -B07 ) were combined to extract SWATH data generated from the HLA peptidome of JYEBV+ cells . Sixteen summed transition groups are shown here for simplicity . ( B , C ) Visualization of two extracted SWATH transition groups corresponding to the self-HLA-A02 peptide , KILPTLEAV and the non-self HLA-A02 EBV peptide , YVLDHLIVV . ( D ) Reproducibility of intensity measurements for technical replicates . ( E ) Dynamic range of transition group intensities following targeted analysis of SWATH-MS HLA peptidomic data generated from various cell types expressing different combinations of HLA alleles . SWATH/DIA data were acquired in four independent international laboratories . DOI: http://dx . doi . org/10 . 7554/eLife . 07661 . 02010 . 7554/eLife . 07661 . 021Figure 3—source data 1 . OpenSWATH analysis . DOI: http://dx . doi . org/10 . 7554/eLife . 07661 . 02110 . 7554/eLife . 07661 . 022Figure 3—figure supplement 1 . OpenSWATH analysis of HLA peptidomic data . ( A ) HLA class I peptides isolated from JY cells were acquired in SWATH/DIA mode using windows of 10 Da ( blue ) or 25 Da ( red ) width each . The graph shows the proportion of peptides that were confidently extracted ( FDR < 0 . 01 ) using OpenSWATH from a merged ( A02+B07 ) or unmerged ( A02 or B07 ) HLA allele-specific assay library . ( B ) pyProphet statistical analysis from a JY HLA class I peptide extract . The histogram plots show the distribution of decoy and target transition groups according to their discriminant score ( dscore ) determined by the pyProphet software . ( C ) HLA class I peptides were isolated form various cell types and analyzed by SWATH-MS using windows of 25 Da width each . The histogram shows the number of HLA peptides that were confidently extracted ( FDR < 0 . 01 ) using OpenSWATH from different HLA allele-specific assay library . DOI: http://dx . doi . org/10 . 7554/eLife . 07661 . 02210 . 7554/eLife . 07661 . 023Figure 3—figure supplement 2 . OpenSWATH analysis and PyProphet statistics of HLA peptidomic data acquired at ETH Zurich , Switzerland . HLA-A02 and HLA-B07 peptides were isolated from JY cells . Graphs showing ROC , d_score performance and d-score distributions were generated automatically using the iPortal workflow . DOI: http://dx . doi . org/10 . 7554/eLife . 07661 . 02310 . 7554/eLife . 07661 . 024Figure 3—figure supplement 3 . OpenSWATH analysis and PyProphet statistics of HLA peptidomic data acquired at ETH Zurich , Switzerland . HLA-A03 , -A26 , -B51 and -B57 peptides were isolated from PBMCs . Graphs showing ROC , d_score performance and d-score distributions were generated automatically using the iPortal workflow . DOI: http://dx . doi . org/10 . 7554/eLife . 07661 . 02410 . 7554/eLife . 07661 . 025Figure 3—figure supplement 4 . OpenSWATH analysis and PyProphet statistics of HLA peptidomic data acquired at University of Oxford , UK . HLA-A03 , -B07 and -B35 peptides were isolated from Jurkat cells . Graphs showing ROC , d_score performance and d-score distributions were generated automatically using the iPortal workflow . DOI: http://dx . doi . org/10 . 7554/eLife . 07661 . 02510 . 7554/eLife . 07661 . 026Figure 3—figure supplement 5 . OpenSWATH analysis and PyProphet statistics of HLA peptidomic data acquired at Monash University , Australia . HLA-B27 peptides were isolated from C1R cells . Graphs showing ROC , d_score performance and d-score distributions were generated automatically using the iPortal workflow . DOI: http://dx . doi . org/10 . 7554/eLife . 07661 . 02610 . 7554/eLife . 07661 . 027Figure 3—figure supplement 6 . OpenSWATH analysis and PyProphet statistics of HLA peptidomic data acquired at Centro National de Biotechnologia , Madrid , Spain . HLA-B40 peptides were isolated from C1R cells . Graphs showing ROC , d_score performance and d-score distributions were generated automatically using the iPortal workflow . DOI: http://dx . doi . org/10 . 7554/eLife . 07661 . 02710 . 7554/eLife . 07661 . 028Figure 3—figure supplement 7 . Visualization and analysis of SWATH-MS HLA peptidomic data in Skyline . Skyline is a free open-source software for targeted data analysis of various types of peptidomic data . It specifically facilitates manual and automated analysis of SWATH data and other data-independently acquired MS data using assay libraries . The software itself can be downloaded from the website: http://skyline . maccosslab . org . ( A ) Skyline-daily or Skyline v2 . 6 was used to import HLA peptide SWATH assay libraries , and to import , extract , and visualize SWATH HLA peptidomic data . ( B ) The ‘Advanced Peak Picking Models’ was used to work with decoy transition groups and for large-scale , automated SWATH data analysis . For more information , see Schubert et al . ( 2015b ) . DOI: http://dx . doi . org/10 . 7554/eLife . 07661 . 028 To further establish the robustness of SWATH-MS for the measurement of HLA-associated peptides , we tested whether the JYEBV+ HLA peptidome could be reproducibly detected across multiple MS injections . For this purpose , we prepared a sample of class I peptides by immunoaffinity purification from JYEBV+ cells and we acquired three datasets in SWATH mode . The datasets were analyzed using OpenSWATH and a combined HLA-A02 and -B07 peptide assay library as described above . At an estimated peptide-level FDR of 1% , a total of 2933 unique HLA class I peptides were identified by SWATH-MS and 2832 peptides ( 97% ) were found in all the SWATH analyses ( Figure 1—figure supplement 1B , Figure 1—source data 1 ) . We then conducted a comparative analysis by acquiring three additional datasets in DDA mode from the same sample of class I peptides using the same chromatographic conditions . In total , 3153 HLA-A and -B peptides were identified at 1% peptide-level FDR and 1261 peptides ( 40% ) were found in all the DDA analyses ( Figure 1—figure supplement 1A , Figure 1—source data 1 ) . Thus , the SWATH method clearly outperformed the DDA approach for the reproducible identification of JYEBV+ HLA class I peptides across several technical replicates . Overall , our results indicate that SWATH-MS has the capability of detecting large numbers of HLA peptides across multiple injections at a high degree of reproducibility . By providing a community resource for the continuous expansion of the library contents and by improving the performance of the OpenSWATH software , it can be expected that additional HLA peptides—including cryptic and mutant peptides—will be reproducibly identified and quantified from the same digital SWATH maps in the future . The life sciences community greatly benefits from robust technologies such as microarrays and RNA-seq . Similarly , robust generation and analysis of quantitative digital maps of HLA peptidomes is expected to have important implications in basic and translational research as these will allow research groups to accurately investigate the dynamics of immunopeptidomes in various immune-related diseases such as autoimmunity , infectious diseases and cancers . For instance , reproducible digital mapping of tumor-specific mutant HLA peptides during cancer progression will facilitate stratification of patients who might best benefit from innovative immunotherapeutic interventions ( Gubin et al . , 2014; Snyder et al . , 2014; Schumacher et al . , 2015 ) . The workflow and the computational and data resources presented in this community-based study is a first step towards highly reproducible and quantitative MS-based measurements of HLA peptidomes across many samples and could therefore be greatly beneficial in the design of personalized immune-based therapies . Moreover , the storage of HLA peptide spectral and assay libraries by class and allele in the SWATHAtlas database provides an initial framework to collect , organize and share HLA peptidomic data , thereby supporting the recently proposed Human Immunopeptidome and Vaccines Projects ( Admon and Bassani-Sternberg , 2011; Koff et al . , 2014 ) .
PBMCs from healthy donors were isolated by density gradient centrifugation . Informed consent was obtained in accordance with the Declaration of Helsinki protocol . HLA typing was carried out by the Department of Hematology and Oncology , Tübingen , Germany . PBMCs were stored at −80°C until further use . JYEBV+ , Jurkat and C1R cells were cultured in RPMI supplemented with 10% fetal bovine serum , 50 IU/ml penicillin , and 50ug/ml streptomycin ( Invitrogen , Life Technologies Europe BV , Zug , Switzerland ) . C1R cells were stably transfected with -B2705 , -B3901 and -B4002 constructs , as described previously ( Marcilla et al . , 2014; Schittenhelm et al . , 2014a ) . The EBV peptide was synthesized by Thermo Fischer Scientific ( Ulm , Germany ) . The collection of 20 , 176 MTB peptides was synthesized by Mimotopes ( Victoria , Australia ) as described ( Lindestam Arlehamn et al . , 2013 ) . HLA class I peptide complexes were isolated by standard immunoaffinity purification as described previously using the pan-HLA class I-specific mAb W6/32 ( Hunt et al . , 1992; Croft et al . , 2013; Kowalewski and Stevanovic , 2013; Marcilla et al . , 2014 ) . For the RT normalization and analysis , the peptides from the iRT Kit ( Biognosys AG , Schlieren , Switzerland ) were added to samples ( see Figure 2—source data 1 ) prior to MS injection according to vendor instructions ( Escher et al . , 2012 ) . All raw instrument data were centroided and processed as described previously ( Collins et al . , 2013; Rosenberger et al . , 2014 ) . The datasets were searched individually using X ! tandem ( Craig et al . , 2004 ) , MS-GF+ ( Kim and Pevzner , 2014 ) and Comet ( Eng et al . , 2012 ) against the full non-redundant , canonical human genome as annotated by the UniProtKB/Swiss-Prot ( 2014_02 ) with 20 , 270 ORFs and appended iRT peptide and decoy sequence . Oxidation ( M ) was the only variable modification . Parent mass error was set to ±5 p . p . m . , fragment mass error was set to ±0 . 5 Da . The search identifications were then combined and statistically scored using PeptideProphet ( Keller et al . , 2002 ) and iProphet ( Shteynberg et al . , 2011 ) within the TPP ( 4 . 7 . 0 ) ( Keller et al . , 2005 ) . All peptides with an iProbability/iProphet score above 0 . 7 were exported in Excel . Assumed charges were also exported , as this information is needed in SpectraST . Length considered was 8–12 residues for class I HLA peptides . FDR was manually estimated based on the target-decoy approach ( Elias and Gygi , 2007 ) . Peptides ( 1% and 5% peptide-level FDR ) were then exported to a . txt file for annotation to their respective HLA allele . Annotation of the identified peptides ( 1% and 5% peptide-level FDR ) to their respective HLA allele was performed automatically by integrating the stand-alone software package of NetMHC 3 . 4 ( Lundegaard et al . , 2008 ) with our in-house software tools ( Supplementary file 1 and Source code 1 ) . An HLA annotation score was computed by the software tools for individual peptides ( Figure 2—figure supplement 1 ) . A predefined cutoff score of 3 was then used to annotate each peptide to their respective HLA allele . A cutoff value of 3 was selected because >90% of the identified peptides with an annotation score above 3 have a predicted IC50 below 1000 nM . FDR was corrected from the list of annotated HLA peptides based on the target-decoy approach ( Elias and Gygi , 2007 ) . The software tools were used to process and visualize the peptidomic datasets . The final lists of HLA-allele specific peptides were exported into a . txt file and used in SpectraST for library generation . This section was adapted from Schubert et al . ( 2015b ) . The parameters below were used for Spectrast ( Lam et al . , 2008 ) . Exact meaning of each parameter can be found in the following link: http://tools . proteomecenter . org/wiki/index . php ? title=Software:SpectraST . Spectrast was used in library generation mode with CID-QTOF settings ( -cICID-QTOF ) for the Triple-TOF 5600+ or CID ( default ) settings for the Orbitrap-XL and Orbitrap-ELITE . Retention times were normalized against the iRT Kit peptide sequences ( -c_IRTiRT . txt -c_IRR ) . Only HLA-allele specific peptide ions were included for library generation ( -cT ) : spectrast -cNSpecLib_celltype_allele_fdr_iRT -cICID-QTOF -cTReference_celltype_allele_fdr . txt -cP0 . 7 -c_IRTiRT . txt -c_IRR iprophet . pep . xml A consensus library was then generated: spectrast -cNSpecLib_cons_celltype_allele_fdr_iRT -cICID-QTOF -cAC SpecLib_celltype_allele_fdr_iRT . splib HLA-allele specific consensus libraries were merged: spectrast -cNSpecLib_cons_celltype_alleles_fdr_iRT -cJU -cAC SpecLib_celltype_allele1_fdr_iRT . splib SpecLib_celltype_allele2_fdr_iRT . splib SpecLib_celltype_allele3_fdr_iRT . splib SpecLib_celltype_allele4_fdr_iRT . splib The script spectrast2tsv . py ( msproteomicstools 0 . 2 . 2; https://pypi . python . org/pypi/msproteomicstools ) was then used to generate the HLA-allele specific peptide assay library with the following recommended settings: spectrast2tsv . py -l 350 , 2000 -s b , y -x 1 , 2 -o 6 -n 6 -p 0 . 05 -d -e -w swaths . txt -k openswath -a SpecLib_cons_celltype_alleles_fdr_iRT_openswath . csv SpecLib_cons_celltype_alleles_fdr_iRT . sptxt The _openswath . csv file was then converted into a . tsv file and opened in Excel . Reference coordinates for the 11 iRT peptides were confirmed and any remaining decoy sequences were removed . The file was then saved in . txt format and then converted back in . csv format . The OpenSWATH tool ConvertTSVToTraML converted the TSV/CSV file to TraML: ConvertTSVToTraML -in SpecLib_cons_celltype_alleles_fdr_iRT_openswath . csv -out SpecLib_cons_celltype_alleles_fdr_iRT . TraML Decoys were appended to the TraML assay library with the OpenSWATH tool OpenSwathDecoyGenerator as described before ( Rosenberger et al . , 2014; Röst et al . , 2014; Schubert et al . , 2015b ) in reverse mode with a similarity threshold of 0 . 05 Da and an identity threshold of 1: OpenSwathDecoyGenerator -in SpecLib_cons_celltype_alleles_fdr_iRT . TraML -out SpecLib_cons_celltype_alleles_fdr_iRT_decoy . TraML -method shuffle -append -exclude_similar The library was then uploaded into the iPortal workflow for SWATH data analysis ( see below ) . For SWATH-MS data acquisition , the same mass spectrometer and LC-MS/MS setup was operated essentially as described before ( Collins et al . , 2013; Rosenberger et al . , 2014 ) using 32 windows of 25 Da effective isolation width ( with an additional 1 Da overlap on the left side of the window ) and with a dwell time of 100 ms to cover the mass range of 400–1200 m/z in 3 . 3 s . Before each cycle , an MS1 scan was acquired , and then the MS2 scan cycle started ( 400–425 m/z precursor isolation window for the first scan , 424–450 m/z for the second . . . 1 , 174–1200 m/z for the last scan ) . The collision energy for each window was set using the collision energy of a 2+ ion centered in the middle of the window with a spread of 15 eV . Four independent international laboratories acquired their own SWATH maps using the settings described above: ( 1 ) Antony Purcell , Monash University; ( 2 ) Nicola Ternette , University of Oxford; ( 3 ) Miguel Marcilla , Spanish National Biotechnology Center; ( 4 ) Ruedi Aebersold , ETH-Zurich . The iPortal workflow was used for data analyses ( Kunszt et al . , 2014 ) . The OpenSWATH analysis workflow ( OpenSWATHWorkflow ) ( http://www . openswath . org ) was implemented in the iPortal workflow . The parameters were selected analogously to the ones described before ( Röst et al . , 2014 ) : min_rsq: 0 . 95 , min_coverage: 0 . 6 , min_upper_edge_dist: 1 , mz_extraction_window: 0 . 05 , rt_extraction_window: 600 , extra_rt_extraction_window: 100 . pyprophet ( https://pypi . python . org/pypi/pyprophet ) was run on the OpenSwathWorkflow output adjusted to contain the previously described scores ( xx_swath_prelim_score , bseries_score , elution_model_fit_score , intensity_score , isotope_correlation_score , isotope_overlap_score , library_corr , library_rmsd , log_sn_score , massdev_score , massdev_score_weighted , norm_rt_score , xcorr_coelution , xcorr_coelution_weighted , xcorr_shape , xcorr_shape_weighted . yseries_score ) ( Röst et al . , 2014 ) . Assay libraries were loaded into Skyline and SWATH traces were analyzed as described previously ( Schubert et al . , 2015b ) . Advanced protocols for analysis of SWATH/DIA data can be downloaded from the website: http://skyline . maccosslab . org . | The cells of the immune system protect us by recognizing telltale molecules produced by damaged and diseased cells , or by infection-causing microorganisms ( which are also called pathogens ) . To help with this process , the cells in our bodies display small fragments of proteins ( called peptides ) on their surface that are then checked by the immune cells . Collectively , these peptides are referred to as the ‘immunopeptidome’ , and deciphering the complexity of the human immunopeptidome is important for both basic research and medical science . Such an achievement would help to guide the development of next-generation vaccines and therapies against autoimmune disorders , infectious diseases and cancers . In the past , immune peptides were mostly identified using a technique that is commonly called ‘shotgun’ mass spectrometry . However , this approach doesn't always provide reproducible results . In 2012 , researchers reported the development of a new approach—which they called ‘SWATH’ mass spectrometry—that could yield more reproducible data . Now , Caron et al . —including many of the researchers involved in the 2012 study—have developed a large collection of standardized tests that use SWATH mass spectrometry to analyze the human immunopeptidome . The workflow and the computational and data resources developed as part of this international effort are the first steps toward highly reproducible and measurable analyses of the immunopeptidome across many samples . Moreover , the large repository of assays generated by the project has been made public and will serve a large community of researchers , which should enable better collaborations . In the future , SWATH mass spectrometry could be used as a robust technology for the reproducible detection and measurement of pathogen-specific or cancer-specific immune peptides . This could greatly help in the design of personalized immune-based therapies . | [
"Abstract",
"Introduction",
"Results",
"and",
"discussion",
"Materials",
"and",
"methods"
] | [
"computational",
"and",
"systems",
"biology",
"tools",
"and",
"resources",
"immunology",
"and",
"inflammation"
] | 2015 | An open-source computational and data resource to analyze digital maps of immunopeptidomes |
Ploidy-increasing cell cycles drive tissue growth in many developing organs . Such cycles , including endocycles , are increasingly appreciated to drive tissue growth following injury or activated growth signaling in mature organs . In these organs , the regulation and distinct roles of different cell cycles remains unclear . Here , we uncover a programmed switch between cell cycles in the Drosophila hindgut pylorus . Using an acute injury model , we identify mitosis as the response in larval pyloric cells , whereas endocycles occur in adult pyloric cells . By developing a novel genetic method , DEMISE ( Dual-Expression-Method-for-Induced-Site-specific-Eradication ) , we show the cell cycle regulator Fizzy-related dictates the decision between mitosis and endocycles . After injury , both cycles accurately restore tissue mass and genome content . However , in response to sustained growth signaling , only endocycles preserve epithelial architecture . Our data reveal distinct cell cycle programming in response to similar stimuli in mature vs . developmental states and reveal a tissue-protective role of endocycles .
Throughout development , cell cycle regulation is altered to build tissues and organs . Examples include the lack of gap phases to rapidly increase cell number in embryos of many species ( Edgar and O'Farrell , 1990; Graham and Morgan , 1966; McKnight and Miller , 1977; Newport and Kirschner , 1982 ) , ploidy-reducing meiotic divisions to produce haploid gametes ( Kleckner , 1996 ) , or ploidy-increasing cycles that enable rapid post-mitotic tissue growth ( Calvi , 2013; Fox and Duronio , 2013; Hua and Orr-Weaver , 2017; Edgar et al . , 2014 ) . After development is completed , a diversity of cell cycle regulation is also found in mature adult tissues during injury repair . In many stem cell-based tissues , or in highly regenerative organs/organisms , mitotic cell cycles restore pre-injury cell number and size ( Jiang et al . , 2009; Mascré et al . , 2012; Mauro , 1961; Poss et al . , 2002; Ryoo et al . , 2004; Yan et al . , 2012 ) . In contrast , we and others previously defined injury responses in the adult Drosophila hindgut and abdomen , tissues that lack mitotic divisions ( Fox and Spradling , 2009; Losick et al . , 2013; Sawyer et al . , 2017 ) . In these adult tissues , injury leads to an increase in cellular ploidy through endocycles ( G/S cycles without M phase , see cell cycle nomenclature section of Materials and methods ) . These Drosophila responses have clear parallels in the hypertrophic tissue injury repair of mammals . Injured mammalian hearts alter their cell cycle programming from mitotic to ploidy-increasing cell cycles during defined periods in development ( Porrello et al . , 2011 ) . As a result , cardiac cells typically undergo hypertrophy instead of hyperplasia in response to injury or sustained tissue growth signals such as from the Ras/Raf pathway ( Hunter et al . , 1995; Porrello et al . , 2011; Wu et al . , 2011; Yu et al . , 2015 ) . In the liver , injury can cause either mitotic or ploidy-increasing cell cycle responses ( Gentric et al . , 2015; Miyaoka et al . , 2012; Nagy et al . , 2001 ) . Recently , the mouse kidney was discovered to endocycle in response to acute injury ( Lazzeri et al . , 2018 ) . Thus , both during development and in post-development injury contexts , diverse cell cycle responses can occur . Little is known about the molecular programming or functional consequence of distinct cell cycles used in injured adult tissues . One technical limitation to studying this question is the ability to conduct carefully targeted injury experiments while simultaneously performing genetic studies . Here , we introduce a new system termed Dual-Expression-Method-for-Induced-Site-specific-Eradication ( DEMISE ) , which enables us to finely control and independently manipulate both injury and genetics in our system . Using this system , we uncover developmental regulation and functional differences between two injury-induced cell cycle programs in the Drosophila hindgut pyloric epithelium . The pyloric epithelium is the only segment of the hindgut to persist throughout the lifespan of the fly . Taking advantage of this persistence , we uncover that when injured the same way , pyloric cells undergo mitotic cycles in larvae but undergo endocycles in mature adults . Further , using this tissue model and our new genetic system , we demonstrate that active inhibition of mitotic cyclins by the conserved Anaphase Promoting Complex/Cyclosome ( APC/C ) regulator Fizzy-related ( Fzr ) underlies the alteration in injury-induced cell cycle programs in the pyloric epithelium . We identify that by blocking entry into mitosis , Fzr-mediated endocycles protect the adult pylorus against disruptions in epithelial architecture and permeability under conditions of sustained tissue growth signaling . Together , our results suggest that in some mature tissues , endocycles may represent a tradeoff between loss of regenerative capacity and preservation of tissue architecture .
We previously demonstrated that the adult Drosophila hindgut pyloric epithelium ( hereafter- pyloric cells ) provides an accessible model to study tissue injury repair through endocycles ( Fox and Spradling , 2009; Losick et al . , 2013; Sawyer et al . , 2017 ) . Unlike many adult intestinal cells , pyloric cells are also a constituent segment of the larval hindgut . During metamorphosis , pyloric cells act as facultative progenitor cells , as they remodel the hindgut by undergoing mitotic cell division to both expand the larval pylorus into its adult form while also producing cells of the adult ileum , which replace the histolysed larval ileum ( Figure 1A , Fox and Spradling , 2009; Robertson , 1936; Sawyer et al . , 2017; Takashima et al . , 2008 ) . Thus , pyloric cells are capable of distinct cell cycles- mitotic cycles during organ remodeling ( at metamorphosis ) and endocycles during tissue injury repair ( at adulthood ) . We tested two possible models for the difference in pyloric cell cycle programs . In one model , pyloric cell cycle program is dictated by the stimulus: that is , induced apoptotic injury promotes endocycles while developmental gut histolysis promotes cell division . In a second alternative model , developmental status of the pylorus may solely govern cell cycle status , regardless of the type of injury . To distinguish between these possibilities , we injured the larval pylorus at the last developmental stage before metamorphosis ( wandering third larval instar , L3 ) and allowed the animals to progress to adulthood . For tissue injury , we used temporal and spatial control of the pro-apoptotic genes head involution defective ( hid ) and reaper ( rpr ) , as before ( Materials and methods , Figure 1B , ‘1’ , Fox and Spradling , 2009; Losick et al . , 2013; Sawyer et al . , 2017 ) . As a comparison , we injured adult pyloric cells using the same scheme ( Figure 1B , ‘2’ ) . In both cases , we confirmed that our injury protocol causes pyloric cell death ( Figure 1—figure supplement 1A–E ) . To clearly demarcate the recovered pylorus , we used reporters of pyloric boundaries ( Figure 1C–E ) . We then compared adult flies recovered from either larval or adult injury to identify any differences in the mode of pyloric recovery . While animals recovered from both larval and adult injury show no obvious defects in recovery of pyloric tissue mass , the response to larval and adult injury is strikingly different . When compared against uninjured animals ( Figure 1C–C’ ) , adult pyloric cells recovered from a larval injury ( Figure 1D–D’ ) show no change in cell ploidy and remain diploid ( Figure 1G ) . Additionally , larval injury does not change the number or size of cells recovered in adults ( Figure 1H , Figure 1—figure supplement 1F–F’ , H ) . The ability to produce an adult gut of normal cell number and ploidy was impressive given that we eliminated a high percentage of larval pyloric cells by injury ( Figure 1—figure supplement 1E ) . In contrast , comparable injury to the adult pylorus persistently increases ploidy , decreases cell number , and increases cell size , as we previously reported ( Figure 1E , G , H , Losick et al . , 2013 ) . However , it remained possible that the number of surviving cells following tissue injury dictates the pyloric response . To test this idea , we took advantage of the ability to finely tune injury level in our system ( Materials and methods ) and used varying durations ( 12–48 hr ) of apoptotic gene expression to produce a linear decrease in cell survival ( Figure 1F , H ) . Regardless of the severity of adult injury , cell number does not recover , whereas cell ploidy and cell size increases ( Figure 1F , G , Figure 1—figure supplement 1G–I ) . Thus , developmental stage and not injury severity dictates the pyloric cell cycle response . Our quantitation of adult pyloric ploidy after injury revealed a proportional tissue injury response- ploidy increases closely track with the degree of cell number loss ( Figure 1G–H , R2 = 0 . 97 , p<0 . 002 ) . These results suggested that regardless of developmental stage , the pylorus remains capable of closely replacing the number of genomes lost to injury . To test this model , we quantified the total genome content per pylorus ( number of cells x ploidy ) following injury recovery ( Materials and methods ) . Our data show that regardless of injury severity or developmental stage , the pre-injury pyloric total genome content is fully recoverable ( Figure 1I ) . These results indicate that through developmentally distinct cell cycle injury responses , the pylorus accurately restores genomic content in proportion to the injury stimulus ( Figure 1J ) . We next analyzed the cell cycle responses of the two distinct pyloric injury recovery modes . Our cell ploidy and cell number quantitation suggest that larval pyloric cells are capable of producing new cells to regenerate the injured gut during metamorphosis , whereas adult pyloric cells have lost this regenerative ability . To assess this , we first traced the lineage of uninjured and injured larval hindgut cells during metamorphosis ( Figure 2A , Figure 2—figure supplement 1A ) . We previously used a low-background , low-frequency clonal marking system to demonstrate that single larval pyloric cells produce , on average , either ~5 adult pyloric cells or two adult ileal cells during metamorphosis , and afterwards cease to divide ( Fox and Spradling , 2009 ) . We reproduced these data in uninjured animals ( Materials and methods , Figure 2C , Figure 2—figure supplement 1B–E ) . By comparison , in adults recovered from larval injury , clone size increases by ~3 fold on average ( Figure 2D–D’ , G , H , uninjured: 5 . 389 ± 0 . 5725 SEM , injured: 16 . 54 ± 1 . 863 SEM , p<0 . 0001 ) . Our clonal data coupled with our cell counts suggest that approximately 75% of larval cells were eliminated using our injury protocol and were then recovered by compensatory proliferation . Clone size in the ileum also increases in response to injury , consistent with the model that these cells derive from the larval pylorus ( Figure 2—figure supplement 1C–E ) . Further , these data show that remaining larval pyloric cells ( approximately 25% ) have the capability of increasing their mitotic capacity to completely regenerate the adult hindgut pylorus and ileum following acute injury . In contrast to larvae , inducing clones in adult pylori ( Figure 2B ) produces only single labeled cells ( 27/27 clones , Figure 2E ) . Our marking system does not require a cell division to generate single-labelled cells ( Figure 2—figure supplement 1A ) . Using this system , we did not see any expansion of single labeled cells into mitotic clones even after 20–30 days of recovery from a severe injury ( Figure 2F–H , uninjured: 1 . 00 ± 0 SEM , injured: 1 . 07 ± 0 . 07 SEM , p=0 . 99 and Figure 2—figure supplement 1F–H ) . In further support of our lineage data , the M-phase marker Phospho-Histone H3 ( PH3 ) does not label adult pyloric cells ( data not shown , Fox and Spradling , 2009; Sawyer et al . , 2017 ) , while it frequently labels larval pyloric cells ( Figure 2—figure supplement 1I ) . However , both larvae and adult pyloric cells incorporate the S-phase marker EdU following injury ( Figure 2—figure supplement 1J , K ) . Taken together , our cell cycle marker , lineage , and cell ploidy/number quantitation show that only larval pyloric cells divide while adult pyloric cells instead endocycle after injury . Having determined the two distinct post-injury pyloric cell cycle responses , we next analyzed the pre-injury cell cycle status of both larval and adult pyloric cells . For this analysis , we used the Fly Fluorescent Ubiquitin-based Cell Cycle Indicator ( Fly-FUCCI ) system ( Zielke et al . , 2014 ) . In the pylorus of uninjured larvae , the FUCCI components E2F11-230-GFP and Cyclin B1-266-mRFP are expressed in patterns consistent with the presence of G1 , S , and G2/M cells ( Figure 2I–J ) . Coupled with our M phase data ( Figure 2—figure supplement 1I ) , we find that both before and after injury , larval pyloric cells undergo mitotic cycles , although injury leads to additional rounds of mitotic cycles ( Figure 2L ) . In contrast , 100% of uninjured adult pyloric cells are E2F11-230-GFP positive and Cyclin B1-266-mRFP negative ( Figure 2K ) , indicative of a quiescent/arrested G0/G1 state of these cells prior to their entry into injury-activated endocycles . Thus , pyloric injury at distinct developmental stages induces distinct cell cycle changes ( Figure 2L ) . Having identified developmental stage as a determinant of the distinct pyloric cell cycle injury responses , we next sought to identify molecular regulation of these responses . Our apoptotic injury system takes advantage of the temperature sensitive Gal80ts repressor to control Gal4-mediated expression of apoptotic genes ( Brand and Perrimon , 1993; McGuire et al . , 2004 ) . While this enables fine control over the developmental stage , tissue location , and degree of injury , it also prevents use of the numerous Gal4-mediated gene knockdown tools such as the large collection of transgenic UAS-RNAi and now CRISPR lines ( Dietzl et al . , 2007; Lin et al . , 2015; Ren et al . , 2013; Wu et al . , 2011 ) . This is because any transgene-expressing cell would be dead using our standard apoptotic method . An ideal system would instead allow for separate control over injury and transgene expression . We thus developed such a system , which we term DEMISE ( Dual-Expression-Method-for-Induced-Site-specific-Eradication ) . DEMISE combines the Gal4/Gal80ts and FLP/FRT systems to induce developmentally-timed , tunable tissue-specific injury in a mosaic fashion while independently expressing any UAS-driven transgene of interest in the same tissue . Compared to our previous injury system , DEMISE adds an FRT-Stop-FRT cassette between a UAS promoter and the apoptotic rpr gene ( Figure 3A ) . Thus , cell death and transgene expression are still under control of a tissue-specific Gal4 and temperature inactivation of Gal80ts ( Figure 3A’ ) but cell death also requires FLP-mediated removal of the Stop cassette . Temporal control of injury , as well as injury strength , is then achieved by timing expression of FLP under heat shock ( hs ) control ( Figure 3A’’ ) . In parallel , Gal4/Gal80ts drives expression of a transgene of interest independent of FLP-mediated cell death ( Figure 3A’ , A’’ ) . The result is that all cells of a given tissue can express a transgene ( such as a gene knockdown RNAi ) , while a fraction of these cells ( determined by degree of FLP induction ) will be eliminated by injury ( Figure 3A’’’ ) . This experimental design enables one to use the vast UAS-driven Drosophila transgene collection while also inducing a precise injury . To test the efficacy of DEMISE , we generated a series of UAS-FRT-Stop-FRT-reaper insertions in the Drosophila genome ( Materials and methods ) . We identified FLP/DEMISE combinations with both low background cell death ( lack of leaky Stop cassette excision ) along with robust apoptosis induction upon heat shock ( Table 1 ) . We began by examining larval wing imaginal discs , a frequently used model to study tissue regeneration in response to apoptotic injury ( Halme et al . , 2010; Harris et al . , 2016; Smith-Bolton et al . , 2009 ) . We drove DEMISE expression in the posterior wing disc compartment with engrailed ( en ) > Gal4 . Using temperature shifts and a UAS-GFP reporter , transgene expression is induced in the expected pattern ( Figure 3—figure supplement 1A–C ) . Our transgene expression conditions do not induce cell death , as assayed by TUNEL ( Figure 3—figure supplement 1B ) . However , upon heat shock to induce FLP expression , robust cell death occurs in a mosaic pattern specifically in the engrailed domain ( Figure 3—figure supplement 1C–D , Materials and methods ) . We then returned to the hindgut system and performed similar transgene expression controls ( Figure 3B–D ) . Again , using our most efficacious DEMISE/FLP combination in the hindgut , we can achieve a low background level of cell death ( Figure 3C , D , F ) , followed by robust induction of cell death upon heat shock that occurs independently of transgene expression ( Figure 3C–F , Materials and methods ) . Our results establish DEMISE as a new method for inducing temporal and site-specific injury while maintaining independent control of transgene expression . After establishing DEMISE as a novel system to study molecular regulators of injury responses , we sought to identify genes that dictate which pyloric cell cycle program is induced by injury . Our pre-injury FUCCI data ( Figure 2I–K ) provided a potential clue , as larval but not adult uninjured pyloric cells co-express the GFP-E2F11-230 and RFP-CycB1-266 FUCCI reporters , indicative of a G2/M state ( Zielke et al . , 2014 ) . This led us to hypothesize that inhibition of mitotic cyclins in the adult pylorus may prevent entry into mitotic cycles after injury . Fizzy-related ( Fzr ) , the Drosophila homolog of mammalian FZR1/CDH1 , is a binding partner of the APC/C , which facilitates degradation of mitotic cyclins including Cyclin B . Fzr is also a well-known regulator of the endocycle in many cell types ( Lehner and O'Farrell , 1989; Schaeffer et al . , 2004; Sigrist and Lehner , 1997; Stormo and Fox , 2016 ) and fzr mutant cells can ectopically undergo mitotic cycles instead of endocycles ( Schaeffer et al . , 2004; Schoenfelder et al . , 2014; Sigrist and Lehner , 1997 ) . Thus , increased fzr expression in the adult pylorus is a plausible mechanism for the altered injury responses we observe . We first asked whether fzr is upregulated in the adult pylorus . Indeed , using two independent fzr enhancer traps , fzr expression is high in the adult pylorus , whereas fzr is undetectable in the larval pylorus ( Figure 4A–B’ , Figure 4—figure supplement 1L–M’ ) . We next asked if elimination of fzr is sufficient to revert the adult pyloric injury response ( endocycles ) to the larval injury response ( mitotic cycles ) . Using DEMISE , we expressed fzr RNAi in the hindgut throughout development ( Figure 4C ) . Without injury , fzr knockdown did not noticeably alter hindgut development with the exception of the rectal papillae , where we previously identified a role for fzr in the pre-mitotic endocycles ( Schoenfelder et al . , 2014 , Figure 4—figure supplement 1A–B ) . Following apoptotic injury induction , both control and fzr RNAi-expressing animals contain both TUNEL positive nuclei and pycnotic nuclei ( Figure 4D–F ) , which are more prevalent at the anterior pylorus as previously described ( Fox and Spradling , 2009; Sawyer et al . , 2017 ) . Further , lack of fzr does not alter the number of adult pyloric cells entering the cell cycle as seen by BrdU staining ( Figure 4—figure supplement 1C ) . Strikingly , following 5 days of recovery from adult injury , the adult pylorus of fzr flies completely restores pre-injury cell number and ploidy ( Figure 4G–J ) . Further , during injury recovery , mitotic cells are visible in the adult fzr pylorus ( 5/7 animals ) , but not in control animals ( 0/17 animals , as assayed by PH3 and the centrosomal marker Centrosomin- Cnn , Figure 4K–L’’’ ) . We noted one additional cell cycle alteration in fzr animals . Endocycles frequently exhibit under-replication of late-replicating sequences , which in Drosophila cluster in a DAPI ‘bright-spot’ ( Belyaeva et al . , 1998; Endow and Gall , 1975; Fox and Duronio , 2013; Gall et al . , 1971; Nordman et al . , 2011; Schoenfelder and Fox , 2015; Edgar et al . , 2014 ) . In control adults , pyloric cells undergoing injury-induced endocycles only exhibit early replication patterns ( as assayed by BrdU , Figure 4—figure supplement 1D ) , consistent with the S-phase pattern of many endocycling cells that undergo under-replication . In agreement with this idea , DNA FISH shows that a satellite DNA repeat that is commonly under-replicated in endocycling cells ( Endow and Gall , 1975 ) does not increase in intensity in proportion to the obvious nuclear size increase induced by adult pyloric injury ( Figure 4—figure supplement 1N–O ) . In contrast , fzr animals instead exhibit both early and late-S patterns after injury ( Figure 4—figure supplement 1E ) . This pattern suggests that fzr loss may also enable progression through late S-phase , possibly due to a role for the fzr target cyclin A in promoting late replication ( SalleSallé et al . , 2012 ) . Taken together , we find the two distinct injury-induced cell cycle programs in the pylorus are dictated by the action of Fzr , a conserved negative regulator of mitotic cyclins ( Figure 4M ) . Previous work on developmentally programmed switches from mitotic cycles to endocycles in Drosophila follicle cells also implicated a role for Fzr ( Schaeffer et al . , 2004 ) . In this context , fzr is regulated upstream by Notch signaling and its transcriptional target Hindsight . Further , Notch frequently promotes endocycles in development and in adult tissue homeostasis ( Deng et al . , 2001; Guo and Ohlstein , 2015; Von Stetina et al . , 2018 ) . Therefore , we sought to determine if Notch is also used in an acute injury context to regulate the pyloric cell cycle program . Either with or without injury , adult pyloric cells do not express Hindsight , suggesting that this Notch target is not involved in the pyloric cell cycle injury response ( Figure 4—figure supplement 1F–G ) . Further , expressing dominant negative Notch receptor throughout pupation using DEMISE does not prevent injury-mediated endocycles in the adult pylorus ( Figure 4—figure supplement 1H–I ) , despite the fact that we observe expected Notch phenotypes in the rectal papillae ( Figure 4—figure supplement 1J–K , Fox et al . , 2010; Schoenfelder et al . , 2014 ) . Thus , unlike previous studies on developmental endocycles , our results suggest a Notch-independent , fzr-dependent alteration in injury cell cycle responses in the Drosophila pylorus . Our studies of fzr animals showed that the normally diploid adult pylorus is primed to endocycle rather than maintain cellular ploidy under injury conditions . Both mitotic cycles and endocycles are capable of restoring pre-injury genome number throughout the tissue . This raises the question of whether there are any benefits to the switch away from mitosis and towards endocycles following cell cycle re-entry in the adult pylorus . One possibility that we considered is that , in response to pro-growth signaling , endocycles may be better at preserving tissue architecture than mitotic cycles . Activation of the pro-growth signaling Ras pathway , which is well-known to drive cell number increases in many contexts , is also linked to ploidy increases in cardiac tissue in flies and mammals ( Hunter et al . , 1995; Wu et al . , 2011; Yu et al . , 2015 ) . We thus examined if constitutive activation of the small GTPase Ras ( RasV12 ) could be used as an experimental tool to ask whether a tissue undergoing endocycles or mitotic cycles responds differently under conditions of sustained tissue growth signaling . Using similar methods to our injury induction protocol ( Figure 5A ) , we expressed RasV12 in larvae and then examined the resulting adults ( Figure 5B , ‘1’ ) . The pylorus of these animals is extremely expanded in size ( Figure 5C , D ) . Cell number counts and size measurements show that larval RasV12 expression nearly triples the adult pyloric cell number without a substantial change in nuclear size ( Figure 5G , H ) . This larval RasV12 response parallels the mitotic cell cycle response to injury in larvae , with the exception of a tissue overgrowth phenotype , which is likely due to sustained RasV12 expression overriding normal growth suppression signals . We next drove RasV12 expression in the adult ( Figure 5B , ‘2’ ) , again in the absence of injury . As with larval RasV12 expression , the adult pylorus expands in size following RasV12 expression ( Figure 5C , E ) . However , and in parallel with our injury results , we observe no increase in cell number , while instead the nuclear size increases ( Figure 5G , H ) . These data are consistent with RasV12 inducing endocycles in the mature adult pylorus . Thus , sustained growth signaling through RasV12 expression can be used to mimic both the larval and adult injury cell cycle responses found in the pylorus . We next used RasV12 expression as a model to examine whether endocycles or mitotic cycles confer any tissue-level difference in the adult pylorus under conditions of prolonged tissue growth signaling . To ask this question , we compared RasV12 expressing adults to adults expressing both RasV12 and fzr RNAi ( Figure 5B ‘3’ ) . As in our injury studies , fzr suppresses nuclear size expansion in RasV12 expressing adults ( Figure 5C , F , H ) . These results are consistent with a requirement of fzr for RasV12-driven endocycles in this tissue . Further , the overall pyloric cell number increases ~20% above controls in fzr RNAi + RasVV12 animals , consistent with a role of fzr in preventing aberrant cellular hyperplasia in the adult pylorus ( Figure 5C , E–G ) . In further examination of RasV12 vs . fzr RNAi +RasVV12 animals , we noticed an important difference with regards to intestinal epithelial architecture . Normally , the pyloric epithelia consists of a single cylindrical layer surrounding the hindgut lumen ( Figure 5I–I’’ , Figure 5—figure supplement 1A–A’ ) . The tissue expansion caused by RasV12 expression does not substantially distort this epithelial architecture ( Figure 5J–J’ ) . In contrast , in fzr RNAi +RasVV12 animals , we noticed severe recurring abnormalities . Specifically , 23 . 3% ( ±3 . 3% SEM ) of fzr RNAi +RasVV12 animals examined show severe distortions of the intestinal epithelial architecture , either into or away from the intestinal lumen ( Figure 5K–K’ , L , Figure 5—figure supplement 1B–C’’’ ) . These distortions were not caused by increased tissue area in fzr RNAi +RasVV12 animals , as adult pyloric area actually decreases in these animals relative to animals expressing RasV12 alone ( Figure 5—figure supplement 1D , see Discussion for proposed mechanism ) . One possible consequence of the altered intestinal epithelial architecture that we observe is a compromised epithelial barrier . We thus tested whether RasV12 and/or fzr RNAi expression alters intestinal barrier function . Using the well-established ‘Smurf’ assay ( Rera et al . , 2011; see Materials and methods ) we observed that only fzr RNAi +RasVV12 animals show substantial permeability of the epithelial barrier . We note that the frequency of fzr RNAi +RasVV12 animals with a permeable epithelial barrier ( 17 . 33% ± 3 . 18% SEM , Figure 5M ) is very similar to the proportion of animals of this same genotype with epithelial architecture distortions ( Figure 5L vs . M ) , implying that these defects may be related . Thus , under conditions of prolonged growth signaling , endocycles protect the pyloric epithelium against hyperplasia , distortions in epithelial lumen architecture , and gut barrier permeability ( Figure 5N ) .
To date , many studies have viewed ploidy increases following tissue injury as either maladaptive or neutral . In the vertebrate heart , ploidy increases cause organ hypertrophy and overgrowth ( González-Rosa et al . , 2018; Li et al . , 1996; Senyo et al . , 2013; Steinhauser and Lee , 2011 ) , a disease phenotype that increases heart wall thickness while decreasing valve size ( Bonow et al . , 2011 ) . Unlike the vertebrate heart , tissue overgrowth does not commonly occur in the injured mammalian liver , which undergoes differing degrees of hepatocyte ploidy increases upon injury . However , likely due to competing contributions from mitosis of diploid hepatocytes or from liver stem cells , as well as an inherent flexibility in the cell cycles used to regenerate the liver ( Chen et al . , 2012; Diril et al . , 2012; Lazzeri et al . , 2018; Miyaoka et al . , 2012; Nevzorova et al . , 2009; Pandit et al . , 2012 ) a role for ploidy in the repairing liver remains unclear . Relative to other tissue models involving ploidy increases , the adult pyloric response is simpler , as it does not occur in parallel to cell-cell fusion ( Losick et al . , 2013; Losick et al . , 2016 ) , divisions of diploid stem cells ( Lin et al . , 2018; Wang et al . , 2015 ) and does not occur in cells that are already programmed to become polyploid regardless of injury ( Tamori and Deng , 2013 ) . Using our simpler system , we were able to more directly ask whether polyploidy has any advantage in tissue repair and tissue overgrowth , or rather is an aberrant response to injury . Our work here suggests that polyploidy can represent a regulated and potential beneficial tissue injury response . First , using the simplicity of our system , we discovered that ploidy increases in the repairing pylorus are perfectly tuned to replace the pre-injury number of genomes . This result suggests that the number of post-injury endocycles is tightly regulated and can be responsive to injury severity . While diploid mitotic cycles replace pre-injury genomes in the larva , endocycles do the same in the adult . Further , reverting adult endocycles to mitotic cycles in the injured adult also leads to the correct number of pre-injury cells with no ploidy increase , suggesting that both mitotic cycles and endocycles are able to accurately replace lost tissue mass/genome content . Future work can determine whether endocycling is triggered by mechanical stress , as suggested by recent work in the repairing zebrafish epicardium ( Cao et al . , 2017 ) . Additionally , it will be interesting to determine whether there is any advantage to skipping late S-phase replication during endocycles , which occurs in wild-type but not in fzr animals during repair . Our results also mirror findings in the injured Drosophila abdomen where ploidy matching was also observed . However , our analysis is not complicated by parallel cell-cell fusion events ( Losick et al . , 2016 ) . Second , using a RasV12 model , we present evidence that endocycles enable the pylorus to resist tissue malformation and permeability phenotypes under conditions of sustained tissue growth signaling . Our findings in the post-developmental pylorus may mimic the finding in developing glia of the Drosophila blood-brain barrier , in which endocycles preserve tissue integrity during growth ( Unhavaithaya and Orr-Weaver , 2012; Von Stetina et al . , 2018 ) . As activated Ras/Raf signaling reproduces the injury responsive cell cycles in our system and in injured cardiac tissues , this model may suggest that endocycles are employed in some tissues to maintain epithelial integrity in the face of stresses such as injury or excess growth signaling . One form of tissue stress may come from cell shape changes during mitosis , which depend on regulated changes in cell adhesion ( Kunda et al . , 2008; Lancaster et al . , 2013 ) . Future work can determine if such mitotic shape changes are incompatible with preserving pyloric tissue architecture . We note that the pylorus has a potentially important similarity to other tissues exhibiting injury-induced ploidy increases such as the mammalian heart , liver , and kidney: these tissues are all normally quiescent or exhibiting very low cell turnover . In such tissues , stem cell-based divisions are not quickly rejuvenating the cell population , and thus cells with de novo mutations ( such as dominant Ras mutations ) may accumulate during aging . As a result , these long-lived tissues may require a strategy to minimize mitosis of any cell . While such quiescent tissues are then less reliant on mitosis to restore tissue mass , they can still employ ploidy-increasing cycles to accomplish the same goal . In line with this idea , we note that of any organ , the heart has one of the lowest incidences of cancers ( Bisel et al . , 1953; Leja et al . , 2011 ) , and a recent study also showed a tumor-protective role of polyploidy in the liver ( Zhang et al . , 2017 ) . Future work can determine whether tumor protection or preservation of epithelial integrity is a general property of tissues prone to injury- or growth signal-induced ploidy increases . Our results show that , while both injury-induced and developmental endocycles rely on Fzr , the upstream regulation in the injury context does not involve regulation of Notch , a well-known developmental endoreplication regulator in flies and mammals ( Cornejo et al . , 2011; Deng et al . , 2001; Domanitskaya and SchupbachSchüpbach , 2012; Mercher et al . , 2008; Micchelli and Perrimon , 2006; Ohlstein and Spradling , 2006; Poirault-Chassac et al . , 2010; Shcherbata et al . , 2004; Sun and Deng , 2007; Von Stetina et al . , 2018 ) . Future work in our system can determine what factors converge on fzr regulation and the control of entry into mitosis or endocycles following pyloric injury . One candidate is the ecdysone steroid hormone receptor , which peaks in activity during metamorphosis , close to when the pyloric cell cycle injury response changes . It is somewhat surprising that loss of fzr alone is sufficient to restore pyloric mitosis , as the cyclin dependent kinase 1 ( Cdk1 ) activator Cdc25/String is often required ( Schaeffer et al . , 2004; Von Stetina et al . , 2018 ) . Thus , in the pylorus , Cdk1 may be primed for activity but is kept inactive without its cyclin binding partners , which are negatively regulated by high levels of APC/Fzr . More broadly , further study of the mechanisms that alter cell cycle programming after tissue injury may also improve therapeutic efforts to regenerate injured organs . Another question raised by our work is whether there are functional benefits to the altered pyloric cell cycle injury response . As our data suggest , in the normally quiescent adult stage , induced cell proliferation may be more detrimental for tissue integrity than induced endocycles . However , building new tissues during intestinal development necessitates the use of cell division . For poorly understood reasons , the Dipteran intestine is completely remodeled during metamorphosis , and cell divisions from the pylorus are the source of the majority of new hindgut cells ( Aghajanian et al . , 2016; Bodenstein , 1950; Fox and Spradling , 2009; Robertson , 1936; Sawyer et al . , 2017; Takashima et al . , 2008 ) . This need for wholescale organ remodeling necessitates the ability to divide during larval/early pupal stages . Once organ remodeling is complete , the pylorus , which lacks stem cells , then ceases to divide ( Fox and Spradling , 2009; Sawyer et al . , 2017 ) . Our data suggest that at this point , high fzr levels prevent any future cell division . Following injury , there may be an increased negative regulation of mitotic entry , as we and others have shown that polyploid cell division causes genomic instability ( Davoli et al . , 2010; Duncan et al . , 2010; Fox et al . , 2010; Hassel et al . , 2014; Schoenfelder et al . , 2014; Storchová et al . , 2006 ) . Future work can determine if mitotic integrity is compromised during mitotic re-entry in the adult pylorus . In addition to the new tissue injury biology presented here , we also introduce DEMISE as a method for dual control over tissue injury and transgene expression . We show that this system is amenable to injury studies in the imaginal disc , a widely used tissue injury system ( Harris et al . , 2016; Smith-Bolton et al . , 2009 ) . Other tissues with injury responses that we have not tested but that would benefit from DEMISE as a tool include the stem cell-based midgut ( Apidianakis and Rahme , 2009; Buchon et al . , 2009; Chatterjee and Ip , 2009; Jiang et al . , 2009 ) , the brain ( Moreno et al . , 2015 ) , the abdomen ( Losick et al . , 2013; Losick et al . , 2016 ) , the muscle ( Chaturvedi et al . , 2017 ) and ovarian follicle cells ( Tamori and Deng , 2013 ) . Beyond studies of tissue injury , our inducible reaper transgene can be used in studies of apoptotic signaling , such as the regulation of ‘undead’ cells which contain active caspase signaling but persist and influence the behavior of neighboring cells ( Deveraux et al . , 1998; Fan and Bergmann , 2014; Hay et al . , 1994; Ryoo et al . , 2004 ) . While many dual transgene systems in flies combine Gal4 with either LexA or Q systems ( Kockel et al . , 2016; Lai and Lee , 2006; Potter et al . , 2010 ) , there are vastly more currently available Gal4 and partner UAS lines than for these other systems , making our system more immediately employable for the most cell types/transgenes . Additionally , the use of a single Gal4 promoter for both injury and transgene activation provides tissue specificity and ensures both injury and transgenes are expressed in the same population of cells . However , use of the same promoter to drive injury and transgenes also creates a limitation to our system if one wants to study the response of cell population A on injury to population B . In such case , combining LexA/Q system injury with Gal4 transgene induction ( or vice versa ) would be necessary . Beyond Drosophila , we note that our system could be adapted for use in other organisms , by adapting existing FLP and Cre-mediated injury and gene knockout models . In summary , this study highlights the utility of the Drosophila pylorus in identifying regulation and purposes of specific cell cycle programs induced by tissue injury . Future work in this system can continue to illuminate the role and regulation of variant cell cycles and polyploidy in tissue injury biology .
Full genotypes are described at flybase . org . Except where indicated , flies were raised at 25C on standard Drosophila media ( Archon Scientific , Durham ) . For larval experiments , animals were collected at wandering 3rd instar stage ( L3 ) . All adults dissected were older than 4 days . The following publicly available stocks were used in the study: wg > LacZ ( #BS 1567 ) , act >> LacZ ( #BS 6355 ) , hsFLP12;Sco/CyO ( #BS 1929 ) , UAS fzr RNAi ( #VDRC 25550 ) , fzrG0418 ( #BS12297 ) , fzrG0326 ( BS#12241 ) and UAS-Ras85DV12 ( #BS ) where BS = Bloomington Stock Center and VDRC = Vienna Drosophila Resource Center . The following fly stocks were generously gifted to us: byn >Gal4 ( Shigeo Takashima- UCLA , Singer et al . , 1996 ) , fz3RFP ( Andrea Page-McCaw , Vanderbilt ) , P ( Vha16-1CA06708 ) /CyO ( Carnegie Protein Trap Collection ) , UAS-hid , UAS-reaper ( Zhou et al . , 1997 , Toshie Kai , Temasek Laboratory ) , UAS N RNAi and UAS N-DN ( Sarah Bray , University of Cambridge ) . All UAS transgenes were induced by byn > Gal4 or en > Gal4 . All temporally controlled UAS-transgene experiments involved culturing flies with a Gal80ts repressor driven by the tubulin promoter in the genetic background at 18C except during the desired period of expression , during which they were transferred to 29C as described in the appropriate figure panel and accompanying text . For adult injury , after eclosion , all animals were aged 4–7 days at 18C before injury as done previously ( Fox and Spradling , 2009; Losick et al . , 2013; Sawyer et al . , 2017 ) . For larval injury , we shifted animals at L3 stage to 29C for 16 hr , a sub lethal level of injury induction . Heat shock conditions to induce FLP were performed at 37C by heat shocking either three times for 45 min each with a 1–2 hr recovery in between ( DEMISE experiments ) or a single time for 25 min ( cell lineage tracing experiments ) . The DEMISE plasmid was generated by restriction digest of pUAST > Stop > mCD8-GFP plasmid ( addgene #24385 , Potter et al . , 2010 ) using XhoI and StuI . An insert containing a rpr-cDNA , SV40NLS and SV40PolyA sequences was then synthesized to make a final vector pUAST-FRT-Stop-FRT-rpr ( Bio Basic Inc , New York ) . The construct was then injected to embryos and six fly lines were selected to include the construct on either the 2nd or 3rd chromosome . DEMISE flies ( pUAST-FRT-Stop-FRT-rpr/CyO; byn >Gal4 , Tub >Gal80/TM6 ) were raised at 25C unless otherwise noted . Flies were crossed to hsFLP[12] to induce cell death . Apoptotic death was observed in the wing disc in less than 48 hr , while in the hindgut it was observed in less than 24 hr . Markers for ileal cells ( Vha16[Ca06708] ) and the hindgut-midgut boundary ( wg-LacZ or fz3-RFP ) were used to specify the pylorus from adjacent regions of the gut . BrdU labeling and colcemid ( Sigma ) feeding were performed as in Sawyer et al . ( 2017 ) . For cell death quantifications , TUNEL was performed using in situ cell death detection kit ( Roche , Basel , Switzerland ) as described previously ( Schoenfelder et al . , 2014 ) . For antibody staining , tissues were dissected in 1X PBS and immediately fixed in 1XPBS , 3 . 7% paraformaldehyde , 0 . 3% Triton-X for 30–45 min . Immunostaining was performed as described in Sawyer et al . ( 2017 ) . The following antibodies were used in this study: Fasciclin III ( FasIII , DSHB , 7G10 , 1:50 ) , Beta-Galactosidase ( Abcam , ab9361 , 1:1000 ) , DCP1 ( Cell Signaling , Asp261 , 1:1000 ) , BrdU ( Serotec , 3J9 , 1:200 ) , Phospho-Histone H3 ( Cell Signaling , #9706 , 1:1000 ) , Centrosomin ( generous gift from Nasser Rusan lab , NIH/NHLBI , 1:10 , 000 ) . All secondary antibodies used were Alexa Fluor dyes ( Invitrogen , 1:500 ) . Tissues were mounted in Vectashield ( Vector Laboratories Inc . ) . Images were acquired with the following: an upright Zeiss AxioImager M . 2 with Apotome processing ( 10X NA 0 . 3 EC Plan-Neofluar air lens or 20X NA 0 . 5 EC Plan-Neofluar air lens ) or inverted Leica SP5 ( 40X NA 1 . 25 HCX PL APO oil immersion ) . Image analysis was performed using ImageJ ( Schneider et al . , 2012 ) , including adjusting brightness/contract , Z projections , cell counts , cell area and integrated density quantification . DNA FISH was performed as in Beliveau et al . ( 2014 ) . Cy5-labeled oligo-probes to the AACAC repeat were synthesized by Integrated DNA Technologies . For ploidy measurements , guts were dissected in 1X PBS and prepared as described previously ( Losick et al . , 2013 ) . Total tissue ploidy was calculated by timing the average animal ploidy per injury condition with the average recovered cell number ± STERR ( DNA content * cell count ) . For measuring epithelial barrier integrity , we adapted the established Smurf assay ( Rera et al . , 2011 ) . Flies were raised at 18C on standard Drosophila media ( Archon Scientific , Durham ) until 4–7 days post eclosion . Flies were then shifted to vials containing standard Drosophila media mixed with 0 . 5% Bromophenol blue , and were then kept at 29C and scored for gut permeability every 24 hr for 14 days . ( C ) refers to the haploid DNA content . We define a ‘polyploid’ cell as a somatic cell that contains more than the diploid number of chromosome sets . We define the ‘endocycle’ as any programmed cell cycle in which the genome reduplicates without any sign of mitotic entry . We define ‘endoreplication’ as a broader term encompassing any truncated cycle that generates polyploid cells . We note the use of other terms in the literature that refer to similar processes , and also note that such terms are not always used consistently . We thus adopt the terms used most consistently in the current literature ( Edgar et al . , 2014 ) . Statistical analysis was performed using GraphPad Prism 7 . Statistical tests and adjustments of P-values for multiple comparisons are detailed in figure legends . For all tests , P value reporting is as follows: ( p>0 . 05 , ns ) ; ( p<0 . 05 , * ) ; ( p<0 . 01 , ** ) ; ( p<0 . 001 , *** ) ; ( p<0 . 0001 , **** ) . Regression analysis for ploidy measurements and cell number was done using the formula log2 ( ploidy ) x cell number , and a Pearson correlation analysis . | How does an injured organ replace cells that have died as a result of the damage ? Making new cells might seem like the best option , because this would restore the organ to how it looked before the injury . To make new cells , existing cells in the organ divide . But not all organs make new cells to repair damage . Instead , some organs make their remaining cells bigger – a process known as hypertrophy – to fill the space created by the injury . Cohen et al . have now developed a technique to investigate which method of repair a damaged organ uses . The technique uses genetic engineering to create an injury in a specific tissue in fruit flies , while also altering the activity of other genes that might affect how the tissue responds to the injury . Using the technique to study injuries to part of the gut that remains the same throughout a fly's life revealed that fly larvae repair this damage by creating new cells . However , adult flies repair the same injuries using hypertrophy . Cohen et al . found that a gene known as ‘fizzy-related’ helps to control how the organ repairs damage . The fizzy-related gene produces a protein that stops cells dividing , which forces the cells to grow to repair any injuries to the organ . Adult flies that lacked the gene repaired their guts through cell division instead of by hypertrophy . This did not affect how well minor injuries to the gut were repaired . However , under conditions of more extreme tissue injury cell division distorted the gut and led to leakiness of gut contents . Hypertrophy has been seen in injured human organs , including the heart , liver and kidneys . This was thought to be an abnormal response , but the results presented by Cohen et al . suggest that hypertrophy may instead help to protect the organs during repair . Further research into the role of hypertrophy could reveal ways to regenerate damaged organs , perhaps by targeting the activity of the fizzy-related gene . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"developmental",
"biology"
] | 2018 | Fizzy-Related dictates A cell cycle switch during organ repair and tissue growth responses in the Drosophila hindgut |
Interferon ( IFN ) inhibits HIV replication by inducing antiviral effectors . To comprehensively identify IFN-induced HIV restriction factors , we assembled a CRISPR sgRNA library of Interferon Stimulated Genes ( ISGs ) into a modified lentiviral vector that allows for packaging of sgRNA-encoding genomes in trans into budding HIV-1 particles . We observed that knockout of Zinc Antiviral Protein ( ZAP ) improved the performance of the screen due to ZAP-mediated inhibition of the vector . A small panel of IFN-induced HIV restriction factors , including MxB , IFITM1 , Tetherin/BST2 and TRIM5alpha together explain the inhibitory effects of IFN on the CXCR4-tropic HIV-1 strain , HIV-1LAI , in THP-1 cells . A second screen with a CCR5-tropic primary strain , HIV-1Q23 . BG505 , described an overlapping , but non-identical , panel of restriction factors . Further , this screen also identifies HIV dependency factors . The ability of IFN-induced restriction factors to inhibit HIV strains to replicate in human cells suggests that these human restriction factors are incompletely antagonized . Editorial note: This article has been through an editorial process in which the authors decide how to respond to the issues raised during peer review . The Reviewing Editor's assessment is that all the issues have been addressed ( see decision letter ) .
The HIV-1 pandemic resulted from a series of successive cross-species transmissions of primate lentiviruses . Simian Immunodeficiency Virus ( SIV ) transmission from African Old World primates to chimpanzees yielded the recombinant virus SIVcpz , which ultimately crossed into humans ( Sharp and Hahn , 2011 ) . Successful replication of lentiviruses in a new host species required adaptation to restriction factors in the new host ( Etienne et al . , 2015; Etienne et al . , 2013; Kirmaier et al . , 2010 ) . Restriction factors that target primate lentiviruses include TRIM5alpha , MxB , Tetherin , SAMHD1 , the APOBEC3 family of cytidine deaminases ( Malim and Bieniasz , 2012 ) and more recently described factors such as SERINC3/5 , Zinc Antiviral Protein ( ZAP ) , GBP5 , SLFN11 , LGALS3BP ( 90K ) , the HUSH complex ( Chougui et al . , 2018; Krapp et al . , 2016; Li et al . , 2012; Lodermeyer et al . , 2013; Rosa et al . , 2015; Takata et al . , 2017; Usami et al . , 2015; Yurkovetskiy et al . , 2018 ) , as well as nearly 200 other proposed factors ( reviewed in ( Gélinas et al . , 2018 ) . HIV-1 has evolved accessory proteins that degrade many host restriction factors ( Duggal and Emerman , 2012 ) . Further , mutations preventing recognition by restriction factors , such as evolution of low CG dinucleotide content in the HIV-1 genome ( Takata et al . , 2017 ) or mutations in capsid ( Kirmaier et al . , 2010; Wu et al . , 2013 ) , represent another mechanism of escape . Many restriction factors that target HIV-1 are induced by type I Interferon ( IFN ) and are therefore Interferon-Stimulated Genes ( ISGs ) . Interferon has been implicated in at least partial control of HIV replication in chronically-infected individuals treated with IFN ( Asmuth et al . , 2010; Azzoni et al . , 2013 ) as well as in SIV-infected rhesus macaques ( Sandler et al . , 2014 ) . In contrast , IFN levels have also been correlated with higher viral load and decreased CD4 T cell counts in HIV-infected individuals ( Hardy et al . , 2013 ) . Further , it appears that ISG expression exerts changing selective pressure on HIV evolution in vivo since transmitted/founder ( T/F ) strains are relatively resistant to IFN compared to viruses isolated later in infection ( Fenton-May et al . , 2013; Iyer et al . , 2017; Parrish et al . , 2013 ) . It remains to be determined if one dominant ISG mediates all or most of the IFN inhibition , or if a multitude of antiviral ISGs together limit viral replication in response to IFN . The HIV-1 LAI strain ( HIV-1LAI ) was isolated from a chronically-infected individual ( Wain-Hobson et al . , 1991 ) and is sensitive to type I IFN . Specifically , potent IFNα inhibition of HIV-1LAI can be observed in the THP-1 monocytic cell line ( Goujon and Malim , 2010 ) . MxB , an interferon-induced GTPase that binds to and blocks lentiviral capsids , was identified as an IFNα-induced factor in THP-1 cells ( Goujon et al . , 2013; Haller et al . , 2015; Kane et al . , 2013; Liu et al . , 2013 ) , although the role of MxB in the IFNα-induced inhibition of HIV infection in these cells has been questioned ( Opp et al . , 2015 ) . Restriction factors have previously been discovered through cDNA library screening or by comparing expression of transcript levels in permissive versus non-permissive cells ( Goujon et al . , 2013; Kane et al . , 2013; Neil et al . , 2008; Sheehy et al . , 2002; Stremlau et al . , 2004 ) . More high-throughput approaches to find HIV restriction factors have focused on either overexpression screens to identify broad antiviral ISGs ( Schoggins et al . , 2011 ) or HIV-specific antiviral ISGs ( Kane et al . , 2016 ) . Further , one screen for HIV restriction factors was also performed by transfection of siRNA pools ( Liu et al . , 2011 ) . However , a more complete understanding of the constellation of restriction factors that inhibit HIV in human cells and a more tractable , high-throughput method to discover restriction factors remains to be described . Here we describe a CRISPR/Cas9-mediated gene knockout functional screening method in which lentiviral genomes encoding CRISPR sgRNAs are packaged into budding HIV virions , allowing robust identification of HIV restriction factors and dependency factors in a high-throughput manner . Cas9 endonuclease and sgRNA are delivered to cells in a vector that is modified to be transcribed and subsequently packaged in trans by the infecting HIV virus . Deep sequencing of packaged HIV-CRISPR RNA in nascent HIV virions released from pooled KO cells serves to proxy the efficiency of HIV replication in each genetic knockout . Thereby , our approach allows for targeted gene knockout and a functional assay simultaneously across thousands of genes in a heterogeneous population of cells . Furthermore , as read-out of the functional assay is done of at the level of newly budded viruses , the approach allows for screening of restrictions factors affecting the full HIV life cycle . We find a small panel of ISGs to mediate IFN inhibition of HIV-1 in THP-1 cells , including MxB , TRIM5alpha , IFITM1 and Tetherin . We find that restriction factor sensitivity is , in part , strain-dependent . Further , this approach can as be used to identify HIV dependency factors as we identify CD169 , SEC62 and TLR2 as host factors in THP-1 cells . The results presented here suggest that adaptation of primate lentiviruses to humans is incomplete as the same host restriction factors that block cross-species transmission also play a role in limiting the replication of highly-adapted HIV-1 in IFN-stimulated cells .
IFNα inhibits HIVLAI replication in THP-1 cells 10-fold ( Goujon et al . , 2013 ) . To identify the factor ( s ) mediating the IFNα-induced inhibition of HIV , we designed a novel HIV-based CRISPR screen in which the virus itself serves as a reporter . Cells which lack a dependency factor due to CRISPR-mediated gene knockout will release less virus , whereas cells which lack a restriction factor will produce more virus as compared to control cells which containing single-guide RNA ( sgRNA ) sequences that do not target any human genes , Non-Targeting Controls ( NTCs ) . We engineered a Cas9 and sgRNA-encoding lentiviral vector such that sgRNA-encoding genomic RNA can be packaged in trans by budding HIV virions . Therefore , the normalized abundance of Cas9/sgRNA-encoding genomes themselves are the direct readout for the functional activity of each gene knockout on viral replication . Importantly , this approach allows for assay of effects of gene knockout on a complete round of viral replication . Standard Self-Inactivating ( SIN ) lentiviral vectors contain a deletion in the Long-Terminal Repeat ( LTR ) sequence that encodes transcription factor binding sites in wild type HIV sequences . The lentiCRISPRv2 lentiviral vector contains a SIN LTR and , therefore , full-length genomic lentiviral RNA is not transcribed after integration ( Shalem et al . , 2014a ) . To create a version of the lentiCRISPR vector that can be mobilized after HIV infection , we repaired the 3’ LTR with a complete HIV-1 LTR . Thus , the HIV-CRISPR vector maintains complete HIV-1 LTR sequences on integration . We call this transcription- and packaging-competent construct HIV-CRISPR ( Figure 1A and B ) . Importantly , in addition to internal promoters driving Cas9 and sgRNA transcription , full-length HIV-CRISPR genomic RNA is also produced after transcription from the LTR . More robust replication will occur specifically in cells that have been deleted of an antiviral factor . The enhanced release of antiviral factor-targeting HIV-CRISPR genomes occurs at a higher frequency than control sgRNAs or most sgRNAs in the library that target genes with no effect on HIV replication . Therefore , any sgRNA that targets a gene that inhibits viral replication will be enriched in the virions . Infecting HIV viruses serve to readout efficiency of HIV replication in each knockout cell through deep sequencing of packaged HIV-CRISPR RNA in secreted HIV virions . To target genes mediating the IFN inhibition of HIV-1 , we curated a list of potential ISGs from existing microarray and RNA-seq datasets from cell types relevant to HIV-1 infection , including PBMCs , primary CD4+ T cells , monocyte-derived macrophages ( MDMs ) , monocytes and the THP-1 monocytic cell line ( Figure 1—figure supplement 1A and Figure 1—source data 1 ) . Thus , the library is also enriched in genes that are specifically expressed in HIV target cells . For each of the 1905 ISGs present in our library , we selected a total of 8 sgRNA sequences from existing human whole-genome CRISPR/Cas9 libraries ( Figure 1—figure supplement 1B and Figure 1—source data 2 ) ( Doench et al . , 2016; Hart et al . , 2015; Sanjana et al . , 2014a; Shalem et al . , 2014b; Shalem et al . , 2015; Wang et al . , 2015; Wang et al . , 2014 ) . 200 Non-Targeting Control ( NTC ) sgRNA sequences that are not predicted to target any loci in the human genome were also included ( Shalem et al . , 2015 ) ( Figure 1—figure supplement 1B and Figure 1—source data 2 ) . In total 15 , 348 unique sgRNA sequences were assembled into the HIV-CRISPR backbone to create the Packageable ISG Knockout Assembly or PIKAHIV library ( Figure 1B ) . The enrichment or depletion of sgRNA sequences in the viral RNA ( vRNA ) as compared to the genomic DNA ( gDNA ) of the cells is quantified through sequencing of sgRNA sequences both in released HIV particles and integrated into the cellular genomic DNA . sgRNAs that target antiviral genes ( restriction factors ) are overrepresented in viral supernatants due to more robust viral replication specifically in these KO cells . Conversely , sgRNAs that target dependency factors are depleted in viral supernatants due to decreased viral replication specifically in these KO cells . To perform the screen , 8 × 106 THP-1 cells were transduced with the PIKAHIV library at an MOI <1 ( MOI = 0 . 6 ) to create a population of cells with single HIV-CRISPR integrations at >500X coverage . THP-1/PIKAHIV cells were split in two independent replicates and left untreated or treated with IFNα overnight . Each replicate was then infected with HIV-1 at a dose that infects 50% of cells without IFNα treatment . Secreted virus was collected 3 days after infection , and sgRNA sequences encoded by HIV-CRISPR genomic RNA packaged into budding HIV virions were amplified by RT-PCR and quantitated through deep sequencing ( Figure 1C ) . THP-1/PIKAHIV cells were collected in parallel at the time of viral supernatant harvest and the genomic DNA ( gDNA ) was also deep sequenced . We compared the relative enrichment of HIV-CRISPR sgRNA sequences in the viral RNA ( vRNA ) to the genomic DNA ( gDNA ) to find enriched and depleted sgRNA sequences ( Figure 1D and Figure 1—source data 3 ) . Relative to the NTCs ( Figure 1D , gray circles ) , there were a number of sgRNA sequences that were either enriched ( Figures 1D and 500 most enriched sgRNAs in cyan ) or depleted ( Figures 1D and 500 most depleted sgRNAs in orange ) in the viral supernatant as compared to the NTCs . Since each gene in the PIKAHIV library is targeted by eight individual sgRNAs , we analyzed the enrichment across all sgRNAs for a gene using the MAGeCK package across both duplicates ( Li et al . , 2014 ) . We identified the type I IFN pathway genes , STAT1 , IFNAR1 , STAT2 and IRF9 as the highest-scoring hits ( magenta in Figure 1E and Figure 1—source data 4 ) . Therefore , the PIKAHIV screen functioned as designed: cells in which IFN signaling is compromised exhibited increased viral production and , therefore , enriched HIV-CRISPR representation of sgRNAs in the secreted HIV virions . After the IFN pathway genes , the Zinc Antiviral Protein ( ZAP ) and its modifier TRIM25 were the next to highest scoring hits . ZAP is an antiviral effector that has potent activity against alphaviruses as well as moderate activity against retroviruses ( Bick et al . , 2003; Gao et al . , 2002; Kerns et al . , 2008; Takata et al . , 2017 ) . TRIM25 is a gene known to modify ZAP’s antiviral activity ( Li et al . , 2017 ) . More recently , it was shown that ZAP blocks virus replication by degrading transcripts with a high CG dinucleotide content ( Takata et al . , 2017 ) . We also find NEDD4 Binding Protein 1 ( N4BP1 ) , a poorly-characterized inhibitor of the E3 ligase ITCH in mice ( Oberst et al . , 2007 ) that has not been previously known for antiretroviral activity ( Figure 1E; cyan ) . N4BP1 encodes RNA binding domains and is proposed to have RNase activity ( Anantharaman and Aravind , 2006 ) . To ask if ZAP or N4BP1 inhibit HIV replication we generated ZAP and N4BP1 knockout ( KO ) cell lines by electroporating crRNA/Cas9 complexes ( crRNPs ) into THP-1 cells followed by single-cell cloning . Knockout of either gene only very modestly increased infection by HIV-1LAI ( Figure 2—figure supplement 1A; Figure 2—figure supplement 1B; Figure 2—figure supplement 1C; Figure 2—figure supplement 1D ) . Therefore , the IFN-induced restriction factors that potently inhibit HIV in THP-1 cells were not identified in this initial screen . Analysis of the CG dinucleotide content across the HIV-CRISPR genome shows high levels of CG dinucleotides , particularly in the Cas9 and Puromycin resistance ORFs , that are potential targets for ZAP-mediated RNA degradation ( Figure 2A ) . Given its role in degradation of RNA with high CG content , we hypothesized that ZAP could inhibit the full-length HIV-CRISPR genomic RNA that is packaged into budding virions rather than the wt HIV genome . Thus , we determined whether or not ZAP KO allows for increased packaging of the HIV-CRISPR vector in viral particles released from cells by measuring both wild type HIV-1LAI genomes ( HIV-1LAI; black in Figure 2B ) and HIV-CRISPR genomes ( HIV-CRISPR; gray in Figure 2B ) with a ddPCR assay . Indeed , we found enhanced packaging of HIV-CRISPR genomes relative to wild type HIV-1LAI genomes in the viral supernatant in cell clones with no ZAP expression ( Figure 2B – 10 . 5% in wt THP-1 cells; 24 . 8% and 31 . 6% in the ZAP-KO clonal lines ) . Therefore , to circumvent the inhibitory effects of ZAP on the HIV-CRISPR vector , we repeated the PIKAHIV screen in two ZAP-KO THP-1 clonal cell lines ( THP ZAP-KO #9* and #11* in Figure 2—figure supplement 1A ) . As expected for a screen in ZAP-KO cells , ZAP was no longer a significantly-scoring hit in the screen ( rank # 1647/3812 in combined ZAP-KO screen data; Figure 2—source data 1 ) . In addition , there was also no enrichment of N4BP1 or TRIM25 in the ZAP-KO screens ( rank # 3789/3812 and 3090/3812 in combined ZAP-KO screen data; Figure 2—source data 1 ) , suggesting that the inhibitory activity of N4BP1 and TRIM25 in the HIV-CRISPR screen are ZAP-dependent . To ask if N4BP1 is a novel modifier of ZAP antiviral activity we screened ZAP-KO and N4BP1-KO cell lines for changes in susceptibility to Semliki Forest Virus ( SFV ) infection . SFV is an alphavirus that is highly-susceptible to ZAP-mediated inhibition ( Bick et al . , 2003 ) . While ZAP KO rescued SFV replication significantly , we saw no effect of N4BP1 KO on SFV replication ( Figure 2—figure supplement 1E ) . Therefore , N4BP1 may be a novel modifier of ZAP antiviral activity against lentiviruses . However , it does not appear to play a role in ZAP inhibition of alphaviruses . Without ZAP-mediated inhibition of the HIV-CRISPR vector in ZAP-KO THP-1 cells we would expect the PIKAHIV screen to more efficiently identify ISGs inhibiting HIV infection . Since ZAP-KO results in increased HIV-CRISPR expression and/or packaging ( Figure 2B ) we would expect increased HIV-CRISPR representation in viral supernatants and , therefore , better sampling and therefore better correlation in read counts across duplicate infections . We assessed whether or not ZAP knockout improves performance of the HIV-CRISPR screen by analyzing read counts across duplicates in the two independent ZAP-KO THP-1 clonal lines and compared the results to the screen performed in wild-type THP-1 cells . Indeed , there was better correlation in sgRNA representation across replicates performed in ZAP-KO THP-1 cells as compared to control THP-1 cells ( Figure 2C; r2 = 0 . 92 and 0 . 94 for the ZAP-KO screens as compared to r2 = 0 . 87 for the screen in wild type THP-1 cells ) . Further , an analysis specifically across the four genes that are well-described components of the type I IFN pathway also showed increased Gene Scores in the ZAP-KO THP-1 clonal lines ( Figure 2D and Figure 1—source data 4 and Figure 2—source data 1 ) . Therefore , deletion of ZAP-mediated inhibition from THP-1 cells improves performance of the HIV-CRISPR screen . By multiplying gene scores from both ZAP-KO screens ( Figure 2E; MAGeCK score on x-axis ) we identified a list of candidate hits . To ask which genes are most likely to contribute specifically to the IFN-mediated inhibition of HIV-1 , we calculated the level of IFN induction of each of the top hits from an existing THP-1 microarray dataset ( Figure 2E; IFN log2FC on y-axis and Figure 2—source data 1 ) . No hit scored as highly as the type I IFN pathway genes ( magenta in Figure 2E ) . Therefore , multiple genes , rather than a single ISG , are responsible for the IFN-mediated inhibition of HIV infection in THP-1 cells . Further , a small subset of genes , including MxB , IFITM1 , Tetherin , TRIM5alpha , UBE2L6 , LGALS3BP/90K and SAMD9L , are candidate restriction factors mediating the IFN inhibition of HIV-1 in THP-1 cells . MxB , IFITM1 , Tetherin and TRIM5alpha are the most significantly-scoring hits in the PIKAHIV screen that are also highly-induced by IFN ( Figure 2E ) . All have well-described anti-lentiviral functions ( Goujon et al . , 2013; Kane et al . , 2013; Liu et al . , 2013; Lu et al . , 2011; Malim and Bieniasz , 2012; Neil et al . , 2008; Stremlau et al . , 2004 ) . Thus , the PIKAHIV screen identified IFN-induced restriction factors in a massively-parallel approach assaying all gene targets simultaneously in pools of knockout cells . To determine the relative importance of MxB to the IFN-induced block to infection , we created MxB KO THP-1 cells . MxB was deleted from THP-1 cells by transduction with a lentiCRISPRv2 MxB-targeting construct followed by single-cell cloning . Deletion of MxB expression was confirmed through western blot of IFN-treated clonal MxB-KO lines ( Figure 3A ) . On creating clonal populations of THP-1 cells , we observed substantial heterogeneity across clonal lines of THP-1 cells ( compare infection levels in NTC clonal lines in Figure 3B ) . Therefore , we infected many clonal NTC and MxB-KO cell lines in parallel . Infection of MxB-KO cells confirms that MxB plays a major role in the IFN block to infection as there was rescue of the IFN effect as compared to controls ( Figure 3B and C; the Fold Inhibition in MxB-KO cells is close to 1 ) . Therefore , MxB is a dominant , early-acting ISG inhibiting HIV replication in THP-1 cells . The role of MxB in restriction of HIV has recently been questioned ( Opp et al . , 2015 ) . However , in that study restriction against VSV-G pseudotyped HIV was measured rather than against a virus with a wildtype HIV envelope . Viral entry is a key target of potent IFN-mediated restriction , specifically by ISGs such as IFITMs , a family of 5 membrane-resident antiviral genes in humans with broad antiviral effects ( Shi et al . , 2017 ) . IFITMs restrict viruses that enter cells by fusion at the plasma membrane or in the endosome . We hypothesized that sensitivity to MxB restriction may be dependent on the viral envelope since our previous work has shown that restriction of lentiviruses using distinct entry pathways are differentially affected by ISGs ( Roesch et al . , 2018 ) . Therefore , we tested the MxB-KO cells for IFN-mediated restriction of VSV-G pseudotyped HIV-1 . We found that while the IFN inhibition in the MxB-KO clonal lines is significantly lower than that of control clonal lines ( Figure 3D; p = 0 . 014 unpaired t test ) , there is still a large inhibition of replication of VSV-G pseudotyped HIV-1 by IFNα ( Figure 3D; 53-Fold ) . Thus , one or more ISGs induced by IFNα potently block VSV-G mediated entry in THP-1 cells independent of MxB . To ask what factors mediate this block , we repeated the HIV-CRISPR screen with VSV-G pseudotyped HIV-1 in THP-1 cells that were knocked out for ZAP . In contrast to the screen with wildtype virus , the antiviral proteins IFITM3 is the most significantly-scoring hit , along with IFITM1 , IFITM2 , TRIM5alpha and Tetherin ( Figure 3E ) . MxB ranks as hit #74 in this screen ( Figure 3—source data 1 ) suggesting that MxB does still inhibit VSV-G pseudotyped virus but that its effect is largely masked by more potent inhibition of the VSV-G envelope by IFITM3 and other ISGs . This suggests that IFITMs are the dominant IFN-induced blocks to replication when HIV-1 is pseudotyped with the VSV-G envelope . Significant overlap in sequence across IFITM orthologues complicates interpretation of the screen data in terms of which IFITMs are most important , as some sgRNAs in our library likely target multiple IFITM loci . However , these results show that while MxB does play a role in the IFN-mediated inhibition of VSV-G pseudotyped HIV-1 viruses ( compare MxB-KO cells to NTCs in Figure 3D ) , this effect is masked by dominant IFITM inhibition of these VSV-G pseudotyped viruses ( Figure 3E ) . Similar results have been reported recently elsewhere ( Xu et al . , 2018 ) . Therefore , viral entry route impacts restriction factor sensitivity and the role of MxB in IFN inhibition of HIV-1 in THP-1s is partially-masked by other ISGs when HIV-1LAI is pseudotyped by VSV-G . We were surprised to find TRIM5alpha and Tetherin in this screen as HIV-1 is thought to be highly-adapted to these human restriction factors . To assay the contribution of each of these ISGs to IFN inhibition of HIV in THP-1 cells , we measured viral replication in THP-1 KO pools . Significant knockout of each gene target was assayed by ICE editing analysis of amplified genomic loci ( Figure 4A , left panel ) or through cell surface staining for Tetherin ( Figure 4A , right panel ) . Pretreating cells with IFNα shows ~ 7 fold inhibition of infection in the control NTC cell pools ( Figure 4B and C ) while IFN-mediated inhibition of HIV was significantly lower in MxB , TRIM5alpha and IFITM1 KO lines than in NTCs pools ( Figure 4B; MxB_1 = 2 . 6 fold , MxB_2 = 2 . 5 fold , TRIM5_1 = 3 . 9 fold , TRIM5_2 = 4 . 8 fold , IFITM1_1 = 4 . 7 fold , IFITM1_2 = 6 fold and IFITM1_3 = 4 . 3 fold in Figure 4C; p < 0 . 05 ) . The largest rescue we observed was in the MxB knockout pools ( Figure 4C ) , confirming the key role of MxB in the IFN phenotype . However , TRIM5alpha and IFITM1 also contribute to IFNα inhibition ( Figure 4B and C ) . We find no effect of Tetherin KO on early steps of HIV replication as expected given its role as a late-acting HIV restriction factor ( Figure 4B; Tetherin_1 = 6 fold , Tetherin_2 = 6 . 4 fold in Figure 4C ) ( Neil et al . , 2008; Van Damme et al . , 2008 ) . We also observed a significant IFNα-mediated block to the late stages of the HIV lifecycle ( after translation of the viral Gag protein used to detect infection in Figure 4B ) in both control and MxB-KO cells ( Figure 4D ) . While MxB-KO clonal lines show a decreased IFN effect compared to NTC clonal lines ( compare controls to MxB-KOs in Figure 4E ) , there is still a 4 . 8-fold inhibition of virus released from MxB-KO clonal lines ( magenta in Figure 4E ) . Since Tetherin is a well-characterized late-acting restriction factor and was also a hit in our PIKAHIV screen , we asked if Tetherin is responsible for the late ISG block we observed . The HIV-1 Vpu accessory gene antagonizes Tetherin-mediated restriction ( Neil et al . , 2008 ) . Deletion of Vpu sensitizes HIV-1 to Tetherin restriction . We assayed virus release from control and Tetherin KO cell pools when IFN was added 16 hr after infection to bypass early-acting ISGs ( Figure 4F; NTCs in gray , Tetherin KO pools in dark blue ) . Infection with Vpu-deficient HIV-1 ( HIV-1LAIΔvpu ) in IFN-treated Tetherin-KO cells shows increased virus release as compared to control cells ( Tetherin_1 = 20 . 5 fold , Tetherin_2 = 14 fold in Figure 4F – left panel ) , confirming the late inhibition of Vpu-deficient HIV by Tetherin . Infection of these cell pools with wt HIV-1LAI also shows significantly-increased virus release , suggesting that HIV-1LAI Vpu does not completely antagonize IFN-induced Tetherin in THP-1 cells ( Tetherin_1 = 4 . 4 fold , Tetherin_2 = 2 . 14 fold in Figure 4F – right panel ) . Therefore , Tetherin is a late-acting ISG contributing to IFN inhibition of HIV-1LAI in THP-1 cells . Although we designed our screen specifically to find IFN-induced factors restricting HIV-1 in THP-1 cells , HIV-CRISPR screening can also identify HIV dependency factors . The sgRNA sequences of genes that HIV uses for enhanced viral replication will be depleted in viral supernatants as the virus will be less well able to replicate specifically in these cells ( see Figure 1C ) . Analysis of the negative MAGeCK Gene Scores , representing genes for which sgRNAs were depleted in HIV supernatants , identifies a panel of candidate host factors targeted by the PIKAHIV library that are important for HIV replication ( Figure 5A and Figure 5—source data 1 ) . The top hit is the HIV-1 co-receptor CXCR4 ( Figure 5A ) which is required for entry by HIV-1LAI ( note: sgRNAs targeting the receptor , CD4 , are not present in the PIKAHIV library ) . The next highest scoring hit is SIGLEC1/CD169 ( Figure 5A ) , an HIV attachment factor that has been characterized to facilitate trans infection of CD4 +T cells by DCs through binding to sialylated glycosphingolipids on the HIV particle ( Izquierdo-Useros et al . , 2012; Puryear et al . , 2013 ) . CD169 is upregulated by IFNα in THP-1 cells ( Figure 5B – gray = untreated , purple =+IFN , left panel ) . Our screen only assays cell-autonomous effects suggesting that CD169 also plays a role in cis-infection of monocytic cells , consistent with recent work showing enhanced infection of THP-1 cells by CD169 , specifically in the presence of IFNα ( Akiyama et al . , 2017 ) . Indeed , when CD169 expression is knocked-down ( Figure 5B , right panel ) these cells are less susceptible to infection both in the presence and absence of IFN pretreatment ( Figure 5C ) , although this effect is stronger in presence of IFNα ( 6 . 5-fold vs 4 . 7-fold; Figure 5C ) . Thus , we find that Siglec-1/CD169 is an IFN-induced , HIV dependency factor in THP-1 cells . TLR2 , a toll-like receptor characterized to recognize bacterial PAMPs ( Akira et al . , 2006 ) is the next highest-scoring hit in our dependency factor analysis . We generated TLR2-KO and CXCR4-KO THP-1 cell pools by transduction with lentiCRISPRv2 sgRNA constructs and verified KO through ICE analysis of the CXCR4 genomic locus ( Figure 5D , top ) or cell surface staining of TLR2 ( Figure 5D , bottom; NTCs in gray and TLR2-KO pools in green ) . Infection of these cells with wildtype HIV-1LAI demonstrates lower infection as compared to the controls , although this effect is not as extreme as for CXCR4 KO ( Figure 5E left panel; 31-fold decreased infection for CXCR4-KO lines in orange , 3-fold for TLR2-KO lines in green ) . Of note , infection with VSV-G pseudotyped HIV-1 shows a loss of infectivity similar to wild-type HIV-1 ( Figure 5E bottom right panel ) . Therefore the effect of TLR2 on enhanced infection is independent of viral entry . Conversely , the CXCR4-KO lines are efficiently infected by VSV-G pseudotyped HIV-1 ( orange in Figure 5E , right panel ) as entry occurs independent of the CXCR4 co-receptor . Finally , we investigated the effect of SEC62 , on HIV replication . SEC62 is a component of the protein translocation machinery in the ER membrane . Knockdown of SEC62 by transducing THP-1 cells with two lentiviral shRNA constructs targeting SEC62 showed significant loss of expression as measured by Western Blot ( Figure 5F ) . Infection of these cells with wildtype HIV-1LAI showed decreased levels of infection ( Figure 5G , left panel ) . Therefore , SEC62 is a dependency factor for HIV replication in THP-1 cells . As SEC62 is a component of the machinery that mediates translocation of transmembrane proteins into the ER membrane for targeting to the cell surface , we reasoned that SEC62 knockdown may be affecting cell-surface expression of HIV receptors , co-receptors or other cell-surface markers mediating attachment and/or entry of HIV . Consistent with this hypothesis , infection via an alternative entry pathway via pseudotyping HIV-1 particles with VSV-G , demonstrates equivalent infection in control and SEC62 knockdown cells ( Figure 5G , right panel ) . Analysis of the cell surface expression of the HIV-1 receptor , CD4 , shows that levels of CD4 on the cell surface are decreased in SEC62 knockdown cells ( Figure 5H , left panel ) . Interestingly , we do not observe decreased cell-surface expression of CXCR4 ( Figure 5H , right panel ) , suggesting that the effect of SEC62 knockdown on cell surface proteins is not global but specific to certain transmembrane proteins . To ask if the same restriction factor and dependency factor genes are important for replication of a primary HIV isolate , we repeated the PIKAHIV screening with a CCR5-tropic Clade A HIV-1 virus derived from a molecular clone containing sequences of two primary HIV isolate , HIV-1Q23 and HIV-1BG505 , here called HIV-1Q23 . BG505 ( Haddox et al . , 2018; Poss and Overbaugh , 1999 ) . In order to do this screen , we engineered the THP-1 ZAP KO cells to also express the HIV co-receptor CCR5 ( see Materials and methods ) . Similar to our initial screening with HIV-1LAI the type I IFN pathway genes are the most significantly-scoring genes for HIV-1Q23 . BG505 ( Figure 6A and Figure 6—source data 1 ) . We also find MxB , TRIM5alpha , IFITM1 and UBE2L6 to be among the top hits in this screen in common with HIV-1LAI ( Figure 6A and Figure 6—source data 1 ) . To directly visualize the similarities and differences for restriction factor hits from the screens between the two different HIV strains , we replotted the data of the MaGeCK scores for the top 30 scoring genes for HIV-1LAI on the X axis and for the top 30 scoring genes for HIV-1Q23 . BG505 on the Y axis ( Figure 6B ) . We find IFI16 , a cytoplasmic DNA sensor known to inhibit HIV ( Jakobsen et al . , 2013; Monroe et al . , 2014 ) , as a restriction factor specific to HIV-1Q23 . BG505 ( Figure 6B and Figure 6—source data 1 ) . In contrast , Tetherin is no longer a significantly scoring hit in the HIV-1Q23 . BG505 screen ( rank #1897/3812; Figure 6—source data 1 ) , suggesting that Tetherin plays less of a role in restriction of this primary isolate as compared to HIV-1LAI . Finally , SLC35A2 and NFKB2 , while not IFN-induced in THP-1s , are highly-scoring genes for both viruses ( Figure 6B ) . SLC35A2 , a nucleotide sugar transporter ( reviewed in Hadley et al . , 2014 ) , and NFKB2 , a component of the NFKB transcription factor complex ( reviewed in Vallabhapurapu and Karin , 2009 ) , may both negatively impact HIV replication in an IFN-independent manner . Next , we examined the HIV-1Q23 . BG505 data to identify HIV dependency factors important for infection of this virus ( Figure 6C and Figure 6—source data 1 ) . Similar to our HIV-1LAI screen we find the coreceptor ( in this case CCR5 ) to be one of the most significantly-scoring depleted genes in the HIV-1Q23 . BG505 screen ( Figure 6C and Figure 6—source data 1 ) . SIGLEC1/CD169 is the highest hit in the HIV-1Q23 . BG505 screen ( Figure 6C and Figure 6—source data 1 ) suggesting that this HIV attachment factor is important for infection also by a Clade A primary isolate . In common with HIV-1LAI we also find SEC62 , TLR2 , MYD88 and NFKB1 again as important HIV dependency factors for HIV-1Q23 . BG505 ( Figure 6B and Figure 6—source data 1 ) . In addition , we analyzed the VSV-G pseudotyped HIV-1LAI infection to compare requirements across viral entry routes ( Figure 6—figure supplement 1 ) . The role of SEC62 in CD4 receptor expression is further supported by our finding of SEC62 as a hit for both HIV envelope-containing viruses but not for the VSV-G pseudotyped infection ( compare Figures 5A and 6C and Figure 6—figure supplement 1 ) .
While HIV has evolved antagonism or evasion strategies for restriction factors that limit replication in cells from other species , the results here imply that HIV is not able to completely antagonize or escape some host encoded restriction factors . This is true even for well-adapted HIV strains , such as HIVLAI , that were isolated from chronically-infected individuals with high viral load . Such incomplete antagonism may be due to conflicting evolutionary pressures acting on the HIV genome . HIV proteins must evade restriction factor recognition while maintaining other protein functions , including evasion of other restriction factors and binding to required cellular host factors . Our finding of significant TRIM5alpha restriction in human cells suggests that HIV is still partially-sensitive to human TRIM5alpha-mediated restriction . Similarly , we find that IFN-induced MxB restricts infection in THP-1 cells , consistent with previous work ( Goujon et al . , 2013; Kane et al . , 2013; Liu et al . , 2013 ) . A similar example of incomplete antagonism of human restriction factors can be found in HIV-infected patients in which a signature of APOBEC3 G-to-A hypermutation in integrated proviruses can be observed ( Cuevas et al . , 2015; Sadler et al . , 2010 ) despite the fact that HIV encodes an antagonist , Vif , that targets APOBEC3 proteins for degradation . Both TRIM5alpha and Tetherin are rapidly-evolving genes in primates with described consequences for host adaptation by primate lentiviruses ( Lim et al . , 2010; Liu et al . , 2005 ) . Capsids from diverse primate lentiviruses have adapted to TRIM5alpha alleles in various primates and are variably-restricted by TRIM5alpha orthologs ( Kirmaier et al . , 2010 ) . Selection for capsid mutations that evade TRIM5alpha restriction is a key adaptive step that HIV and related SIVs must make to successfully replicate in a particular primate species ( Wu et al . , 2013 ) . Consistent with a role for TRIM5alpha in humans , TRIM5alpha is active against HIV in Langerhans cells ( Ribeiro et al . , 2016 ) and TRIM5alpha polymorphisms are associated with the clinical course of HIV-1 infection ( van Manen et al . , 2008 ) . Further , CA mutations in HIV-infected individuals have been associated with sensitivity to TRIM5alpha restriction ( Battivelli et al . , 2010; Onyango et al . , 2010 ) . Also of note , most studies of HIV have not been done in the context of IFN despite evidence that TRIM5alpha is highly-IFN upregulated in HIV target cells ( Carthagena et al . , 2009 ) . Thus , our finding of IFN-mediated TRIM5alpha inhibition of HIV represents a potentially-important role of TRIM5alpha particularly during acute infection when IFN levels are high . More stable artificial variants of human TRIM5alpha can inhibit HIV-1 ( Richardson et al . , 2014 ) , suggesting that increased TRIM5alpha levels , such as after IFN induction , may play a role in restricting HIV replication . Evasion of TRIM5alpha restriction may come at the cost of loss of fitness due to other requirements for capsid function within host cells such as uncoating , nuclear import and integration . Further , capsid sequences also mediate evasion of other restriction factors , including MxB , or escape from host CTL responses . Primary isolates of HIV-1 have increased sensitivity to TRIM5alpha that is proposed to be driven by CTL escape variants ( Battivelli et al . , 2011 ) . Therefore , we speculate that this TRIM5alpha sensitivity may underscore the requirement of HIV proteins to balance multiple functions simultaneously to infect human cells . Like TRIM5alpha , escape from IFITMs also appears to be subject to conflicting evolutionary pressures . IFITMs may exert significant selective pressure in vivo as HIV evolves increased susceptibility to IFITMs over the course of infection ( Foster et al . , 2016 ) . In addition to basic required functions of HIV proteins , we further propose that further adaptation of HIV for efficient replication in human cells may not be possible due to constraints on viral evolution imposed simultaneously by multiple , independent restriction factor barriers . MxB , TRIM5alpha , UBE2L6 and IFITM1 are strong hits in both the HIV-1LAI and the HIV-1Q23 . BG505 screens ( cyan in Figure 6B ) suggesting that these ISGs inhibit a range of HIV isolates . UBE2L6 has recently been shown to inhibit HIV in an overexpression assay and is antagonized by Vpu ( Jain et al . , 2018 ) . In contrast , several other ISGs scored highly in the HIV-1LAI PIKAHIV screen but not the HIV-1Q23 . BG505 screen including LGALS3BP ( also known as 90K or M2BP ) , SAMD9L and Tetherin/BST2 ( dark blue in Figure 6B ) . UBE2L6 and 90K inhibit HIV in over-expression assays ( Jain et al . , 2018; Lodermeyer et al . , 2013; Wang et al . , 2016 ) and SAMD9L was recently shown to be an IFN-induced restriction factor for poxviruses ( Meng et al . , 2018 ) . Tetherin/BST2 , LGALS3BP/90K/M2BP and SAMD9L therefore show differential , strain-dependent restriction . Finally , the HIV-1Q23 . BG505 screen reveals one factor , IFI16 , that is a strong hit in the HIV-1Q23 . BG505 screen but not the HIV-1LAI screen ( Green in Figure 6B ) . IFI16 is a dsDNA sensor that has been implicated in sensing and inhibition of HIV infection ( Jakobsen et al . , 2013; Monroe et al . , 2014 ) . Our results here suggest that IFI16 restricts HIV-1 in THP-1 cells in a strain-dependent manner . Similar to TRIM5alpha , our finding of inhibition of wt HIV by Tetherin despite intact Vpu expression may suggest a functional tradeoff in which HIV Vpu is unable to completely antagonize host cell Tetherin activity . It has previously been shown that primary HIV isolates better antagonize Tetherin as compared to HIV-1 NL4-3 ( Pickering et al . , 2014 ) which may explain why Tetherin is a hit in the screen using HIV-1LAI , but not HIV-1 Q23 . BG505 . Adaptation HIV-1 group M to the unique form of human Tetherin allele required evolution of the viral protein Vpu to antagonize Tetherin ( Kmiec et al . , 2016; Lim et al . , 2010; Sauter et al . , 2009 ) . Consistent with conflicting evolutionary constraints , IFN treatment in HCV- and HIV- coinfected patients resulted in evolution of Vpu variants with stronger Tetherin antagonism when ISGs are expressed in vivo ( Pillai et al . , 2012 ) . Perhaps more complete antagonism of Tetherin by Vpu would compromise some of the other functions of Vpu in cells ( Apps et al . , 2016; Margottin et al . , 1998; Schubert et al . , 1998; Shah et al . , 2010 ) . Further , a moderate level of Tetherin antagonism could be selected for if cell-to-cell transmission is enhanced by Tetherin restriction ( Gummuluru et al . , 2000; Jolly et al . , 2010 ) , such as is observed for MoMLV in mice ( Liberatore et al . , 2017 ) . Interestingly , Vpu counteracts UBE2L6 ( Jain et al . , 2018 ) which we describe as an inhibitor of both HIV-1LAI and HIV-1Q23 . BG505 suggesting that Vpu from both viruses are unable to fully counteract UBE2L6 in IFN-treated THP-1 cells . We found that ZAP mediates a small , but detectable inhibition of HIV replication as we find enhanced infection of ZAP-KO cells both in the presence and absence of IFN pretreatment ( Figure 2 - Supplemental Figure 1A and 1B ) . Similar to the effect of ZAP , N4BP1 ( Nedd4-binding protein 1 ) also has a modest effect on HIV replication both after IFN pretreatment and when constitutively-expressed ( Figure 2 - Supplemental Figure 1C and 1D ) . In our screen , the anti-lentiviral function of N4BP1 appears to be genetically linked to ZAP activity as N4BP1 is no longer a hit in the ZAP-KO screen ( Figure 2 – Supplemental Figure 1E ) . Therefore , N4BP1 may modify or enhance ZAP-mediated antiviral activity similar to the modification of ZAP activity described for TRIM25 ( Li et al . , 2017 ) . However , we did not find a role for N4BP1 in ZAP-mediated inhibition of SFV ( see Figure 2 – Supplemental Figure 1E ) . Therefore , the role of N4BP1 in ZAP’s antiviral activity varies depending on the viral target . Other high-throughput screens with siRNA pools have focused on identifying dependency factors that the virus takes advantage of to infect cells ( Brass et al . , 2008; König et al . , 2008; Zhou et al . , 2008 ) . Importantly , all of these high-throughput approaches still require individual gene knockdowns or overexpression in individual wells . The HIV-CRISPR screening approach represents a significant advance in screening for host factors that affect HIV replication in several ways , including: ( 1 ) we can simultaneously screen thousands of gene targets in a single experiment , ( 2 ) we can use any virus strain , ( 3 ) we do not need any type of reporter to assay infections as virus replication itself provides the assay readout and ( 4 ) we can capture host factors that affect all stages of the HIV life-cycle including entry , nuclear import , integration , transcription , nuclear export , translation , packaging , budding and release . After finding ZAP-mediated inhibition of the HIV-CRISPR vector used in our screening approach , we modified our PIKAHIV screen to avoid this inhibition by specific KO of ZAP expression and rescreening in ZAP-KO THP-1 clonal lines . Genetic deletion of ZAP resulted in enhanced performance of the HIV-CRISPR screen and allowed for our identification of ISGs contributing to the IFN block in THP-1 cells . Further the data presented here demonstrates that the screen is sensitive enough to find key factors in just a single round of viral replication , even when multiple factors together mediate potent inhibition . Despite targeting less than 10% of the genes in the human genome by our PIKAHIV library , we were able to identify and validate a small panel of HIV dependency factors that HIV usurps to enhance infection in THP-1 cells . We demonstrate that the HIV attachment factor , SIGLEC1/CD169 , plays a role in enhancing infection in THP-1 cells in cis rather than the more fully described role of SIGLEC1 to mediate infection from dendritic cells to T cells in trans ( Izquierdo-Useros et al . , 2012; Puryear et al . , 2013 ) . Further , we find that TLR2 mediates enhanced infection of THP-1 cells by HIV-1 regardless of viral entry pathway used , as it impacted infection through both the HIV envelope and the VSV-G glycoprotein ( Figure 5E ) . Recent work in CD4 +T cells has similarly demonstrated enhanced infection and/or viral production in T cells on stimulation of TLR2 ( Bolduc et al . , 2017; Ding and Chang , 2012; Ding et al . , 2010; Equils et al . , 2003; Henrick et al . , 2015 ) . Of note , MYD88 , a downstream effector for TLR2 activation of transcription , is also a strong hit in our dependency factor screening ( Figure 5A ) suggesting that it is the downstream signaling functions of TLR2 that enhance infection . In addition to CD169 and TLR2 , our identification of SEC62 as a novel HIV dependency factor that correlates with CD4 receptor cell surface expression highlights the ability of the HIV-CRISPR screening approach to find genes that function in pathways ( such as CD4 receptor expression ) important for HIV infection . Finally , as would be expected , NFKB1 is a dependency factor ( Figure 5A ) as it is a precursor to a downstream activator of the HIV LTR that is important for HIV infection , although somewhat surprisingly NFKB2 is a restriction factor in the screen ( Figure 6B ) . Thus , HIV-CRISPR screens have the capability of identifying pathways of both positive and negative HIV regulation . A recent CRISPR knockout screen demonstrated that a pooled CRISPR approach to gene knockout could identify HIV dependency factors ( Park et al . , 2017 ) . The Park et al . study differs from our study in several key aspects . The screen by Park et al . relied on Tat-driven LTR-GFP reporter gene expression as well as many rounds of spreading infection across multiple weeks in culture in a T cell line . In contrast , PIKAHIV screening is performed over a single round of infection in three days in the THP-1 monocytic cell line with the screen relying on virus replication itself to enrich for gene targets of interest rather than selecting on a reporter gene . Further , Park et al screened with a whole-genome library while our screening has been performed with an ISG-specific sgRNA library ( PIKAHIV ) . Moreover , we screened in the presence of IFN with a specific focus on finding HIV restriction factors , whereas Park et al . specifically screened for HIV dependency factors . Both screens do identify the appropriate HIV co-receptor ( CXCR4 or CCR5 in our study and CCR5 in the Park et al . study ) . Of note , 3 of the five genes identified by Park et al . ( TPST2 , SLC35B2 and CD4 ) are not represented in the PIKAHIV library and , therefore , could not be identified in our screen . The final gene identified by Park et al . , ALCAM , is not found as a hit in our screening and may reflect important differences in cell type or other technical aspects of the screening approaches . For example , the genes identified by Park et al . are the HIV receptor or co-receptor ( CD4 , CCR5 ) , genes that affect CCR5 expression ( TPST2 and SLC35B2 ) or genes that affect the ability of cells to complex with other cells ( ALCAM ) . It is possible that the Park et al . screen may , therefore , reflect the loss of the ability of these KO cells to support syncytia formation rather than a loss of HIV infection per se . Some hits identified in our PIKAHIV screening were not found in the Park et al . study , including SEC62 , TLR2 and SIGLEC1/CD169 . We hypothesize that differences in cell type , viral strains and screening approach likely explain these differences . Further studies using a whole genome HIV-CRISPR library should more comprehensively identify further HIV dependency factors that are not present in the PIKAHIV library . In summary , here we describe a novel screen that is highly sensitive to detect restriction factors for HIV-1 . This new tool shows that the IFN inhibition of HIV-1 in a monocytic cell line is due the combined function of fewer than eight different genes . Our results demonstrate that IFN-mediated inhibition of HIV-1 in THP-1 cells is mediated by restriction factors for which HIV has described mechanisms of antagonism and/or escape . The increased IFN sensitivity of specific HIV strains , such as those isolated during chronic HIV infection , may be due to relaxation of constraints on the virus that would otherwise limit virus replication during transmission events . We propose that conflicting functional constraints acting on HIV may result in incomplete antagonism or escape from host ISGs during chronic infection .
1905 human ISGs were selected from gene expression datasets of type I IFN-stimulated cells ( Goujon et al . , 2013; Hung et al . , 2015; Linsley et al . , 2014 ) or from previously assembled ISG overexpression ( Schoggins et al . , 2011 ) or shRNA libraries ( Li et al . , 2013 ) . These included all the genes from the previously assembled ISG libraries ( Li et al . , 2013; Schoggins et al . , 2011 ) as well as additional ISGs as defined here . For the GSE46599 dataset ( Goujon et al . , 2013 ) , raw probe-level signal intensities from Illumina HumanHT-12 V4 . 0 expression BeadChip data were retrieved from GEO , then background-corrected , quantile-normalized and log2-transformed using the Bioconductor package lumi ( Du et al . , 2008 ) . Fold changes ( FC ) in expression between type I IFN-treated and untreated samples were calculated for untreated and PMA-treated THP-1 cells , primary CD4+ T cells and primary macrophages . For THP-1 cells , genes with FC ≥2 were selected . For primary cells , genes with a donor-specific FC ≥2 in at least 2 out of 3 donors were selected . For the GSE60424 dataset ( Linsley et al . , 2014 ) , TMM normalized RNA-seq read count data ( Illumina HiScan ) were retrieved from GEO . FC in expression in whole blood , isolated CD4 +T cells and monocytes of a Multiple Sclerosis patient , pre- and post-treatment with AVONEX ( IFNβ ) , were calculated and genes with FC ≥2 were selected . For the GSE72502 dataset ( Hung et al . , 2015 ) , de novo identification of differentially-expressed genes in IFNα treated PBMCs was performed from the raw RNA sequencing data ( Illumina Genome Analyzer ) . SRA files were retrieved from GEO and converted to FASTQ format using NCBI’s SRA toolkit . Reads were mapped to the human reference genome ( hg19 ) using GSNAP ( Wu et al . , 2016 ) and quantified using HTSeq ( Anders et al . , 2015 ) . Differentially-expressed ( DE ) genes were identified using the Bioconductor edgeR package ( Robinson et al . , 2010 ) . DE genes were defined at an FDR threshold of 0 . 05 . The glmTreat function was used to detect genes with a FC significantly greater than one between the IFN-treated and control samples . Finally , non-coding RNAs and pseudogenes were removed from the list . Inspection of the curated list of genes showed that overlap between the different datasets was limited and many genes ( >2000 ) were only present in 1 of the 10 datasets/libraries . As such , a second selection round was performed in which the expression threshold for genes present in only one of the datasets was raised to FC ≥3 . For genes present in at least two datasets , the initial cut-off of FC ≥2 was kept . Finally , 35 additional genes identified through RNA sequencing gene expression analysis as being responsive to both type I/type III IFN and IL-1β were also included ( M . Gale , personal communication , Anders et al . , 2015 ) . For analysis of IFN induction specific to THP-1 cells , raw signal intensities were downloaded from GEO ( GSE46599 ) and the data was quantile normalized using the Bioconductor package lumi . For a given probe , both samples from at one least condition were required to have a detection p-value<=0 . 05 . The Bioconductor package limma was used to identify significantly differentially expressed probes . A false discovery rate ( FDR ) method was employed to correct for multiple testing ( Reiner et al . , 2003 ) , with differential expression defined as |log2 ( ratio ) |≥0 . 585 ( ±1 . 5 fold ) with the FDR set to 5% . The THP-1 monocytic cell line ( ATCC; RRID: CVCL_0006 ) was cultured in RPMI ( Invitrogen ) with 10% FBS , Pen/Strep , 10 mM HEPES , 0 . 11 g/L sodium pyruvate , 4 . 5 g/L D-Glucose and Glutamax . 293T ( ATCC CRL-3216; RRID: CVCL_0063 ) and TZM-bl cells ( 8129; RRID: CVCL_B478 ) were cultured in DMEM ( Invitrogen ) with 10% FBS and Pen/Strep . For some validation studies , THP-1 cells with single-cell sorted into 96-well plates to create individual clonal lines ( BD FACS Aria II – Fred Hutch Flow Cytometry Core ) . Universal Type I Interferon Alpha was obtained from PBL Assay Science ( Catalog No . 11200–2 ) , diluted to 105 Units/mL in sterile-filtered PBS/1% BSA according to the activity reported by manufacturer and frozen in aliquots at −80°C . All Puromycin selections were done at 0 . 5–1 ug/mL . The identity of THP-1 cells was confirmed by STR profiling ( Fred Hutch Research Cell Bank ) . Mycoplasma was detected in some THP-1 cultures after the completion of experiments . lentiCRISPRv2 plasmid was a gift from Feng Zhang ( Addgene #52961 ) . pMD2 . G and psPAX2 were gifts from Didier Trono ( Addgene #12259/12260 ) . lentiCRISPRv2 constructs targeting genes of interest were cloned into BsmBI-digested lentiCRISPRv2 by annealing complementary oligos ( Supplementary file 1 ) with overhangs that allow directional cloning into lentiCRISPRv2 . Stable LKO SEC62 shRNA lentiviral vectors were obtained from Sigma . SEC62_1: CCGGCCAGGAAATCATGGAACAGAACTCGAGTTCTGTTCCATGATTTCCTGGTTTTTG ( TRCN0000289739 ) . SEC62_2: CCGGGAAATGAGAGTAGGTGTTTATCTCGAGATAAACACCTACTCTCATTTCTTTTTG ( TRCN0000289833 ) . Scramble shRNA ( CCTAAGGTTAAGTCGCCCTCGCTCGAGCGAGGGCGACTTAACCTTAGG ) was a gift from David Sabatini ( Addgene #1864 ) . The CD169 shRNA ( Sigma TRCN155147 ) ( CCGGGTGTGGAGATTCACAACCCTTCTCGAGAAGGGTTGTGAATCTCCACACTTTTTTG ) was a gift from Rahm Gummuluru . The CD169 shRNA sequence was subcloned into pLKO . 1neo ( Addgene #13425 ) using EcoRI and AgeI sites . HIV-CRISPR was constructed ( Genscript ) by inserting a synthesized 433 bp sequence from HIV-1LAI into the deleted 3’ LTR U3 sequence of lentiCRISPRv2 . HIV-1LAI LT insert: ATCCTTGATCTGTGGATCTACCACACACAAGGCTACTTCCCTGATTGGCAGAACTACACACCAGGGCCAGGGGTCAGATATCCACTGACCTTTGGATGGTGCTACAAGCTAGTACCAGTTGAGCCAGATAAGGTAGAAGAGGCCAATAAAGGAGAGAACACCAGCTTGTTACACCCTGTGAGCCTGCATGGAATGGATGACCCTGAGAGAGAAGTGTTAGAGTGGAGGTTTGACAGCCGCCTAGCATTTCATCACGTGGCCCGAGAGCTGCATCCGGAGTACTTCAAGAACTGCTGACATCGAGCTTGCTACAAGGGACTTTCCGCTGGGGACTTTCCAGGGAGGCGTGGCCTGGGCGGGACTGGGGAGTGGCGAGCCCTCAGATGCTGCATATAAGCAGCTGCTTTTTGCCTGTACTGGGTCTCTCTGGTTA . The wild type ( HIV-1LAI ) , env-deleted ( HIV-1LAI VSV-G ) and vpu-deficient ( HIVLAIΔvpu = VpuFS/Rap5 ) HIV-1LAI proviruses were previously described ( Bartz and Vodicka , 1997; Gummuluru et al . , 2000; Peden et al . , 1991 ) . The Clade A HIV-1Q23 . BG505 molecular clone was previously described ( Haddox et al . , 2018; Poss and Overbaugh , 1999 ) . The pHIV-zsGreen/CCR5 construct was created by cloning the human CCR5 CDS into pHIV-zsGreen ( Addgene #18121 ) with BamHI and NotI . Four sgRNA sequences were selected randomly from the Brunello library for each gene target ( Doench et al . , 2016 ) and four additional non-identical sgRNAs were subsequently selected randomly from the Genome-scale CRISPR Knock-Out ( GeCKO v2 ) library ( Sanjana et al . , 2014b; Shalem et al . , 2014a ) . For genes for which eight unique sgRNAs could not be obtained from these libraries , additional sgRNAs were added from the Moffat ( Hart et al . , 2015 ) and Sabatini/Lander libraries ( Wang et al . , 2015; Wang et al . , 2014 ) . 12 genes contained no sgRNAs in any of the libraries and for those genes eight new sgRNAs were designed using the sgRNA Designer from the Broad Institute ( http://portals . broadinstitute . org/gpp/public/analysis-tools/sgrna-design ) . A total of 15 , 348 unique sgRNA sequences were synthesized . The sgRNAs were split in two pools ( Zhang and Brunello sgRNAs ) for synthesis ( four per gene in each pool ) and two independent sets of 200 Non-Targeting Control ( NTC ) sgRNAs obtained from the GeCKOv2 library were added in duplicate to each pool . The PIKAHIV ISG-targeting sgRNA library was synthesized ( Twist Biosciences ) and cloned into HIV-CRISPR . Oligo pools were amplified using Phusion HF ( Thermo ) using 1 ng of pooled oligo template , primers ArrayF and ArrayR ( ArrayF primer: TAACTTGAAAGTATTTCGATTTCTTGGCTTTATATATCTTGTGGAAAGGACGAAACACCG and ArrayR primer: ACTTTTTCAAGTTGATAACGGACTAGCCTTATTTTAACTTGCTATTTCT AGCTCTAAAAC ) , an annealing temperature of 59°C , an extension time of 20 s , and 25 cycles . Following PCR amplification , a 140 bp amplicon was gel-purified and cloned into BsmBI digested vectors using Gibson assembly ( NEB ) . Each Gibson reaction was carried out at 50°C for 60 min in a thermocycler . 1 μl of the reaction was used to transform 25 μl of electrocompetent cells ( Stellar Competent Cells; Clontech ) according to the manufacturer’s protocol using a GenePulser ( BioRad ) . To ensure adequate representation , sufficient parallel transformations were performed and plated onto ampicillin containing LB agarose 245 mm x 245 mm plates ( Thermo Fisher ) at 200-times the total number of oligos of each library pool . After overnight growth at 37°C , colonies were scraped off , pelleted , and used for plasmid DNA preps using the Endotoxin-Free Nucleobond Plasmid Midiprep kit ( Takara Bio #740422 . 10 ) . The PIKAHIV plasmid library was sequenced and contains 15 , 339 of the 15 , 348 total sgRNAs included in the synthesis ( Figure 1—figure supplement 2 ) . 293 T cells ( ATCC ) were plated at 2 × 105 cells/mL in 2 mL in 6-well plates one day prior to transfection using TransIT-LT1 reagent ( Mirus Bio LLC ) with 3 μL of transfection reagent per μg of DNA . For lentiviral preps , 293Ts were transfected with 667 ng lentiviral plasmid , 500 ng psPAX2 and 333 ng MD2G . For HIV-1 production , 293Ts were transfected with 1 ug/well proviral DNA . One day post-transfection media was replaced . Two- or three- days post-transfection viral supernatants were clarified by centrifugation ( 1000 g ) and filtered through a 20 μm filter . For PIKAHIV library preps , supernatants from 40 × 6 well plates were combined and concentrated by ultracentrifugation . 30 mL of supernatant per SW-28 tube were underlaid with sterile-filtered 20% sucrose ( 1 mM EDTA , 20 mM HEPES , 100 mM NaCl , 20% sucrose ) and spun in an SW28 rotor at 23 , 000 rpm for 1 hr at 4°C in a Beckman Coulter Optima L-90K Ultracentrifuge . Supernatants were decanted , pellets resuspended in DMEM over several hours at 4°C and aliquots frozen at −80°C . All viral and lentiviral infections and transductions were done in the presence of 20 μg/mL DEAE-Dextran ( Sigma; D9885 ) . Semliki Forest Virus ( SFV ) stocks were created by co-transfection of SFV replicon ( pSMART-LacZ ) and helper ( pSCA-helper ) plasmids in 293Ts as previously described ( DiCiommo and Bremner , 1998; DiCiommo et al . , 2004 ) . SFV particles were harvested 48 hr post-transfection and activated by treatment with α-Chymotrypsin ( Sigma #C4129 ) . Large-scale preps of the PIKAHIV lentiviral library were titered by a colony-forming assay in TZMbl cells and used to transduce THP-1 cells at an MOI of 0 . 7 . Cells were selected in Puromycin ( 0 . 5 μg/mL ) for 10–14 days . 8 × 106 cells per replicate ( >500X coverage of the PIKAHIV library ) were treated or not with IFNα and infected the following day at a viral dose determined to allow approximately 50% of cells in culture to be infected by spinoculation at 1100xg for 30 min with 20 μg/mL DEAE-Dextran . After overnight incubation , cells were resuspended in media with or without IFNα at 5 × 105 cells/mL . Cells and supernatants were collected 3 days post infection . Genomic DNA was extracted from cell pellets with a QIAamp DNA Blood Midi Kit ( Qiagen #51183 ) and genomic DNA eluted in water . Viral supernatants were spun at 1100xg to remove cell debris , filtered through a 0 . 2 μm filter , overlaid on a 20% sucrose cushion and concentrated in SW28 rotor for 1 hr at 4°C . After resuspension in PBS , viral RNA was extracted ( QIAamp viral RNA Kit , Qiagen , 52904 ) . sgRNA sequences present in the genomic DNA and viral supernatants were amplified by PCR and RT-PCR , respectively , using primers specific for the HIV-CRISPR construct ( Supplementary file 1 ) ( Toledo et al . , 2015 ) . Libraries were then barcoded/prepared for Illumina sequencing by a second round of PCR ( Supplementary file 1 ) . Each amplicon was then cleaned up through double-sided SPRI ( Agencourt AMPure XP Beads – Beckman Coulter #A63880 ) , quantitated with a Qubit dsDNA HS Assay Kit ( Q32854 – ThermoFisher ) and pooled to 2 nm for each library . Pooled , multiplexed libraries were then sequenced on a single lane of an Illumina HiSeq 2500 in Rapid Run mode ( Fred Hutch Genomics and Bioinformatics Shared Resource ) . Raw sequence read data from screens are available as a GEO DataSet #GSE118631 . Following demultiplexing of libraries to assign sequences to each sample ( allowing no mismatches ) , reads were trimmed and aligned to the PIKAHIV sgRNA library , using Bowtie ( Langmead et al . , 2009 ) . NTC sgRNA sequences were iteratively binned to create an NTC sgRNA set as large as the ISG gene set in the PIKAHIV library . Relative enrichment or depletion of sgRNAs and genes were analyzed using the MAGeCK statistical package ( Li et al . , 2014 ) . For the VSV-G screen a single IFITM1-targeting sgRNA sequence ( AGCATTCGCCTACTCCGTGA ) with complete homology to IFITM3 was removed from the analysis . An Excel ( Microsoft ) worksheet was created to analyze the CG dinucleotide content of the HIV-CRISPR-NTC1 transcript from the beginning of the 5’R region to the end of the 3’R region . The HIV-CRISPR sequence was broken into fragments of 3 nucleotides ( codons ) , and at each position the number of CG dinucleotides within or between two adjacent codons was determined . The CG counts at each position over the length of the sequence were then summed within a sliding window of 67 codons ( 201 nucleotides ) and plotted against the nucleotide position of the transcript in GraphPad Prism . Wild-type or ZAP-knockout THP-1 cells were transduced with a pooled library of HIV-CRISPR encoding 39 distinct gRNAs at an MOI of 0 . 5 and selected with puromycin for 15 days as described above . Cells were infected with HIV-1LAI at an MOI of 1 . Three days post-infection , viral supernatants were cleared by centrifugation , filtered through a 0 . 4 μm filter , and viral RNA was extracted from 140 μL of supernatant using the QIAamp viral RNA Kit , with subsequent aliquoting and freezing at −80⁰C . cDNA was synthesized from viral RNA with random hexamers ( OneStep RT-ddPCR Advanced Kit for Probes , BioRad , 1864021 ) , and the number of copies of either HIV-1LAI or HIV-CRISPR genomes per μL of cDNA was quantified by ddPCR using the QX200 Droplet Digital PCR System ( Bio-Rad , Hercules , CA ) . HIV was detected using previously published primers and probe directed towards pol ( Benki et al . , 2006 ) . To specifically detect HIV-CRISPR , we used primers ddPCR-cPPT-F ( GTA CAG TGC AGG GGA AAG ) , ddPCR-U6-R ( ATG GGA AAT AGG CCC TCG ) , and probe cPPT-probe ( 6-FAM/ZEN- AGA CAT AAT AGC AAC AGA CAT ACA AAC -IBFQ ) ( Integrated DNA Technologies , Skokie , IL ) . Both sets of reactions were set up according to the manufacturer’s protocols with an annealing temperature of 60⁰C . The HIV-CRISPR primers were found to be specific , as no amplification was detected in untransduced cells infected with HIV-1 . Control reactions on viral RNA without reverse transcriptase revealed that carry-over plasmid contamination from viral preps accounted for only a low level ( <50 copies/μL ) of amplification . ZAP knockout cell pools were created by electroporating THP-1 cells with a custom ZAP-targeting crRNA ( ATGTGGAGTCTTGAACACGG; IDT ) . 1 μL crRNA was resuspended at 160 μM in 10 mM Tris pH 7 . 4 and complexed at an equimolar ratio with 1 μL 160 μM tracrRNA ( IDT #1072534 ) and incubated 30 min at 37°C followed by addition of 2 μL of 40 μM Cas9-NLS ( UC Berkeley MacroLab ) and further incubation at 37°C for 15 min to create the ZAP-targeting crRNP complexes . 3 . 5 μL crRNP was added to 5 × 105 THP-1 cells resuspended in Amaxa SG Cell Line 96-well Nucleofector Kit ( Lonza #V4SC-3096 ) and electroporated according to the manufacturer’s protocol ( Lonza 4D Nucleofector ) . 80 μL of prewarmed media was added , followed by incubation for 30 min recovery in the 37°C incubator . Cells were then resuspended at 2 . 5 × 105 cells/mL in 500 μL in a 24-well plate for 48 hr before single cell sorting into 96 well U-bottom plates containing RPMI media supplemented with 20% FBS ( BD FACS Aria II – Fred Hutch Flow Cytometry Core ) . MxB-KO clonal lines were generated by transduction with lentiCRISPRv2 containing MxB-targeting sgRNA sequences ( see Supplemental Table S5 for sgRNA sequences ) followed by single-cell cloning and puromycin selection . lentiCRISPRv2 KO cell pools targeting N4BP1 , STAT1 , STAT2 , IRF9 , MxB , TRIM5alpha , IFITM1 , Tetherin , CXCR4 and TLR2 as well as two Non-Targeting Controls ( NTC_1 and NTC_2 ) were created through transduction with lentivirus and selection in Puromycin ( see Supplemental Table S5 for sgRNA sequences ) . Both KO cell pools and individual KO cell lines were validated using Western blotting , flow cytometry and/or genomic editing analysis as described below . shRNA knockdown cell pools were made by transducing wildtype THP-1 cells with lentiCRISPRv2 shRNA constructs and selected in RPMI containing 1 μg/mL Puromycin for two weeks prior to validating via Western blotting or flow cytometry . CCR5-overexpressing ZAP-KO THP-1 cells were created by transducing ZAP-KO THP-1 cells with pHIV-zsGreen/CCR5 lentiviral supernatants and sorting a zsGreen +population 48 hr post-transduction . Knockout cells were harvested and either lysed in Epicentre QuickExtract DNA Extraction Solution ( Lucigen QE09050 ) for direct PCR amplification or genomic DNA was extracted ( QIAamp DNA Blood Mini Kit – Qiagen #51185 ) . Edited loci were amplified from cell pool DNA using primers specific to each targeted locus ( Supplementary file 1 ) as previously published ( Hultquist et al . , 2016 ) . PCR amplicons were sequenced ( Fred Hutch Shared Resources Genomics Core – sanger sequencing ) and analyzed by ICE ( Synthego ) to determine the percent of alleles edited at each locus in the cell population ( Hsiau et al . , 2018 ) . Editing was confirmed at each locus ( Figure 4—source data 1; KO scores varied across pools from 48% to 93% ) . For Western blotting the following antibodies were used as follows: MxB ( Santa Cruz sc-271527; RRID: AB_10649506 ) at 1:200 , Sec62 ( Abcam ab168843 ) at 1:2000 , ZAP/ZC3HAV1 ( Proteintech #16820–1-AP; RRID: AB_2728733 ) at 1:5000 , N4BP1 ( Cohesion Biosciences #CPA2415 ) at 1:1000 , tubulin ( Sigma T6199; RRID: AB_477583 ) at 1:1000 and actin ( Sigma A2066; RRID: AB_476693 ) at 1:5000 . Secondary antibodies were used as follows: 1:5000 donkey anti-goat IgG-HRP ( Santa Cruz Biotechnology sc-2020; RRID: AB_631728 ) and 1:5000 goat anti-rabbit IgG-HRP ( Santa Cruz Biotechnology sc-2004; RRID: AB_631746 ) . For flow cytometry , antibodies were used as follows: CD4 ( BD Pharmingen 555349; RRID: AB_398593 ) 1:50 , CXCR4 ( eBioscience 17-9999-42; RRID: AB_1724113 ) 1:50 , CD-169 ( BioLegend 346003; RRID: AB_2189038 ) 1:50 , TLR2 ( BioLegend 309707; RRID: AB_314777 ) 1:100 , Tetherin ( BioLegend 348410; RRID: AB_2067121 ) 1:50 . For intracellular Gagp24 ( p24 ) staining , cells were harvested and fixed in 4% paraformaldehyde for 10 min and diluted to 1% in PBS . Cells were permeabilized in 0 . 5% Triton-X for 10 min and stained with 1:300 KC57-FITC ( Beckman Coulter 6604665; RRID: AB_1575987 ) . Cells were read on a BD FACSCANTO II ( Fred Hutch Flow Cytometry Core ) and analyzed in FlowJo . For cell surface marker staining , cells were washed twice in PBS , stained in PBS/1% BSA , incubated at 4°C for 1 hr , washed twice in PBS , and analyzed on the Canto two flow cytometer ( Fred Hutch Flow Cytometry Core ) . Cells were lysed in 2X SDS-SB lysis buffer ( 10% glycerol , 2% BME , 6% SDS , 62 . 5 mM Tris-HCl pH 6 . 8 ) , boiled at 95*C , sonicated for one minute and resolved by NuPAGE 4–12% Bis-Tris Gel ( Invitrogen ) . Following transfer to a PVDF membrane and blocking in PBS/5%milk for 1 hr , blots were probed with antibodies for 1 hr or overnight , washed in PBST , probed with HRP secondary , washed in PBST and bands visualized with SuperSignal West Femto Maximum Sensitivity Substrate ( ThermoFisher #34095 ) . Blots were visualized on a BioRad Chemidoc MP . Cells were pre-stimulated with IFNα 24 hr prior to infection . Virus and 20 μg/mL DEAE-Dextran in RPMI were added to cells , spinoculated for 20 min at 1100xg , and incubated overnight at 37°C . Cells were washed the next day and re-suspended in RPMI supplemented with IFNα . For experiments to assay ISGs affecting late steps in viral replication , cells were spinoculated at 1100xg for 20 min with HIV-1LAI or Vpu-deficient HIV-1LAI ( HIVLAIΔvpu ) at an MOI of 0 . 4 , incubated at 37°C for 16 hr , and then treated with 1000 mU/mL IFNα . 24 hr post infection , cells were washed of virus and re-suspended in interferon containing media with 1 μg/mL T-20 entry inhibitor ( NIH AIDS Reagent Program , Division of AIDS , NIAID , NIH: Enfuvirtide #12732 ) . For SFV infections THPs were infected overnight with SFV-LacZ stocks and LacZ activity assayed the following day with the Gal-Screen β-Galactosidase Reporter Gene Assay System ( ThermoFisher #T1029 ) . p24 ELISA on cell culture supernatants was performed with a HIV-1 p24 Ag Assay ( ABL #5421 ) . Reverse transcriptase activity in viral supernatants was measured using the RT activity assay as described ( Roesch et al . , 2018; Vermeire et al . , 2012 ) . A stock of HIV-1LAI virus was titered multiple times , aliquoted at −80°C and used as the standard curve in all assays . | The different strains of the human immunodeficiency virus ( or HIV ) can infect a variety of cells in the human body . When a cell senses being attacked , it can defend itself using molecules called restriction factors , which are created under the control of a signal known as interferon . Researchers have already identified several restriction factors , using techniques that are relatively laborious and time-consuming , but many questions remain about these proteins . Here , Ohainle et al . created a new method to screen for restriction factors; by harnessing the CRISPR/Cas9 technique , HIV was tricked into revealing its own weaknesses . The method allowed Ohainle et al . to make precise , targeted changes to thousands of genes that are turned on by interferon , and deactivate them . The experiments revealed that HIV multiplied better in human cells in which several specific genes had been neutralized . This suggests that these genes encode restriction factors that are activated by interferon to combat HIV . The combined action of a few of these proteins can fight the virus , even if it cannot completely eradicate it . Further experiments found that a different , but overlapping set of restriction factors defended cells against a different strain of HIV . The method developed by Ohainle et al . is a useful tool to identify new restriction factors . By dissecting the role of these proteins in keeping different HIV strains under control , we may understand how the virus has become dangerous for humans by evading some of these defenses . Ultimately , this could help with finding better ways to fight this deadly disease . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"research",
"communication",
"microbiology",
"and",
"infectious",
"disease"
] | 2018 | A virus-packageable CRISPR screen identifies host factors mediating interferon inhibition of HIV |
Rapid diagnostic tests ( RDTs ) have transformed malaria diagnosis . The most prevalent P . falciparum RDTs detect histidine-rich protein 2 ( PfHRP2 ) . However , pfhrp2 gene deletions yielding false-negative RDTs , first reported in South America in 2010 , have been confirmed in Africa and Asia . We developed a mathematical model to explore the potential for RDT-led diagnosis to drive selection of pfhrp2-deleted parasites . Low malaria prevalence and high frequencies of people seeking treatment resulted in the greatest selection pressure . Calibrating our model against confirmed pfhrp2-deletions in the Democratic Republic of Congo , we estimate a starting frequency of 6% pfhrp2-deletion prior to RDT introduction . Furthermore , the patterns observed necessitate a degree of selection driven by the introduction of PfHRP2-based RDT-guided treatment . Combining this with parasite prevalence and treatment coverage estimates , we map the model-predicted spread of pfhrp2-deletion , and identify the geographic regions in which surveillance for pfhrp2-deletion should be prioritised .
Efforts to control malaria globally have made substantial progress in the last 15 years ( World Health Organization , 2015a ) . This progress reflects the impact made by reinvigorated political commitment that has yielded a twenty-fold increase in international funding for the control and elimination of malaria ( World Health Organization , 2015a ) . The World Health Organisation ( WHO ) Global Technical Strategy for Malaria 2016–2030 sets ambitious goals to further reduce incidence and mortality rates by 90% by 2030 ( World Health Organization , 2015b ) . Central to achieving these goals is the need to test , treat and track all malaria ( World Health Organization , 2010 ) . In sub-Saharan Africa ( SSA ) , diagnostic testing of suspected malaria cases has risen from 36% to 60% between 2005 and 2014 ( World Health Organization , 2015a ) . Microscopy was historically the most common method for diagnosis; however rapid diagnostic tests ( RDTs ) accounted for 71% of all diagnostic testing of suspected cases in 2014 ( World Health Organization , 2015a ) . The most widely used RDTs target histidine-rich protein 2 ( HRP2 ) , which is expressed by the Plasmodium falciparum ( Pf ) specific gene pfhrp2 , with over 85% of RDTs tested in the WHO Foundation for Innovative New Diagnostics ( FIND ) Malaria RDT Evaluation Programme targeting PfHRP2 ( World Health Organization , 2012a ) . False-negative RDT results due to a partial or complete deletion of the pfhrp2 gene have been reported in areas of South America since 2010 , resulting in the recommendation against the use of PfHRP2-based RDTs in these areas ( Akinyi et al . , 2013; Abdallah et al . , 2015; Cheng et al . , 2014 ) . These pfhrp2-deleted mutants may still possess a functioning pfhrp3 gene; however , the cross reactivity between PfHRP2-based RDT antibodies and PfHRP3 epitopes is such that a positive result may only occur at very high parasitaemia ( Baker et al . , 2005 ) . Confirmed pfhrp2-deleted mutants are rarer in Africa , with the first cases reported in Mali in 2012 ( Koita et al . , 2012 ) . However , recently confirmed occurrences in Ghana , ( Amoah et al . , 2016 ) Zambia , ( Laban et al . , 2015 ) the Democratic Republic of Congo ( DRC ) , ( Parr et al . , 2016 ) Rwanda ( Kozycki et al . , 2017 ) and Eritrea ( Berhane et al . , 2017 ) ( Table 1 ) have prompted the WHO to host Technical Consultations on pfhrp2/3-deletions and to issue interim guidance for malaria control programs ( World Health Organization , 2017; World Health Organization , 2016a; World Health Organization , 2016b ) . This raises the concern that pfhrp2-deleted mutants may be selected for by RDT-guided treatment decisions – which if confirmed would be one of the first example of selection of a pathogen through diagnostic testing . Here we use mathematical modelling to characterise the impact of introducing PfHRP2-based RDTs on the emergence and spread of pfhrp2-deleted parasites . We adapt a previously published transmission model ( Griffin et al . , 2016 ) , incorporating the transmission of pfhrp2-deleted mutants and the contribution of PfHRP3 cross-reactivity to identify settings in which the selective pressure favouring pfhrp2-deleted strains is greatest . In addition , we conduct sensitivity analyses to characterise the influence of assumptions within our model concerning adherence to RDT-guided treatment decisions , the use of microscopy-based diagnostic testing , fitness costs associated with the mutant parasite and the impact of non-malarial fevers upon the selective advantage of pfhrp2 gene deletions . We continue by using a nationally representative cross-sectional study of pfrhp2-deletion in the DRC ( Parr et al . , 2016 ) to estimate the prevalence of pfhrp2-deleted mutants prior to RDT introduction . This , in turn , allows us to map predicted geographical regions across SSA where pfhrp2-deletion surveillance should be focused . These mapped predictions are explored across a range of estimates of the prevalence of pfhrp2-deleted mutants prior to RDT introduction .
Using our newly adapted model incorporating the transmission of pfhrp2-deleted mutants , we first explored the potential for RDT-guided treatment decisions to exert an evolutionary pressure on the prevalence of the mutant . Figure 1 shows the predicted proportion of strains that are pfhrp2-deleted within the population after 10 years . Within all settings that explored different transmission intensities and starting frequencies of pfhrp2-deletion , RDT introduction is predicted to increase the proportion of pfhrp2-deleted mutants . The strength of selection is predicted to be greatest at low PfPR ( Figure 1a ) ; however , a selective pressure is still predicted at both high PfPR and at low starting pfhrp2-deletion frequencies ( Figure 1b ) . The variance in the selection pressure exerted by RDTs is also predicted to be greatest at low PfPR ( Figure 1c ) . A more gradual but analogous trend is predicted in the proportion of the population that were only infected with pfhrp2-deleted mutants ( Figure 1d ) . The prevalence of malaria within Figure 1a was also observed to increase after RDT introduction ( Figure 1—figure supplement 1 ) , with the greatest increase in lower transmission settings resulting from untreated infections due to false-negative RDT results . Within the sensitivity analyses , a selective pressure is observed to exist at comparative fitness costs of greater than 90% ( see Figure 1—figure supplement 2 ) , however below this the pfhrp2-deletion allele is quickly lost . Both the introduction of additional diagnosis with microscopy-based methods and non-adherence to RDT results decreased the selective pressure , slowing the rate of pfhrp2-deletion emergence ( see Figure 1—figure supplement 3 ) . The introduction of non-malarial fevers , however , increased the rate of pfhrp2-deletion emergence ( see Figure 1—figure supplement 4 ) , even at 25% below the mean estimated rate of non-malarial fever . When these opposing factors were combined , RDT introduction is still predicted to increase the proportion of pfhrp2-deleted mutants ( Figure 1—figure supplement 5 ) . The proportion of clinical cases seeking treatment ( assumed here to be treated on the basis of an RDT result ) is also predicted to exert a strong selection pressure for pfhrp2-deletion ( Figure 2 ) . A consistent relationship was seen across comparable PfPR ranges , with the lowest treatment seeking rates ( fT = 0 . 2 ) yielding the slowest increase in the proportion of infections due to only pfhrp2-deleted mutants . Again , the lower PfPR categories show the greatest selection pressures for pfhrp2-deletion , with treatment seeking rates >30% and PfPR <25% leading to 20% of infections due to only pfhrp2-deleted mutants in fewer than five years ( Figure 2a ) . The selection pressure favouring pfhrp2-deletions is predicted to be weaker when PfHRP3 epitopes are assumed to cause positive RDT results ( Figure 2b ) . In settings where PfHRP3 epitopes are assumed to cause a positive RDT result in 25% of cases ( ε = 0 . 25 ) , there are four fewer prevalence categories that reach 20% of infections due to only pfhrp2-deleted mutants in fewer than five years . A similar effect is observed in the mean final frequency of pfhrp2-deletion , with 64% frequency recorded after 20 years when no PfHRP3 epitope effect is assumed in comparison to 56% when ε is equal to 0 . 25 ( Figure 2—figure supplement 1 ) . To estimate the starting frequency of pfhrp2-deleted mutants , we used estimates of the proportion of pfhrp2-deleted mutants from a national study in DRC ( Parr et al . , 2016 ) to calibrate the model . The calibration incorporated both the PfPR levels and estimates of the treatment rates in the 26 Divisions Provinciales de la Santé ( DPS ) that would drive selection of the mutant . We estimate a starting frequency of pfhrp2-deleted P . falciparum of 6% in the DRC prior to any introduction of RDTs . The observed relationship between the proportion of infections due to pfhrp2-deleted mutants and PCR PfPR among children 6–59 months of age ( Figure 3a ) displays a similar trend to the simulations , however with a notably steeper increase at lower prevalence . Of note , the same relationship was not predicted in the absence of selection pressure due to RDT-based treatments ( i . e . purely on the basis of the variation in monoclonal infections ) ( Figure 3b ) . Finally , using the baseline frequency estimate of 6% prior to RDT introduction , we explored 1000 different prevalence and treatment seeking rates spanning the range of estimates of the PfPR ( Bhatt et al . , 2015 ) and treatment levels across sub-Saharan Africa ( SSA ) in 2010 ( Cohen et al . , 2012 ) ( Figure 4—figure supplement 2 ) . The model output was aligned with these estimates by first administrative units ( Figure 4—figure supplement 1 ) , which enabled us to project the potential increase of the mutant strain and its impact on RDT-guided treatment ( Video 1 ) . Our results suggest that 160 of 850 first-administrative regions may have over 20% of all infections due to only pfhrp2-deleted mutants by 2016 ( Figure 4c ) . These areas , which we term of ‘high HRP2 concern’ , are largely located in areas where PfPR2-10 in 2010 was less than 25% ( Figure 4a ) . A number of other regions , classified as ‘moderate HRP2 concern’ have high treatment rates , and hence potential selective pressure , despite having comparatively higher transmission ( Figure 4b ) . Our results also illustrate that regions with low transmission may have low HRP2 concern if the frequency of people seeking treatment is very low .
Our results demonstrate that the key drivers of pfhrp2-deletion selection are low malaria transmission and a high frequency of people seeking treatment and being correctly treated on the basis of diagnosis with a PfHRP2-based RDT . Based on Africa-wide estimates of parasite prevalence and treatment-seeking behaviour at the time of RDT-introduction , we identified 160 first-administrative units which we classify as ‘high HRP2 concern’ . These are areas where the pfhrp2-deleted strain is expected to increase in frequency over a relatively short timescale , and hence where further surveillance efforts should be concentrated . Our results are based on calibration to a large representative survey of malaria across DRC . Due to its size and location in the centre of SSA , the DRC is arguably one of the most representative countries for endemic malaria in Africa . That the model was able to predict the observed relationship in the DRC , despite variability at a province level , provides support for the hypothesis that the variability in pfhrp2-deletion frequency with transmission is driven by selection . However , in contrast to other reported surveys , the samples in this survey were primarily drawn from asymptomatic infections , and hence may not be representative of other reports of pfhrp2-deletion in symptomatic cases with higher parasite density . However , it is interesting to note that our results show broad agreement with published data sets from Zambia ( Laban et al . , 2015 ) and Ghana ( Amoah et al . , 2016 ) ( Table 1 ) . In particular , our predictions confirm that the HRP2 concern would be greater in Ghana than in Southern Zambia . However , one study in Senegal found a lower prevalence of pfhrp2-deletion than we predict ( Wurtz et al . , 2013 ) . A key uncertainty in predicting the potential spread of pfhrp2-deletion due to RDT-induced selective pressure is the extent of use of , and adherence to , RDT results and the availability of appropriate treatment . On the one hand , if adherence to RDT results is poor ( for example , with patients who show continued clinical symptoms of malaria in the absence of a positive test ) or additional microscopy-based detection is used ( Figure 1—figure supplement 3 ) , if appropriate treatment is not available ( for example , due to stock-outs ) , or if treatment is not fully curative ( for example , due to patient non-adherence , drug resistance or fake drugs ) then the spread of these deletions will be slower than predicted . On the other hand , in areas in which active case detection occurs , or in which treatment is sought for non-malaria fevers ( Figure 1—figure supplement 4 ) , RDT-based treatment may also selectively clear non-deleted asymptomatic infections and hence increase the rate of spread of the deletion . However , when these factors , along with potential fitness costs associated with pfhrp2-deletion , were investigated together we still observed an increase in pfhrp2-deletion ( Figure 1—figure supplement 5 ) , which showed a similar rate of increase to that predicted by our model . Further data on RDT usage and adherence , as well as on non-malarial fevers and the precise fitness cost of pfhrp2-deletion , however , could help to refine mapping of areas of HRP2 concern . A second uncertainty relates to the underlying prevalence of the mutation prior to RDT introduction . There is considerable variability in the estimates that have been measured both before and after RDT introduction , and it is entirely plausible that the presence of mutations could vary geographically at a range of spatial scales . However , estimating this variation is difficult given the lack of a sampling framework in reports mainly based on clinical cases and given the relatively small sample sizes . Thus our results should be interpreted not as predictions of the absolute levels of the gene deletion , but rather indicative of geographical areas in which surveillance should be focused . Similarly , these results should not be interpreted as predictions of the precise negative impact on malaria prevalence as a result of increased gene deletions ( Figure 1—figure supplement 1 ) , but illustrative of the potential impact of false-negative test results upon malaria prevalence and the importance of alternative diagnostic methods ( Figure 1—figure supplement 3 ) . At the same time , further data collated in the coming months and years can be incorporated to iteratively update and refine our projections . As with any modelling exercise , there are a number of important limitations . Firstly , we did not capture seasonality or any fitness cost associated with pfhrp2-deletion . At a given transmission level , highly seasonal locations are likely to have a lower frequency of pfhrp2-deletion in comparison to regions with perennial transmission . Seasonality could however cause substantial bottlenecks which may result in repetitive founder effects that could affect selection , resulting in either a decreased chance of pfhrp2-fixation or an accelerated fixation if it occurred ( Aguilée et al . , 2009 ) . In simulations incorporating a fitness cost the selection pressure was found to be considerably weaker ( Figure 1—figure supplement 2 ) . The exact fitness cost , despite being unknown , is likely subtle as our modelled fitness penalty would cause the strain to be eliminated at less than 90% comparative fitness . In addition , current theories concerning the role of PfHRP2 indicate a more minor role in heme detoxification than previously thought . Strains lacking PfHRP2 have been shown to be viable ( Papalexis et al . , 2001 ) , with heme detoxification more dependent on the recently characterised haem detoxification protein ( HDP ) ( Jani et al . , 2008 ) . Furthermore , in South America the first cases of pfhrp2-deleted P . falciparum were confirmed prior to the introduction of RDTs ( Gamboa et al . , 2010 ) . This suggests that these mutants may possess sufficiently high fitness such that the frequency of pfhrp2-deletion is maintained in the absence of a selective advantage exerted through the use of HRP2-based RDTs . Secondly , our results depend on assumptions made regarding the contribution of PfHRP3 epitope cross-reactivity and the potential for false-positive RDT results . We found that increased cross-reactivity with PfHRP3 epitopes decreases selection for pfhrp2-deletion and was investigated due to confirmed observations of PfHRP2-based RDTs detecting PfHRP3 epitopes at high parasitaemia ( Baker et al . , 2010 ) . In simulations with no epitope effect , the model predicts the pattern in DRC well ( Figure 3—figure supplement 1 ) and predicts a higher overall estimate of HRP2 concern ( Figure 4—figure supplement 1 ) , although the same regional patterns are identified ( Figure 4—figure supplement 2 ) . Furthermore , false-positive RDT results would decrease the strength of the selection pressure , with pfhrp2-monoclonal infections being treated . However , false-positivity rates observed within round 6 of WHO RDT product testing were found to be low , with the median false-positive rate on both clean negative samples and samples containing other infectious agents equal to 0% , and the overall false-positive rate on samples containing immunological factors equal to 0 . 9% ( World Health Organization , 2015c ) . Thirdly , in the absence of systematic country introduction data , we assumed introduction of RDTs in all countries from 2010 in accordance with the WHO recommendation of testing in 2010 ( World Health Organization , 2010 ) . The precise date from region to region is undoubtedly more complex , however 2010 is a sensible estimate given the reported years at which RDTs were available at the community level in SSA by the WHO ( Figure 4—figure supplement 5 ) ( World Health Organization , 2012b ) . However , the ratio of testing via microscopy versus RDT is likely to have decreased over this period , and hence our estimate of RDT use ( which our model assumes is 100% from introduction ) is likely too high . The sensitivity of the output to this parameter is demonstrated in the data from the DRC , in which higher levels of pfhrp2-deletion are observed in Kivu , an area in which RDT introduction likely occurred earlier than elsewhere in the country ( Médecins Sans Frontières , 2007 ) . Fourthly , extrapolating the starting frequency of pfhrp2-deletion strains from the DRC across the rest of SSA is a clear oversimplification; however , in the absence of similar datasets , we feel it provides a reasonable first estimate . To assess the implications of this estimate , we also considered how the pattern of geographical areas that we have recommended for priority surveillance changes under different assumed starting frequencies of pfhrp2-deletion ( Figure 4—figure supplement 4 ) . Despite the expected changes in the final frequency of pfhrp2-deletion in these settings , the overall pattern of areas with the highest selection-driven increase in pfhrp2-deletion remains the same . A final limitation is that we assumed that treatment rates and transmission of malaria remain constant from 2010 . This is clearly not the case , with 30 countries in SSA reporting a decline in prevalence from 2010 to 2015 ( Bhatt et al . , 2015 ) . These combined effects , however , would presumably cause an increase in monoclonal infections and subsequent false-negative RDTs due to pfhrp2-deleted parasites . In summary , our modelling predicts that an increased emergence of pfhrp2-deleted mutants may be explained by the introduction of testing by PfHRP2-based RDTs . If this is indeed the case , this would be , to our knowledge , one of the first examples of the emergence of resistance of a pathogen to a diagnostic test . The use of these RDTs will result in the greatest selection pressure in regions that have low malaria transmission and a high frequency of RDT-based treatment of clinical cases . Rapid and accurate diagnosis of P . falciparum infection , however , is essential for continued reduction in malaria transmission . In light of this , it may be sensible for public health bodies who are responding to reports of pfhrp2 gene deletions to focus surveillance in the regions we have identified as having a high HRP2 concern . This work should proceed alongside further improvement of non-HRP2-based RDTs , such as those that detect lactate dehydrogenase , and the development of new alternative diagnostics .
An individual-level stochastic model was developed to simulate the transmission dynamics of Plasmodium falciparum . The model is based upon previous modelling efforts ( Griffin et al . , 2016; Griffin et al . , 2014; Griffin et al . , 2015 ) , and is described in full here before describing the extensions made with regards to PfHRP2 dynamics , and defining the parameters used and their sources . The model is implemented as stochastic individual-based model with a fixed daily time step , incorporating the necessary delay terms where mentioned , which is described in greater detail later . In overview , the transmission model considers people to exist in one of six infection states ( Figure 5 ) : susceptible ( S ) , clinical disease ( D ) , clinically diseased and receiving treatment ( T ) , asymptomatic infection ( A ) , protective state of prophylaxis ( P ) , and subpatent infection ( U ) . Individuals begin life susceptible to infection ( state S ) . At birth , individuals possess a level of maternal immunity that decays exponentially over the first 6 months . Each day individual i is probabilistically exposed to infectious bites governed by their individual force of infection ( Λi ) . Λi is dependent on their pre-erythrocytic immunity , biting rate ( dependent on both their age and their individual relative biting rate due to heterogeneous biting patterns in mosquitoes ) and the mosquito population’s size and infectivity . Infected individuals , after a latent period of 12 days ( dE ) , develop either clinical disease ( state D ) or asymptomatic infection ( state A ) . This outcome is determined by their probability of acquiring clinical disease ( ϕ ) , which is dependent on their clinical immunity . Individuals that develop disease have a fixed probability ( fT ) of seeking treatment ( state T ) , and a variable probability ( εi ) that the clinical case yields a positive diagnostic result and subsequently receives treatment . εi is dependent on the assumed role of PfHRP3 epitopes , and the strain profile of infected individual i with respect to pfhrp2-deleted mutants . Treated individuals are assumed to always recover , i . e . fully-curative treatment , and then enter a protective state of prophylaxis ( state P ) at rate rT , before returning to susceptible at rate rS . Individuals that did not receive treatment recover to a state of asymptomatic infection at rate rD . Asymptomatic individuals progress to a subpatent infection ( stage U ) at rate rA , before clearing infection and returning to susceptible at rate rU . Additionally , superinfection is possible for all individuals in states D , A and U . Superinfected individuals who receive treatment will move to state T . Individuals who are superinfected but do not receive treatment in response to the superinfection will either develop clinical disease , thus moving to state D , or develop an asymptomatic infection and move to state A ( except for individuals who were previously in state D , who will remain in state D ) . The introduction of a fixed time step translates the waiting times at which individuals move from one infection state to another into a daily probability that this event occurs , with the probability drawn from the related exponential distribution . Thus the probability of a transition from state A to state B with hazard rate λ is given by:ProbA →B: 1-exp-λ The set of state transitions for individuals and their associated hazard rates are given below . Here state I denotes an infection state which is not tracked but which leads to either clinical disease ( D ) , treated clinical disease ( T ) or asymptomatic infection ( A ) . In the original model the probability of entering these states is determined by drawing a sequence of Bernoulli trials for each infected individual as:ProbClinical Disease: BernoulliϕiProbTreated Clinical Disease | Clinical Disease: BernoullifT For our model here , in which treatment is guided by RDT-based diagnostics , we introduce…ProbTreated Clinical Disease | Clinical Disease: BernoullifTεi We assume that each person has a unique biting rate , which is the product of their relative age dependent biting rate , ψi , given byψi ( a ) =∑i=1nψi ( a ) n ( 1−ρexp−aa0 ) and an assumed heterogeneity in biting patterns of mosquitoes , ζi , which we assume persists throughout their lifetime and is drawn from a log-normal distribution with a mean of 1 , logζi~ N-σ22 , σ2 where 1-ρ is the relative biting rate at birth when compared to adults and a0 represents the time-scale at which the biting rate increases with age . The product of these biting rates is subsequently used to calculate an individual’s entomological inoculation rate , hi , and subsequently their force of infection , which are given byhi= αkIMζiψiΛi= hibi where αk is the daily rate at which a mosquito takes a blood meal , IM is the size of the infected mosquito population , and bi is the probability of infection given an infectious mosquito bite . The human population was assumed to have a maximum possible age of 100 years , with an average age of 21 years within the population yielding an approximately exponential age distribution typical of sub-Saharan countries . When an individual dies , they are replaced with a new-born individual whose individual biting rate due to heterogeneity in biting patterns is drawn again from a log-normal distribution with a mean of 1 . We consider three stages at which immunity may impact transmission: Maternal clinical immunity is assumed to be at birth a proportion , PM , of the acquired immunity of a 20 year-old and to decay at rate 1dM . The probabilities of infection , detection and clinical disease are subsequently created by transforming each immunity function by Hill functions . An individual’s probability of infection , bi , is given bybi=b0b1+1-b11+IBIB0κB where b0 is the maximum probability due to no immunity , b0b1 is the minimum probability and IB0 and κB are scale and shape parameters respectively . An individual’s probability of clinical disease , ϕi , is given byϕi=ϕ0ϕ1+1-ϕ11+ICA+ICMIC0κC where ϕ0 is the maximum probability due to no immunity , ϕ1ϕ0 is the minimum probability and IC0 and κC are scale and shape parameters respectively . An individual’s probability of being detected by microscopy when asymptomatic , qi , is given byqi=d1+1-d11+IDID0κDfD where d1 is the minimum probability due to maximum immunity , and ID0 and κD are scale and shape parameters respectively . fD is dependent only on an individual’s age is given bydfDda=1-1-fD01+aaDγD where fD0 represents the time-scale at which immunity changes with age , and aD and γD are scale and shape parameters respectively . Lastly , αA and αU are parameters that determine the probability that an individual in states A and U are detectable by PCR , which are given by qαA and qαU respectively . The contribution made by each infected individual towards the overall infectiousness of the human population towards mosquitoes is proportional to both their infectious state and their probability of detection , with a lower probability of detection assumed to correlate with a lower parasite density . Individuals who are in state D ( clinically diseased ) , state U ( sub-patent infection ) and state T ( receiving treatment ) contribute cD , cU and cT . In state A , infectious contribution , cA , is given by cU+cD-CUqγI where q is the probability of being detected by microscopy when asymptomatic , and γI is a parameter that controls how quickly infectiousness falls within the asymptomatic state . Given the definitions above , the full stochastic individual-based human component of the model can be formally described by its Kolmogorov forward equations . As before , let iindex individuals in the population . Then the state of individual i at time t is given by j , k , tk , l , tl , m , tm , cm , a , t , where a is age , j represents infection status ( S , D , A , U , T or P ) , k is the level of infection-blocking immunity and tk is the time at which infection blocking immunity was last boosted . Similarly , l and tl denote the level and time of last boosting of clinical immunity , respectively , while m and tm do likewise for parasite detection immunity , and cm represents maternal immunity . Let δp , q denote the Kronecker delta ( δp , q=1 if p=q and 0 otherwise ) and δx denote the Dirac delta function . Defining Pi ( j , k , tk , l , tl , m , tm , cm , a , t ) as the probability density function for individual i being in state j , k , tk , l , tl , m , tm , cm , a , t at time t , the time evolution of the system is governed by the following forward equation: ( 1 ) ∂Pi ( j , k , tk , l , tl , m , tm , cm , a , t ) ∂t+∂Pi ( j , k , tk , l , tl , m , tm , cm , a , t ) ∂a= ( 2 ) δj , S[rPPi ( P , k , tk , l , tl , m , tm , cm , a , t ) +rUPi ( U , k , tk , l , tl , m , tm , cm , a , t ) ] ( 3 ) +δj , ArDPiD , k , tk , l , tl , m , tm , cm , a , t ( 4 ) +δj , UrAPiA , k , tk , l , tl , m , tm , cm , a , t ( 5 ) +δj , PrTPiT , k , tk , l , tl , m , tm , cm , a , t ( 6 ) + ( 1−bi ) hi ( t−dE ) [δj , S+δj , D+δj , A+δj , U]Ob⋄Pi ( j , k , tk , l , tl , m , tm , cm , a , t ) ( 7 ) +bihi ( t−dE ) [δj , A ( 1−ϕi ) +δj , Dϕi ( 1−εifT ) +δj , TϕiεifT] Ob⋄Oc⋄Ob⋄∑j′∈{S , A , U}Pi ( j′ , k , tk , l , tl , m , tm , cm , a , t ) ( 8 ) +bihi ( t−dE ) [+δj , Dϕi ( 1−εifT ) +δj , TϕiεifT] Ob⋄Oc⋄Od⋄Pi ( D , k , tk , l , tl , m , tm , cm , a , t ) ( 9 ) +[rBk∂∂k+rCAl∂∂l+rIDm∂∂m+rCMcm∂∂cm]Pi ( j , k , tk , l , tl , m , tm , cm , a , t ) ( 10 ) +μ δ ( a ) δ ( tk+Tbig ) δ ( tl+Tbig ) δ ( tm+Tbig ) δj , Sδk , 0δl , 0δm , 0∑j′Pi ( j′ , k , tk , l , tl , m , tm , cm , a , t ) ( 11 ) −[μ+rPδj , P+rUδj , U+rDδj , D+rAδj , A+rTδj , P+hi ( t−dE ) [δj , S+δj , D+δj , A+δj , U]]Pi ( j , k , tk , l , tl , m , tm , cm , a , t ) Here 𝒪b , 𝒪c and 𝒪d are commutative integral operators with the following action on an arbitrary density fj , k , tk , l , tl , m , tm , cm , a , t:Ob⋄f= δ ( t−tk ) ∫0∞f ( j , k−1 , t−uB−τ , l , tl , m , tm , cm , a , t ) dτ+ θ ( t−tkuB ) f ( j , k , tk , l , tl , m , tm , cm , a , t ) Oc⋄f= δ ( t−tl ) ∫0∞f ( j , k , tk , l−1 , t−uC−τ , m , tm , cm , a , t ) dτ+ θ ( t−tluC ) f ( j , k , tk , l , tl , m , tm , cm , a , t ) Od⋄f= δ ( t−tm ) ∫0∞f ( j , k , tk , l , tl , m−1 , t−uD−τ , cm , a , t ) dτ+ θ ( t−tmuD ) f ( j , k , tk , l , tl , m , tm , cm , a , t ) . Finally , θx is an indicator function such that θx=1 if x<1 and 0 otherwise . These functions allow the fixed periods of time in which immunities are not boosted after a previous boost to be included within the stochastic equations , while also allowing superinfection events to be incorporated . For simulation , a discrete time approximation of this stochastic model was used , with a time-step of 1 day . For each individual k , l and m are set to zero at birth , while tk , tl and tm are set to a large negative value -Tbig ( to represent never having been exposed or infected ) . Each immunity term increases by 1 for an individual whenever that individual receives an infectious bite ( k ) , or is infected ( l and m ) , if the previous boost to k , l and m occurred more than uB , uC and uD days earlier , respectively . Immunity levels decay exponentially at rate rB , rCA and rID , where rB , rCA and rID are equal to 1dB , 1dCA and 1dID respectively . The stochastic model equations detailed above can be explained as follows . The first line is the total time derivative of Pi ( j , k , tk , l , tl , m , tm , cm , a , t ) . The next four lines describe the flows into states S , A , U and P due to progression through infection states . The sixth line describes exposure to malaria that boosts pre-erythrocytic immunity but does not lead to an infection . The first term within the commutative integral operator 𝒪b here considers the density of individuals who are in immunity class k-1 and whose last boost to their pre-erythrocytic immunity was more than uB days earlier , and thus will be flowing into the considered density , k , from a lower pre-erythrocytic immunity . The second term in the integral will equate to 1 when considering individuals who are in immunity class k and whose last boost to their pre-erythrocytic immunity was less than uB days earlier and thus do not see their immunity boosted and hence remain in class k . This is needed to represent the current density of individuals in the considered density . There is no term for individuals in immunity class k whose last boost to their pre-erythrocytic immunity occurred more than uB days earlier as they would move out of the considered density ( into class k + 1 ) and hence the indicator function will equate to 0 for these individuals . The seventh line describes exposure events occurring to individuals in states S , A and U which do result in patent ( blood-stage ) infection , resulting in transition into states A , D or T . The force of infection acting on the density in state D is not included here but rather in the eighth line since these individuals may only move to states T or D and not A . In both the seventh and eighth lines , the commutative operators here function as described earlier . This tracks the density of individuals in immunity states one lower whose last boost to any of the three immunity types occurred a sufficient number of days earlier to flow into the considered density , while also considering the individuals already at the same immunity as the considered density to remain in their current combined infection/immunity state ( when the indicator function equates to 1 ) or to move to a new infection/immunity state ( when the indicator function equates to 0 ) . The ninth line ( effectively a first order wave equation ) represents deterministic exponential decay of the four different types of immunity . The tenth line represents the birth process . We assume a constant population size , so upon death , individuals flow into the state with no immunity and last immunity boosting times are set to -Tbig , chosen to be sufficiently early to allow immediate boosting upon exposure to infection ( i . e . zero immunity other than maternal at birth ) . The last line shows the removal of individuals from the population through death , balancing the inflow from the previous line . The adult stage of mosquito development was modelled in a compartmental formulation . Susceptible adult mosquitoes ( SM ) become infected at a rate which is proportional to the infectiousness of the human population lagged by dg days , which represents the delay from emergence of asexual blood-stage parasites to sexual gametocytes that contribute towards onward infectivity . The force of infection towards mosquitoes on a given day , ΛM , is represented by the sum of the contributions from each infected human , delayed by dg , towards the overall infectiousness of the human population , which is given byΛM=αkN ( ∑i=1ΣDζiψicD+∑i=1ΣTζiψicT+∑i=1ΣAζiψicA+∑i=1ΣUζiψicU ) ( t−dg ) Infected mosquitoes then pass through a latent stage ( EM ) of duration dEM , before becoming infectious to humans ( IM ) . Infectious mosquitoes remain infectious until they die . The differential equations governing the adult stage of mosquitoes are given bydSMdt=μMMv-μMSM-ΛMSMdEMdt=ΛMSM−μMEM−ΛM ( t−dEM ) SM ( t−dEM ) exp−μMdEMdIMdt=ΛM ( t−dEM ) SM ( t−dEM ) exp−μMdEM−μMIM where μM is the daily death rate of adult mosquitoes , and Mv is the total mosquito population , i . e . SM+EM+IM . Individuals that are newly infected receive either a pfhrp2-deleted mutant or a wild type , determined probabilistically by the ratio of pfhrp2-deleted mutants in the contribution to onwards infectiousness governed by the human infectious population delayed by dEM . An individual with clinical disease ( state D ) , who possesses an equal number of wild type and pfhrp2-deleted mutants will for example contribute ½cD to both the wild type and the mutant profile of the onwards infectiousness to mosquitos . In simulations incorporating a fitness cost associated with pfhrp2-deletion the contribution terms ( cD , cT , cA and cU ) required to calculate the contribution to the human infectious reservoir made by the deletion strains are comparatively decreased relevant to the wild type strains in order to represent an assumed decrease in parasitaemia and onward transmission . This effect would also capture if the pfhrp2-deleted strain is comparatively less fit within the mosquito . This would be of importance when considering blood meals taken by mosquitoes feeding on polyclonally infected individuals , in which we would expect the fitter wild-type parasite to be probabilistically more likely to be onwardly transmitted . If a newly infected individual is only infected with pfhrp2-deleted mutants the probability that they enter the treated class is εfT , where ε is equal to the cross reactivity contribution of PfHRP3 epitopes . If , however , they contain any wild type strains , ε is assumed to always equal 1 . Additionally , if a subpatent individual is superinfected resulting in clinical disease , ε is assumed to equal the cross reactivity contribution of PfHRP3 epitopes if the acquired strain from superinfection is pfhrp2-deleted . Analogously , ε is assumed to always equal 1 if the acquired strain was wild type . This is to reflect the inability of RDTs to detect any of the strains that were previously present within the subpatent individual . Individuals that clear infection lose all strains , and infected individuals clear a random strain at rate nrC , where n is the total number of strains and rC is the rate at which one strain is cleared in a monoinfected individual , that is 1dA + dU . This introduces a carrying capacity on the number of strains an individual can be infected with , which scales with the transmission intensity . The multiplicity of infection and strain profile of an infected individual have no effect on the disease outcome except when the use of RDTs is introduced . The model code was developed using the R language ( RRID:SCR_001905 ) , ( R Core Team , 2016 ) and is available with shape files and plotting scripts through an open source MIT license at https://github . com/OJWatson/hrp2malaRia ( Watson , 2017 ) . A copy is archived at https://github . com/elifesciences-publications/hrp2malaRia . The model is also written out in full as a pseudocode model using mathematical syntax ( Supplementary file 1 ) . Parameter estimates used within the model were taken from Griffin et al . ( 2014 ) , ( 2015 ) and ( Griffin et al . , 2016 ) however have been included in Table 2 for clarity . The rate of pfhrp2-deleted mutant emergence after the introduction of RDTs was examined across a range of malaria transmission intensities ( 10% , 25% and 60% parasite prevalence across all ages [PfPR] ) and starting proportions of pfhrp2-deleted mutants ( 2% , 8% and 12% mutants ) . For all simulations conducted , the proportion of clinically diseased cases seeking treatment was equal to 40% ( fT = 0 . 4 ) . In all simulations ten stochastic realisations of 100 , 000 individuals were simulated for 60 years to reach equilibrium first , before exploring different parameter settings . RDTs were then introduced and the proportion of strains that are pfhrp2-deleted recorded over the following 10 years . The effect of transmission intensities ( 0%–90% PfPR ) was explored further by recording both the proportion of strains that are pfhrp2-deleted and the proportion of individuals only infected with pfhrp2-deleted mutants at 5 years after RDT introduction . In these simulations PfHRP3 epitopes were assumed to never yield a positive RDT result ( ε = 0 . 0 ) . The rate of emergence was further examined under different assumptions about the proportion of people seeking treatment ( fT = 0 . 2–0 . 6 ) , recording the time taken for the proportion of individuals only infected with pfhrp2-deleted mutants to reach 20% . For each simulation we assumed 8% of strains carried pfhrp2-deleted mutants prior to RDT introduction . We also considered the role that PfHRP3 antigens may have in the performance of PfHRP2-based RDTs , assuming that 25% of individuals only infected with pfhrp2-deleted mutants receive treatment due to the presence of PfHRP3 epitopes ( ε = 0 . 25 ) , based on an estimate of PfHRP2-based RDT cross-reactivity ( Baker et al . , 2005 ) . A series of analyses were additionally conducted to characterise the impact of a number of assumptions within the model . These sensitivity analyses were conducted at 20% PCR PfPR across all ages , with the proportion of clinically diseased cases seeking treatment equal to 40% ( fT = 0 . 4 ) , and an assumed starting frequency of pfhrp2-deletion equal to 10% . As before , in all simulations ten stochastic realisations of 100 , 000 individuals were simulated for 60 years to reach equilibrium first , before exploring different parameter settings . We initially assessed the impact upon the strength of selection of a range of assumed comparative fitness costs associated with pfhrp2-deletion ( 5%–100% ) . Secondly , we explored the impact on selection of introducing additional microscopy-based diagnosis , occurring in 30% of cases in alignment with the 71% use of RDTs in 2014 ( World Health Organization , 2015a ) , with and without an assumed non-adherence to RDT results in 10% of cases in alignment with estimated improving levels of adherence to RDT results ( D’Acremont et al . , 2013 ) . Lastly , we investigated the impact of non-malarial fevers ( NMF ) , introducing an estimate for the annual NMF rate . This estimate was sourced by first finding household Demographic Health Surveys in SSA that surveyed whether individuals had been previously sick with a fever in the last 2 or 4 weeks and if and where they sought treatment for that fever . The resultant six surveys ( Institut de Statistiques et d’Études Économiques du Burundi - ISTEEBU et al . , 2013 ) , ( Institut de Statistiques et d’Études Économiques du Burundi - ISTEEBU et al . , 2013 ) Liberia ( 2009 [National Malaria Control Program - NMCP/Liberia , Ministry of Health and Social Welfare/Liberia et al . , 2012] and 2011 [National Malaria Control Program - NMCP/Liberia , Ministry of Health and Social Welfare et al . , 2009] ) , Mali ( 2010 ) and Nigeria ( 2010 [National Population Commission - NPC/Nigeria et al . , 2012] and 2015 [National Malaria Elimination Programme - NMEP/Nigeria et al . , 2016] ) were then subset by those that had sufficiently representative fever data across all ages , which yielded five surveys with Mali failing to be sufficiently representative at higher age ranges . The survey from Burundi was chosen for further analysis as both surveys from Liberia and Nigeria reported substantial treatment sought from drug peddlers and drug hawkers respectively , complicating inference on the clinical outcome of any treatment sought for the fever . From this survey an age-bracketed annual rate of fever that led to treatment being sought was calculated , with smaller age brackets used at younger ages to capture the rapid change in fever rates at younger ages . This annual estimate was then scaled by 57% to represent the likely NMF rate , as estimated from a large scale estimate across Africa ( Gething et al . , 2010 ) . We subsequently incorporated this rate to explore the impact of NMF upon selection . Within these simulations , we assume that individuals currently treated or in prophylaxis will not receive further antimalarial treatment when presenting with a NMF . Susceptible and subpatent individuals who seek treatment due to a NMF will only receive treatment due to non-adherence to test results . Lastly , asymptomatic and diseased individuals who seek treatment due to a NMF will always receive treatment , unless they are monoinfected with pfhrp2-deleted parasites in which case they will only be treated due to potential PfHRP3 epitope contributions , non-adherence to RDT results or if they were diagnosed with microscopy-based diagnosis . To estimate the current and future proportion of pfhrp2-deleted mutants across SSA , we require a starting frequency of pfhrp2-deletion . We used estimates of the proportion of pfhrp2-deleted mutants ( Parr et al . , 2016 ) from the 2013–2014 DRC Demographic and Health Survey ( DHS ) ( Meshnick et al . , 2015 ) to infer the starting frequency before RDTs were introduced in 2010–2011 , ( Meshnick et al . , 2015 ) using the weighted PCR prevalence of malaria in children aged 6–59 months ( PCR PfPR 6–59 months ) and the reported frequency of people seeking treatment in the 26 Divisions Provinciales de la Santé ( DPS ) . The DHS survey was a nationally representative cross-sectional study of 7137 children aged 6–59 months and 783 subjects with RDT-/PCR+ results were tested using PCR assays to detect and confirm pfhrp2-deletion . We explored 50 starting frequencies between 0 . 1%–10% , with an assumed probability of a clinical case seeking treatment , who is only infected with pfhrp2-deleted mutants , producing a positive RDT result ( ε ) equal to 0 . 25 . RDTs were assumed to be introduced in 2010 except for North- and South-Kivu where the use of RDTs occurred from 2007 in the refugee camps . ( Médecins Sans Frontières , 2007; United Nations High Commissioner for Refugees , 2013 ) For each starting frequency , ten stochastic realisations of 100 , 000 individuals were simulated for each DPS at malaria prevalence levels aligned to the observed weighted PCR prevalence of malaria in children aged 6–59 months for these provinces . These simulations were run for 60 years prior to the introduction of RDTs to ensure equilibrium was reached . The output from each set of simulations at a given starting frequency was smoothed using a local regression ( LOESS ) model , and the starting frequency identified as the set of simulations with the smallest residual sum of squares when compared to the recorded relationship from the DHS survey . It is important to highlight that due to the non-spatial nature of the model , each geographical region simulated occurs independently to neighbouring regions , i . e . there is no spatial spread of parasites between regions . Additionally , novel mutation emergence was not modelled explicitly and thus stochastic loss of the pfhrp2-deletion genotype would always yield a final pfhrp2-deletion frequency of 0% . The estimated starting frequency was then used to simulate trends in the prevalence of pfhrp2-deleted mutants across SSA , exploring a range of treatment coverages and transmission intensities , with ε = 0 . 25 . These simulations considered populations of 100 , 000 individuals that were simulated for 20 years from 2010 to 2030 , with the introduction of RDTs assumed across all regions in 2010 . These outputs were matched to the mean microscopy-based PfPR in 2–10 year olds ( PfPR2-10 ) in 2010 by first administrative unit and estimates of the proportion of cases seeking treatment from previously modelled estimates using the DHS and the Malaria Indicator Cluster Surveys ( Cohen et al . , 2012 ) . The time taken for the proportion of infections due to only pfhrp2-deleted mutants to reach 20% was recorded and classified to map areas of HRP2 concern under four qualitative classifications shown in Table 3 . | Since the turn of the millennium , a large increase in international funding has helped to reduce the public health impact of malaria . The introduction of rapid diagnostic tests has played a central role in these efforts , particularly in remote areas that are heavily affected by the disease . These tests analyse human blood samples for specific proteins that are produced by malaria parasites . The most common rapid diagnostic tests for malaria detect a protein called HRP2 , which is produced by the deadliest malaria parasite , Plasmodium falciparum . Recently , however , cases have emerged where the tests have failed to detect these malaria infections . The first occurred in South America , and were found to be because some malaria parasites no longer possessed the gene that produces HRP2 . Since then , malaria parasites that lack this gene have been found in several locations in Africa . This raises the question of whether using the tests favours the survival and spread of parasites that cannot produce the HRP2 protein . Using mathematical modelling techniques , Watson et al . now present evidence that suggests that the use of HRP2-detecting rapid diagnostic tests over the past 10 years could have favoured the evolution of malaria parasites that lack this protein . Furthermore , the models suggest that the conditions that are most likely to cause such selection are places where malaria infections are not common but people seek treatment at high rates . Using this information , Watson et al . created a map of 160 locations in Africa most at risk of rapid diagnostic test-driven selection against the gene that produces HRP2 . Public health authorities could use these maps to determine where they should more closely monitor malaria parasites to see if they lack this gene . Future genetic investigations will be required in the high-risk areas to confirm and refine the predictions . The development of rapid diagnostic tests that detect other malaria proteins will also be essential if malaria parasites that lack HRP2 continue to spread . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"epidemiology",
"and",
"global",
"health"
] | 2017 | Modelling the drivers of the spread of Plasmodium falciparum hrp2 gene deletions in sub-Saharan Africa |
Cocaine is an addictive drug that acts in brain reward areas . Recent evidence suggests that cocaine stimulates synthesis of the endocannabinoid 2-arachidonoylglycerol ( 2-AG ) in midbrain , increasing dopamine neuron activity via disinhibition . Although a mechanism for cocaine-stimulated 2-AG synthesis is known , our understanding of 2-AG release is limited . In NG108 cells and mouse midbrain tissue , we find that 2-AG is localized in non-synaptic extracellular vesicles ( EVs ) that are secreted in the presence of cocaine via interaction with the chaperone protein sigma-1 receptor ( Sig-1R ) . The release of EVs occurs when cocaine causes dissociation of the Sig-1R from ADP-ribosylation factor ( ARF6 ) , a G-protein regulating EV trafficking , leading to activation of myosin light chain kinase ( MLCK ) . Blockade of Sig-1R function , or inhibition of ARF6 or MLCK also prevented cocaine-induced EV release and cocaine-stimulated 2-AG-modulation of inhibitory synapses in DA neurons . Our results implicate the Sig-1R-ARF6 complex in control of EV release and demonstrate that cocaine-mediated 2-AG release can occur via EVs .
The sigma-1 receptor ( Sig-1R ) is a small protein that resides at the endoplasmic reticulum ( ER ) -mitochondrion interface ( mitochondrion-associated ER membrane; MAM ) ( Hayashi and Su , 2007; Hayashi et al . , 2009; Mori et al . , 2013 ) , where it constrains type-3 inositol 1 , 4 , 5-trisphosphate receptors ( IP3R3 ) to facilitate Ca2+ signaling from ER to mitochondria ( Hayashi and Su , 2007; Hayashi et al . , 2009 ) . In addition , the Sig-1R binds a wide range of molecules , including psychotropic drugs and psychostimulants , such as cocaine and methamphetamine ( Largent et al . , 1987 ) , and can translocate to other cellular regions to associate with organelles , proteins , plasma membranes , and the nuclear envelope to control trafficking of other molecules , such as ion channels and receptors in neurons ( Su et al . , 2016; Yasui and Su , 2016 ) . These diverse signaling roles for the Sig-1R highlight a widespread influence on cellular function that is incompletely understood . Substantial data suggest that the Sig-1R is also a target of the abused psychostimulant cocaine ( Hayashi and Su , 2007; Sharkey et al . , 1988; Hayashi and Su , 2003; Kourrich et al . , 2013; Tsai et al . , 2015; Chen et al . , 2007 ) . In the mouse nucleus accumbens ( NAc ) , cocaine decreases the excitability of GABAergic medium spiny neurons by strengthening an association between the Sig-1R and Kv1 . 2 potassium channels , contributing to behavioral sensitization to the drug ( Kourrich et al . , 2013 ) . Moreover , the Sig-1R is also involved in cocaine reward ( Romieu et al . , 2002 ) . Given the diverse demonstrated roles for the Sig-1R in cellular signaling , its regulation by cocaine has the potential to affect many unknown cellular properties . Extracellular vesicles ( EVs ) are a diverse group of membranous entities of endosomal origin that are secreted from a broad range of cell types ( van Niel et al . , 2018 ) . The EV classification broadly includes exosomes and microvesicles that range in size from 30 to 150 nm , and 100–1000 nm , respectively . Exosomes are formed by invagination of the endosomal membrane to form multivesicular bodies that are released into the extracellular space via budding of the cellular membrane , whereas microvesicles are formed by budding of plasma membrane ( van Niel et al . , 2018; Radhakrishna et al . , 1996; Huang-Doran et al . , 2017 ) . It is increasingly apparent that EV formation occurs through highly regulated cellular processes ( Abels and Breakefield , 2016 ) , that permit their participation in intercellular communication via delivery of cargos of RNAs , microRNAs , proteins , and bioactive lipids such as prostaglandins ( van Niel et al . , 2018; Huang-Doran et al . , 2017; EL Andaloussi et al . , 2013 ) . This implicates EVs in a wide range of physiological and pathological processes . EV motility can be controlled by signaling molecules such as the guanine-nucleotide binding protein , ADP-ribosylation factor 6 ( ARF6 ) ( Abels and Breakefield , 2016; Muralidharan-Chari et al . , 2009; D'Souza-Schorey and Chavrier , 2006 ) . As a small GTPase , ARF6 exists in GTP- or GDP-bound forms ( ARF6-GTP or ARF6-GDP ) , and stimulation of ARF6 by neurotransmitters or growth factors recruits guanine nucleotide exchange factors ( GEFs ) to convert ARF6-GDP to the active ARF6-GTP ( EL Andaloussi et al . , 2013 ) . Although ARF6 itself has GTPase activity , ARF6-GTP requires GTPase-activating proteins ( GAPs ) to hydrolyze to its inactive ARF6-GDP form . ARF6-GTP influences a wide variety of cellular events including endocytosis , actin cytoskeleton reorganization and phosphoinositide metabolism in many types of cells . Importantly , ARF6-GTP is involved in EV release from plasma membranes ( Muralidharan-Chari et al . , 2009; Than et al . , 2017 ) , and exosome budding into multivesicular bodies ( Ghossoub et al . , 2014; Friand et al . , 2015; Imjeti et al . , 2017 ) . Thus , GEFs and GAPs regulate ARF6 activity to then modulate EV secretion ( D'Souza-Schorey and Chavrier , 2006 ) . Also , since ARF6 is a GTPase , it is noteworthy that another GTPase , Rac-GTPase , forms a complex with the Sig-1R ( Natsvlishvili et al . , 2015 ) , suggesting the possibility that the Sig-1R may interact with other molecules of this class . Collectively , these points of regulation position EVs and ARF6 as important participants in diverse physiological and pathological processes ( van Niel et al . , 2018 ) . Endocannabinoids ( eCB ) are lipid signaling molecules that activate CB1 or CB2 cannabinoid receptors . Of these , CB1Rs are expressed at high levels on neuronal axon terminals where they inhibit fast neurotransmitter release ( Misner and Sullivan , 1999; Hoffman and Lupica , 2000; Katona et al . , 1999 ) . The eCBs are typically synthesized in postsynaptic structures , such as dendrites , to then retrogradely activate CB1Rs on axon terminals ( Wilson and Nicoll , 2001 ) . Moreover , eCBs are not released via canonical mechanisms of calcium-dependent synaptic vesicle exocytosis , but rather through poorly understood processes . Recent evidence gathered using cell cultures suggests that the eCB N-arachidonoylethanolamine ( AEA , anandamide ) is found in EVs , suggesting a possible mechanism to release these messengers and permit retrograde eCB signaling ( Gabrielli et al . , 2015a; Gabrielli et al . , 2015b ) . Another eCB , 2-arachidonoylglycerol ( 2-AG ) , is released from neurons in an activity-dependent fashion , or via neurotransmitter stimulation of phospholipase-regulating G-protein coupled receptors ( GPCRs ) ( Kano et al . , 2009; Maejima et al . , 2005; Alger and Kim , 2011 ) . Recent evidence shows that inhibition of catecholamine uptake by cocaine leads to activation of GPCRs that stimulate 2-AG synthesis in the rodent ventral tegmental area ( VTA ) ( Wang et al . , 2015 ) . Moreover , as VTA GABAergic axons express CB1Rs , the cocaine-stimulated increase in 2-AG inhibits GABA release via these receptors ( Wang et al . , 2015; Riegel and Lupica , 2004 ) , and this can be used as a sensitive measure of eCB function . Although measurements like these are used to detect eCBs throughout the CNS , the mechanisms through which these lipids cross the extracellular space to bind to presynaptic CB1Rs remain poorly understood . Given that cocaine stimulates 2-AG synthesis , can act as a Sig-1R agonist , and that the Sig-1R interacts with Rac-GTPase , we hypothesize that it may also control other GTPases such as ARF6 , a known EV release modulator ( Muralidharan-Chari et al . , 2009; Ghossoub et al . , 2014; Natsvlishvili et al . , 2015; Tsai et al . , 2009 ) , and this might regulate 2-AG release . Through convergent experiments we demonstrate that Sig-1Rs can control EV release via interaction with ARF-6 , and that cocaine stimulates this process . Moreover , the cocaine-evoked 2-AG release required intact Sig-1Rs , ARF-6 , and cytoskeletal function , implicating EVs as a mechanism for 2-AG release in the VTA .
To investigate whether Sig-1Rs are involved in EV function , we first conducted studies in NG-108 cells to permit manipulation of signaling pathways . The integrin β1 ( Iβ1; CD29 ) protein mediates transcellular interaction of EVs with target membranes , and is a useful marker of EVs isolated through differential sequential sucrose-gradient centrifugation ( van Niel et al . , 2018; EL Andaloussi et al . , 2013; Muralidharan-Chari et al . , 2009; Imjeti et al . , 2017; Benmoussa et al . , 2019; Momen-Heravi et al . , 2013 ) . We prepared membrane fractions enriched in EVs in effluent from NG-108 cells and measured Iβ1 using western blots . Cocaine ( 10 µM ) caused a time- and concentration-dependent increase in the accumulation of Iβ1 in isolated fractions from these NG-108 cells ( Figure 1A and B ) , suggesting that cocaine increased EV release . Because our previous studies show that cocaine interacts with Sig-1Rs , we next investigated their involvement in cocaine-stimulated EV release . We found that the Sig-1R agonists PRE-084 , or fluvoxamine , both increased the Iβ1-marker of EV release from NG-108 cells in the absence of cocaine ( Figure 1B ) , and that pretreatment with either of the Sig-1R antagonists , BD1063 ( Figure 1C ) or NE100 ( Figure 1D ) , prevented the effect of cocaine . We also found that the knock-down of Sig-1Rs with siRNA alone significantly increased Iβ1 and abolished the stimulatory effect of cocaine ( Figure 1—figure supplement 1A ) , and that overexpression of Halo-tagged Sig-1Rs decreased EV release from NG-108 cells , but also blocked the effect of cocaine ( Figure 1—figure supplement 1B ) . These data support a mechanism in which Sig-1Rs tonically inhibit EV release , and this inhibition is relieved in the presence of cocaine . Having established that Sig-1Rs are involved in the stimulatory effect of cocaine on EV release in NG-108 cells , we next investigated the role of additional other signaling molecules known to also regulate EV secretion ( van Niel et al . , 2018; Muralidharan-Chari et al . , 2009; Imjeti et al . , 2017 ) . Cytohesins are a family of GEFs that activate ARFs by catalyzing a shift from GDP- to GTP-bound forms ( D'Souza-Schorey and Chavrier , 2006; Frank et al . , 1998; Hafner et al . , 2006 ) , and this can trigger EV release from LOX cells ( D'Souza-Schorey and Chavrier , 2006; Than et al . , 2017 ) . To determine whether ARF6 is similarly involved in cocaine-induced release of EVs in NG108 cells , we used the GEF inhibitor secinH3 ( SH3 , 10 µM ) ( Hafner et al . , 2006 ) and found that it prevented the cocaine-stimulated increase in Iβ1 levels in the EV fractions ( Figure 1E ) . To next determine the nature of the association between ARF6 and Sig-1R proteins in NG-108 cells , we overexpressed ARF6 mutants that mimic either the active , GTP-bound ( Q67L ) , or the inactive GDP-bound ( T27N ) forms of this protein , and performed co-immunoprecipitation experiments with a Halo-tagged Sig-1R ( Halo-Sig-1R ) ( Radhakrishna et al . , 1996; Muralidharan-Chari et al . , 2009 ) . We found that the Halo-Sig-1R co-immunoprecipitated much more strongly with the GDP-bound form of ARF6 ( ARF6-T27N ) , compared to either wild-type ARF6 , or the GTP-bound form ( ARF6-Q67L ) ( Figure 1F , Figure 1G ) . This suggests that the Sig-1R more strongly binds the inactive GDP-ARF6 , rather than the active GTP-ARF6 . As previous studies show that the Sig-1R C-terminus region contains a chaperone domain that interacts with MAM proteins ( Hayashi and Su , 2007; Su et al . , 2016; Ortega-Roldan et al . , 2013 ) , we also performed experiments with mutant Sig-1Rs to determine the regions of interaction with ARF6-GDP ( Figure 1—figure supplement 1C ) . NG-108 cells were transfected with plasmids expressing Halo-tagged N- or C-termini on the full-length Sig-1R ( Halo-Sig-1R and Sig-1R-Halo , respectively ) , or on truncated forms of the Sig-1R ( Sig-1R-1–60-Halo or Halo-Sig-1R-61–223 ) that contained chaperone ( Hayashi and Su , 2007 ) , or ligand binding motifs ( Chen et al . , 2007; Pal et al . , 2008 ) , respectively . We then examined whether the Halo-tagged receptors co-immunoprecipitated with either the active or the inactive ARF6 mutants described above . The inactive form of ARF6 ( ARF6-T27N ) co-precipitated with Sig-1R-61–223-Halo , but not with Sig-1R-1–60-Halo ( Figure 1—figure supplement 1C ) , suggesting that the C-terminus , chaperone region of the Sig-1R interacts with GDP-bound ARF6 . Interestingly , co-immunoprecipitation also revealed that ARF6-T27N interacted with the Halo-Sig-1R , but not the Sig-1R-Halo ( Figure 1—figure supplement 1C ) , suggesting that the C-terminus tag interferes with the interaction between Sig-1R and ARF6 . Taken together , our data in NG-108 cells support a model in which the chaperone region of the Sig-1R binds to the inactive form of ARF6 ( GDP-ARF6 ) to tonically inhibit EV release . Therefore , we next examined the co-localization of ARF6 and Sig-1Rs and their ability to regulate EV release in the mouse midbrain to determine the functional relevance of this interaction . Mice received single injections of cocaine ( 15 mg/kg , i . p . ) , followed by removal and processing of the midbrain for EV content ( Figure 2—figure supplement 1 ) . In agreement with previous reports ( Perez-Gonzalez et al . , 2012; Polanco et al . , 2016 ) , a membrane fraction 3 ( fr3 ) , obtained by sequential sucrose-gradient centrifugation , was isolated and found to be enriched with several markers of EVs , such as Iβ1 , alix , and flotillin-1 ( Figure 2A ) . Moreover , high concentrations of ARF6 and tyrosine hydroxylase ( TH ) were found in the EV enriched fr3 ( Figure 2A ) . However , because of the stringency of the EV isolation procedure , only a small amount of material could be obtained for analysis from these fractions . Therefore , in several experiments , we also utilized a total EV membrane fraction preparation ( tEV ) that was not subjected to a stepwise sucrose gradient , but nevertheless contained the same EV markers as fr3 ( Figure 2—figure supplement 1 ) . The mean size of the midbrain tEVs was 154 ± 1 . 41 nm ( Figure 2C ) , and midbrain tEVs contained higher levels of Iβ1 , ARF6 , and TH , compared to tEVs isolated from cortex and hippocampus ( Figure 2B ) . The topology of TH , Iβ1 , and ARF6 in midbrain tEV preparations was next examined using the broad-spectrum serine protease , proteinase-K ( PK ) ( Wang et al . , 2017; de Jong et al . , 2016 ) . In tEVs not treated with Triton X detergent , PK decreased only Iβ1 ( Figure 2D ) , which is consistent with its location on the plasma membrane ( van Niel et al . , 2018; EL Andaloussi et al . , 2013; Muralidharan-Chari et al . , 2009; Imjeti et al . , 2017 ) . In contrast , all three proteins were degraded by PK in tEV preparations treated with Triton X ( Figure 2D ) , suggesting that , unlike Iβ1 , TH and ARF6 are located within EVs , rather than on their membranes . Because they were found in EV-rich preparations of midbrain , TH , Iβ1 , and ARF6 were used as markers to evaluate the effect of cocaine on tEVs . Like NG-108 cells , cocaine ( 15 mg/kg ) increased Iβ1 ( and TH ) levels in midbrain tissue within 30 min of an intraperitoneal ( i . p . ) injection ( Figure 2E ) , and this returned to control levels 60 min following cocaine treatment ( Figure 2E ) . However , ARF6 levels were not significantly altered by cocaine ( Figure 2E ) . As in NG-108 cells , the cocaine-stimulation of tEV markers in midbrain was also prevented by the Sig-1R antagonist , BD1063 ( Figure 2F ) . Moreover , cocaine failed to increase any of the tEV markers ( Figure 2G ) in midbrain preparations from mice lacking the Sig-1R gene ( Sigmar1 ) , suggesting that Sig-1Rs are essential for cocaine-induced tEV release in mouse midbrain . To determine cellular locations of the Sig-1R we used immunofluorescence confocal microscopy in the mouse ventral midbrain . We found that Sig-1R ( Mavlyutov et al . , 2016 ) and TH fluorescence signals were colocalized ( Figure 3A ) , and as TH is a marker for DA neurons in the ventral midbrain , the data suggest that Sig-1Rs are found in DA neurons . However , the Sig-1R signal was also found associated with the vesicular GABA transporter ( vGAT ) , a marker of GABA neurons in the mouse ventral midbrain ( Figure 3A ) . Therefore , the Sig-1R is likely expressed in both DA and GABA neurons in the midbrain . Immunofluorescence confocal microscopy also revealed co-localization of Sig-1R and ARF6 in TH-positive neurons in the mouse ventral midbrain ( Figure 3C ) , and these proteins co-immunoprecipitated in midbrain samples from wild-type , but not Sig-1R knockout mice ( Figure 3D ) . Also , the Sig-1R immunohistochemical signal was absent in Sig-1R knockout mice ( Figure 3—figure supplement 1 ) . The subcellular distribution of ARF6 in the mouse midbrain was next compared with Sig-1Rs in a fractionation assay allowing detection of the MAM ( Figure 3B ) , where Sig-1Rs are abundant ( Hayashi and Su , 2007; Lewis et al . , 2016 ) . Both the Sig-1R and ARF6 were found in this MAM fraction ( Figure 3B ) , but another ARF GTPase , ARF-1 , was not detected ( Figure 3B ) . Together , our results indicate that Sig-1Rs and ARF6 colocalize with GABA and DA neuron markers and are associated with the MAM in the mouse midbrain . To determine whether , like in NG-108 cells , cocaine-stimulation of EV secretion occurred through Sig-1R- and ARF6-dependent mechanisms , we manipulated signaling by these proteins , followed by preparation of midbrain tEV fractions . We found that an injection of cocaine ( 15 mg/kg , i . p . ) significantly attenuated the co-immunoprecipitation of ARF6 and Sig1R in the mouse midbrain ( Figure 4A ) , and this was prevented by a preceding subcutaneous ( s . c . ) injection of the Sig-1R antagonist , BD1063 ( 10 mg/kg ) ( Figure 4B ) . This suggests that the cocaine facilitates activation of the Sig-1R , and this triggers Sig-1R dissociation from ARF6 . Next , we determined whether in vivo cocaine treatment altered the intracellular localization of ARF6 , using the MAM fractionation assay . We found that , unlike that observed in the P3 fraction where ARF6 levels remained unchanged , 10 min after cocaine injection the level of MAM-associated ARF6 was decreased ( Figure 4C ) . Moreover , Sig-1R levels were not significantly altered in either the P3 or the MAM fractions ( Figure 4C ) . These results suggest that the Sig-1R is activated by cocaine while associated with the MAM and this facilitates dissociation of the Sig-1R from ARF6 . As ARF6-GTP modulation by the GEF inhibitor SH3 altered EV secretion in NG-108 cells ( Figure 1E ) , we measured its effect ( s . c . , 10 mg/kg ) on cocaine-stimulated tEV secretion in mouse midbrain . Consistent with NG-108 cell data , SH3 significantly inhibited the cocaine-induced increase of TH and Iβ1 in mouse midbrain ( Figure 4D ) . Existing data also support the involvement of cytoskeletal myosin and actin in EV release and show that ARF6 exerts its effects on EV release through phosphorylation of myosin light-chain kinase ( MLCK ) ( van Niel et al . , 2018; Muralidharan-Chari et al . , 2009 ) . Therefore , we examined MLCK involvement in the cocaine-simulated EV release in midbrain tissue and found that the MLCK inhibitor ML7 ( 2 µM ) prevented the increase in EV release , as measured by Iβ1 , or TH in EV-rich fractions ( Figure 4D ) . In consideration of these data , we propose the following model; 1 ) the Sig-1R forms a stable complex with the inactive ARF6-GDP at the MAM , 2 ) cocaine , through interaction with the Sig-1R , causes dissociation of the ARF6-GDP/Sig-1R complex , 3 ) free ARF6-GDP is then converted to the active ARF6-GTP by GEFs , and 4 ) ARF6-GTP translocates to the plasma membrane where it stimulates EV release into the extracellular space ( Figure 4E ) by activating MLCK , and permitting EV mobility . Using this model of EV secretion , we next sought to determine its functional relevance to synaptic modulation by eCBs in the mouse midbrain . A recent study found that microvesicle-enriched fractions from primary microglia cultures contained the eCB anandamide ( Gabrielli et al . , 2015a ) , and work from our laboratory showed that cocaine promotes the release of eCB 2-AG in the midbrain ( Wang et al . , 2015 ) . However , the potential involvement of EVs in 2-AG function has not been assessed . To determine whether 2-AG is found in EV fractions from mouse midbrain , we used Fourier transform mass spectrometry ( FTMS ) . We found that the levels of 2-AG were higher in midbrain homogenates than in cerebral cortex , and were approximately fivefold larger than those observed in tEV fractions from these brain regions ( Figure 5A ) . The concentration of 2-AG in midbrain tEV fractions ( 206 . 9 ± 70 . 2 pmol/mg , Figure 5A ) was also higher than that measured in the cerebral cortex ( 121 . 4 ± 16 . 1 pmol/mg , Figure 5A ) , suggesting regional differences in concentrations of 2-AG . We also found that cocaine significantly increased 2-AG levels in midbrain tissue ( Figure 5B ) . However , when cocaine-stimulation of 2-AG levels in tEV fraction were measured using FTMS in pooled samples of mouse midbrain , we observed considerable variability in baseline saline-injected controls ( n = 15 mice in three experiments; Figure 5C ) , and in cocaine-stimulated levels of the eCB ( n = 15 mice in three experiments ) . Thus , although a clear trend toward increased 2-AG in these tEV fractions was observed , and cocaine significantly increase midbrain tissue levels of 2-AG ( Figure 5B ) , the effect of cocaine on 2-AG content in the tEV fractions was not significant ( t8 = 1 . 61 , p=0 . 147 , unpaired Student’s t-test; Figure 5C ) . Recent studies show that fatty acid binding proteins can act as intracellular carriers for 2-AG ( Kaczocha et al . , 2009 ) , and one of these , fatty acid binding protein 5 ( FABP5 ) , was involved in mediating extracellular 2-AG release in the mouse brain ( Haj-Dahmane et al . , 2018 ) . To determine whether this carrier of 2-AG could also be localized to midbrain EVs , we isolated EV fractions from mouse midbrain and used western blots to measure FABP5 and other EV markers . These EV fractions contained FABP5 as well as the EV markers TH , Iβ1 , and flotillin-1 ( Figure 5D ) . This suggests that the FABP5 protein is associated with EVs to perhaps mediate 2-AG signaling in the CNS . There is strong evidence that 2-AG is synthesized in rodent midbrain VTA neurons , where it can modulate synaptic neurotransmitter release ( Riegel and Lupica , 2004; Melis et al . , 2004; Parsons and Hurd , 2015; Labouèbe et al . , 2013 ) . Moreover , 2-AG function is increased during heightened DA neuron activity ( Riegel and Lupica , 2004; Melis et al . , 2004 ) , or when phospholipases are activated by certain Gαq11-containing GPCRs , such as the α1-noradrenergic ( α1R ) , or type-I metabotropic glutamate receptors ( mGluRIs ) ( Wang et al . , 2015; Haj-Dahmane and Shen , 2014 ) . These data also show that cocaine’s ability to increase VTA 2-AG function occurs via its inhibition of the norepinephrine transporter ( NET ) , causing activation of α1Rs on VTA DA neurons and 2-AG synthesis from membrane phospholipids ( Wang et al . , 2015 ) . Based on this previous work , and our data showing cocaine interactions with midbrain Sig-1Rs , ARF6 and EV release , we evaluated the possibility that 2-AG function in the VTA occurs via EV- and Sig-1R-dependent mechanisms in mouse midbrain DA neurons . Local 2-AG function can be measured with high temporal fidelity through its activation of CB1Rs leading to local inhibition of synaptic transmission ( Alger , 2002 ) . This functionally relevant endogenous 2-AG reduces inhibitory postsynaptic currents ( IPSCs ) mediated by synaptic GABA release onto GABAB receptors ( GABABRs ) located on DA neuron dendrites ( Wang et al . , 2015; Riegel and Lupica , 2004 ) . Similar to previous data from rat VTA DA neurons ( Wang et al . , 2015 ) , we found that cocaine ( 10 µM ) inhibited IPSCs recorded in mouse DA neurons ( Figure 5E and F ) . The IPSC inhibition by cocaine was prevented by the CB1R antagonist , AM251 ( 1 µM; Figure 5H-Figure 5—figure supplement 1 ) and was absent in mice lacking the CB1R ( Zimmer et al . , 1999 ) ( Figure 5E and F ) . The inhibition of IPSCs by cocaine was also reduced by tetrahydrolipostatin ( THL , 2 µM ) , an inhibitor of the enzyme diacylglycerol lipase-α ( DGLα ) , preventing conversion of diacylglycerol ( DAG ) to 2-AG ( Figure 5—figure supplement 1A1 , Figure 5—figure supplement 1B ) . Cocaine-mediated 2-AG release was also absent in mutant mice lacking expression of DGLα in DA neurons ( Shonesy et al . , 2014 ) ( Daglaflox/flox x DATCre mice; Figure 5G and H ) . These experiments confirm that inhibition of GABA release onto DA neurons by cocaine occurs via stimulation of 2-AG function in the mouse VTA . We next examined Sig1-R involvement in cocaine-dependent 2-AG release in mouse VTA DA neurons . Each of two Sig-1R antagonists ( BD1063 or NE100; 2 µM ) significantly reduced the cocaine ( 10 µM ) simulation of 2-AG release in VTA DA neurons ( Figure 6A–C and E ) . This effect of cocaine was also significantly reduced in DA neurons from Sig-1R knockout mice , particularly 5–10 min after beginning cocaine application ( Figure 6D and E ) . Importantly , the inhibition of IPSCs by the synthetic CB1R agonist , WIN55 , 212–2 ( 1 µM ) , was not reduced by Sig-1R antagonism , or by genetic deletion of this receptor ( Figure 6—figure supplement 1 ) . This indicates that Sig-1Rs are linked to cocaine-stimulated 2-AG function in the CNS , and that CB1R signaling is not diminished by altered Sig-1R function or expression . To examine whether Sig-1Rs are involved in facilitating 2-AG release derived from direct GPCR activation , we determined whether α1R and mGluRI co-activation could stimulate 2-AG function in mouse VTA , and whether this is altered in Sig-1R knockout mice . Consistent with our previous report ( Wang et al . , 2015 ) , co-application of the α1R agonist phenylephrine ( PE , 100 µM ) and the mGluRI agonist , DHPG ( 1 µM ) inhibited GABAB IPSCs in wildtype mouse VTA DA neurons , and this was blocked by AM251 ( Figure 7B and C ) . However , it is also important to note that the properties of the IPSC inhibition produced by DHPG+PE differed from that seen with cocaine . Thus , the response to DHPG+PE was much slower to reach maximum and lacked the early fast component observed with cocaine ( Figure 7—figure supplement 1 ) in wildtype mice . Therefore , in comparison , the effect of DHPG+PE primarily consisted of the delayed slow component ( Figure 7—figure supplement 1C ) . Also , in DA neurons from Sig-1R knockout mice , the slow response to DHPG+PE was significantly smaller ( Figure 7A–7C , Figure 7—figure supplement 1A ) , which contrasts with that seen with cocaine where the early fast inhibition was absent , but the later inhibition was less affected in Sig-1R knockout mice ( Figure 6D , Figure 7—figure supplement 1B ) . These differences could indicate reliance upon distinct signaling pathways that convergence upon Sig-1Rs to permit 2-AG release via EVs . To determine whether the effects of 2-AG derived from a non-GPCR source are also altered in the Sig-1R knockout mouse , we measured tonic 2-AG release that is observed without GPCR activation ( either indirectly by cocaine or directly by DHPG+PE ) ( Wang et al . , 2015 ) . The tonic inhibition of GABAB IPSCs mediated by this basal level of endogenous 2-AG is revealed when CB1Rs are blocked by antagonists , resulting in an increase in these synaptic currents ( Wang et al . , 2015; Riegel and Lupica , 2004 ) . We found that DA neurons from both wildtype and Sig-1R knockout mice exhibited similar significant IPSC increases when the CB1R antagonist AM251 was applied ( Figure 7A and B ) . Therefore , the data suggest that only 2-AG derived from GPCR stimulation is dependent upon intact Sig-1R function , and additionally that 2-AG synthesis itself is not disrupted in Sig-1R knockout mice . Our NG-108 experiments indicated that Sig-1Rs stabilize the inactive GDP-bound form of ARF6 , and that cocaine activates GTP-bound ARF6 through an interaction with Sig-1Rs , thereby permitting EV release . Moreover , our FTMS experiments identified 2-AG in midbrain tEV fractions ( Figure 5A , Figure 5C ) . Therefore , involvement of ARF6 in the 2-AG-dependent inhibition of GABA release by cocaine was tested in wild-type mouse VTA DA neurons . Manipulation of ARF6 activation with the GEF inhibitor , SH3 ( Figure 8A and E ) , or , direct inhibition of ARF6 with NAV2729 ( both at 10 µM ) ( Yoo et al . , 2016 ) , significantly inhibited cocaine-induced 2-AG function in midbrain DA neurons ( Figure 8B and E ) . Also , like that observed with Sig-1R antagonists or knockouts ( Figure 6 ) , the reduction in the cocaine inhibition of IPSCs by both SH3 and NAV2729 was more prominent within the first 10 min of cocaine application ( Figure 8A , Figure 8B ) . As inhibition of MLCK significantly reduced EV release in midbrain tissue experiments , we examined its involvement in the synaptic effects of cocaine-simulated 2-AG function in DA neurons . We found that the MLCK inhibition by ML7 ( 2 µM ) also significantly reduced the effect of cocaine on 2-AG release in this electrophysiological assay of eCB function ( Figure 8C and E ) . Together these data demonstrate that EV release is controlled by the Sig-1R , ARF6 , and MLCK , and that cocaine’s interaction with the Sig-1R can recruit this signaling cascade . The data further demonstrate that disruption of these signaling mechanisms leads to reduced synaptic 2-AG function in the midbrain , thereby implicating these proteins and EVs in the release of eCBs .
Previous studies show that a cocaine binds to Sig-1Rs ( Sharkey et al . , 1988; Chen et al . , 2007; Hiranita et al . , 2011 ) , and that blockade of this interaction reduces effects of the psychostimulant ( Romieu et al . , 2002; Hiranita et al . , 2011; Lever et al . , 2014; Fritz et al . , 2011 ) . Additionally , cocaine’s actions at Sig1-Rs alters its ability to influence voltage-gated potassium channel function , and this can reduce its behavioral effects ( Kourrich et al . , 2013; Romieu et al . , 2002; Lever et al . , 2014; Fritz et al . , 2011 ) . The present data demonstrate that the Sig-1R also regulates EV secretion in cultured cells and in the mouse midbrain , and that cocaine modulates this process through interaction with the Sig-1R . We also show that the interactions among Sig1-Rs , cocaine , and EVs can regulate synaptic transmission in the brain via the control of 2-AG release and its inhibition of GABAergic input to DA neurons in the mouse VTA . Therefore , our study identifies novel mechanisms for Sig-1R control of EV function and implicates EVs in eCB release in the CNS . EVs are increasingly recognized as a highly regulated mechanism to permit exchange of signaling molecules , such as lipids , nucleic acids , organelles , and proteins , among cells ( van Niel et al . , 2018 ) . As such , regulatory control points for EV formation , budding , translocation , and cargo release have been delineated in many cell types during normal cellular function , and in disease states ( van Niel et al . , 2018; Huang-Doran et al . , 2017; EL Andaloussi et al . , 2013; Muralidharan-Chari et al . , 2009; Wang et al . , 2017; Yoo et al . , 2016 ) . Here , we show that cocaine treatment of NG108 cells , or of mouse midbrain after in vivo injection , stimulates EV release , and that this is mimicked by agonists of Sig-1Rs , and prevented by antagonists or genetic elimination of these receptors . Moreover , using co-immunoprecipitation assays , we provide evidence for an association between ARF6 , an established regulator of EV secretion ( D'Souza-Schorey and Chavrier , 2006; Yoo et al . , 2016 ) , and the Sig-1R in TH-positive VTA neurons , and find that blockade of ARF6 activation prevents cocaine-induced EV release in both NG-108 cells and midbrain . We also report that in vivo cocaine causes the ARF6/Sig1R complex to dissociate , and this is prevented by Sig-1R antagonism . These data suggest that Sig-1Rs bind ARF6 proteins to hold them in an inactive GDP-bound form , and that cocaine facilitates the dissociation of these proteins to permit conversion of ARF-GDP to the active ARF6-GTP . Our data also suggest that this interaction between ARF6 and Sig-1Rs occurs at the MAM , and that cocaine enables translocation of ARF6-GTP to the plasma membrane . This mechanism is notable because ARF6 is implicated in EV secretion via regulation of cytoskeletal actin function in a wide range of mammalian tissues ( D'Souza-Schorey and Chavrier , 2006; Yoo et al . , 2016 ) , and this is supported by our observation that inhibition of MLCK also prevents the cocaine-induced increase in EV levels in mouse midbrain . Previous work shows that anandamide is found in EV-containing membrane fractions of rodent microglia cultures , and that these fractions exhibit cannabinoid agonist properties when applied to hippocampal brain slices ( Gabrielli et al . , 2015a ) . Here , we show using FTMS that 2-AG is found in acute mouse midbrain preparations that are enriched in tEVs , and that 2-AG levels are significantly increased in midbrain homogenates after in vivo exposure to cocaine . In contrast , although 2-AG could be measured in tEV fractions using FTMS in mouse midbrain , and tEV markers were significantly increased after in vivo cocaine treatment , the increase in 2-AG levels produced by cocaine in the tEV preparation did not reach statistical significance despite a clear trend . As these preparations are technically demanding and yield small amounts of material , it is possible that the between-groups ex vivo design and variability among samples in both saline control and cocaine injected mice contributed to this outcome . Alternatively , it is possible that cocaine causes an increase in 2-AG-containing EV release , but that the amount of 2-AG per vesicle does not change , and this increase in vesicle release could be sufficient to locally activate CB1Rs on GABAergic axon terminals . The observation that cocaine increased midbrain levels of 2-AG provides biochemical support for our finding of cocaine-increased 2-AG function in mouse ( this study ) and rat VTA DA neurons in vitro ( Wang et al . , 2015 ) . In this regard , we demonstrate that cocaine stimulates a 2-AG-dependent inhibition of GABAB receptor-mediated synaptic responses that is absent in mice lacking the CB1R , or the 2-AG biosynthetic enzyme , DGLα , in DA neurons . Based upon present data and our published work ( Wang et al . , 2015 ) , we propose that 2-AG synthesis is stimulated when cocaine blocks norepinephrine uptake in the VTA , resulting in activation of G-protein-αq-coupled α1Rs , which , together with Gq-coupled mGluRIs stimulated by endogenous glutamate , activate phospholipases and liberate 2-AG from precursor membrane lipids ( Figure 8—figure supplement 1 ) ( Kano et al . , 2009; Maejima et al . , 2005; Alger and Kim , 2011; Wang et al . , 2015; Haj-Dahmane and Shen , 2014; Mátyás et al . , 2008 ) . Although this model of 2-AG synthesis is supported by our studies , the mechanism of 2-AG is release is unknown . Here , using this 2-AG-sensitive synaptic response , we find that the same manipulations that blocked EV release in NG-108 cells and midbrain EV assays also reduced or eliminated cocaine-stimulated 2-AG effects on synaptic transmission in the mouse VTA . These manipulations include the disruption of Sig-1R signaling , the inhibition of ARF6 function , and the inhibition of MLCK . Moreover , we also found that the IPSC inhibition produced by a synthetic CB1R agonist was not altered by antagonism or genetic deletion of Sig-1Rs , suggesting that Sig-1Rs regulate 2-AG signaling but not CB1R function . The involvement of Sig-1Rs in the GPCR-dependent 2-AG release was supported by experiments showing that co-activation of mGluRIs and α1Rs by DHPG+PE could increase the release of this eCB , and that this was significantly reduced in Sig-1R KO mice . Moreover , another form of tonic 2-AG release that occurs under basal conditions in the absence of GPCR stimulation was unaltered in Sig-1R KO mice . Therefore , the data suggest that Sig-1Rs and EVs mediate only GPCR-dependent 2-AG release , and not that generated by other cellular pathways . Based on our biochemical and electrophysiological data , we propose a model ( Figure 8—figure supplement 1 ) in which cocaine initiates 2-AG synthesis via inhibition of the NET , leading to activation of α1Rs coupled to Gq proteins controlling phospholipases and the liberation of the 2-AG precursor DAG . DAG is then converted to 2-AG via DGLα and then packaged in EVs through an unknown process . 2-AG release from EVs is triggered when cocaine binds to Sig-1Rs to liberate ARF6-GDP and permit its conversion to the active ARF6-GTP , which can then act at MLCK to initiate EV fusion with the cellular membrane and release of 2-AG . Although these mechanisms are supported by the present data , our finding that the inhibition of IPSCs by 2-AG release by DHPG+PE is absent cells from Sig-1R KO mice suggests that cocaine binding to the Sig-1R is not necessary to initiate EV release . However , fundamental differences in the characteristics of the inhibition produced by these methods were noted . Thus , the kinetics of the 2-AG-mediated inhibition of GABA release caused by cocaine differ from DHPG+PE in that the effect onset and the peak response to cocaine occurred more rapidly than that seen with DHPG+PE ( Figure 6—figure supplement 1 ) . Also , the cocaine effect reached a maximum within approximately the first 5 min after application , and this early phase was completely blocked when Sig-1R , ARF6 or MLCK function was disrupted ( Figure 8 ) , whereas the smaller late phase of inhibition was resistant to these manipulations ( Figure 8 , Figure 7—figure supplement 1 ) . Despite this , data showing that both the early and late phases of cocaine inhibition are prevented by AM251 ( Figure 5—figure supplement 1 ) and absent in mice lacking the CB1R or DGLα ( Figure 5E–H ) , indicate that both inhibitory phases depend upon 2-AG and CB1Rs . In contrast to the effect of cocaine , DHPG+PE does not produce a robust early phase of IPSC inhibition ( Figure 6—figure supplement 1 ) and the delayed inhibition produced by the agonists is smaller , but not absent in Sig-1R KO mice ( Figure 7—figure supplement 1A ) . These differences suggest that although cocaine and DHPG+PE initiate 2-AG-dependent inhibition of synaptic GABA release , they may involve distinct upstream mechanisms that converge on Sig-1Rs and their control of EV release . Thus , the faster time-course of the cocaine effect may result from its direct binding to Sig-1Rs ( Sharkey et al . , 1988; Chen et al . , 2007; Hiranita et al . , 2011 ) to more rapidly stimulate EV release , resulting in their depletion during the late phase . In contrast , the slower and more sustained effect of DHPG+PE on 2-AG release may reflect coupling of EV release to a signaling pathway that relies upon intracellular release of an endogenous Sig-1R agonist . In support of this , several putative endogenous Sig-1R agonists have been identified ( Monnet and Maurice , 2006; Ramachandran et al . , 2009; Fontanilla et al . , 2009 ) , and a more recent study shows that agonists of Gq-coupled receptors that stimulate phospholipases can increase intracellular levels of choline , which then acts as an agonist at Sig-1Rs to enhance their calcium signaling properties ( Brailoiu et al . , 2019 ) . Therefore , we speculate that the distinct phases of 2-AG-dependent inhibition are related the ability of the cocaine to act as a direct agonist at Sig-1Rs , compared to potential indirect effects of DHPG+PE that may be mediated by an intracellular signaling molecule having agonist properties at sig-1Rs . Future experiments will test this hypothesis . Fatty acid binding proteins ( FABPs ) can bind and transport lipid molecules within and between cells ( Kaczocha et al . , 2009; Ertunc et al . , 2015 ) . One of these , adipocyte fatty-acid binding protein 4 ( aP2 ) , is secreted from adipocytes via EVs ( Ertunc et al . , 2015 ) , and several FABPs are found in brain ( Owada et al . , 1996 ) . Recent studies show that one of these proteins , FABP5 , has high affinity for 2-AG , and its inhibition or genetic deletion impairs 2-AG-mediated signaling and plasticity at glutamate synapses in the dorsal raphe nucleus ( Haj-Dahmane et al . , 2018; Owada et al . , 1996; Kaczocha et al . , 2012 ) . Based on these results , and our present observation that FABP5 is co-localized with the EV markers Iβ1 and flotillin-1 in EV fractions from the mouse midbrain , it is possible that 2-AG release may occur via binding to FABPs that are transported to the extracellular space via EVs , and therefore subject to mechanisms regulating EV secretion , such as Sig-1Rs , ARF6 , and MLCK . Future studies will more closely examine this possibility to more completely understand the mechanisms of EV-dependent eCB release in the brain .
1-[2- ( 3 , 4-Dichlorophenyl ) ethyl]−4-methylpiperazine dihydrochloride ( BD 1063 dihydrochloride , Cat#: 0883 , Tocris ) , and cocaine hydrochloride were dissolved in 0 . 9% NaCl . N-[4-[5- ( 1 , 3-Benzodioxol-5-yl ) −3-methoxy-1H-1 , 2 , 4-triazol-1-yl]phenyl]−2- ( phenylthio ) acetamide ( SecinH3 , Cat#: 2849 , Tocris ) was dissolved in DMSO , and then diluted with 25% DMSO/75% glucose solution ( 5 w/v% ) . Group membership was determined by genotype where transgenic mice were used . In in vitro electrophysiology studies , recordings from untreated control brain slices were interleaved with recordings from drug pre-incubated brain slices from the same animal . In cell biology experiments , mice were chosen for experiments depending upon date of arrival from the supplier . In this way , mice were assigned to groups according availability and to the experimental procedures to be performed that day . In most cases , brain tissue from each mouse was used in both control and treatment conditions . NG-108 cell culture dishes were selected randomly from those available in the tissue incubator . Mice were killed with CO2 gas , and brains were removed , and rinsed in ice-cold Hank’s balanced salt solution ( Thermo Fisher Scientific ) . Midbrain samples were isolated by cutting coronal sections containing the VTA using mouse brain matrices ( Roboz ) , and the cortex and a hippocampus dissected free ( Figure 2—figure supplement 1 ) . For vesicle fractions from brain tissue we used an established protocol with minor modifications ( Perez-Gonzalez et al . , 2012; Polanco et al . , 2016 ) . Briefly , following dissection , midbrain slices from two wildtype male C57BL/6J mice were chopped and then incubated in 1 . 5 ml of 0 . 125% collagenase ( Sigma-Aldrich ) in Neurobasal medium ( Thermo Fisher Scientific ) for 30 min at 37°C ( see Figure 2—figure supplement 1 for a graphic summary of Ev isolation procedures ) . To stop the digestion , 4 . 5 ml of ice-cold phosphate-buffered saline ( PBS ) was added and the temperature maintained at 4°C throughout subsequent steps . The tissue was then gently disrupted by multiple passes through a 200 µL pipette tip , followed by a series of differential centrifugations at 300 x g for 10 min , 2000 x g for 10 min , and 7500 x g for 30 min . The pellets resulting from these spins , containing cells , membranes , and cellular debris , respectively , were then discarded . For EV purification , the 7500 x g supernatant was syringe filtered at 1 . 0 μm ( Whatman Puradisc Syringe Filters , GE Healthcare Life Sciences , Cat . #6780–2510 ) and centrifuged at 100 , 000 x g for 70 min to obtain a pellet containing EVs . The 100 , 000 x g pellet was washed with PBS and spun again at 100 , 000 x g for 60 min to obtain a total EV ( tEV ) pellet . For EV purification , the tEV sample was resuspended in 0 . 5 mL of 0 . 95 M sucrose in 20 mM HEPES ( pH 7 . 4 ) before addition to a sucrose-step gradient column . The column consisted of 6 × 0 . 5 mL fraction running from the bottom 2 . 0 M , 1 . 65 M , 1 . 3 M , 0 . 95M , 0 . 6 M , to 0 . 25 M at the top . Similarly , sucrose step gradients were centrifuged for 16 hr at 200 , 000 x g , after which the six fractions were collected . EVs settled typically at 0 . 95 M sucrose . The original six 0 . 5 mL fractions were collected and resuspended in 6 mL of ice-cold PBS , followed by a 100 , 000 x g centrifugation for 70 min at 4°C . Finally , the pellets were resuspended in 30 μL of filtrated-PBS when EVs were used for cell assays or 15 µl of lysis buffer ( 50 mM Tris pH7 . 4 , 150 mM NaCl , 1% Triton-X and protease inhibitor ( Sigma-Aldrich ) when EVs were intended for western blots . For western blotting , EV lysates in lysis buffer were quantified for protein content with a Micro BCA Protein Assay Kit ( Thermo Fisher Scientific ) . We also prepared brain lysate sample ( BL ) in lysate buffer using the midbrain tissues from the 300 x g pellets obtained in the courses of the EV isolations , which were used as positive controls for the western blots and to normalize tEVs sample amount between each treatment . Drugs were injected i . p . at a volume of 5 ml/kg . Regimen 1 ( for Figure 2E ) : Thirty and 60 min after i . p . injections with cocaine ( 15 mg/kg ) , midbrain slices were collected . Regimen 2 ( for Figures 2F and 4E ) : Injections with BD1063 ( 10 mg/kg , s . c . ) , SecinH3 ( 10 µmol/kg , s . c . ) , ML7 ( 5 mg/kg ) , or vehicle ( inj 1 ) were performed 20 min prior to injections with saline or cocaine ( 15 mg/kg , i . p . ; inj 2 ) . Thirty min after inj 2 , midbrain slices were collected . Regimen 3 ( for Figure 4A ) : 10 , 20 and 30 min after i . p . injections with cocaine ( 15 mg/kg ) , midbrain slices were collected . Regimen 4 ( for Figure 4B ) : Injections with BD1063 ( 10 mg/kg , s . c . ) , SecinH3 ( 10 µmol/kg , s . c . ) , or vehicle ( s . c . ) ( inj 1 ) were performed 20 min prior to injections with saline or cocaine ( 15 mg/kg , i . p . ; inj 2 ) . Ten min after inj 2 , midbrain slices were collected . Regimen 5 ( for Figure 4C ) : 20 min after i . p . injections with cocaine ( 15 mg/kg ) or vehicle , midbrain slices were collected . Regimen 6 ( for Figure 5B ) : 30 min after i . p . injections with cocaine ( 15 mg/kg ) or vehicle , midbrain slices were collected . For western blotting of extracellular vesicles from NG108 cells , the cells on 10 cm dishes were washed with prewarmed Hanks' Balanced Salt Solution ( HBSS ) twice and incubated in HBSS at 37°C in the presence of cocaine . In brief , western blotting was performed with protein samples separated using a 12% sodium dodecyl sulfate-polyacrylamide gel electrophoresis ( SDS-PAGE ) , and then transferred onto a Immobilon FL Transfer polyvinylidene difluoride ( PVDF ) membrane ( Mollipore ) in the Tris/Glycine buffer ( Bio-Rad Laboratories ) without methanol . After incubation with 5% blotting-grade blocker ( Bio-Rad Laboratories ) or 5% bovine serum albumin ( BSA , Sigma-Aldrich ) in TBST buffer ( 10 mM Tris . pH 8 . 0 , 150 mM NaCl , and 0 . 5% Tween 20 ) for 1 hr , membranes were incubated with the primary antibodies at 4°C overnight . Membranes were washed for 10 min four times by using TBST buffer and incubated with a 1:10 , 000 dilution of secondly antibodies ( LI-COR Biosciences ) at room temperature for 1 hr . Blots were washed for 10 min four times by using TBST buffer and the signal intensity was determined using Odyssey Imaging System ( LI-COR Biosciences ) . Resultants were analyzed using an Image Studio Lite ( LI-COR Biosciences ) . Total EV ( tEV ) samples were isolated in filtered ( at 1 µm ) -PBS from WT and Sig1R KO mouse midbrain , 30 min after treatment with either saline or cocaine ( 15 mg/kg , i . p . ) , and sent to Systems Biosciences ( Palo Alto , CA ) for metric analysis of tEVs . MAM was isolated from mouse mid brain as previously reported ( Hayashi and Su , 2007; Kourrich et al . , 2013 ) . Briefly , following homogenization of the brain tissue , nuclear , crude mitochondrial , and microsomal fractions were prepared by differential centrifugation . Supernatants were collected as the cytosolic fraction . The crude mitochondrial fraction in the isolation buffer ( 250 mM mannitol , 5 mM HEPES , 0 . 5 mM EGTA , pH 7 . 4 ) was subjected to a Percoll gradient centrifugation for separation of the MAM from mitochondria . Immunofluorescence staining was performed as described previously . In brief , after blocking , the sections were incubated with the first antibodies in 5% BSA/0 . 1% Triton X-100 PBS overnight at 4°C . Bound antibodies were detected with Alexa Fluor 405-conjugated anti-mouse IgG ( 1:200 , Thermo Fisher Scientific ) , Alexa Fluor 488-conjugated anti-chicken IgG ( 1:200 , Thermo Fisher Scientific ) , and Alexa Fluor 568-conjugated anti-Rabbit IgG antibodies ( 1:200 , Thermo Fisher Scientific ) in 5% BSA PBS . An UltraView confocal microscopic system ( PerkinElmer ) was used for imaging . For the immunostaining of Sig-1R , rabbit anti-serum against Sig-1R , a gift from Dr . Arnold Ruoho ( University of Wisconsin , USA; Ramachandran et al . , 2007 ) , was used . When compared to several commercially available products , the affinity-purified antibody from this antiserum , is very specific for the sigma-1 receptor in the mouse dorsal root ganglia ( Mavlyutov et al . , 2016 ) . We established the following procedures to allow for the best specific detection of the Sig-1R in mouse brain slices , using the antiserum from Dr . Ruoho . Deeply anesthetized animals were transcardially perfused with filtered 0 . 1 M Phosphate buffer ( PB; pH 7 . 4 ) followed by 4% paraformaldehyde ( w/v ) in 0 . 1 M PB . After perfusion , whole brains were isolated and post-fixed in the same fixatives overnight at 4°C with rotation . Subsequently , they were dehydrated with 20% sucrose in 0 . 1 M PB ( w/v ) and then 30% sucrose in 0 . 1 M PB ( w/v ) at 4°C with rotation . The brain samples were then embedded in O . C . T . compound ( Sakura Finetek , Torrance , CA ) on dry ice and stored in −80°C . Thirty-µm sections were cut on a cryostat and mounted on Tissue Path Superfrost Plus Gold Microscope Slides ( Fisher Scientific , Hamilton , NH ) dried overnight . Sections were blocked with 5% bovine serum albumin ( BSA , w/v ) in PBS containing 0 . 1% Triton-X100 ( v/v ) for 1 hr at room temperature . The sections were then incubated with the sigma-1 receptor anti-sera diluted at 1:1000 in the blocking solution overnight at 4°C . Following 10 min PBS washing for three times , sections were incubated with Alexa Fluor ( 488 for green/568 or 594 or 546 for Red ) -conjugated goat anti-rabbit IgG ( 1:500 , Invitrogen , Carlsbad , CA ) in 5% BSA in PBS for 90 min at room temperature . The sections were washed with PBS for 5 min three times , then counterstained with 4’ , 6’-diamino-2-phenylindole ( DAPI , Invitrogen , 1 µg/mL in MilliQ; Millipore , Billerica , MA ) by 10 min incubation at room temperature . Sections were washed with PBS for 5 min three times , mounted on coverslips with Prolong Diamond Antifade Mountant ( Life technologies , Carlsbad , CA ) for imaging . The specificity of this antiserum in labeling the Sig-1R is demonstrated in brain slices from wildtype mice , where strong staining is shown , and in and Sig-1R knockout mice , where staining is absent ( Figure 3—figure supplement 1 ) . NG108 cells were cultured at 37°C and 5% CO2 in High glucose Dulbecco's Modified Eagle Medium ( DMEM , Thermo Fisher Scientific ) containing L-glutamine , 10% Fetalgro Bovine Growth Serum ( RMBIO ) , HAT supplement ( Thermo Fisher Scientific ) , 100 mg/ml Penicillin-Streptomycin ( Thermo Fisher Scientific ) . Transfection of cells with expression vectors was done by using PolyJet DNA In Vitro Transfection Reagent ( Signagen Laboratories , Rockville , MD ) according to manufacturer’s instructions . Sources of vectors are provided above . Twelve-week-old WT C57BL6 , Cnr1-/- ( CB1R knockout ) , or Sigmar1+/- Sig-1R KO mice were decapitated , and their brains rapidly removed and transferred to an oxygenated ( 95% O2 , 5% CO2 ) ice-cold solution containing ( in mM ) 93 N-Methyl-D-glucamine ( NMDG ) , 2 . 5 KCl , 1 . 2 NaH2PO4 , 30 NaHCO3 , 20 HEPES , 25 Glucose , 3 Sodium pyruvate , 10 MgCl2 , 0 . 5 CaCl2 , 5 . 6 Ascorbic acid . Horizontal slices ( 220 µm ) containing the VTA were sectioned using a Leica VT1200S vibratome ( Leica Biosystems ) and transferred to a holding chamber at room temperature ( RT ) filled with oxygenated solution containing ( in mM ) 109 NaCl , 4 . 5 KCl , 1 . 2 NaH2PO4 , 35 NaHCO3 , 20 HEPES , 11 Glucose , 1 MgCl2 , 2 . 5 CaCl2 , 0 . 4 Ascorbic acid . After incubation for at least 1 hr in the holding chamber at RT , slices were transferred to a recording chamber perfused with oxygenated aCSF containing ( in mM ) 126 NaCl , 3 KCl , 1 . 2 NaH2PO4 , 26 NaHCO3 , 11 Glucose , 1 . 5 MgCl2 , 2 . 4 CaCl2 , maintained at 35–36°C using an inline solution heater ( Warner Instruments , Hamden , CT ) . Cells were visualized with an upright microscope ( Olympus BX51WI ) equipped with infrared interference-contrast optics . Recorded neurons identified in the lateral VTA , medial to the terminal nucleus of the accessory optic track ( MT ) and anterior to the third cranial nerve . Dopamine neurons were identified in the lateral VTA using electrophysiological criteria in cell-attached mode . Only cell demonstrating regular pacemaker firing ( >3 Hz ) and action potential widths > 2 . 5 ms were chosen for further recording ( Ungless and Grace , 2012 ) . Whole-cell voltage-clamp recordings from DA neurons were acquired using an Axopatch 200B amplifier ( Molecular Devices , San Jose , CA ) . Recording pipettes ( 3 . 5–5 MΩ ) were pulled with a P-97 horizontal micropipette puller ( Sutter Instruments , Novato , CA ) and filled with internal solution containing ( in mM ) 140 K-gluconate , 2 NaCl , 1 . 5 MgCl2 , 10 HEPES , 10 Tris-phosphocreatine , 4 Mg-ATP , 0 . 3 Na-GTP , 0 . 1 EGTA ( pH 7 . 2 , 290 mOSM ) . DNQX ( 20 µM ) , DL-AP5 ( 40 µM ) , picrotoxin ( 100 µM ) and strychnine ( 1 µM ) were present in the aCSF to block AMPA , NMDA , GABAA and glycine receptors , respectively . Electrophysiological identification of DA neurons was performed in cell-attached mode to select only cells exhibiting pacemaker firing and action potential widths < GABAB IPSCs were evoked using electrical stimulation with bipolar tungsten stimulating electrodes with tip separation of 300–400 µm . A train of 6 stimuli of 100µs duration were delivered at 50 Hz every 30 s . Stimulation protocols were generated , and signals acquired using the electrophysiology software WinLTP . Control GABAB currents were recorded for 10 min before the appropriate drug was applied for an additional 30 min . Data was analyzed using WinWCP software ( Courtesy of Dr . John Dempster , Strathclyde University , Glasgow , UK ) . Figures were generated , and statistics analyzed using GraphPad Prism6 ( v6 . 07; LaJolla , CA ) . Data are presented as the change in percent from control . The experiments were designed using estimates of effect size and standard error derived from prior experience and pilot experiments . These values were then used in power analysis calculations using the program G-Power ( version 3 . 1 . 9 . 4 , University of Dusseldorf , Germany ) to determine sample sizes . Means ± s . e . m . are used throughout to report measures of centricity and dispersion . A spreadsheet ( Source data 1 ) describing means , significance levels and 95% confidence intervals for each experiment is included with this report . Statistical tests were determined by the number of groups and treatments to be compared . An omnibus test was used when necessary statistical assumptions could be met . Thus , in experiments where repeated measures could be obtained from the same subjects , samples , or cells ( e . g . time course data ) , a repeated-measures ANOVA was used . When repeated measures were not performed , and group size was >2 , a one-way ANOVA was used . Post-hoc analyses ( Tukey’s , Dunnett’s , or Bonferroni’s multiple comparison tests ) were determined by the type of omnibus test , as well as the nature of the multiple comparisons ( pairwise rows and columns , comparison to control columns , main effects versus interactions ) . When only two groups of data were compared , a Student’s t-test was used . In all cases , a two-tailed p value of 0 . 05 was considered the minimum for significance . Actual p values are reported for all omnibus tests , unless p<0 . 0001 , and the statistical information is reported in the figure captions . In immunoprecipitation experiments , co-localization was determined from observed association on Western blots , and therefore , statistical tests were not used ( Figure 1F and G; Figure 3B and D; Figure 5D; Figure 1—figure supplement 1C ) . | The cannabis plant contains hundreds of different chemicals , including more than sixty types of cannabinoids . By binding to specific sites on brain cells , cannabinoids change how cells communicate with one another . This in turn triggers widespread alterations in brain activity , which can affect mood , appetite , coordination and perception . But not all cannabinoids come from plants . The brain also produces its own versions , known as endocannabinoids ( or eCBs for short ) . These bind to the same sites on brain cells as the plant-derived chemicals . Changes in endocannabinoid activity have been implicated in various brain disorders . These include Alzheimer's disease , epilepsy and stress disorders . They may also have a role in drug addiction . Exposing rats to cocaine causes endocannabinoid levels to increase in areas of the brain that process pleasurable sensations . This suggests that the release of endocannabinoids may contribute to cocaine addiction . But how cocaine triggers this release has been unclear . By studying brain tissues and cells kept alive in petri dishes , Nakamura , Dryanovski et al . show that cocaine drives cells to release endocannabinoids via a process called extracellular vesicle release . In essence , cocaine causes cells to make endocannabinoids that are then enclosed inside membrane-bound packages . These packages – or extracellular vesicles – can then fuse with the cell’s outer membrane . Multiple proteins must interact with each other for cells to assemble and release extracellular vesicles . Nakamura , Dryanovski et al . show that disrupting these interactions prevents vesicles from forming , and also prevents cocaine from triggering endocannabinoid release . Blocking extracellular vesicle release prevents cocaine from altering communication between brain cells . Cocaine thus drives endocannabinoid release in the brain’s pleasure centers via the assembly of extracellular vesicles . Using other drugs to manipulate the protein interactions that underlie vesicle assembly could provide a new way to counter cocaine addiction . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"cell",
"biology",
"neuroscience"
] | 2019 | Cocaine-induced endocannabinoid signaling mediated by sigma-1 receptors and extracellular vesicle secretion |
Perception of external objects involves sensory acquisition via the relevant sensory organs . A widely-accepted assumption is that the sensory organ is the first station in a serial chain of processing circuits leading to an internal circuit in which a percept emerges . This open-loop scheme , in which the interaction between the sensory organ and the environment is not affected by its concurrent downstream neuronal processing , is strongly challenged by behavioral and anatomical data . We present here a hypothesis in which the perception of external objects is a closed-loop dynamical process encompassing loops that integrate the organism and its environment and converging towards organism-environment steady-states . We discuss the consistency of closed-loop perception ( CLP ) with empirical data and show that it can be synthesized in a robotic setup . Testable predictions are proposed for empirical distinction between open and closed loop schemes of perception .
Closed loops are systems in which every signal eventually affects its source; open loops are systems in which signals cannot affect their sources . Clearly , brains contain closed-loops at all levels , some of which have been implicated in relation to perceptual processing ( Ahissar and Kleinfeld , 2003; Edelman , 1993; Martin , 2002; Pollen , 1999 ) . Yet , whether perceptual acquisition is considered an open-loop or closed-loop process does not depend on the existence of closed loops within the chain of processing , but on whether the entire chain of processing is closed ( as a loop ) or open . Thus , a perceptual process that starts at the sensory organ and ends somewhere in the brain , whether containing local loops or not , is termed here an open-loop perceptual ( OLP ) process ( Figure 1A ) , whereas a perceptual process that includes the sensory organ but has no starting nor ending point , is termed a closed-loop perceptual ( CLP ) process ( Figure 1B ) . 10 . 7554/eLife . 12830 . 003Figure 1 . Possible perceptual schemes . ( A ) An open-loop scheme ( in the motor-sensory sense ) – perception begins with an interaction ( uni- or bi-directional ) between the object and the sensory organ ( an eye in this illustration ) and ends somewhere in the brain where a relevant neuronal representation ( NR ) is formed . ( B ) A closed-loop scheme ( in the motor-sensory sense ) – perception is a circular process , with no starting or ending points , which contains the sensory organ . DOI: http://dx . doi . org/10 . 7554/eLife . 12830 . 003 The OLP doctrine holds that external objects and features are perceived in an open-loop manner , in the motor-sensory sense ( Baars , 2002; Dehaene et al . , 1998; Hochstein and Ahissar , 2002; Riesenhuber and Poggio , 2000; Tononi and Koch , 2008; Ullman , 2007 ) . Thus , for example , an apple activates retinal receptors , which in turn initiate a stream of activations in the brain , some of which may depend on internal loops , i . e . , loops that do not include the sensory organ . An activity pattern that is repeatedly evoked in a given neuronal network in response to a presentation of the apple , and/or when such an apple is perceived , is often termed a neuronal correlate or neuronal representation ( NR ) of that apple . NRs are representations that are not necessarily consistent or unique , i . e . , they may appear in only some of the cases in which the apple is presented or perceived , and may appear also when other objects are presented or perceived . If a specific NR is evoked in a given brain for each and every perceived appearance of the apple , is invariant to changes in internal and environmental conditions , and is unique to the apple , it can be termed “the” invariant representation ( IvR ) of the apple in that specific brain . Assuming OLP , IvRs should be invariant to the acquisition mode . Visual IvRs of the apple , for example , should be the same in passive and active acquisition modes , i . e . , when the eye is stationary and the object moves or flashes ( passive mode ) and when the object is stationary and the eye moves ( active mode ) . The search of NRs that are also IvRs , during the last 6–7 decades , yielded several key findings . Among those is the characterization of NRs of various external features along the relevant sensory streams . For example , NRs of brief presentations of visual elements , such as dots and bars , were characterized among retinal , thalamic and cortical neurons ( Hartline , 1938; Hubel and Wiesel , 1962 ) . NRs of more complex visual patterns were characterized in various cortical areas ( Creutzfeldt and Nothdurft , 1978; Fujita et al . , 1992; McMahon et al . , 2014 ) . Crucially , however , although partial invariance had been demonstrated for portions of the proposed NRs in some of the cases , none of these NRs was shown so far to be “the” IvR of a specific external object or feature , namely an NR that is ( at least substantially ) invariant to changes in the most relevant conditions of perception . Moreover , none of these studies provides information that can discriminate between OLP and alternative hypotheses . Consider , for example , studies exhibiting single neurons that increase their firing rate significantly and selectively for a given object ( e . g . , a face ) out of several presented objects , and for several variations of that object ( McMahon et al . , 2014; Quiroga et al . , 2005; Viskontas et al . , 2009 ) . The critical factor here is that such a neuron cannot be considered as describing the IvR of that object , neither as describing a reliable projection of the IvR . Based on combinatorial considerations and response variations the assumption in such cases is that the elevated firing rate of such a neuron is a ( tiny ) component of the relevant NR , and not the NR itself . The question is , then , would the assumed NR be invariant to a sufficiently large portion of all relevant variations of object presentation and context . Given that these neurons are not completely invariant even to the limited sample of variations presented to them ( as is evident from the substantial trial-by-trial variability of their responses ) and their tiny contribution to the actual NR , it is impossible to infer the level of invariance of the actual NR out of the firing patterns measured from these neurons . Studying the passive mode of sensation also revealed various forms of internal transformations between NRs , such as , for example , transformation from NRs of static dots to NRs of static bars ( Hubel , 1996; Reid , 2001 ) , from temporal-code based NRs to rate-code based NRs ( Ahissar et al . , 2000 ) or from rate-code based NRs to temporal-code based NRs ( Cleland , 2010 ) . Clearly , these mechanisms can function within both OLP and CLP schemes of perception . Passive-mode experiments were also instrumental in describing the minimal exposure times required for generating meaningful perceptual reports . Across a large set of stimuli it was found that , depending on practice , exposure times as short as a few tens of milliseconds already allow a categorization of the presented stimulus , at least in a binary manner . As will be shown below , these findings are consistent with both OLP and CLP schemes . As described above , the OLP doctrine allowed an invaluable characterization of various components of the perceptual systems of mammals , using a set of reductionist steps . In order to verify that these specific reductions of the perceptual process are scientifically valid , one has to reconstruct perception by combining back the individual identified components . Succeeding in doing so will not only validate the specific reductionist approaches used , but , more importantly , show that OLP can be considered as a valid ( i . e . , self-consistent ) theory of perception . At this stage we can ask whether OLP is consistent with the data collected so far . We describe here several major findings that appear to be inconsistent with OLP and thus significantly challenge the validity of OLP as a mechanism for natural perception in mammals .
Here we propose a closed-loop scheme of perceptual acquisition , and suggest to refer to it as a possible alternative to the OLP doctrine . Within the scope of this paper we describe the acquisition of information about the organism’s immediate environment and do not address the interactions between perceptual acquisition and perceptual report . The CLP scheme is consistent with the same data challenging OLP , primarily because it considers sensor motion as an integral part of perception rather than as a factor that needs to be corrected for . We propose to continue comparing the two alternative schemes on equal grounds against accumulating data , and for aiding such a comparison we list potentially discriminative experiments towards the end of this article . One natural choice of a mathematical framework for CLP is the framework of dynamical systems ( Kelso , 1997; Port and Van Gelder , 1995 ) . Within this framework each MSM-loop is modeled as a dynamical system that includes motor , sensory and neuronal variables , as well as the differential equations which describe their relations . The following is a general mathematical description of such a model ( Figure 6A ) :s¯=f ( m¯ , u ) n˙¯=g ( n¯ , s¯ ) m˙¯=h ( m¯ , n¯ ) The bars above the letters indicate that they represent a vector ( of one variable or more ) . g and h are functions describing the intrinsic dynamics of the variables ( n¯ and m¯ respectively ) and their dependency on the variables in the preceding stations of the loop ( s¯ and n¯ respectively ) . The sensory variables ( s¯ ) do not depend on their intrinsic dynamics in this formalization , which assumes short sensory time constants; they are determined by the motor variables ( m¯ ) and the state of the environment ( u ) , according to the function f . The function f encapsulates the physical laws governing the sensory organ-environment interactions and the transduction of physical signals to neuronal ones . The state of the system is defined as the vector containing all the variables ( m¯ , s¯ , n¯ ) . Perception is achieved through the convergence of the system to a steady-state within this state-space . The information of the perceived feature is contained in the values of the dynamic variables ( the system’s state ) at this steady-state . High-level functions such as integration of the general context or a report mechanism are not included in this model of single-feature acquisition . 10 . 7554/eLife . 12830 . 008Figure 6 . Synthesis of closed-loop perception in a robotic setup . ( A ) A sketch of the MSM-loop model template . m , motor variable; SO , sensory organ; s , sensory variable; n , neuronal variable; h , f , g , transfer functions; u the environment dynamics . The arrows depict the direction of information flow within the loop . ( B ) The SYCLOP robotic platform . A sketch of the robot with its different components: Pan-Tilt control unit ( PTCU , only the pan axis was used here ) ( 1 ) , DVS camera ( 2 ) , and desktop computer ( 3 ) . The computer sends commands to the PTCU which controls the camera’s rotations in the azimuth ( θ ) and elevation ( ε ) axes . The DVS camera sends visual ‘on’ and ‘off’ events to the computer . ( C-F ) Implementation of a specific contrast perceiving CLP model ( see text ) . ( C , D ) n1 , the integrated difference between ‘ON’ and ‘OFF’ events , and ω , sensor angular velocity along the pan axis , ( C and D , respectively ) as a function of time in two different runs of the CLP algorithm , one facing a contrast of 0 . 9 ( red ) and one facing a contrast of 0 . 5 ( green ) . n1 is scaled in units of 1000 events . ( E ) System’s trajectories in the 2D n1-ω state space . Same data as in C and D . ( F ) Example of emergent smooth-pursuit like behavior when using a moving edge as a stimulus . The trajectory of the system in the n1-ω plane ( gray line ) overlaid on a heat map where the color of each segment corresponds to the amount of time in seconds the system spent within this segment . The smooth pursuit periods are represented by the white and light red squares . While in a smooth pursuit , the camera was moving with a constant angular velocity – smoothly tracking the edge . DOI: http://dx . doi . org/10 . 7554/eLife . 12830 . 008 One way to test such CLP models and demonstrate their basic behavior is to implement them using a synthetic agent . We built a simple robot for this purpose; the robot ( SYCLOP: SYnthetic Closed-LOop Perceiver ) includes two motors , one sensor and their bilateral connections ( Figure 6B ) . This platform allows the implementation of minimal MSM-loops based models ( one motor DOF and one sensor ) . The SYCLOP uses a biomimetic camera ( DVS128 , iniLabs Ltd Zurich , Switzerland , Lichtsteiner et al . , 2008 ) as its sensor; this camera , like a retina , sends signals only upon luminance intensity changes . The camera is mounted on a pan-tilt control unit ( PTU-46-17 , DirectedPerception , CA , USA ) . The motor-to-sensory connection is implemented by moving the camera along the pan-tilt axes while the sensory-to-motor connection is implemented by a computer that implements the model's equations . The SYCLOP platform was used , for example , to implement and test the behavior of a single MSM-loop model which was designed to perceive a visual contrast . The stimulus , in this case , was presented on a computer screen: half of the screen was kept dark and on the other half a uniform grayscale surface was displayed . The grayscale values ranged from dark to white . We defined two sensory variables ron and roff - the rate of ‘ON’ events ( single-pixel events in which the luminance intensity increased ) and the rate of ‘OFF’ events ( single-pixel events in which the luminance intensity decreased ) - integrated over the entire camera’s field . The characterization of the dependency of these two sensory variables on the chosen motor variable ( sensor angular velocity along the pan axis , ω ) and the external feature ( contrast , γ ) resulted in the following equation:ron−roff=C1γω Where C1 represents a constant and noise is ignored . The MSM-loop model is completed by the addition of two transfer functions that define two differential equations , sensory-to-neuronal ( g ) and neuronal-to-motor ( h ) :{n˙1=g ( n1 , ron , roff ) =C2 ( ron−roff ) =C2C1γωw˙=h ( ω , n1 ) =1C3 ( μ ( 1−C4n12 ) C3ω−n1 ) Where n1 is defined as the ( single ) neuronal variable , which integrates the difference between ron and roff ( Demb and Singer , 2012 ) , ω is the ( single ) motor variable defined as the sensor’s angular velocity and C2 , C3 and C4 represent constants . The functions g and h were chosen such that the resulting dynamical system would be equivalent ( up to constants multiplications , assuming all constants and parameters are positive ) to a Van der Pol oscillator ( Kanamaru , 2007 ) . This specific system was chosen due to its known dynamics: the system converges to a single closed trajectory within its 2D phase plane ( i . e . a limit cycle ) independently of the initial values of the variables . After convergence each of the dynamic variables is a periodic function of time ( e . g . , Figure 6C and D ) . Clearly , other dynamical systems could fit as well . This model was implemented on the SYCLOP platform with the aid of a c program running on the computer incorporated in the platform ( Figure 6B , item 3 ) . The program received the ON and OFF events from the DVS camera , computed the ron and roff sensory variables and used them to compute the values of n1 and ω by integrating the two differential equations described above . The value of ω was then sent by the program to the pan-tilt controller and modified the camera’s pan velocity . This implementation illustrates a simple CLP convergence process ( Figure 5 ) and shows how different precepts can be differentiated in CLP . The convergence dynamics involves different dynamics of the sensory ( ron and roff ) , neuronal ( n1 , Figure 6C ) and motor ( ω , Figure 6D ) variables . Yet , the variables are strongly linked , as demonstrated by the phase diagram of the neuronal and motor variables ( Figure 6E ) ; these two variables quickly converge to a limit cycle ( i . e . , a constant closed trajectory in the phase plane ) . Similar behavior is observed in the other phase planes ( sensory-motor and sensory-neuronal , not shown ) . Importantly , in all these phase planes the limit cycle depends on the external contrast ( γ ) ; while maintaining all loop parameters constant , a monotonic change in γ results in a corresponding monotonic change of the limit cycle ( green and red trajectories in Figure 6E for contrasts of 0 . 5 and 0 . 9 , respectively ) . Hence , the image’s contrast can be inferred from the asymptotic behavior of the system or , in other words , the motor-sensory-neuronal trajectory that is uniquely associated with ( or , equivalently , the CLP’s IvR of ) a given contrast can be “retrieved” by the presentation of that contrast to the perceiver . The behavior of the SYCLOP is described here in order to demonstrate how a possible implementation of our CLP model would look like . Interestingly , however , it is worth mentioning that the SYCLOP also exhibits behaviors that it was not intentionally designed to exhibit – for example , a smooth pursuit behavior . When presented with a moving image ( back and forth horizontal movement of the contrast image at a constant speed ) SYCLOP tended to track the image smoothly in each direction ( as indicated by the “dwelling spots” at ω≈5; and −5 deg/s Figure 6F ) .
CLP suggests that perception of the external environment is a process in which the brain temporarily ‘grasps’ external objects and incorporates them in its MSM-loops . Such objects become virtual components of the relevant loops , hardly distinguishable , as long as they are perceived , from other components of the loop such as muscles , receptors and neurons . What primarily distinguishes external objects from body parts are inclusion duration and state; short and transient inclusions mark external objects while long and steady inclusions mark body parts ( see also Uexkull , 1926 ) . Interestingly , the perceptual dynamics suggested by this hypothesis reconciles a conflict between objective scientific observations and the subjective everyday experience of perceiving objects with no mediation ( see also A philosophical angle below ) . Everyday perception of a given external object , CLP suggests , is the dynamic process of inclusion of its features in MSM-loops . This process starts with a perturbation , internal or external , and gradually converges towards a complete inclusion - approaching , although never reaching , a state of “direct” perception . A laboratory-induced flashed stimulus , according to this model , probes the initiation of a perceptual process , whereas dreaming and imagining evoke internal components of the process . We consider here all versions of OLP , i . e . , all versions of hypotheses in which perception does not depend on the integrity of the MSM-loop and its closed-loop dynamics within individual perceptual epochs . We consider here two major OLP classes: in one , sensory OLP ( sOLP ) , the movement of the sensory organ is not an essential component of perception , and in the other , motor-sensory OLP ( msOLP ) , it is ( Figure 7 ) . sOLP thus assumes that IvRs are confined to the brain ( i . e . , they are specific NRs ) and can be fully retrieved by sensory activations alone when the sensor is passive . msOLP , in contrast , postulates that IvRs are not confined to the brain , and can form the basis for perception only if they include the relevant MS-contingencies ( Figure 7 ) . According to msOLP , IvRs cannot be retrieved with passive sensory organs . Importantly , however , msOLP does not assume a motor-sensory-motor loop; that is , its scheme includes a motor-to-sensory arc but not a sensory-to-motor arc ( Figure 7 ) . Hence , with msOLP , movements of the sensory organ are predetermined for each perceptual epoch and are not affected by the ongoing sensory input during that epoch . In contrast to the OLP hypotheses , using the same representational terminology , CLP postulates that IvRs can form the basis for perception only if they contain the dynamics and state of the entire MSM-loop including the relevant features of the object ( Figure 7 ) . Thus , the minimal set of variables that must be included in the IvR of each object , or feature , is different for each hypothesis ( Figure 7 , bluish ellipses ) : internal-only sensory variables in sOLP , internal sensory variables and MS-contingencies in msOLP and the entire perceiving loop in CLP . 10 . 7554/eLife . 12830 . 009Figure 7 . Functional connectivity and essential elements of perceptual schemes . The essential elements in each scheme are indicated by solid curves and blue titles . MSM , motor-sensory-motor; MS , motor-sensory; S , sensory; NR , neuronal representation; green curves , re-afferent related pathways; magenta curves , ex-afferent related pathways . Note that re-afferent related pathways can form closed-loops with their sensory organs also in OLP schemes ( dashed curves ) . Arrows indicate optional whisker ( black ) or object ( magenta ) movement; solid arrows indicate movements that are essential for perception; in the sOLP scheme none of the movements is essential in itself , but it is essential that at least one of them will occur in order to activate the receptors . Appropriate experimental paradigms are indicated by green titles; CLP and msOLP schemes can be studied only via active sensing paradigms . The minimal sets for invariant representations ( IvRs ) of external features , i . e . , the components that must be included in any IvR according to each perceptual scheme , are marked by the bluish ellipses . sOLP: internal , sensory only NRs . msOLP: sensory NRs + motor-object-sensory contingencies . CLP: entire motor-object-sensory-motor loops . DOI: http://dx . doi . org/10 . 7554/eLife . 12830 . 009 Perhaps the first question that comes to mind when considering msOLP and CLP is whether paralyzed subjects perceive stationary ( i . e . , not flashing or moving ) objects similarly to non-paralyzed subjects . If they do - here go the msOLP and CLP hypotheses . Unfortunately , however , this is not a trivial test . Note that the paralysis must include the relevant sensory organ and the object must be entirely stationary . In the case of touch it should be evident that while contacts may be detectable , no object perception is possible with paralyzed hands – we are not aware of any study contradicting this conjecture . In contrast , our intuition regarding hearing is that action is not a fundamental requirement for hearing . Yet , two important points are relevant here . First , our intuition may be misled by the fact that we cannot be aware of motor activation of the outer hair cells and the muscles of the middle ear – we are not aware of perceptual experiments in which these activations were blocked , or measured . Second , no stationary object exists in audition . Acoustic waves are always dynamic and always activate the inner hair cells . This makes auditory sensation less dependent on self-motion , a fact that indeed may put audition in a motor-sensory regime that is distinct from those of touch and vision . Regarding vision , we are aware of only one study analysing visual perception in a congenital ophthalmoplegic patient , a patient who had no eye movements since birth; in this case , the patient developed a pattern of head movements that resembled that of natural eye movements , only on a slower rhythm ( Gilchrist et al . , 1997 ) . This adaptation clearly indicates the need in active sensation for visual perception , at least in that patient . Natural employment of active vision is indicated by the "weird , confusing and indescribable" forms of perceptions reported during acute partial paralysis of the ocular muscles ( Stevens et al . , 1976 ) . These data are certainly not consistent with sOLP . Yet , these data , as well as part of the OLP-challenging data presented above , may still be consistent with msOLP . The distinction between msOLP and CLP hypotheses is thus more demanding , and requires specifically designed experiments . We describe here examples of potentially discriminative experiments in three categories . Ideally , the comparison of the behaviors predicted by CLP and OLP , related to the inter-dependencies of motor , object , sensory and report variables , should be done in natural conditions . Practically , as the scientific method enforces reductionist steps , it is important to notice what reductions are allowed , as behavioral predictions of CLP or msOLP , regarding natural perception , cannot be tested in paradigms in which their basic assumptions are “reduced out . ” Clearly , if eye or whisker motion is prevented , critical predictions of CLP or msOLP cannot be tested . Experiments in which eye or whisker motion is allowed but head motion is restrained have a limited discriminative power - conclusions in these cases should take into account the possibility that head-restrained animals develop unique compensatory active strategies which may not be indicative for the head-free condition . When MSM-loops are not given enough time to converge , as is the case with passive sensing ( e . g . , visual flashes ) for example , discrimination between CLP and OLP is usually not possible ( as both predict partial perception , Figure 5 ) . For at least four centuries the philosophical community , and during the last century also the neuroscience community , have been puzzled by the contrast between objective scientific observations that relate to perception and the everyday subjective experience of perception . What feels direct and immediate to every human perceiver appears indirect and mediated when physical constrains are taken into account ( Crane , 2005; Kelso , 1997; Port and Van Gelder , 1995; Ullman , 1980 ) . Our CLP hypothesis proposes a reconciliation of objective scientific observations and subjective everyday experience via closed-loop dynamics between the perceiver and the perceived . Such closed-loops converge gradually to a state in which the perceiver and the perceived are inseparable . The idea is that , although the loops never actually reach an ideal steady-state , they get closer and closer to these states during a perceptual epoch and typically quit the convergence process when the distance from a steady state is barely sensible . Being close enough to the steady state can give rise to the feeling of direct and immediate perception . In practical terms , this article proposes to open the discussion about the phenomenology and mechanisms of perception , and in particular to confront open- and closed-loop schemes . We hope that the set of predictions listed here will serve as a starting point for informative experimental confrontation . | How do we perceive the world around us ? Today the dominant view in brain research is that sensory information flows from the environment to our eyes , fingers and other sense organs . The input then continues on to the brain , which generates a percept . This process is referred to as “open-loop perception” because information flows through the system predominantly in one direction: from the environment , to the sense organs , to the brain . Open-loop perception struggles to account for a number of key phenomena . The first is that sensation is an active process . Our eyes and hands constantly move as we interact with the world , and these movements are controlled by the brain . According to Ahissar and Assa , a more accurate view of perception is that the brain triggers the movement of the sense organs , and thereby alters the sensory information that these organs receive . This information is relayed to the brain , triggering further movement of the sense organs and causing the cycle to repeat . Perception is therefore a “closed loop”: information flows between the environment , sense organs and brain in a continuous loop with no clear beginning or end . Closed-loop perception appears more consistent with anatomy and with the fact that perception is typically an incremental process . Repeated encounters with an object enable a brain to refine its previous impressions of that object . This can be achieved more easily with a ‘circular’ closed-loop system than with a linear open-loop one . Ahissar and Assa show that closed-loop perception can explain many of the phenomena that open-loop perception struggles to account for . This is largely because closed-loop perception considers motion to be an essential part of perception , and not an artifact that must be corrected for . The open- and closed-loop hypotheses should now be compared systematically . One approach would be to construct an artificial perceiver ( or robot ) based on each hypothesis and examine its behavior . Another would be to perform experiments in which the two hypotheses make opposing predictions . Paralyzing a sensory organ without affecting the flow of sensory information , for example , would impair perception according to the closed-loop hypothesis , but would have no effect according to the open-loop hypothesis . | [
"Abstract",
"Introduction",
"Hypothesis",
"and",
"Results",
"Discussion"
] | [
"neuroscience"
] | 2016 | Perception as a closed-loop convergence process |
Characterising the longevity of immunological memory requires establishing the rules underlying the renewal and death of peripheral T cells . However , we lack knowledge of the population structure and how self-renewal and de novo influx contribute to the maintenance of memory compartments . Here , we characterise the kinetics and structure of murine CD4 T cell memory subsets by measuring the rates of influx of new cells and using detailed timecourses of DNA labelling that also distinguish the behaviour of recently divided and quiescent cells . We find that both effector and central memory CD4 T cells comprise subpopulations with highly divergent rates of turnover , and show that inflows of new cells sourced from the naive pool strongly impact estimates of memory cell lifetimes and division rates . We also demonstrate that the maintenance of CD4 T cell memory subsets in healthy mice is unexpectedly and strikingly reliant on this replenishment .
The diversity and longevity of T cell memory are shaped by influx , cell division and cell death . A major challenge is to understand how these processes are regulated in health and how they respond to perturbations such as chronic infections . To understand the evolution of immune repertoires within a host therefore requires quantifying homeostatic processes , identifying the rules of replacement within memory subsets , and characterising any distinct homeostatic niches that lie within them . The dominant approaches to studying T cell population dynamics at steady state involve the adoptive transfer of cells labelled with inert dyes such as CFSE and using its rate of dilution to infer rates of proliferation ( De Boer et al . , 2006; Choo et al . , 2010 ) , or tracking the uptake and dilution of labels incorporated into the DNA of dividing cells , the kinetics of which contain information regarding both division and loss of cells ( De Boer and Perelson , 2013 ) . In both cases , mathematical models are needed to interpret the data . However , while for some T cell subsets in mice and humans there is broad agreement regarding basic parameters such as population-averaged cell lifetimes , discrepancies remain and defining homeostatic dynamics in detail is difficult . Gaps in our understanding of how memory compartments are structured , and how the processes of cell division and death are correlated , generate uncertainty in how to formulate the dynamical models to describe data from labelling studies ( De Boer et al . , 2003 ) and indeed these data may not be sufficiently rich in information to allow us to discriminate between these models ( De Boer et al . , 2012 ) . One major difficulty is measuring the contribution that any influx of new cells originating from naive precursors makes to the maintenance of immune memory . While it is clear that new memory T cells are generated during infections and by the seeding of empty peripheral compartments , either early in ontogeny ( Le Campion et al . , 2002 ) or following reconstitution in irradiation chimeras ( Surh and Sprent , 2008 ) , it is unclear whether there continue to be significant contributions from the naive pool in the steady state in normal healthy hosts , in the absence of infection . Labelling experiments can be inconclusive in this regard . In early studies , flows into cell populations of interest – for example , entry of recent thymic emigrants into the mature naive T cell pools , or cells moving from the naive pools into memory through antigen-driven expansion – were invoked to explain observations that average rates of division and death were estimated to be unequal , despite the populations being apparently stable in size ( Mohri et al . , 1998 , 2001; Bonhoeffer et al . , 2000 ) . However , in some cases the nature of these source terms was puzzling . First , the required magnitude of the source in studies of naive T cell turnover vastly exceeded what was expected from the thymus ( Asquith et al . , 2002; Borghans and de Boer , 2007 ) . Second , sources were also required to be largely unlabelled during label administration ( Mohri et al . , 2001; Asquith et al . , 2002 ) , but memory precursors might be expected to have divided – and thus incorporated the label – shortly before entering the memory pool . An alternative explanation of these observations reflects heterogeneity within T cell compartments . A key study ( Asquith et al . , 2002 ) pointed out that differences in division and death rates inferred from the accrual and loss of label need not derive from influx and instead may arise if the behaviour of labelled ( and so recently-divided ) cells does not reflect the population average . Such heterogeneity can be classified into two non-exclusive types; kinetic ( Asquith et al . , 2002; Macallan et al . , 2003; Asquith et al . , 2006; Borghans and de Boer , 2007; Vrisekoop et al . , 2008; Asquith et al . , 2009; Ganusov et al . , 2010; Ganusov and De Boer , 2013; De Boer et al . , 2012; De Boer and Perelson , 2013; Westera et al . , 2013 ) and temporal ( Grossman et al . , 1999; Bonhoeffer et al . , 2000; Ribeiro et al . , 2002; De Boer et al . , 2003 , 2012; De Boer and Perelson , 2013 ) . A kinetically heterogeneous population comprises two or more subpopulations with different rates of division and/or death , and short-term labelling experiments will tend to over-sample those dividing fastest . Temporal heterogeneity reflects the idea that cells within a single population may display different rates of division or turnover ( loss ) at different times . For example , quiescent and dividing or recently-divided cells may be differentially susceptible to death . This form of heterogeneity has been invoked in models of T cell clonal expansion ( Bonhoeffer et al . , 2000; Ribeiro et al . , 2002; De Boer et al . , 2003 , 2012 ) and homeostasis ( De Boer et al . , 2012 ) . Discriminating between kinetic and temporal heterogeneity with DNA labelling alone is challenging ( De Boer et al . , 2012; De Boer and Perelson , 2013 ) , but doing so has the potential to give us mechanistic insights into T cell homeostasis on different levels . Kinetic heterogeneity likely reflects phenotypic substructure , and suggests the existence of distinct homeostatic or ecological niches if these subpopulations are stably maintained . In contrast , the extent of temporal heterogeneity – that is , how division and death are coupled – provides clues as to how numbers are regulated at the single-cell level . In short , our understanding of memory T cell homeostasis is limited because the effects of external sources of cells and heterogeneity in population dynamics may mimic or mask one another and it is difficult to distinguish them with conventional approaches . In this study we aimed to disentangle these processes , focusing on memory CD4 T cell subsets in mice . We use a temporal fate mapping method to directly estimate the constitutive rates of flow of cells into different memory CD4 T cell compartments . We then generate fine-grained timecourses of DNA labelling combined with measurements of cell cycle status , a strategy that when paired with the independent estimates of memory influxes yields sufficient information to discriminate between alternative models of population dynamics . This combined approach allows us to estimate for the first time the contributions at steady state of de novo production of memory cells and production through division of existing memory cells , as well as yielding insights into the cellular mechanisms regulating memory CD4 T cell subsets in mice .
It is an immunological paradigm that activation of naive T cells by foreign antigens ultimately gives rise to persistent populations of memory cells . However , in healthy individuals , not deliberately infected , it is unclear whether generation of memory is an event restricted to first encounter with environmental antigens , such as in the establishment of the T cell compartment in neonates , or whether generation of memory cells occurs constitutively throughout life . Whether there is continual differentiation of naive cells into memory in the steady state has not previously been assessed , and knowledge of this quantity is critical for quantitative analysis of memory homeostasis . To characterise the fluxes into memory subsets , we took advantage of a temporal fate mapping approach described previously ( Hogan et al . , 2015 ) that allows visualisation of tonic reconstitution processes within different haematopoetic compartments . Briefly , we condition young adult CD45 . 1 hosts with the chemotherapeutic drug busulfan that ablates haematopoetic stem cells ( HSC ) but leaves compartments of committed lineages intact , including thymic and peripheral T cells . Conditioned hosts are then reconstituted with CD45 . 2 bone marrow ( Figure 1A ) . Total numbers of thymocytes remain normal , and by 6 weeks the CD45 . 1:CD45 . 2 ratio equilibrates in all thymic compartments ( Hogan et al . , 2015 ) . We see no trend in thymic chimerism across treated animals out to a year post-BMT ( Hogan et al . , 2015 ) , indicating that chimerism among T cell precursors is stably maintained . Chimeric mice also exhibit normal numbers of peripheral CD4 naive and CD4 memory T cells ( Figure 1B ) , and both populations display normal levels of proliferation as assessed by Ki67 expression ( Figure 1C ) . Together these data indicate that busulfan treatment and the generation of bone marrow chimerism have no meaningful impact on lymphocyte homeostasis , as previously described ( Vezys et al . , 2006; Hogan et al . , 2015 ) . By 8 weeks post-BMT , donor-derived cells are readily detectable not only in the naive but also to a striking extent in the CD44hi memory compartment ( Figure 1D ) , revealing that well into adulthood newly generated naive T cells continue to differentiate into memory in clean healthy mice . We observe a steady but ultimately incomplete replacement of the host-derived CD4 memory cells with donor cells over the course of a year ( Figure 1E ) , while the total CD4 memory compartment remains relatively stable in size ( Figure 1B ) . 10 . 7554/eLife . 23013 . 003Figure 1 . New donor T cells differentiate into memory compartments in the absence of deliberate infection . ( A ) Outline of experimental protocol . Host CD45 . 1 mice aged 8 weeks were treated with two doses of 10 mg/kg busulfan , followed by injection of 107 T cell-depleted bone marrow cells from CD45 . 2 donors . The numbers of donor and host cells in the thymus and peripheral lymphocyte compartments were evaluated by flow cytometry at various time points up to one year post bone marrow transplantation ( BMT ) . ( B ) Numbers of naive and memory CD4 T cells ( host + donor ) recovered from spleen and lymph nodes of busulfan chimeras made at age 8 weeks , compared to numbers in WT CD45 . 1 controls . ( C ) Ki67 expression in naive and memory CD4 cells in chimeras ( 14 weeks post-BMT ) compared to age-matched WT controls; 11 mice per group . ( D ) Identification of host and donor-derived cells in a representative mouse 8 weeks post-BMT . ( E ) Timecourses of normalised peripheral chimerism ( defined as the proportion of the population that is donor-derived , divided by the proportion of the DP1 population that is donor-derived ) in naive and memory CD4 T cell populations , showing steady but incomplete replacement of host cells in both . Fitted curves are empirically determined to show trends only . DOI: http://dx . doi . org/10 . 7554/eLife . 23013 . 00310 . 7554/eLife . 23013 . 004Figure 1—source data 1 . Comparing naive and memory cell numbers and Ki67 expression in busulfan chimeras and wild-type controls ( panels B and C ) . DOI: http://dx . doi . org/10 . 7554/eLife . 23013 . 00410 . 7554/eLife . 23013 . 005Figure 1—source data 2 . Timecourses of infiltration of donor-derived T cells into the naive and memory compartments in busulfan chimeras ( panel E ) . DOI: http://dx . doi . org/10 . 7554/eLife . 23013 . 005 Since the continuous generation of new memory T cells was readily detectable , we examined the underlying dynamics more closely , paying particular attention to the feeding of canonically defined memory subsets . Specifically , we modelled the kinetics of replacement of host cells by donor cells within the CD4+CD25−CD44hi CD62Lhi and CD4+CD25−CD44hiCD62Llo populations , termed CD4 central memory ( CD4 TCM ) and CD4 effector memory ( CD4 TEM ) respectively ( Figure 2A ) , for over a year post-treatment . We previously demonstrated that the kinetics of lymphocyte replacement in the busulfan chimeras are a rich source of information regarding homeostatic turnover and population substructure ( Hogan et al . , 2015 ) . 10 . 7554/eLife . 23013 . 006Figure 2 . Estimating constitutive rates of generation of CD4 T cell memory . ( A ) Gating strategy for CD4 central and effector memory subsets . ( B ) Describing the kinetics of the source . Fits of empirical descriptor functions to the timecourses of naive CD4 counts and chimerism , with 95% uncertainty envelopes ( see Materials and methods ) . Similar curves ( not shown ) were used to describe CD4 TCM numbers and chimerism when modelled as the source for CD4 TEM . Estimates of the parameters defining the source functions are in Appendix 1—table 1 . ( C ) Timecourses of total ( host+donor ) numbers of CD4 TCM and TEM and of chimerism , modelled from 6 weeks post-BMT ( age 14 weeks/98 days ) . The resistant memory models with naive source described both the CD4 TCM and TEM data well ( left-hand and central panels ) . Also shown are the statistically poorer fits to CD4 TEM kinetics using a model in which they are fed exclusively by CD4 TCM ( ΔAIC = 11 ) . Both models contained five free parameters; estimates are in Appendix 1—table 2 . ( D ) Projections of how the rates of memory replacement change with age , assuming a naive source . Replacement is shown both as a fraction of the total pool , and as a fraction of the displaceable subset only . DOI: http://dx . doi . org/10 . 7554/eLife . 23013 . 00610 . 7554/eLife . 23013 . 007Figure 2—source data 1 . Timecourses of numbers and chimerism within the naive , effector memory and central memory CD4 T cell compartments in busulfan chimeras ( Figure 2 panels B and C , and Figure 2—figure supplement 1 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 23013 . 00710 . 7554/eLife . 23013 . 008Figure 2—source data 2 . Source code used to analyse flows between naive , CD4 TEM and CD4 TCM populations . DOI: http://dx . doi . org/10 . 7554/eLife . 23013 . 00810 . 7554/eLife . 23013 . 009Figure 2—figure supplement 1 . Early kinetics of peripheral replacement in busulfan chimeras made at age 8 weeks , showing that the generation of CD4 TEM cells lags that of CD4 TCM ( three mice per timepoint; mean and s . e . m . ) . DOI: http://dx . doi . org/10 . 7554/eLife . 23013 . 00910 . 7554/eLife . 23013 . 010Figure 2—figure supplement 2 . Estimated sizes of memory populations resistant to displacement . Proportions of CD4 TEM and TCM predicted to be numerically stable , self-renewing cells resistant to displacement , with age . Shaded regions indicate 95% confidence intervals . DOI: http://dx . doi . org/10 . 7554/eLife . 23013 . 010 Assuming host and donor cells behave similarly , the rate of accumulation of donor cells in each subset is a constant fraction of the total rate of influx of cells from the naive pool , presumably following clonal expansion , and/or through differentiation from other memory subsets . We used simple mathematical models to describe these flows ( see Materials and methods ) . Our choice of models was guided by two key observations . First , the appearance of donor CD4 TEM cells lagged that of both naive and TCM cells ( Figure 2C and Figure 2—figure supplement 1 ) , suggesting that CD4 TCM are sourced predominantly from naive precursors while CD4 TEM may be sourced either directly from the naive compartment and/or via CD4 TCM . Second , donor cells displayed a more restricted capacity for populating memory relative to the naive pool ( Figure 1E ) . We considered two explanations of this observation; either that there exist self-renewing populations of host-derived memory cells that resist displacement by newly recruited cells ( the ‘resistant memory’ model ) , or that the per-cell rates of entry into each memory subset wane over time ( the ‘declining recruitment’ model ) . We then assessed the abilities of these models to describe the kinetics of the sizes and host/donor compositions of the CD4 TEM and CD4 TCM pools in healthy , chimeric laboratory mice aged between 14 and 60 weeks . For both memory subsets we fitted each combination of model and source population to the timecourses of the total numbers and donor chimerism of cells recovered from spleen and lymph nodes . The donor chimerism was normalised to that at the early DP1 stage of thymic development in each animal . Doing this controlled for varying degrees of depletion of HSC across animals with busulfan treatment . The size and donor/host composition of the putative source populations ( naive for CD4 TCM , and either naive or TCM for CD4 TEM ) were not modelled explicitly but instead described by empirical functions fitted to the observations ( Figure 2B; Materials and methods ) . Together these steps allowed us to describe the data from multiple mice with single predictors reflecting the population-average parameters ( Figure 2C; Materials and methods ) . For both models explaining the apparent capping in host memory cell replacement , we compared the variant models in which CD4 TEM was fed either directly from naive or from CD4 TCM . Comparing these fits , we found the strongest statistical support for a dominant naive CD4 → TEM recruitment pathway over CD4 TCM →TEM ( ΔAIC = 11 ) . Although the model fits were visually similar ( Figure 2C ) , the conclusion favouring a naive source derives largely from substantial differences in the quality of the fits during the early stages of infiltration of donor cells into memory , which are relatively data-rich and well defined . Thus , naive T cell numbers , and not CD4 TCM numbers , provide the strongest predictor of CD4 TEM accumulation over long timescales . For CD4 TEM we found comparable statistical support for the resistant memory and declining recruitment models ( ΔAIC = 0 . 16 ) . For CD4 TCM we found stronger support for the resistant memory model ( ΔAIC = 8 . 3 ) . We cannot rule out a combination of resistant memory and declining recruitment , and by parsimony we favour the resistant memory model for both populations ( fits shown in Figure 2C ) . We estimate that in 14 week-old mice , 6 . 3% of CD4 TEM ( 95% confidence interval 4 . 4–8 . 6 ) and 9 . 8% ( 6 . 5–16 . 9 ) of CD4 TCM are displaced per week by new memory cells from the source . The declining recruitment model yielded comparable parameter estimates ( Table 1 ) . Indeed , consistent rates of production of new memory cells could be derived simply from the growth rate of chimerism in memory and the difference in chimerism between it and its source , irrespective of the details of the mechanism limiting memory replacement ( see Materials and methods ) . 10 . 7554/eLife . 23013 . 011Table 1 . Estimated rates of replacement of CD4 effector and central memory through influx of new cells in 14 week-old mice . We quote both absolute influx ( cells per day as a percentage of the pool size ) and percent replaced per week . The latter is slightly less than 7 × the daily rate of influx because immigrant cells are assumed to be lost at the same rate as existing displaceable cells ( see Methods ) . AIC differences ( ΔAIC ) are quoted relative to the best-fitting model for each cell type . These differences reflect the relative support for two models , with exp ( -ΔAIC/2 ) being the relative probability that it is the model with the lower penalised likelihood ( larger AIC value ) that minimises the information lost in describing the data . For CD4 TEM the two models have equal support , but for CD4 TCM the resistant memory model is favoured ( ΔAIC = 8 . 3 , exp ( -ΔAIC/2 ) =0 . 02 ) . Models considered most plausible are highlighted . DOI: http://dx . doi . org/10 . 7554/eLife . 23013 . 011Resistant memory modelDeclining recruitment modelSource% Input/day% Replaced/wkΔAIC% Input/day% Replaced/wkΔAICCD4 TEMNaive1 . 0 ( 0 . 7 , 1 . 4 ) 6 . 3 ( 4 . 4 , 8 . 6 ) 0 . 161 . 0 ( 0 . 7 , 1 . 6 ) 6 . 4 ( 4 . 3 , 9 . 7 ) 0CM4 . 1 ( 2 . 6 , 7 . 5 ) 23 . 0 ( 16 , 35 ) 113 . 6 ( 2 . 1 , 6 . 2 ) 21 ( 13 , 30 ) 11CD4 TCMNaive1 . 5 ( 1 . 0 , 2 . 8 ) 9 . 8 ( 6 . 5 , 17 ) 02 . 3 ( 1 . 3 , 4 . 1 ) 13 . 5 ( 8 . 4 , 22 ) 8 . 3 If this seeding of new memory occurs through recruitment of naive cells followed by clonal expansion and differentiation , one would expect there to be a delay in the transition between source and memory . To explore this we extended the models to allow for lags of 1–7 days before changes in the source population were reflected in changes in the rate of entry into memory . During this period the transitioning cells would likely disappear from the naive and memory T cell populations as we defined them ( see Materials and methods ) , through expression of the IL-2 receptor α-chain , CD25 . These extensions yielded rates of replacement that were very similar to the zero-lag models , with weaker statistical support , although the timecourses lack the resolution required to examine this transition process in detail . The models can also be used to predict how the rates of replacement of CD4 TEM and CD4 TCM change with age ( Figure 2D ) , although these predictions derive from relatively uncertain projections of the sizes of the populations beyond a year of age ( Figure 2C ) . We predict that between 14 weeks and 1 year of age , the proportion of cells replaced each week by new memory falls from 6 . 3% to 1% for CD4 TEM and 9 . 8% to 6% for CD4 TCM . For CD4 TEM this decline stems from a combination of the fall in naive T cell numbers ( the putative source population ) with age , and a predicted slow increase in CD4 TEM numbers ( Figure 2C ) . The declining recruitment model predicts a steeper drop in rates of replacement with age , due to the multiplicative effect of the fall in both naive T cell numbers and the per capita rate of recruitment from the naive pool with age ( Figure 2D , blue shaded regions ) . For CD4 TCM the proportional replacement remains relatively steady with age , because the drop in the size of the naive source population is balanced by the predicted slow decline in CD4 TCM numbers . Finally , we estimate that between 14 weeks and 1 year of age the resistant , numerically stable memory populations make up 16% to 40% of CD4 TCM and 96% to 46% of CD4 TEM , though with some uncertainty ( Figure 2—figure supplement 2 ) . Throughout this period approximately 10% of the remaining displaceable CD4 TCM subpopulation is replaced each week . For CD4 TEM , because the resistant population at 14 weeks of age is estimated to be a large proportion of the pool and the source is substantial , we predict that as much 65% of displaceable CD4 TEM are replaced per week . This rate falls to 1 . 5 %/week in year-old mice as the displaceable population grows and the rate of immigration falls in tandem with naive T cell numbers ( Figure 2D , right-hand panels ) . In summary , we find clear evidence for substantial tonic flows of cells from the naive T cell pool into both CD4 central and effector memory . For central memory we favour a model in which this flow remains high well into the second year of life , but displaces only a subset of cells . The remainder are generated before 8 weeks of age and analogous to the apparently stable ‘incumbent’ populations of naive CD4 and CD8 T cells that also resist replacement ( Figure 2B , right-hand panel; and Hogan et al . ( 2015 ) ) . We estimate that CD4 effector memory is replaced at a rate comparable to that of central memory in young adult mice , but that the rate of assimilation of new effector memory cells declines more strongly with age . This kinetic can be explained equally well by the existence of a resistant CD4 TEM subset or simply by a waning force of recruitment from the naive pool . Having identified and measured the contributions to CD4 memory subsets from naive sources , we wanted to measure cell lifetimes and division rates within these subsets in normal healthy mice and to test alternative models of homeostatic dynamics . Resolving different types of heterogeneity in these dynamics requires dissecting the fates of quiescent and dividing or recently-divided cells . Doing so is difficult with DNA labelling alone because for anything other than very short pulse-chase experiments the labelled fraction contains cells with a wide range of times since their last division . We therefore measured the division-linked uptake of the nucleoside analogue 5-bromo-2’-deoxyuridine ( BrdU ) in the context of Ki67 expression . Ki67 is a nuclear protein that is expressed during cell division but subsequently lost by non-dividing cells on a timescale of a few days ( Pitcher et al . , 2002; Younes et al . , 2011; De Boer and Perelson , 2013 ) . As such , it is a marker of active and recent division . The frequency of cells expressing Ki67 is expected to be constant in a population at steady state , but when combined with time courses of BrdU labelling , Ki67 acts as a timestamp allowing us to distinguish the fates of recently divided Ki67high BrdU+ cells and their quiescent Ki67low BrdU+ progeny ( Figure 3A ) . 10 . 7554/eLife . 23013 . 012Figure 3 . Quantifying the homeostatic dynamics of effector and memory CD4 T cells by combining BrdU labelling with measurements of Ki67 expression . ( A ) Representative data from flow cytometric analyses of BrdU uptake and Ki67 expression in a pulse-chase experiment . Cells were recovered from lymph nodes . ( B ) Outline of experimental design . ( C ) A schematic of the core multi-compartment model used to describe the flows between the BrdU×-/+ Ki67low/high populations during and after labelling . Shown here is a model of temporal heterogeneity , in which either effector or central memory CD4 T cells are modelled as a single population entering division stochastically at per capita rate α; with quiescent ( Ki67low ) and recently divided ( Ki67high ) cells dying at rates δ- and δ+ respectively; an external source of cells feeding the BrdU± Ki67high populations at rates S+ and S- , where S++S- is a constant , S; and cells transitioning from Ki67high to Ki67low at rate β . This basic model was refined to account for multiple subpopulations ( kinetic heterogeneity ) , different distributions of Ki67 expression times , inefficient BrdU uptake , and post-labelling dilution of BrdU within both labelled cells and within the source ( S+/- ( t ) ) . See Materials and methods and Appendix 1 for details of the model formulation . DOI: http://dx . doi . org/10 . 7554/eLife . 23013 . 012 We performed three pulse-chase experiments in which mice were fed BrdU for either 4 , 7 or 21 days , with a chase period of 8–14 days following withdrawal of the label from drinking water . Groups of mice were analysed for co-staining of Ki67 and BrdU at different times during these experiments to define the population dynamics of CD4 TEM and CD4 TCM in detail ( Figure 3B; for the experimental protocols see Materials and methods M1 ) . Over the relatively short time courses of the experiments we saw no substantial changes in either the absolute sizes of memory T cell subsets or in the fraction of cells expressing Ki67 ( Appendix 1—figure 1 ) . With these constraints the dynamics of the system can be characterised by two quantities – the proportions of cells within the Ki67high and Ki67low populations that are BrdU+ . In this equilibrium , loss ( turnover ) of memory is balanced by production of new cells by division and input from external sources . To assess the support for different homeostatic mechanisms , we used mathematical models to describe the fluxes of cells between the BrdU+/− × Ki67high/low populations within the CD4 TEM and CD4 TCM subsets ( Figure 3C and Materials and methods; detailed in Appendix 1 ) . In a model of pure temporal heterogeneity ( TH ) , each memory subset is assumed to comprise one population of cells undergoing single stochastic divisions characteristic of T cell homeostasis ( Yates et al . , 2008; Choo et al . , 2010; Hogan et al . , 2013 ) , but with potentially different rates of loss of quiescent ( Ki67low ) and recently divided ( Ki67high ) cells . We also considered a model of pure kinetic heterogeneity ( KH ) in which each memory subset is assumed to comprise two sub-populations maintained independently , each at constant size and with their own rates of division and loss , and with Ki67high and Ki67low cells within each subpopulation having equal susceptibility to death . In both TH and KH models , any external source is assumed to feed the Ki67high subpopulation ( s ) exclusively . We aimed to estimate the rate ( s ) of division and loss in both models , together with a minimal set of additional parameters representing key biological quantities . In the KH models these included the relative sizes of the two subpopulations and the allocation of the source into each . Both TH and KH models also required parameters quantifying the efficiency of BrdU uptake per cell division and the gradual decline of the BrdU+ fraction once BrdU feeding stops . The latter can result from three non-exclusive processes; ( i ) differences in the death rates of BrdU− and BrdU+ cells , ( ii ) dilution of the labelled population by unlabelled cells from the source ( Tough and Sprent , 1994; Bonhoeffer et al . , 2000; Debacq et al . , 2002; De Boer et al . , 2003 ) , and ( iii ) within-cell dilution of BrdU through division post-administration ( Tough and Sprent , 1994; Parretta et al . , 2008; Ganusov et al . , 2010 ) . The first process is captured in the basic KH/TH model structure . The second requires a description of the dilution of label within the source post-administration . We found the strongest support for a simple model in which the BrdU content of the souce drops rapidly from 100% to zero after a delay that is estimated from the data ( see Appendix 1 ) . For the third , we found that the best-fitting models required two divisions to drive cells from BrdU+→ → BrdU- , consistent with another BrdU labelling study in mice ( Parretta et al . , 2008 ) . Finally , we explored different distributions of times spent in the Ki67high state post-mitosis , by assuming cells progress through a variable number of intermediate states before transitioning to Ki67low . Best-fitting models for both KH and TH require more than 12 such states , meaning that there is very little variance in the time cells spend in the Ki67high state . We note that these kinetics and the estimated mean residence time in Ki67high ( 1/β ) reflect the Ki67 gating strategy as well as the cell-intrinsic rate of loss of Ki67 post-mitosis . Exploiting the constraints that there were no significant changes in the numbers and the proportions of cells that were Ki67high within both CD4 TEM and TCM during the labelling experiments ( see Appendix 1—figure 1 ) , four free parameters remained for the TH model and six for the KH model . A detailed description of the model formulation and the strategy for parameter estimation is given in Appendix 1 . Strikingly , despite this freedom in parameterisation , the data were sufficiently rich to discriminate between the models and showed unequivocal support for kinetic over pure temporal heterogeneity within the CD4 effector and central memory pools ( Figure 4 , ΔAIC = 110 ( CD4 TEM ) , 251 ( CD4 TCM ) ) . For both memory subsets the BrdU/Ki67 timecourses were consistent with the existence of two subpopulations roughly equal in size but with highly distinct kinetics ( Table 2 and Figure 5A , at dashed vertical lines ) . CD4 TCM appear to comprise a population dividing and dying roughly every 3 days , and a slower population with mean lifetime of 38 days , dividing every 170 days , with the source feeding the slow and fast populations in roughly a 2:1 ratio . For CD4 TEM the fast population appears to be essentially self-renewing , dividing and dying every 6 days , with the slower population fed by the source ( mean lifetime 43 days , interdivision time 140 days ) . 10 . 7554/eLife . 23013 . 013Figure 4 . BrdU/Ki67 dynamics in memory CD4 T cell subsets are best described by a model of kinetically distinct subpopulations . Data and best fit predictions for two classes of model describing BrdU uptake and loss – kinetic heterogeneity ( left panels ) and temporal heterogeneity ( right panels ) – for CD4 TEM ( upper panels ) and CD4 TCM ( lower panels ) . Fits were generated using the best-fit estimates of the influx into each population ( for CD4 TEM , 7 . 0% of the pool size per week at 14 weeks of age; for CD4 TCM , 10 . 6% per week; these figures are 7 × the daily influx quoted in Table 1 ) . Colours denote different BrdU feeding timecourses and shaded regions represent 95% confidence envelopes on the fits , calculated by resampling the parameters from their bootstrap distributions . The inability of the TH model to describe both the timecourses well stems from the tight coupling between the BrdU+Ki67low cells and their BrdU+Ki67high precursors , with little freedom to fit the timecourses of both simultaneously; whereas in the KH model , those two populations are enriched for the slow and fast subpopulations respectively , which are parameterised independently . DOI: http://dx . doi . org/10 . 7554/eLife . 23013 . 01310 . 7554/eLife . 23013 . 014Figure 4—source data 1 . Timecourses of the BrdU+ fractions within the Ki67high and Ki67low populations during the BrdU labelling/delabelling experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 23013 . 01410 . 7554/eLife . 23013 . 015Figure 4—source data 2 . Source code used to generate and fit models to BrdU/Ki67 timecourses . DOI: http://dx . doi . org/10 . 7554/eLife . 23013 . 01510 . 7554/eLife . 23013 . 016Figure 5 . Quantifying CD4 TEM and CD4 TCM homeostasis assuming kinetic heterogeneity . ( A ) Key kinetic parameters for CD4 TEM and TCM estimated for different levels of memory influx . Grey points represent population average parameters; for interdivision times these are offset for clarity . Vertical dashed lines and shaded areas represent the best estimates of influx with 95% confidence intervals . These estimates are the weekly influxes as a fraction of the pool size ( i . e . , 7 × the daily influxes quoted in Table 1; 0 . 07 of pool/week for CD4 TEM , 0 . 11 for TCM ) . ( B ) Estimated mean duration of Ki67 expression post-mitosis . ( C ) Estimated proportions of cells that are Ki67high within fast and slow subpopulations . The weighted averages of these proportions for each of CD4 TEM and CD4 TCM were constrained to be the observed level of expression ( mean + s . e . m . ) averaged over the course of the BrdU labelling experiments ( Appendix 1—figure 1 ) . ( D ) Stratifying Ki67high expression within CD4 TEM and CD4 TCM by host and donor , in six busulfan chimeras that were 8 weeks post-BMT and of comparable ages to the mice used in the BrdU labelling experiments , indicating that fast/slow cells cannot be exclusively identified as donor/host-derived . DOI: http://dx . doi . org/10 . 7554/eLife . 23013 . 01610 . 7554/eLife . 23013 . 017Figure 5—source data 1 . Ki67 expression in host and donor CD4 TEM and TCM cells in busulfan chimeras 8 weeks post-BMT ( panel D ) . DOI: http://dx . doi . org/10 . 7554/eLife . 23013 . 01710 . 7554/eLife . 23013 . 018Figure 5—figure supplement 1 . Comparing mean lifetimes and interdivision times obtained with the KH model when adding temporal heterogeneity . We fitted extensions of the basic KH model in which death rates of Ki67high cells were set to be 1/10 or 10 times that of Ki67low cells . Orange/blue denote the fast/slow subpopulations respectively . Mean lifetimes for each population ( EM/CM , fast/slow ) are weighted averages of the lifetimes of the Ki67high and Ki67low subsets . Vertical bars indicate 95% confidence intervals calculated by bootstrapping residuals and resampling from the bootstrap estimates of the magnitude of the source . DOI: http://dx . doi . org/10 . 7554/eLife . 23013 . 01810 . 7554/eLife . 23013 . 019Table 2 . Parameters describing homeostasis of murine CD4 memory subsets , using a two-population model of kinetic heterogeneity and the best estimates of the magnitudes of the influx into each subset from the naive pool . Pool-average lifetimes and interdivision times are defined to be the mean of the corresponding quantities for the fast and slow subpopulations weighted by their size estimates . The best-fitting models were those that assumed no difference in death rates of Ki67high and Ki67low cells ( indicated by rows in which ratio of loss rates = 1 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 23013 . 019ParameterRatio of loss ratesCD4 effector memoryCD4 central memoryKi67high:Ki67lowEstimate95% CIEstimate95% CIPool-average lifetime ( days ) 1 10 0 . 129 41 55 ( 28 , 30 ) ( 39 , 42 ) ( 54 , 56 ) 21 27 44 ( 19 , 24 ) ( 26 , 35 ) ( 39 , 51 ) Pool-average interdivision time ( days ) 1 10 0 . 188 63 84 ( 83 , 158 ) ( 53 , 353 ) ( 80 , 108 ) 86 59 51 ( 47 , 144 ) ( 35 , 74 ) ( 38 , 84 ) Ki67 lifetime ( days ) 1 10 0 . 13 . 28 3 . 23 3 . 28 ( 3 . 14 , 3 . 39 ) ( 3 . 03 , 3 . 30 ) ( 3 . 18 , 3 . 41 ) 3 . 59 3 . 74 3 . 32 ( 3 . 47 , 3 . 70 ) ( 3 . 57 , 3 . 82 ) ( 3 . 31 , 3 . 52 ) Efficiency of BrdU uptake ( % ) 1 10 0 . 176 75 78 ( 74 , 79 ) ( 73 , 78 ) ( 76 , 81 ) 77 78 78 ( 76 , 79 ) ( 76 , 79 ) ( 78 , 81 ) Delay before source switches to BrdU− post-labelling ( days ) 1 10 0 . 12 . 5 2 . 6 2 . 1 ( 1 . 7 , 3 . 0 ) ( 1 . 9 , 3 . 5 ) ( 1 . 3 , 2 . 3 ) 0 . 085 0 . 006 0 . 14 ( 0 . 002 , 1 . 16 ) ( 0 . 003 , 1 . 43 ) ( 0 . 003 , 0 . 66 ) Source contribution to peripheral production ( fraction ) 1 10 0 . 10 . 12 0 . 11 0 . 15 ( 0 . 12 , 0 . 13 ) ( 0 . 099 , 0 . 11 ) ( 0 . 15 , 0 . 16 ) 0 . 092 0 . 085 0 . 12 ( 0 . 088 , 0 . 096 ) ( 0 . 08 , 0 . 088 ) ( 0 . 11 , 0 . 12 ) Fraction of source enteringslow subpopulation1 10 0 . 11 1 1 ( 0 . 98 , 1 ) ( 0 . 95 , 1 ) ( 0 . 98 , 1 ) 0 . 69 1 0 . 46 ( 0 . 38 , 0 . 85 ) ( 0 . 45 , 1 ) ( 0 . 19 , 0 . 62 ) Mean lifetime offast subpopulation ( days ) 1 10 0 . 15 . 7 17 18 ( 5 . 5 , 6 . 8 ) ( 15 , 20 ) ( 18 , 21 ) 3 . 3 7 . 1 6 . 4 ( 3 . 1 , 3 . 4 ) ( 6 . 2 , 7 . 5 ) ( 6 . 4 , 6 . 9 ) Mean lifetime ofslow subpopulation ( days ) 1 10 0 . 143 61 75 ( 43 , 48 ) ( 60 , 65 ) ( 74 , 77 ) 38 48 76 ( 34 , 44 ) ( 47 , 65 ) ( 68 , 90 ) Mean interdivision time offast subpopulation ( days ) 1 10 0 . 15 . 7 5 . 7 6 . 8 ( 5 . 5 , 6 . 9 ) ( 5 . 1 , 6 . 7 ) ( 6 . 8 , 7 . 6 ) 3 . 4 3 . 3 4 . 2 ( 3 . 2 , 3 . 5 ) ( 3 . 0 , 3 . 4 ) ( 4 . 2 , 4 . 4 ) Mean interdivision time ofslow subpopulation ( days ) 1 10 0 . 1138 113 125 ( 130 , 275 ) ( 90 , 750 ) ( 119 , 169 ) 167 118 92 ( 89 , 280 ) ( 72 , 151 ) ( 69 , 151 ) Size of fast subpopulation ( fraction of total ) 1 10 0 . 10 . 38 0 . 46 0 . 35 ( 0 . 37 , 0 . 44 ) ( 0 . 43 , 0 . 54 ) ( 0 . 34 , 0 . 38 ) 0 . 49 0 . 52 0 . 47 ( 0 . 47 , 0 . 51 ) ( 0 . 49 , 0 . 53 ) ( 0 . 45 , 0 . 48 ) Given the notorious dependence of estimates of lymphocyte division and death rates on model assumptions ( De Boer and Perelson , 2013 ) , we explored the sensitivity of our estimates and predictions to the magnitude of the source . We performed fits to the BrdU/Ki67 timecourses for multiple values of the influx spanning values of zero to 30% of the pool size per week , which included the point estimates and their 95% confidence intervals ( Figure 5A ) . Despite the two additional parameters required to describe the source ( i . e . its partitioning between the fast and slow subpopulations , and the timing of the switch to unlabelled source after withdrawal of BrdU; see Appendix ) , including a source gave superior descriptions of both CD4 TEM and TCM labelling kinetics compared to models of self-renewing memory with no influx ( ΔAIC = 9 and 20 , respectively , at the best estimates of the source ) . The flows into memory impact measures of homeostatic dynamics significantly; if the contribution of the source is neglected , pool-averaged cell lifetimes may be overestimated by 25–50% and , more strikingly , interdivision times underestimated by a factor of 2–3 ( Figure 5A ) . At our best estimates of the influx into CD4 TEM and TCM from the naive pool , we infer that it predominantly feeds the slower subpopulations within each . Since we presume that memory is largely generated from naive cells through antigen-driven proliferation , this result was puzzling . A more restricted model in which the source was constrained to feed exclusively into the faster subpopulation had substantially lower statistical support ( ΔAIC = 18 for CD4 TEM , 9 . 9 for TCM ) , but visually the fits were similar ( Appendix 1—figure 2; parameter estimates in Appendix 1—table 3 ) . Further , if CD4 TEM derive from TCM to any extent , we predict higher levels of influx ( Table 1 ) and the proportion predicted to enter the slow population is then lower ( Figure 5A , upper panels ) . We therefore remain cautious regarding the extents to which the constitutive influx feeds low and fast-dividing memory subsets . Irrespective , at all values of the source we explored , all variants of the KH model were far better descriptions of the kinetics than the TH model ( ΔAIC > 90 for both CD4 TEM and TCM ) . Despite the richness of the BrdU/Ki67 timecourse , given the number of unknowns to be estimated it was not possible to fit a single model containing both forms of heterogeneity . However to look for a signature of temporal heterogeneity we explored variants of the KH model in which the loss rates of Ki67high cells were forced to be either a tenth or ten times that of the Ki67low cells in each subpopulation . For CD4 TEM neither extension improved on the basic KH model’s description of the data ( ΔAIC>4 ) . For CD4 TCM we found almost equal support for a model in which Ki67high cells died 10 times faster than Ki67low cells and for the basic KH model in which death rates are independent of the level of Ki67 expression . This additional level of temporal heterogeneity increased the estimated mean lifetimes of both fast and slow CD4 TCM subsets but had little impact on estimates of interdivision times ( Table 2 and Figure 5—figure supplement 1 ) . We conclude that our data do not provide evidence for substantial differences in the susceptibility to death of recently-divided and quiescent memory CD4 T cells . We estimated the mean post-mitotic duration of Ki67 expression to be between 3 . 1 and 3 . 7 days , consistent with estimates elsewhere ( Pitcher et al . , 2002; Younes et al . , 2011; De Boer and Perelson , 2013 ) . This estimate was model-independent , insensitive to the magnitude of the influx into memory , and closely comparable for CD4 TEM and TCM ( Figure 5B ) . The KH model predicted that the fast subpopulations express Ki67 at levels of approximately 65% ( CD4 TEM ) and 85% ( TCM ) , while the slow populations in both are less than 10% Ki67high ( Figure 5C ) . Ki67high CD4 memory cells are therefore predicted to be highly enriched for the fast dividing subset . Notably , the relative sizes of the fast and slow subsets were similar to the split of donor/host cells observed in memory in the busulfan chimeras ( Figure 2C ) . It was then tempting to speculate that the slowly-dividing cells were the apparently resistant and stable populations of host-derived CD4 TEM and TCM cells in the busulfan chimeras , while the more rapidly dividing cells represent the influx of donor cells into memory subsets . To test this , we measured Ki67 expression in busulfan chimeras 8 weeks after BMT , when hosts were a comparable age to those used in the BrdU feeding experiments . Although donor cells were indeed slightly enriched for Ki67high cells relative to host cells , both subpopulations exhibited substantial levels of Ki67 expression ( Figure 5D ) and did not map to the fast/slow populations inferred from the BrdU labelling analysis . These data therefore strongly suggest that both newly-recruited donor and more established host memory CD4 T cells are kinetically heterogeneous . In summary , we find strong evidence for rapidly- and slowly-dividing populations within both effector and central CD4 memory T cells in uninfected adult mice . We find no strong evidence for recent division impacting susceptibility to cell death . Accounting for the constitutive flows of cells into both memory subsets significantly improves the description of BrdU labelling kinetics , and strongly impacts estimates of rates of memory T cell division and turnover .
To date , our understanding of how CD4 memory is structured and maintained has been limited by uncertainty in the interpretation of labelling data and lack of knowledge of the extent to which newly differentiated cells contribute to renewal . No single method has been able to successfully distinguish between and quantify these different processes . Here , we were able to both measure tonic influx into memory , and dissect memory compartment dynamics in detail by distinguishing turnover in quiescent and recently divided cells . Temporal fate mapping in the busulfan chimeras revealed the surprisingly high rate of de novo generation of memory/effector cells in the CD4 memory compartments from naive cells , with at least 6–10% of cells replaced each week in 14 week old mice . Even at the lower bounds , the extent of this new memory generation from naive sources was surprising given that the hosts were in a clean , regulated environment and not deliberately infected . A recent study of feral mice and those in dirty environments revealed the expansion of CD8 TEM compartments resulting from the increased antigenic load ( Beura et al . , 2016 ) . The authors concluded that expansion was driven by episodic exposure and not constitutive stimuli , as the activation and proliferative status of immune cells were similar to those in cleaner laboratory mice . However , our data strongly indicate the existence of tonic drivers of generation of new memory cells . An obvious mechanism is the continued recruitment of recent thymic emigrants into responses against commensal or environmental antigens . The fact that memory compartments remain remarkably stable in size in the face of this chronic stimulus suggests that these responding cells are regulated differently to those generated in an active infection , perhaps due to the absence of overt inflammatory stimulus . Whether inflammatory stimuli modulate these responses will be the subject of future study . We also explored the differentiation pathways underlying the flow of cells from naive to different memory compartments over timescales of weeks to months . Previous studies suggest that regulation of CD62L expression by activated CD4 T cells is both heterogenous and slow , compared with CD8 cells ( Bjorkdahl et al . , 2003; Chao et al . , 1997 ) . Loss of CD62L expression is largely irreversible in CD4 TEM ( Kassiotis and Stockinger , 2004; Bingaman et al . , 2005 ) , suggesting that the CD62L-expressing CD4 TCM derive directly from activation of naive T cells and not from TEM . Consistent with this we clearly observed more rapid and slightly greater replacement of the CD4 TCM than CD4 TEM compartment in busulfan chimeras . For CD4 TEM the situation is less clear , but CD8 TEM may be generated both directly from activation of naive T cells or by subsequent differentiation of CD8 TCM ( Restifo and Gattinoni , 2013 ) . Due to the risk of overfitting it was not possible to quantify the contributions of each of these pathway to CD4 TEM at steady state , and so we considered only the extreme alternatives in which CD4 TEM are sourced entirely from naive or entirely from CD4 TCM . The TCM → TEM model was statistically inferior but gave visually similar fits ( Figure 2C ) , and predicted much higher rates of CD4 TEM replacement ( ∼23%/week , compared to ∼6% for a naive source; Table 1 ) . As it seems likely that the CD4 TEM population is fed by both naive and TCM cells to some extents , we conclude that our estimate of 6% is a lower bound and it is possible that nearly a quarter of CD4 TEM are replaced each week under healthy conditions at 14 weeks of age . Our estimated average lifetimes of lymph-node-derived memory CD4 T cells ( 21d for TCM and 29d for TEM ) are slightly higher than those made previously . Other studies of total CD44hi CD4 T cells using BrdU or deuterated water labelling found kinetics consistent with mean lifetimes of 14-22d ( De Boer and Perelson , 2013; Westera et al . , 2013 ) . One of these studies found that a simple model of two self-renewing , stable populations described the labelling kinetics better than the simplest single-compartment model with no temporal heterogeneity , and that the fast and slow CD4 memory populations were comparable in size ( Westera et al . , 2013 ) . Both studies assumed that memory is self-renewing and constant in size , so mean lifetimes are necessarily equal to average interdivision times . If we make a similar assumption and neglect the memory sources , our estimates of lifetimes increase ( 40d and 30d for CD4 TEM and TCM respectively ) . Thus , studies assuming memory is a self-renewing compartment will tend to overestimate lifetimes and underestimate interdivision times . This issue again highlights the sensitivity of measures of population dynamics to the biology encoded in the model . Our analysis suggests that the rate of recruitment into memory from the naive pool varies with age , and given the relative stability of memory population sizes it is therefore likely that memory turnover is also not constant over the life course . It is possible that the discrepancies between our and other estimates may derive from differences in host age or in commensal colonisation arising in the different housing facilities , both of which may impact the rate of tonic recruitment into memory , the relative sizes of fast and slowly dividing subpopulations , and hence estimates of cell lifetimes and division rates . Our experimental analyses revealed heterogeneous behaviour amongst memory CD4 T cells at multiple levels . Heterogeneity within the CD4 memory compartment as a whole has been recognised for some time , with a distinction drawn between slow-dividing cells driven by cytokines ( Seddon et al . , 2003; Purton et al . , 2007 ) and in which memory to defined antigens is thought to reside , and fast-dividing cells ( Tough and Sprent , 1994; Robertson et al . , 2006; Purton et al . , 2007; Surh and Sprent , 2008; Younes et al . , 2011 ) whose proliferation is dependent on TCR signalling ( Min et al . , 2003; Seddon et al . , 2003; Leignadier et al . , 2008; Younes et al . , 2011 ) . We infer that Ki67 expression levels in the slower-dividing populations are approximately 9–11% ( Figure 5C ) , consistent with direct observations of antigen-specific CD4 memory ( Lenz et al . , 2004; Purton et al . , 2007; Pepper et al . , 2010; Younes et al . , 2011 ) . However , the relation of these kinetically distinct populations to the canonical CD4 TEM and TCM subsets delineated by CD62L expression has been unclear . Here , we find evidence that CD4 TEM and TCM subsets comprise both fast and slow subpopulations , suggesting that they are indeed similar in their homeostatic dynamics and structure . The lineage relationships between the respective fast and slow subpopulations of CD4 TEM and TCM therefore need further investigation . In the absence of infection , it is intuitive that the stimulus driving the continuous recruitment into memory derives from environmental antigens in food and commensal organisms . It also seems intuitive that this stimulus should continue to drive fast-dividing memory subpopulations throughout life . Certainly , we find evidence of these within both new ( donor-derived ) and more established ( host-derived ) memory populations ( Figure 5D ) . However if environmental antigens are the stimulus for tonic recruitment , then it is surprising that newly generated memory cells are not exclusively fast-dividing . These observations could be explained if exposure to environmental antigens is subject to natural fluctuations in load , resulting in episodic but frequent stimuli to divide rather than continuous rounds of division . Indeed bursts of TCR-driven proliferation may be involved in the maintenance of CD4 memory to persistent phagosomal infections ( Nelson et al . , 2013 ) . Such a view would be consistent with the estimates of interdivision times for the fast subpopulations , which are still much longer than the interdivision times of several hours that result from cognate antigen challenge . Episodic fast divisions within both CD4 TEM and TCM could also account for the arguably counter-intuitive model prediction that the source predominantly feeds the slowly-dividing subpopulations . We also note that the fits yielded by the model of discrete fast and slow populations are good , but not perfect , and it seems likely that there is a richer kinetic substructure ( Ganusov et al . , 2010 ) . Testing this hypothesis that the composition of fast and slow populations is dynamic , establishing how the influx into memory is routed to these subpopulations , and identifying the lineage relationships between CD4 TEM and TCM at steady state , will require new approaches . What is clear , however , is that whether subdivided by surface phenotype or age structure , kinetically distinct subpopulations are consistently demonstrable within memory CD4 T cells . Our analysis also provides new insight into the interpretation of Ki67 expression , which is commonly used as a proxy for levels of cell proliferation . There has been a growing awareness that while Ki67 is induced at onset of cell cycle , expression persists following completion of mitosis ( Pitcher et al . , 2002; Younes et al . , 2011; Hogan et al . , 2013; De Boer and Perelson , 2013 ) . Here we explicitly model its expression and estimate that cells take approximately 3 . 5 days to become Ki67low . This figure depends in part on the flow cytometry gating strategy and so is a functional rather than a biochemical measure . However the modelling indicated that the residence time in Ki67high has a low coefficient of variation and so we infer that the post-mitotic loss of Ki67 is essentially deterministic , with very little cell-cell variation . Since cell division only takes between 2–8 hr ( Bruno and Darzynkiewicz , 1992; Hogan et al . , 2013 ) , Ki67 is therefore chiefly a post-mitotic marker . Its extended expression makes it a sensitive measure for detecting cell division occurring at low absolute frequencies . Knowledge of its lifetime is also useful for isolating cell populations . The fast CD4 TCM and TEM subpopulations divide approximately every 3 and 6 days respectively , while their slower counterparts divide only every 140 days or more . Therefore a substantial fraction of the fast subsets ( greater than half for CD4 TCM ) will begin to divide again before losing Ki67 expression . Ki67-bright CD4 memory cells are therefore highly enriched for the fast dividing subsets , and the BrdU+ Ki67low subset is increasingly rich in slowly-dividing cells . These properties can be used as basis for further functional characterisation of these subpopulations . Taken together , our data reveal complexity in the regulation of memory compartments , in which the substantial and tonic de novo generation of memory cells braids into highly dynamic and heterogeneous subpopulations which themselves exhibit an unexpectedly diverse age structure . Despite this complexity in cell dynamics , the compartment sizes are remarkably stable throughout life , indicating tight homeostatic control . Key questions for the future are whether tonic influxes contribute to the erosion of antigen-specific CD4 T cell memory over time ( Homann et al . , 2001 ) , whether the tonic recruitment and turnover of memory cells are modulated during the course of an inflammatory , infectious episode , and whether this backdrop of memory cell activity in any way influences T cell activation and development that occurs during such challenges .
We used a simple framework to describe the kinetics of constitutive renewal of the effector and central memory CD4 compartments . Assume that cells flow into a memory subset M ( t ) at per capita rate γ from a precursor population S ( t ) , and that in the absence of this source , memory is lost to death and/or differentiation at net rate λ . We place no constraints on the growth or decay of memory so λ may be positive or negative . Then if host and donor cells follow identical kinetics , ( 1 ) dMhostdt=γShost ( t ) −λMhost ( t ) ( 2 ) dMdonordt=γSdonor ( t ) −λMdonor ( t ) The rate γ is the product of the per capita rate of egress of cells from the precursor population S and the net effect of any expansion and/or contraction that takes place during the transition into memory . We define the memory chimerism to be the fraction of cells that are donor-derived , χM=MdonorMdonor+Mhost which differs among age-matched animals due to variation in the degree of HSC depletion with busulfan treatment . We normalise the memory chimerism to that in the thymic precursor population DP1 , χDP1 , which is stable by approximately 6 weeks post-BMT . Equations 1 and 2 can then be recast in forms that do not depend on the degree of HSC depletion and so are applicable across mice in the experimental cohort: ( 3 ) dMdt=γS ( t ) −λM ( t ) ( 4 ) dρMdt=γS ( t ) M ( t ) ( ρS ( t ) −ρM ( t ) ) where M=Mhost+Mdonor , and ρM=χM/χDP1 and ρS=χS/χDP1 are the normalised donor chimerism in the memory and source populations respectively . To account for the apparent capping of chimerism in the memory subsets , this model can be extended to allow the per-cell rate of recruitment into memory to vary with the age of the animal , γ ( t ) , and/or a population of host-derived memory cells , Minc , that resists displacement by newer cells . Combining these extensions yields ( 5 ) dMdt=γ ( t ) S ( t ) −λ ( M ( t ) −Minc ) ( 6 ) dρMdt=γ ( t ) S ( t ) M ( t ) ( ρS ( t ) −ρM ( t ) ) −λMincM ( t ) ρM ( t ) where now M=Mhost+Mdonor+Minc . Note that in the text we work with two sub-models – one in which there is a resistant population Minc but the per capita rate of recruitment from the source γ=γ0 is a constant; and another in which all memory is displaceable ( Minc=0 ) but that the per capita rate of recruitment from the source wanes with age , γ ( t ) =γ0exp ( -ϕt ) . Below we show the predictions of the most general model that combines both elements . We can make a conservative estimate of the flows into memory by assuming that both the effector and central memory pools are fed directly from the naive pool . For each , we fitted the model of the dynamics of total cell numbers and the normalised chimerism within the naive CD4 T cell pool ( Equations 5 and 6 ) to the observations , using empirical functions describing changes in the size and chimerism of the source population with time . The following functional forms described both putative source populations ( naive and CD4 TCM ) well; Source population size ( 7 ) S ( t ) =S0e−Rt Source chimerism ( 8 ) ρS ( t ) =ρmax1+e−rt ( ρmax−ρ0 ) /ρ0 Estimates of the parameters governing these functions are given in Appendix 1—table 1 . To fit the resistant memory and declining recruitment models , we maximised the product of the log likelihoods of the two timecourses , as described in the Supporting Information of Hogan et al . ( 2015 ) , using a trust region method implemented in Python ( Figure 2—source data 2 ) . We estimated the net rate of loss/growth of memory , λ and the initial rate of recruitment from the source , γ0; and either ( i ) Minc for the model of resistant memory with constant per capita rate of recruitment from the source; or ( ii ) ϕ for the model in which the per capita rate of recruitment from the source itself falls exponentially at rate ϕ . Confidence intervals on these parameters were generated by simultaneously bootstrapping residuals 3000 times and resampling from the bootstrap estimates of the parameters governing the sources , S ( t ) and ρS ( t ) . One quantity of interest is γ ( t ) S ( t ) /M ( t ) , which is the number of new cells entering memory per unit time as a fraction of the memory pool size . Another is the fraction of memory cells replaced through immigration . Because we assumed that recently recruited memory cells are as susceptible to loss as older displaceable memory cells , the number of memory cells expected to be replaced through immigration during a time ( 0 , t ) is slightly less than the total influx in the same period . Consider a memory population M ( 0 ) that comprises a displaceable population X ( 0 ) and an incumbent self-renewing population Minc , M ( 0 ) =X ( 0 ) +Minc . We defined the fractional replacement to be the proportion of the memory pool comprised of immigrants after a time t; ( 9 ) freplace ( t ) =Y ( t ) X ( t ) +Y ( t ) +Minc where X ( t ) =X ( 0 ) e-λt is the number of displaceable cells present at t=0 which survived to time t; and Y ( t ) is the number of cells that entered memory during ( 0 , t ) and survived to t , which is the solution to dY/dt=γ ( t ) S ( t ) -λY ( t ) given Y ( 0 ) =0 . If we assume γ ( t ) =γ0exp ( -ϕt ) and the empirical form for the source S ( t ) =S0exp ( -Rt ) , this yields ( 10 ) freplace ( t ) =γ0S0 ( eψt−1 ) eψt ( ψ ( M ( 0 ) −Minc ) +γ0S0 ) +ψMince ( ψ+λ ) t−γ0S0 , where we define ψ=R+ϕ-λ . In the text we quote both the daily influx γS/M and the expected weekly fractional replacement , freplace ( t=7 days ) , both at 14 weeks of age , which is 6 weeks post-BMT in these animals ( Table 1 ) . We also quote replacement as a fraction of the displaceable subpopulation only: ( 11 ) freplacedisplaceable ( t ) =Y ( t ) X ( t ) +Y ( t ) =γ0S0 ( eψt-1 ) eψt ( ψ ( M0-Minc ) +γ0S0 ) -γ0S0 . Figure 2D shows how the weekly fractional replacement of both total and displaceable memory is predicted to change with age . The estimates of daily influx are robust to the details of the model of the capping of memory chimerism . This is because information regarding this parameter is largely contained in the gradient of the donor chimerism in memory dρM/dt early in reconstitution , which is well-defined in the data ( Figure 2C , lower panels; animals of age c . 100 days ) . When the chimerism ρM and/or the rate of change of memory cell numbers in the absence of influx ( λ ) are low , then in either model ( 12 ) γ ( t ) S ( t ) M ( t ) ≃dρM/dtρS ( t ) -ρM ( t ) =dχM/dtχS-χM . This expression yields estimates of influx of 2 . 5% of the pool size per day for CD4 TCM and 1 . 2% per day for CD4 TEM , assuming both are sourced by naive cells . Both of these estimates lie within the 95% confidence intervals calculated using the best-fitting models ( Table 1 ) . Equation 12 also explains why the predicted rates of CD4 TEM replacement are much higher if one assumes that the source is TCM rather than naive; this rate scales inversely with the difference in chimerism between the source and target populations , which is lower for a TCM → TEM pathway than for naive → TEM . The model is illustrated in Figure 3C , indicating flows between the BrdU×-/+ Ki67low/high populations following first order kinetics with fixed rate constants . To describe the data we made the following extensions to this basic structure: With these extensions , the model parameterises the following processes: α = rate of entry into division resulting in expression of Ki67 , δ− = rate of loss ( death or differentiation ) for Ki67low cells , δ+ = rate of loss ( death or differentiation ) for Ki67high cells , ϵ = probability of incorporating BrdU per division during label administration , S± = rate of entry of cells into the K+B± compartments from the source , k^ = number of Ki67high intermediate compartments , 1/β = mean duration of Ki67 expression ( i . e . , mean total residence time in Ki67high compartments ) , b^ = number of divisions required for a BrdU+ cell to become BrdU− in the absence of label , τ = time for source to become BrdU− following withdrawal of BrdU . For the basic KH model we set δ+=δ-=δ for each subpopulation . We also examined variants of KH in which the death rate of Ki67high cells in each subpopulation was forced be either 1/10 or 10 times the death rate of of Ki67low cells . Detailed descriptions of the representation of this model as ordinary differential equations and the procedure for parameter estimation are given in Appendix 1 .
The analyses were supported in part through the computational resources and staff expertise provided by Scientific Computing at the Icahn School of Medicine at Mount Sinai . | The immune system protects the body from the infectious bacteria , viruses and other microorganisms present in our everyday environment ( collectively known as pathogens ) . One feature of this system is that it can form long-lasting memories of the pathogens it has previously encountered by creating cells called memory cells . When the same pathogen invades the body again , the memory cells help the immune system to kill off the infection more rapidly and efficiently than before . This process also underlies how vaccines work . By exposing the immune system to a pathogen in a controlled , safe way , memory cells form that can efficiently fight off a future infection . Do immune memories only form when we are sick with infections ? Or does constant exposure to the microbes that are present in the natural environment also stimulate the formation of memory cells ? Also , how does the formation of new memory cells affect the existing memory cells ? To answer these questions , Gossel , Hogan et al . studied laboratory mice that were kept in a clean , controlled environment – and not exposed to pathogens – for a year . This timespan represents about half of a mouse’s normal lifespan . Over the course of the year , new immune memory cells constantly formed in the mice . Furthermore , in young healthy mice up to a tenth of the existing immune memory cells were replaced each week . Despite the constant formation of new memory cells , the overall number of immune memory cells in the mice only doubled over the course of the year , suggesting that some memory cells must also be lost . The discovery that new immune memory cells are constantly made raises new questions to be investigated in future studies . For example , does the constant formation of memory cells make it harder to retain useful memories of pathogens , and does this explain the need for booster vaccinations ? | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Methods",
"and",
"materials",
"Acknowledgments"
] | [
"computational",
"and",
"systems",
"biology",
"immunology",
"and",
"inflammation"
] | 2017 | Memory CD4 T cell subsets are kinetically heterogeneous and replenished from naive T cells at high levels |
Class II HLH proteins heterodimerize with class I HLH/E proteins to regulate transcription . Here , we show that E proteins sharpen neurogenesis by adjusting the neurogenic strength of the distinct proneural proteins . We find that inhibiting BMP signaling or its target ID2 in the chick embryo spinal cord , impairs the neuronal production from progenitors expressing ATOH1/ASCL1 , but less severely that from progenitors expressing NEUROG1/2/PTF1a . We show this context-dependent response to result from the differential modulation of proneural proteins’ activity by E proteins . E proteins synergize with proneural proteins when acting on CAGSTG motifs , thereby facilitating the activity of ASCL1/ATOH1 which preferentially bind to such motifs . Conversely , E proteins restrict the neurogenic strength of NEUROG1/2 by directly inhibiting their preferential binding to CADATG motifs . Since we find this mechanism to be conserved in corticogenesis , we propose this differential co-operation of E proteins with proneural proteins as a novel though general feature of their mechanism of action .
The correct functioning of the vertebrate central nervous system ( CNS ) relies on the activity of a large variety of neurons that can be distinguished by their morphologies , physiological characteristics and anatomical locations ( Zeng and Sanes , 2017 ) . Such heterogeneity is generated during the phase of neurogenesis , once neural progenitors have been regionally specified and are instructed to exit the cell cycle and differentiate into discrete neuronal subtypes ( Guillemot , 2007 ) . Neuronal differentiation and subtype specification are brought together by a small group of transcription factors ( TFs ) encoded by homologues of the Drosophila gene families Atonal , Achaete-Scute , Neurogenins/dTap and p48/Ptf1a/Fer2 ( Bertrand et al . , 2002; Huang et al . , 2014 ) . These TFs represent a subgroup of the class II of helix-loop-helix proteins and all share a typical basic helix-loop-helix ( bHLH ) structural motif , where the basic domain mediates direct DNA binding to CANNTG sequences ( known as E-boxes ) and the HLH region is responsible for dimerization and protein-protein interactions ( Massari and Murre , 2000; Bertrand et al . , 2002 ) . They are generally expressed in mutually exclusive populations of neural progenitors along the rostral-caudal and dorsal-ventral axes ( Gowan et al . , 2001; Lai et al . , 2016 ) . They are typically referred to as proneural proteins , since they are both necessary and sufficient to switch on the genetic programs that drive pan-neuronal differentiation and neuronal subtype specification during development ( Guillemot , 2007 ) . This unique characteristic is also illustrated by their ability to reprogram distinct neural and non-neural cell types into functional neurons ( Masserdotti et al . , 2016 ) . Regulating the activity of these proneural proteins is crucial to ensure the production of appropriate numbers of neurons without prematurely depleting the pools of neural progenitors . In cycling neural progenitors , the transcriptional repressors HES1 and HES5 act in response to Notch signalling to maintain proneural TF transcripts oscillating at low levels ( Imayoshi and Kageyama , 2014 ) . The proneural proteins are also regulated at the post-translational level . Ubiquitination and phosphorylation have been reported to control their stability , modify their DNA binding capacity or even terminate their transcriptional activity ( Ali et al . , 2011; Li et al . , 2012; Ali et al . , 2014; Quan et al . , 2016 ) . Furthermore , the activity of these proneural proteins is highly dependent on protein–protein interactions , and particularly on their dimerization status . It is generally admitted that these TFs must form heterodimers with the more broadly expressed class I HLH/E proteins to produce their transcriptional activity ( Wang and Baker , 2015 ) . In this way , the activity of proneural proteins can be controlled by upstream signals that regulate the relative availability of E proteins . Members of the Inhibitor of DNA binding ( ID ) family represent such regulators . As they lack the basic domain required for direct DNA-binding , ID proteins sequester E proteins through a physical interaction and thereby produce a dominant-negative effect on proneural proteins ( Massari and Murre , 2000; Wang and Baker , 2015 ) . Hence , several sophisticated regulatory mechanisms are available during development to control proneural protein activity and fine-tune neurogenesis . Bone morphogenetic proteins ( BMPs ) contribute to multiple processes during the formation of the vertebrate CNS ( Liu and Niswander , 2005; Le Dréau and Martí , 2013 ) . Yet it is only in the past few years that their specific role in controlling vertebrate neurogenesis has begun to be defined ( Le Dréau et al . , 2012; Segklia et al . , 2012; Choe et al . , 2013; Le Dréau et al . , 2014 ) . During spinal cord development , SMAD1 and SMAD5 , two canonical TFs of the BMP pathway ( Massagué et al . , 2005 ) , dictate the mode of division that spinal progenitors adopt during primary neurogenesis . Accordingly , strong SMAD1/5 activity promotes progenitor maintenance while weaker activity enables neurogenic divisions to occur ( Le Dréau et al . , 2014 ) . This model explains how inhibition of BMP7 or SMAD1/5 activity provokes premature neuronal differentiation and the concomitant depletion of progenitors . However , it does not explain why the generation of distinct subtypes of dorsal interneurons are affected differently ( Le Dréau et al . , 2012 ) , nor how BMP signaling affects the activity of the proneural proteins expressed in the corresponding progenitor domains . Here , we have investigated these questions , extending our analysis to primary spinal neurogenesis along the whole dorsal-ventral axis . As such , we identified a striking correlation between the requirement of canonical BMP activity for the generation of a particular neuronal subtype and the proneural protein expressed in the corresponding progenitor domain . Inhibiting the activity of BMP7 , SMAD1/5 or their downstream effector ID2 strongly impaired the production of neurons by spinal progenitors expressing either ATOH1 or ASCL1 alone , while it had a much weaker effect on the generation of the neuronal subtypes derived from progenitors expressing NEUROG1 , NEUROG2 or PTF1a . We found that this differential responsiveness originates from an E-box dependent mode of co-operation of the class I HLH/E proteins with the proneural proteins . E proteins interact with proneural proteins to aid their interaction with CAGSTG E-boxes , facilitating the ability of ASCL1 and ATOH1 to promote neurogenic divisions and hence , neuronal differentiation . Conversely , E proteins inhibit proneural protein binding to CADATG motifs , consequently restraining the ability of NEUROG1/2 that preferentially bind to these motifs to trigger neurogenic division and promote neuronal differentiation . Similar results were obtained in the context of corticogenesis , suggesting that this differential co-operation of E proteins with the distinct proneural proteins is a general feature of their mode of action .
We previously reported that BMP7 signalling through its canonical effectors SMAD1 and SMAD5 , is differentially required for the generation of the distinct subtypes of dorsal spinal interneurons ( Figure 1A and Le Dréau et al . , 2012 ) . Here , we extend this analysis to the generation of neuronal subtypes produced in the ventral part of the developing chick spinal cord . Inhibiting BMP7 or SMAD1/5 expression by in ovo electroporation of specific sh-RNA-encoding plasmids at stage HH14-15 produced a significant reduction in the generation of p2-derived Chx10+ ( V2a ) and Gata3+ ( V2b ) subtypes 48 hr post-electroporation ( hpe ) , whereas Evx1+ ( V0v ) , En1+ ( V1 ) interneurons and Isl1+ motor neurons were not significantly affected ( Figure 1—figure supplement 1 ) . These results revealed a correlation whereby the requirement of the canonical BMP pathway for the generation of discrete spinal neuron subtypes is linked to the proneural protein expressed in the corresponding progenitor domain ( Figure 1B , C ) . The neuronal subtypes strongly affected by BMP7/SMAD1/5 inhibition ( dI1 , dI3 , dI5: Figure 1B , C ) were generated from spinal progenitors expressing ATOH1 ( dP1 ) or ASCL1 alone ( dP3 , dP5 ) . By contrast , all the neuronal subtypes deriving from spinal progenitors expressing either NEUROG1 alone ( dP2 , dP6-p1 ) or NEUROG2 ( pMN ) were much less severely affected ( Figure 1B , C ) . Intriguingly , the V2a/b interneurons that display intermediate sensitivity to BMP7/SMAD1/5 inhibition are derived from p2 progenitors that express both ASCL1 and NEUROG1 ( Misra et al . , 2014 ) , while the relatively insensitive dI4 interneurons are derived from dP4 progenitors that express PTF1a together with low levels of ASCL1 ( Figure 1B , C , Glasgow et al . 2005 ) . These correlations were particularly interesting in view of recent genome-wide ChIPseq studies that identified the optimal E-box ( CANNTG ) motifs bound by these TFs: ATOH1 and ASCL1 both preferentially bind to CAGCTG E-boxes ( Castro et al . , 2011; Klisch et al . , 2011; Lai et al . , 2011; Borromeo et al . , 2014 ) , whereas the optimal motif for NEUROGs is CADATG ( where D stands for A , G or T: see Seo et al . , 2007; Madelaine and Blader , 2011 ) . Interestingly , most of the E-boxes bound by PTF1a in the developing spinal cord correspond to the CAGCTG motif favored by ASCL1 and ATOH1 , yet PTF1a can bind to the NEUROG-like CAGATG motifs in a significant proportion of its targets genes ( Borromeo et al . , 2014 ) . These observations suggested that the sensitivity of a given progenitor domain to canonical BMP activity originates from the intrinsic DNA-binding preferences of the different proneural bHLH TFs ( Figure 1D ) . In many cell contexts , BMP signaling is mediated by ID proteins ( Hollnagel et al . , 1999; Moya et al . , 2012; Genander et al . , 2014 ) , which physically sequester class I HLH/E proteins to produce a dominant-negative effect on proneural proteins ( Figure 1D ) . While this hypothetical signaling cascade could explain the response of spinal progenitors expressing ASCL1 or ATOH1 to altered canonical BMP activity , it would not explain the relative insensitivity of the progenitors expressing NEUROG1 , NEUROG2 or PTF1a . Therefore , we tested the veracity of these functional relationships to identify the basis of this differential response ( Figure 1D ) . To test whether ID proteins act downstream of the canonical BMP pathway during spinal neurogenesis , we focused on ID2 ( Figure 2A ) , not least because canonical BMP signalling is necessary and sufficient to promote cId2 expression in the developing spinal cord ( Figure 2—figure supplement 1 and Le Dréau et al . , 2014 ) . Moreover , the pattern of cId2 expression closely overlaps that described for the canonical BMP activity during spinal cord development ( Le Dréau et al . , 2012; Le Dréau et al . , 2014 ) . At early stages during neural patterning , cId2 expression is restricted to the dorsal region of the developing spinal cord ( Figure 2B ) . Later on during neurogenesis this pattern spreads ventrally throughout the D-V axis , showing expression within all ventral progenitor domains except the pMN domain ( Figure 2C–D and Figure 2—figure supplement 2 ) . Inhibition of endogenous ID2 activity was triggered by in ovo electroporation of a sh-RNA specifically targeting chick Id2 transcripts ( sh-Id2 , Figure 2—figure supplement 3A–E ) . This ID2 inhibition caused premature cell-autonomous differentiation at 48 hpe similar to that provoked by inhibiting SMAD1/5 ( Figure 2E–K and Le Dréau et al . , 2014 ) . Conversely , overexpression of a murine ID2 construct reduced the proportion of electroporated cells that differentiated into neurons ( EP+;HuC/D+ cells , Figure 2H , K and Figure 2—figure supplement 3F , G ) . ID2 overexpression could also partially impede the premature differentiation caused by both sh-Id2 and sh-Smad5 ( Figure 2I–K ) . Similar results were obtained when measuring the activity of the pTubb3:luc reporter 24 hpe ( Figure 2L ) . We next analysed the consequences of ID2 inhibition on the generation of the different subtypes of spinal neurons and detected a significant dose-dependent reduction in the generation of many neuronal subtypes ( Figure 2M , N ) . The overall phenotype caused by ID2 inhibition was comparable to that triggered by inhibiting BMP7 , SMAD1 or SMAD5: the neuronal subtypes deriving from spinal progenitors expressing either ATOH1 or ASCL1 alone were globally more sensitive to ID2 inhibition than those derived from progenitors expressing NEUROG1 , NEUROG2 or PTF1a ( Figure 2O ) . Together , these results suggest that ID2 acts downstream of the canonical BMP pathway in spinal neurogenesis and that it regulates distinctly the generation of spinal neurons derived from progenitors expressing ASCL1/ATOH1 and NEUROG1/NEUROG2 . We wondered whether ID2 contributes to spinal neurogenesis by sequestering E proteins ( Figure 3A ) . Thus , we analysed the expression of these class I HLH genes during spinal neurogenesis . Transcripts from the cTcf3/E2A gene , which encodes the E12 or E47 alternative splice isoforms ( Murre et al . , 1989 ) , were readily detected in the ventricular zone throughout the dorsal-ventral axis of the developing spinal cord , with apparently no domain-specific pattern ( Figure 3B and Holmberg et al . , 2008 ) . Transcripts from the chicken HEB orthologue cTcf12 were detected in the transition zone , following a dorsal-to-ventral gradient ( Figure 3C ) . Previous studies reported that E2-2 transcripts were barely detected in the developing murine spinal cord ( Sobrado et al . , 2009 ) . The overexpression of E47 or TCF12 both produced a significant increase in the proportion of EP+;HuC/D+ cells ( Figure 3D–F , J ) , a phenotype that was reverted by the concomitant electroporation of ID2 ( Figure 3D–J ) . To inhibit the endogenous activity of E proteins , we took advantage of an E47 construct carrying mutations in its basic domain ( E47bm ) and that acts in a dominant-negative manner over E proteins in vivo ( Zhuang et al . , 1998 ) . Electroporation of E47bm inhibited neuronal differentiation in a cell autonomous manner , and it fully compensated for the premature differentiation caused by both E47 and TCF12 ( Figure 3—figure supplement 1 ) . This E47bm construct also rescued to a large extent the premature differentiation triggered by sh-Id2 ( Figure 3K–O ) . Together , these results appear to confirm that the role played by ID2 during spinal neurogenesis depends on its ability to sequester E proteins . The results we obtained so far suggested that E proteins themselves might co-operate differently with the distinct proneural proteins during spinal neurogenesis ( Figure 4A ) . To test this hypothesis , we first analyzed the consequences of expressing the E47bm mutant on the generation of spinal neuron subtypes . There was a marked reduction ( ≥50% ) in the generation of Lhx2/9+ ( dI1 ) and Tlx3+ ( dI3/dI5 ) interneurons , which derive respectively from progenitors expressing ATOH1 and ASCL1 alone ( Figure 4B , C , F ) . By contrast , electroporation of E47bm affected to a lesser extent ( <25% ) the generation of Lhx1/5+ interneurons ( dI2/dI4/dI6-V1 ) or Isl1+ motor neurons deriving from progenitors expressing NEUROG1 alone ( dP2 , dP6-V1 ) , PTF1a ( dP4 ) or NEUROG2 ( pMN , Figure 4D–F ) . Alternatively , we used another dominant-negative construct of E47: E47Δnls-RFP , inserted in a plasmid with low electroporation efficiency ( see the Materials and methods section ) . This version of E47 fused to RFP is deleted from its nuclear localization signals and thereby impairs the nuclear import of E47 , hence its transcriptional activity ( Mehmood et al . , 2009 ) . As previously reported in vitro ( Mehmood et al . , 2009 ) , the subcellular localization of this E47Δnls-RFP mutant after in ovo electroporation was mostly cytoplasmic ( Figure 4G–H’ ) . As seen with E47bm , this E47Δnls-RFP mutant also impaired neuronal differentiation cell-autonomously ( Figure 4G–I ) . We analyzed the consequences of E47Δnls-RFP electroporation on the generation of spinal neuron subtypes by quantifying the proportions of differentiated electroporated cells ( by focusing on the RFP+ cells that were HuC/D+ or Sox2- ) that express distinct neuronal subtype markers ( Figure 4J–R ) . Compared to a control plasmid , electroporation of the E47Δnls-RFP mutant reduced about half the proportions of differentiated electroporated cells that express Lhx2/9 or Tlx3 ( Figure 4J–M and R ) , indicating that inhibiting E47 activity hindered the differentiation of spinal progenitors expressing ATOH1 or ASCL1 alone . By contrast , progenitors electroporated with the E47Δnls-RFP mutant efficiently differentiated into Lhx1/5+ interneurons or Isl1+ motor neurons ( Figure 4N–R ) . Of note , we could even observe Lhx1/5+ or Isl1+ electroporated cells within the ventricular zone ( stars in Figure 4O–O’ , Q–Q’ ) , indicative of a premature differentiation of NEUROG1/2/PTF1a-expressing progenitors when E47 activity is impaired . Hence , ATOH1 and ASCL1 appear to be much more dependent on the activity of E proteins to promote appropriate neuronal differentiation than are NEUROG1 , NEUROG2 and PTF1a . Next , we evaluated how E47 gain-of-function modulates the neuronal differentiation induced when ASCL1 , ATOH1 , NEUROG1 or NEUROG2 are overexpressed ( Figure 5A–H ) . From 24 hpe onwards , all four proneural bHLH proteins caused premature differentiation in a cell-autonomous and concentration-dependent manner ( Figure 5—figure supplement 1A–C ) . Based on these data , we decided to test how E47 addition would alter the phenotypes caused by sub-optimal concentrations of these four proneural proteins . The addition of E47 accentuated the mild increase in neuronal differentiation provoked by ASCL1 at 24 hpe , and more significantly at 48 hpe ( Figure 4A–B’ ) . Accordingly , E47 provoked a significant reduction in the average number of electroporated cells generated 48 hpe of ASCL1 ( Figure 4I ) . A similar tendency , albeit less pronounced , was observed when E47 was combined with ATOH1 , especially in terms of the reduced average number of EP+ cells generated 48 hpe ( Figure 4C–D’ , I ) . Addition of E47 had the opposite effect when combined with NEUROG1 or NEUROG2: it significantly reduced the proportion of EP+;HuC/D+ cells obtained at 24 hpe and consequently increased the final numbers of EP+ cells observed at 48 hpe ( Figure 5E–I ) . These results suggested that E47 differentially regulates the ability of ASCL1/ATOH1 and NEUROG1/NEUROG2 to promote cell cycle exit . To assess cell cycle exit , a fluorescent cytoplasmic-retention dye that is only diluted on cell division was added at the time of electroporation and its mean fluorescence intensity was measured in FACS-sorted electroporated ( GFP+ ) cells 48 hr later ( Figure 5J ) . This assay demonstrated that E47 itself increased the mean Violet intensity , and further enhanced the mild increase caused by ASCL1 ( Figure 5K ) , indicating that E47 facilitates ASCL1’s ability to promote cell cycle exit . E47 had an opposite effect when combined with NEUROG1 , significantly reducing the strong increase in Violet intensity caused by NEUROG1 ( Figure 5K ) , thereby confirming that E47 restricts NEUROG1’s ability to promote cell cycle exit . We next studied how E47 influences the respective abilities of ASCL1 and NEUROG1 to regulate the balance between the three different modes of division that spinal progenitors can undergo during neurogenesis: symmetric proliferative divisions ( PP ) , asymmetric divisions ( PN ) , and symmetric neurogenic divisions ( NN ) ( Saade et al . , 2013; Le Dréau et al . , 2014 ) . To this end , we took advantage of the pSox2:eGFP and pTis21:RFP reporters that are specifically active during progenitor-generating ( PP +PN ) and neuron-generating ( PN +NN ) divisions , respectively ( Saade et al . , 2013 ) . The effects of E47 , ASCL1 and NEUROG1 on the reporters’ activities were assayed 16 hpe by immunohistochemistry or quantified by FACS ( Figure 5L ) . E47 caused a significant decrease in the proportion of pSox2:eGFP+;pTis21:RFP- ( PP ) cells and a reciprocal increase in the proportion of pTis21:RFP+ ( PN +NN ) neurogenic divisions relative to the controls ( Figure 5M , M’ , P ) . While we did not detect any significant change in the proportions of PP , PN and NN cells in response to ASCL1 alone at this concentration , we did observe an increase in neurogenic divisions at the expense of proliferative divisions when ASCL1 was combined with E47 ( Figure 5N , N’ , P ) . Conversely , E47 significantly restrained NEUROG1’s ability to trigger neurogenic divisions at the expense of PP divisions ( Figure 5O–P ) . Assessing the activity of the pSox2:luc reporter confirmed these results , further showing that E47 facilitates the ability of both ASCL1 and ATOH1 to repress pSox2 activity , whereas it restricts the repressive effects of both NEUROG1 and NEUROG2 ( Figure 5—figure supplement 1D ) . Together , these results revealed that E47 co-operates distinctly with ASCL1/ATOH1 and NEUROG1/NEUROG2 to fine-tune neurogenic divisions during spinal neurogenesis . To identify the molecular mechanisms underlying the different outcomes caused by E proteins’ co-operation with the distinct proneural proteins , we focused on the interaction of E47 with ASCL1 and NEUROG1 . A DNA-binding deficient version of NEUROG1 ( NEUROG1-AQ , Sun et al . , 2001 ) , was unable to transactivate the NEUROG-responsive pNeuroD:luc reporter or to promote neuronal differentiation ( Figure 6—figure supplement 1 ) . Hence , the ability of NEUROG1 to trigger neuronal differentiation during spinal neurogenesis depends on its transcriptional activity , as previously reported for ASCL1 and ATOH1 ( Nakada et al . , 2004 ) . Genome-wide ChIP-seq studies have established that the preferential E-box motifs bound by ASCL1 , E47 and NEUROG1 correspond respectively to CAGCTG ( Castro et al . , 2011; Borromeo et al . , 2014 ) , CAGSTG ( where S stands for C or G: Lin et al . , 2010; Pfurr et al . , 2017 ) and CADATG ( where D stands for A , G or T: Seo et al . , 2007; Madelaine and Blader , 2011 ) . In light of these intrinsic preferences , we tested how E47 modulates the abilities of ASCL1 and NEUROG1 to bind to DNA and activate transcription in different E-box contexts ( Figure 6A and Figure 6—figure supplement 2A ) . E47 acted in synergy with both ASCL1 and NEUROG1 to drive transcription of the pkE7:luc reporter under the control of 7 CAGGTG repeats ( Figure 6B , C ) . By contrast , E47 and ASCL1 only weakly transactivated the pNeuroD:luc reporter , the promoter of which contains 9 CADATG E-boxes and 1 CAGGTG box ( Figure 6D ) . A similar result was obtained with a version of the pNeuroD:luc reporter in which the single CAGGTG motif was destroyed by mutagenesis ( Figure 6—figure supplement 2B ) , reinforcing the idea that both E47 and ASCL1 preferentially bind to CAGSTG sequences . Intriguingly , E47 markedly reduced the ability of NEUROG1 to enhance the activity of both pNeuroD:luc and its mutated version ( Figure 6E and Figure 6—figure supplement 2C ) . Of note , TCF12 was also able to enhance ASCL1-dependent pKE7:luc activity while inhibiting NEUROG1’s ability to induce pNeuroD:luc activity , though with milder capacities than E47 ( Figure 6—figure supplement 2D , E ) . To further define whether the way E47 modulates ASCL1 and NEUROG1’s transcriptional activity directly depends on the E-box content , we used two additional reporters: pDll1-M:luc and pDll1-N:luc . These are based on conserved regulatory elements found in the promoter of the Delta-like1 gene and have been described to respectively respond to ASCL1 and NEUROG2 ( Beckers et al . , 2000; Castro et al . , 2006 ) . When combined with the pDll1-M:luc reporter containing 3 CAGSTG +1 CADATG motifs , addition of E47 to ASCL1 had only an additive effect ( Figure 6—figure supplement 2F ) , compared to the synergistic effect observed on pKE7:luc activity ( Figure 6B ) . When combined with the pDll1-N:luc reporter containing 1 CAGSTG +3 CADATG motifs , addition of E47 still inhibited the activity induced by NEUROG1 ( Figure 6—figure supplement 2G ) , but less potently than when combined with the pNeuroD:luc reporter ( Figure 6E ) . In vitro ChIP assays further demonstrated that E47 can bind to and enhance ASCL1 binding at the 7 CAGGTG-containing promoter region of the pkE7:luc reporter ( Figure 6F ) , consistent with the notion that their heterodimerization is required for optimal binding and subsequent transcriptional activation . By contrast , E47 caused a significant reduction in the amount of NEUROG1 bound to the promoter region of the pNeuroD:luc reporter ( Figure 6G ) . The fact that E47 itself bound to this promoter region suggested that E47 and NEUROG1 compete for binding to CADATG motifs ( Figure 6G ) , although E47 cannot transactivate them as potently as NEUROG1 ( Figure 6E ) . Together , these results revealed that E47 acts in synergy with both ASCL1 and NEUROG1 when binding to its own optimal E-box ( CAGSTG ) , while it somehow impedes NEUROG1 from binding to CADATG motifs . To assess whether these E-box-dependent modulations of proneural proteins’ transcriptional activity by E47 indeed depend on physical interactions , we first prevented E47 from interacting with ASCL1 or NEUROG1 by adding ID2 . As expected , addition of ID2 partially rescued both the synergistic effect of E47 on pKE7:luc activity when combined with ASCL1 and the inhibitory effect of E47 on NEUROG1-induced pNeuroD:luc activity ( Figure 6—figure supplement 2H , I ) . Secondly , we compared the activity of tethered constructs that were designed to produce homodimers of ASCL1 ( A-A ) and NEUROG1 ( N-N ) , or heterodimers with E47 ( A-E , N-E: Figure 6—figure supplement 3A–C ) . Consistent with the results obtained with monomers , A-E heterodimers were significantly more potent than A-A homodimers in driving pkE7:luc activity ( Figure 6—figure supplement 3D ) , while N-N homodimers transactivated pNeuroD:luc much more strongly than N-E heterodimers ( Figure 6—figure supplement 3E ) . A-A and A-E promoted similar neuronal differentiation 48 hpe ( Figure 6H–J , M ) , but the average number of EP+ cells obtained after A-E electroporation was significantly lower than after A-A electroporation ( Figure 6N ) , suggesting that A-E promotes early neurogenic divisions more potently than A-A . This idea was supported by the ability of A-E to repress pSox2:luc activity at 20 hpe , unlike A-A ( Figure 6—figure supplement 3F ) . As for NEUROG1 , N-N was significantly more potent than N-E at promoting neuronal differentiation ( Figure 6K–M ) , at reducing the average number of EP+ cells generated 48 hpe ( Figure 6N ) and at repressing pSox2:luc activity ( Figure 6—figure supplement 3F ) . Thus , the tethered constructs performed like the monomers ( Figure 5 ) , supporting the conclusion that E47 facilitates the ability of ASCL1 and restrains that of NEUROG1 to trigger neurogenic divisions during spinal neurogenesis . We were interested to determine if this differential co-operation of E47 with the distinct proneural proteins could be extended to other regions of the developing CNS . We tested this hypothesis in the developing cerebral cortex , as NEUROG1/2 and ASCL1 all contribute to neurogenesis in this region in mammals ( Huang et al . , 2014 ) . The development of the cerebral cortex in birds actually shows unexpected similarities to mammalian corticogenesis , including the conservation of its temporal sequence of neurogenesis ( Dugas-Ford et al . , 2012; Suzuki et al . , 2012 ) . As in mammals , corticogenesis in the chick embryo originates from a region of the dorsal pallium expressing PAX6 ( Figure 7A , B and Suzuki et al . , 2012 ) . From E3 to E5 , an early phase of corticogenesis produces the first SOX2-;HuC/D+ cortical neurons , which are generated specifically from PAX6+;TBR2- radial glia-like progenitors that divide at the apical surface , as in mammals ( Figure 7C–D ) . Cortical TBR2+ progenitors that divide basally , similar to mammalian intermediate progenitor cells , appear at around E5 ( Figure 7D–D” ) . The cortical neurons produced during this early phase express TBR1 ( Figure 7—figure supplement 1A–A” ) , as well as other markers typically expressed by mammalian deep-layer neurons ( Dugas-Ford et al . , 2012; Suzuki et al . , 2012 ) . From E3 to E5 , cTcf3/cE2A transcripts were detected throughout the whole D-V axis of the developing telencephalon , with apparently no domain-specific pattern ( Figure 7—figure supplement 1B–B” ) . cTcf3/cE2A transcripts were mainly detected in the ventricular zone formed by cortical progenitors ( Figure 7E–E’ ) , as previously reported during mouse corticogenesis ( Li et al . , 2012 ) . At E4 , cNeurog1 and cNeurog2 transcripts were detected in a salt-and-pepper fashion in the cortical PAX6+ region ( Figure 7F , G ) , whereas cAscl1 expression was detected strongly in the sub-pallium and more weakly in the developing cerebral cortex ( Figure 7H ) . These expression patterns seen in early chicken embryos are very similar to what is observed in the developing mammalian telencephalon ( Huang et al . , 2014 ) , suggesting the phylogenetic conservation of the functions played by proneural proteins during corticogenesis . To test how E47 modulates the activity of ASCL1 and NEUROG1/2 in the developing chick cerebral cortex , we electroporated the corresponding proneural constructs in the dorsal telencephalic region in ovo at E3 and analysed their effects on neuronal differentiation 2 days later ( Figure 7I ) . Both NEUROG1 and NEUROG2 triggered significant neuronal differentiation in the developing cerebral cortex in a cell autonomous and dose-dependent manner , whereas ASCL1 overexpression had only a minor effect per se ( Figure 7J–M , R and Figure 7—figure supplement 1C ) . E47 , which itself had no obvious effect at this concentration ( Figure 7N , R ) , significantly increased neuronal differentiation when combined with ASCL1 ( Figure 7O , R ) . Conversely , E47 markedly reduced the ability of NEUROG1 , and to a lesser extent that of NEUROG2 , to promote neuronal differentiation ( Figure 7P–R ) . These results suggest that E47 also modulates in opposite ways the neurogenic activities of ASCL1 and NEUROG1/2 in the context of cortical neurogenesis .
Class I HLH/E proteins are generally described as obligatory and permissive co-factors for proneural proteins , which must form heterodimers to become active and regulate transcription ( Wang and Baker , 2015 ) . The main findings of our study are that the co-operation between E proteins and proneural proteins might be more complex than originally thought . Our results indeed revealed that E proteins can facilitate or restrain the transcriptional activity of the proneural proteins , depending both on their intrinsic DNA-binding preferences and on the E-box content ( Figure 8 ) . On the one hand , our results support a revised model whereby E proteins synergize with proneural proteins specifically at CAGSTG E-boxes , the preferential motifs of E proteins ( Lin et al . , 2010; Pfurr et al . , 2017 ) . Therefore , E proteins facilitate the activity of the proneural proteins that share their preferential binding to CAGSTG motifs , such as ASCL1 and ATOH1 ( Castro et al . , 2011; Klisch et al . , 2011; Lai et al . , 2011; Borromeo et al . , 2014 ) . Inhibiting the activity of E proteins by overexpressing the E47bm or E47Δnls-RFP mutants strongly impaired the generation of interneurons derived from spinal progenitors that express ATOH1 or ASCL1 alone . Conversely , enhancing the expression of E47 reinforced the ability of ATOH1 and more markedly , that of ASCL1 to promote neuronal differentiation . Our results suggest that this results from the capacity of E47 to increase the ability of these proneural proteins to trigger neurogenic divisions at the expense of proliferative ones ( Figure 8 ) . Such co-operation appears to be particularly crucial in the case of ASCL1 , whose overexpression could barely increase neurogenic divisions per se , at least at low concentration . These observations support a growing body of evidence that ASCL1 possesses a mild neurogenic potential . For instance , the broad dP3-dP5 domain of spinal progenitors , in which ASCL1 is expressed alone or in combination with PTF1a , expands at the end of primary neurogenesis before producing large numbers of dILA/B neurons during the second neurogenic wave ( Wildner et al . , 2006; Borromeo et al . , 2014 ) . Later on , ASCL1 is also involved in promoting oligodendrogenesis in both the developing brain and spinal cord ( Huang et al . , 2014 ) . Moreover , recent studies have reported cell cycle promoting-genes among the targets bound by ASCL1 in the ventral telencephalon and that it also sustains the proliferation of adult neural stem cells ( Castro et al . , 2011; Urbán et al . , 2016 ) , suggesting that its mild neurogenic ability might actually be required to sustain long-term production of the neural lineages . Whether the ability of ASCL1 to maintain neural progenitor pools is related to its dependence on the availability of E proteins is an intriguing hypothesis that would be worth testing . On the other hand , our findings demonstrate that E proteins inhibit proneural protein binding to CADATG motifs . In consequence , E proteins restrict the activity of the proneural proteins that preferentially bind to these motifs , such as NEUROG1/2 ( Seo et al . , 2007; Madelaine and Blader , 2011 ) . Spinal progenitors electroporated with the E47Δnls-RFP mutant efficiently differentiated into Lhx1/5+ interneurons or Isl1+ motor neurons . Some of these differentiated cells were even observed within the ventricular zone , evidencing that NEUROG1/2/PTF1a-expressing progenitors are prone to differentiate prematurely when E47 activity is impaired . Since E47 restrains the capacity of NEUROG1/2 to promote neuronal differentiation in the context of both spinal neurogenesis and corticogenesis , this would appear to be a general feature . Depletion of the murine E47 isoform was recently shown to increase the production of both TBR1+ and SATB2+ neurons at mid-corticogenesis ( Pfurr et al . , 2017 ) . In fact , the loss of E47 in early cortical progenitors , for which NEUROG2 constitutes the main proneural protein , causes premature neuronal differentiation . This is consistent with our results and model and it contrasts with the block in neuronal differentiation that would be expected if E47 was essential for NEUROG2 activity . Our results also suggest that NEUROGs do not necessarily need to form heterodimers with E proteins to trigger neuronal differentiation . Indeed , forced NEUROG1 homodimers drive CADATG-dependent transcription and neuronal differentiation more efficiently than NEUROG1-E47 heterodimers . Similarly , NEUROG2 homodimers better transactivate neuronal differentiation genes than NEUROG2-E47 heterodimers ( Li et al . , 2012 ) , and EMSA experiments suggested the existence of multiple combinations of proneural homo- and heterodimers ( Henke et al . , 2009 ) . The physiological relevance of such proneural homodimers is worthy of further study but to date , our attempts to determine whether NEUROG1 homodimers are formed in vivo during spinal neurogenesis remain inconclusive for technical reasons ( data not shown ) . Nevertheless , the strong capacity of NEUROGs to trigger neurogenic divisions independently of E proteins , including self-consuming NN divisions , correlates well with the fact that neural progenitors expressing NEUROG1/NEUROG2 are usually depleted during the neurogenic phase ( Simmons et al . , 2001; Kim et al . , 2011 ) . Together , these results support the notion that E proteins are required to dampen the strong capacity of NEUROGs to trigger neurogenic divisions , thereby avoiding the premature depletion of neural progenitor pools ( Figure 8 ) . The findings that E proteins modulate in opposite ways the activities of ASCL1/ATOH1 and NEUROG1/2 would also explain why modulating canonical BMP activity affects differently the generation of the distinct neuronal subtypes produced during primary spinal neurogenesis . Inhibiting BMP7 or SMAD1/5 would result in the release of E proteins from their complexes with IDs . In turn , this would facilitate ATOH1 and ASCL1 activity , prematurely increasing the proportion of neurogenic divisions undertaken by the corresponding dP1 and dP3/dP5/p2 progenitors , causing their premature differentiation and exhaustion , and ultimately leading to a production of fewer neurons . As NEUROGs are less dependent on E proteins , the inhibition of canonical BMP signalling only mildly impairs the generation of the neuronal subtypes that derive from progenitors expressing NEUROG1/NEUROG2 . In summary , the results presented here led us to propose that E proteins fine-tune neurogenesis by buffering the activity of the distinct proneural proteins . As such , these data add another layer of sophistication to the molecular mechanisms that regulate the activity of proneural bHLH proteins and hence , neurogenesis .
Fertilized white Leghorn chicken eggs were provided by Granja Gibert , rambla Regueral , S/N , 43850 Cambrils , Spain . Eggs were incubated in a humidified atmosphere at 38·C in a Javier Masalles 240N incubator for the appropriate duration and staged according to the method of Hamburger and Hamilton ( HH , ( Hamburger and Hamilton , 1951 ) . According to EU animal care guidelines , no IACUC approval was necessary to perform the experiments described herein , considering that the embryos used in this study were always harvested at early stages of embryonic development ( at E5 at the latest ) . Sex was not identified at these stages . Unilateral in ovo electroporations in the developing chick spinal cord and dorsal telencephalon were performed respectively at stages HH14-15 and HH18 ( 54 and 69 hr of incubation ) . In the telencephalon , corticogenesis was studied specifically in the dorsal-medial-lateral ( dML ) subregion to minimize any possible variability along the medial-lateral axis . Plasmids were diluted in RNAse-free water at the required concentration [0 to 4 µg/µl] and injected into the lumen of the caudal neural tube or the right cerebral ventricle using a fine glass needle . Electroporation was triggered by applying 5 pulses of 50 ms at 22 . 5 V with 50 ms intervals using an Intracel Dual Pulse ( TSS10 ) electroporator . Electroporated chicken embryos were incubated back at 37C and recovered at the times indicated ( 16–48 hr post-electroporation ) . The human embryonic kidney-derived HEK293T cell line was obtained from ATCC ( #CRL-3216; STR profile = CSF1PO: 11 , 12; D13S317: 12 , 14; D16S539: 9 , 13; D5S818: 8 , 9; D7S820: 11; TH01: 7 , 9 . 3; TPOX: 11; vWA: 16 , 19; Amelogenin: X ) . This cell line is not listed in the commonly misidentified cell lines list from the International Cell Line Authentication Committee . Cells were Mycoplasma-free , as routinely assessed using the LookOut Mycoplasma PCR Detection Kit ( Sigma #MP0035-1KT ) . HEK293T cells were grown at sub-confluent density in DMEM supplemented with 10% fetal bovine serum and penicillin/streptomycin at 37°C , 5% CO2 . To facilitate comparisons in gain-of-function experiments , all the constructs used in this study were inserted under the control of a pCAGGS promoter that harbors high activity in chick ( pCAGGS or pCAGGS_ires_GFP , kindly provided by Andy McMahon , Megason and McMahon , 2002 ) , and were electroporated at similar concentrations ( 0 , 0 . 1 , 0 . 5 or 1 µg/µl as specified in the respective figure legends ) . Non-fluorescent pCAGGS plasmids were combined with 0 . 25 µg/µl of pCS2_H2B-GFP for visualization . The pCAGGS:ASCL1 , pCAGGS:NEUROG1 and pCAGGS:NEUROG2 plasmids were kindly provided by François Guillemot . The pCAGGS:ATOH1_ires_GFP plasmid was obtained by subcloning from a pCMV:ATOH1 kindly provided by Nissim Ben-Arie ( Krizhanovsky et al . , 2006 ) . The pCAGGS:E47 and pCAGGS:TCF12 were kindly provided by Jonas Muhr ( Holmberg et al . , 2008 ) . The pCAGGS:E47bm_ires_GFP plasmid was derived from a pGK:E47_CFP plasmid kindly provided by Yuan Zhang ( Zhuang et al . , 1998 ) . The pCAGGS:ID2_ires_GFP was derived from a pCMV:ID2 plasmid , and the pCAGGS_SMAD5-SD_ires_GFP was described previously ( Le Dréau et al . , 2012 ) . Only E47Δnls-RFP and E47-RFP ( in a pFlag-CMV2 vector , kindly provided by Yoshihiro Yoneda , Mehmood et al . , 2009 ) , Somitabun ( pCS2:Somitabun , kindly provided by Jonathan Slack , Beck et al . , 2001 ) and NEUROG1-AQ and its wild-type NEUROG1 version ( pMiW:myc-NGN1 and pMiW:myc-NGN1-AQ , kindly gifted by Jane Johnson , Gowan et al . , 2001 ) were used in a different backbone . HA-tagged versions of ASCL1 ( pCAGGS:HA-ASCL1 , Alvarez-Rodríguez and Pons , 2009 ) and NEUROG1 ( pCAGGS:HA-NEUROG1 ) and a Flag-E47-RFP construct were used for chromatin immuneprecipitation assays . Inhibition of cBmp7 , cSmad1 , cSmad5 or cId2 expression was triggered by electroporation of short-hairpin constructs inserted into the pSuper ( Oligoengine ) or pSHIN ( Kojima et al . , 2004 ) vectors . Electroporation of 2–4 µg/µl of these constructs caused a specific and reproducible 50% inhibition of the target expression ( see Le Dréau et al . , 2012 ) . The pSox2:GFP and pTis21:RFP reporters used to assess the modes of divisions undergone by spinal progenitors were previously described in details ( Saade et al . , 2013 ) . The pSox2:luc derived from the pSox2:GFP ( Saade et al . , 2013 ) . The different versions of the pId2:luc reporter were kindly provided by Yoshifumi Yokota ( Kurooka et al . , 2012 ) and the pkE7:luc by Masashi Kawaichi ( Akazawa et al . , 1995 ) . The pNeuroD:luc , pDll1-M:luc and pDll1-N:luc reporters were kindly provided by François Guillemot ( Castro et al . , 2006 ) . The pNeuroDmut:luc reporter was obtained by site-directed mutagenesis of the single CAGGTG E-box contained in the NeuroD promoter region . The pTubb3:luc reporter was obtained by subcloning the Tubb3 enhancer region present in the pTubb3enh:GFP plasmid ( kindly provided by Jonas Muhr , Bergsland et al . , 2011 ) into the pGL3:luc vector ( Promega ) . Please head for the Key Resources Table for additional information . The tethered bHLH dimers were derived from the pCAGGS:ASCL1-t-E47_ires_GFP kindly provided by François Guillemot ( Geoffroy et al . , 2009 ) . This plasmid and pCAGGS:NEUROG1 were used as templates to generate the ASCL1-t-ASCL1 , NEUROG1-t-E47 and NEUROG1-t-NEUROG1 constructs inserted into pCAGGS_ires_GFP , using a tether peptide AAAGTSAGGAAAGTSASAATGA flanked by SpeI and ClaI restriction sites as described previously ( Henke et al . , 2009 ) . Expression of the tethered bHLH dimers was assessed by western blot after transfection into HEK293 cells . Transient cell transfections were obtained by electroporation applying 2 pulses of 120V , 30 ms ( Microporator MP‐100 , Digital Bio ) . Cells were grown for 24 hr onto poly-L-Lysine-coated 6-well dishes in DMEM/F12 supplemented with 10% fetal bovine serum and 50 mg/L of Gentamicin until reaching 70‐80% confluence . The typical transfection efficiency of this procedure was 40–60% . Cells were lysed in 1X SDS loading buffer ( 10% glycerol , 2% SDS , 100 mM dithiothreitol , and 62 . 5 mM Tris-HCl , pH 6 . 8 ) and DNA was disrupted by sonication . Protein extracts were separated by SDS-PAGE electrophoresis , transferred to Immobilon-FL PVDF membranes ( IPFL00010 , Millipore ) , blocked with the Odyssey Blocking Buffer ( 927–40000 , LI-COR ) , and incubated with antibodies against ASCL1 ( BD Pharmingen , cat#556604 , 1:1000 ) , NEUROG1 ( Millipore , cat#AB15616 , 1:3000 ) or E2A ( Santa Cruz , cat#sc-763 , 1:1000 ) . Detection was performed using fluorescence-conjugated secondary antibodies and an Odyssey Imaging System ( LI-COR ) . For immunohistochemistry experiments , chicken embryos were carefully dissected , fixed for 2 hr at 4°C in 4% paraformaldehyde and rinsed in PBS . Immunostaining was performed on either vibratome ( 40 µm ) or cryostat ( 16 µm ) sections following standard procedures . After washing in PBS-0 . 1% Triton , the sections were incubated overnight at 4C with the appropriate primary antibodies ( see Key Resources Table ) diluted in a solution of PBS-0 . 1% Triton supplemented with 10% bovine serum albumin or sheep serum . After washing in PBS-0 . 1% Triton , sections were incubated for 2 hr at room temperature with the appropriate secondary antibodies diluted in a solution of PBS-0 . 1% Triton supplemented with 10% bovine serum albumin or sheep serum . Alexa488- , Alexa555- and Cy5-conjugated secondary antibodies were obtained from Invitrogen and Jackson Laboratories . Sections were finally stained with 1 μg/ml DAPI and mounted in Mowiol ( Sigma-Aldrich ) . Optical sections of fixed samples ( transverse views of the spinal cord , coronal views for the telencephalon ) were acquired at room temperature with the Leica LAS software , in a Leica SP5 confocal microscope using 10x ( dry HC PL APO , NA 0 . 40 ) , 20x ( dry HC PL APO , NA 0 . 70 ) , 40x ( oil HCX PL APO , NA 1 . 25–0 . 75 ) or 63x ( oil HCX PL APO , NA 1 . 40–0 . 60 ) objective lenses . Maximal projections obtained from 2 µm Z-stack images were processed in Photoshop CS5 ( Adobe ) or ImageJ for image merging , resizing and cell counting . Quantification of endogenous ID2 intensity was assessed using the ImageJ software . Cell nuclei of H2B-GFP + electroporated and neighboring non-electroporated cells were delimitated by polygonal selection , and the mean intensity of ID2 immunoreactivity quantified as mean gray values . Quantifications were performed on at least six electroporated and six non-electroporated cells per image , in at least three different images per embryo . Chicken embryos were recovered at the indicated stage , fixed overnight at 4°C in 4% PFA , rinsed in PBS and processed for whole mount RNA in situ hybridization following standard procedures . Probes against chick cId2 ( #chEST852M19 ) and cNeurog2 ( #chEST387d10 ) were purchased from the chicken EST project ( UK-HGMP RC ) . Probes against cTcf3/E2a , cAscl1 and cNeurog1 were kindly provided by Drs Jonas Muhr , José-Maria Frade and Cristina Pujades . The probe against cTcf12/cHeb was obtained by PCR from genomic DNA of E4 chicken embryonic tissue and the purified 623 nucleotides insert was sub-cloned into the pGEM-T vector ( Promega ) . Hybridized embryos were post-fixed in 4% PFA and washed in PBT . 45µM-thick sections were cut with a vibratome ( VT1000S , Leica ) , mounted and photographed using a microscope ( DC300 , Leica ) . The data show representative images obtained from three embryos for each probe . Transcriptional activity was assessed following electroporation of a luciferase reporter together with a renilla luciferase reporter used for normalization , in combination with the indicated plasmids required for experimental manipulation . Embryos were harvested 24 hr later and GFP-positive neural tubes were dissected and homogenized in a Passive Lysis Buffer on ice . Firefly- and renilla-luciferase activities were measured by the Dual Luciferase Reporter Assay System ( Promega ) . The average number of divisions undergone by electroporated spinal progenitors was assessed in vivo using the CellTrace Violet Cell Proliferation Kit ( Invitrogen ) . The Violet cell tracer ( 1 mM ) , a cytoplasmic retention dye that becomes diluted as cells divide , was injected into the lumen of the neural tube at the time of electroporation . Embryos were recovered 48 hr later , the neural tubes were carefully dissected and recovered and the cells dissociated following a 10–15 min digestion in Trypsin-EDTA ( Sigma ) . The fluorescence intensity of the Violet tracer was measured in viable dissociated electroporated GFP+ cells in the 405/450 nm excitation/emission range on a Gallios flow cytometer ( Beckman Coulter , Inc ) . Chicken embryos were recovered 16 hr after co-electroporation of the pSox2:eGFP and pTis21:RFP reporters together with the indicated bHLH TF-encoding plasmids . Cell suspensions were obtained from pools of 6–8 dissected neural tubes after digestion with Trypsin-EDTA ( Sigma ) for 10–15 min , and further processed on a FACS Aria III cell sorter ( BD Biosciences ) for measurement of eGFP and RFP fluorescences . At least 1 , 000 cells for each progenitor population ( PP , PN and NN ) were analyzed per sample . HEK293T cells were transfected by a standard calcium phosphate co-precipitation protocol with combinations of pCAGGS_ires_GFP , pCAGGS:HA-ASCL1 , pCAGGS:HA-NEUROG1 , pCMV2_Flag-E47-RFP together with the pkE7:luc or pNeuroD:luc reporters , with a total of 10 µg of DNA per 100 mm dish . 24 hr later , cells were collected and 10% of the material was reserved to check transfection by Western blot . For chromatin immunoprecipitation assays , approximately 1 million transfected HEK293T cells were fixed with 1% formaldehyde for 10 min at room temperature . Fixation was quenched by adding 0 . 125M glycine for 5 min . After two washes with PBS , cells were lysed on ice for 20 min in a lysis buffer containing protease inhibitors ( 1% SDS; 10 mM EDTA pH8 . 0; 50 mM Tris-HCl pH8 . 1 ) . Sonication was performed with a Bioruptor sonicator to obtain 200–500 bp shredded chromatin fragments . Chromatin purification was carried out by spinning samples down at maximum speed at 4C during 30 min . Purified chromatin was pre-cleared with protein A agarose ( Millipore #16–125 ) for 30 min . 25 µg of chromatin were immunoprecipitated with 5 µL of anti-RFP serum ( Herrera et al . , 2014 ) , 2 µg of anti-HA ( Abcam , cat#20084 ) , anti-NEUROG1 ( Millipore , cat#15616 ) or unspecific rabbit IgG ( Diagenode , cat#C15410206 ) antibodies . Antibody-chromatin complexes were recovered using magnetic beads ( Magna ChIP , Millipore , cat#16–661 ) and immuno-complexes were washed once with TSE I ( 0 . 1% SDS; 1% Triton-X100; 2 mM EDTA pH8 . 0; 20 mM Tris-HCl pH8 . 1; 150 mM NaCl ) , TSE II ( 0 . 1% SDS; 1% Triton-X100; 2 mM EDTA pH8 . 0; 20 mM Tris-HCl pH8 . 1; 500 mM NaCl ) , TSE III ( 0 . 25M LiCl; 1% NP-40; 1% Sodium Deoxicholate; 1 mM EDTA pH8 . 0; 10 mM Tris-HCl pH8 . 1 ) and twice with TE ( Tris-HCl 10 mM , EDTA 1 mM ) . Reversal of crosslinking was done by incubating samples in elution buffer ( 1% SDS , 0 . 1M NaHCO3 ) overnight at 65C . DNA was purified by phenol-chloroform extraction followed by ethanol precipitation . Quantification of the DNA target regions and negative control ( luciferase ORF ) was assessed by qPCR in a Lightcycler 480 ( Roche ) using specific primers ( see Key Resources Table ) . Proteins extracts were obtained by incubation in a RIPA buffer ( 150 mM NaCl , 1 . 0% NP-40 , 0 . 5% sodium deoxycholate , 0 . 1% SDS and 50 mM Tris , pH 8 . 0 ) supplemented with protease and phosphatase inhibitors for 20 min on ice and centrifugation ( 20 min at maximum speed ) . 30 μg of protein samples were mixed with the Laemmli buffer ( 375 mM Tris pH = 6 . 8 , 12%SDS , 60% glycerol , 600 mM DTT , 0 . 06% bromphenol blue ) , heated to 95°C and then separated on a SDS-PAGE gel in running buffer ( 25 mM Tris base , 190 mM glycine , 0 . 1% SDS , pH = 8 , 3 ) . Proteins were transferred to a nitrocellulose membrane using transfer buffer ( 190 mM glycine , 25 mM Tris , 20% Methanol , 0 . 1% SDS ) for 90 min at 80V . Membranes were blocked for 1 hr with a solution of PBS-5% milk , 1% Tween ( PBST ) and further incubated overnight at 4C with appropriate primary antibodies diluted in PBST: rabbit anti-HA ( Abcam , cat #ab20084 ) , rabbit anti-RFP serum ( Herrera et al , 2014 ) and mouse anti-Tubulin beta ( Millipore , cat #MAB3408 ) . After three washes in PBST , membranes were incubated with Horseradish peroxidase-conjugated anti-rabbit IgG or anti-mouse IgG secondary antibodies ( Sigma-Aldrich , cat#GENA934-1ML and cat#GENA931 ) for 1 hr at room temperature and the signals detected by chemiluminescence using the Immobilon western chemiluminiscent HRP substrate ( Sigma-Aldrich , cat# WBKLS0100 ) . No statistical method was used to predetermine sample size . The experiments were not randomized . The investigators were not blinded to allocation during experiments or outcome assessment . Statistical analyses were performed using the GraphPad Prism six software ( GraphPad Software , Inc . ) . For in vivo experiments , cell counts were typically performed on 2–5 images per embryo and n values correspond to different embryos , except for the assessment of the modes of divisions where n values correspond to pools of embryos . For in vitro chromatin immunoprecipitation assays , n values represent the numbers of independent experiments performed . The n values are indicated in the corresponding figure legends . The normal distribution of the values was assessed by the Shapiro-Wilk normality test . Significance was then assessed with a two-sided unpaired t-test , one-way ANOVA + Tukey’s test or two-way ANOVA + Sidak’s test for data presenting a normal distribution , or alternatively with non-parametric Mann–Whitney or Kruskal-Wallis +Dunn’s multiple comparisons’ tests for non-normally distributed data . n . s: non-significant; *: p<0 . 05 or less , as indicated in individual figures . | The brain and spinal cord are made up of a range of cell types that carry out different roles within the central nervous system . Each type of neuron is uniquely specialized to do its job . Neurons are produced early during development , when they differentiate from a group of cells called neural progenitor cells . Within these groups , molecules called proneural proteins control which types of neurons will develop from the neural progenitor cells , and how many of them . Proneural proteins work by binding to specific patterns in the DNA , called E-boxes , with the help of E proteins . E proteins are typically understood to be passive partners , working with each different proneural protein indiscriminately . However , Le Dréau , Escalona et al . discovered that E proteins in fact have a much more active role to play . Using chick embryos , it was found that E proteins influence the way different proneural proteins bind to DNA . The E proteins have preferences for certain E-boxes in the DNA , just like proneural proteins do . The E proteins enhanced the activity of the proneural proteins that share their E-box preference , and reined in the activity of proneural proteins that prefer other E-boxes . As a result , the E proteins controlled the ability of these proteins to instruct neural progenitor cells to produce specific , specialized neurons , and thus ensured that the distinct types of neurons were produced in appropriate amounts . These findings will help shed light on the roles E proteins play in the development of the central nervous system , and the processes that control growth and lead to cell diversity . The results may also have applications in the field of regenerative medicine , as proneural proteins play an important role in cell reprogramming . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"developmental",
"biology"
] | 2018 | E proteins sharpen neurogenesis by modulating proneural bHLH transcription factors’ activity in an E-box-dependent manner |
Humans and animals can integrate sensory evidence from various sources to make decisions in a statistically near-optimal manner , provided that the stimulus presentation time is fixed across trials . Little is known about whether optimality is preserved when subjects can choose when to make a decision ( reaction-time task ) , nor when sensory inputs have time-varying reliability . Using a reaction-time version of a visual/vestibular heading discrimination task , we show that behavior is clearly sub-optimal when quantified with traditional optimality metrics that ignore reaction times . We created a computational model that accumulates evidence optimally across both cues and time , and trades off accuracy with decision speed . This model quantitatively explains subjects's choices and reaction times , supporting the hypothesis that subjects do , in fact , accumulate evidence optimally over time and across sensory modalities , even when the reaction time is under the subject's control .
Effective decision making in an uncertain , rapidly changing environment requires optimal use of all information available to the decision-maker . Numerous previous studies have examined how integrating multiple sensory cues—either within or across sensory modalities—alters perceptual sensitivity ( van Beers et al . , 1996; Ernst and Banks , 2002; Battaglia et al . , 2003; Fetsch et al . , 2009 ) . These studies generally reveal that subjects' ability to discriminate among stimuli improves when multiple sensory cues are available , such as visual and tactile ( van Beers et al . , 1996; Ernst and Banks , 2002 ) , visual and auditory ( Battaglia et al . , 2003 ) , or visual and vestibular ( Fetsch et al . , 2009 ) cues . The performance gains associated with cue integration are generally well predicted by models that combine information across senses in a statistically optimal manner ( Clark and Yuille , 1990 ) . Specifically , we consider cue integration to be optimal if the information in the combined , multisensory condition is the sum of that available from the separate cues ( see Supplementary file 1 for formal statement ) ( Clark and Yuille , 1990 ) . Previous studies and models share a common fundamental limitation: they only consider situations in which the stimulus duration is fixed and subjects are required to withhold their response until the stimulus epoch expires . In natural settings , by contrast , subjects usually choose for themselves when they have gathered enough information to make a decision . In such contexts , it is possible that subjects integrate multiple cues to gain speed or to increase their proportion of correct responses ( or some combination of effects ) , and it is unknown whether standard criteria for optimal cue integration apply . Indeed , using a reaction-time version of a multimodal heading discrimination task , we demonstrate here that human performance is markedly suboptimal when evaluated with standard criteria that ignore reaction times . Thus , the conventional framework for optimal cue integration is not applicable to behaviors in which decision times are under subjects' control . On the other hand , there is a large body of empirical studies that has focused on how multisensory integration affects reaction times , but these studies have generally ignored effects on perceptual sensitivity ( Colonius and Arndt , 2001; Otto and Mamassian , 2012 ) . Some of these studies have reported that reaction times for multisensory stimuli are faster than predicted by ‘parallel race’ models ( Raab , 1962; Miller , 1982 ) , suggesting that multisensory inputs are combined into a common representation . However , other groups have failed to replicate these findings ( Corneil et al . , 2002; Whitchurch and Takahashi , 2006 ) and it is unclear whether the sensory inputs are combined optimally . Thus , multisensory integration in reaction time experiments remains poorly understood , and there is no coherent framework for evaluating optimal decision making that incorporates both perceptual sensitivity and reaction times . We address this substantial gap in knowledge both theoretically and experimentally . For tasks based on information from a single sensory modality , diffusion models ( DMs ) have proven to be very effective at characterizing both the speed and accuracy of perceptual decisions , as well as speed/accuracy trade-offs ( Ratcliff , 1978; Ratcliff and Smith , 2004; Palmer et al . , 2005 ) ( where accuracy is used in the sense of percentage of correct responses ) . Here , we develop a novel form of DM that not only integrates evidence optimally over time but also across different sensory cues , providing an optimal decision model for multisensory integration in a reaction-time context . The model is capable of combining cues optimally even when the reliability of each sensory input varies as a function of time . We show that this model reproduces human subjects' behavior very well , thus demonstrating that subjects near-optimally combine momentary evidence across sensory modalities . The model also predicts the counterintuitive finding that discrimination thresholds are often increased during cue combination , and demonstrates that this departure from standard cue-integration theory is due to a speed-accuracy tradeoff . Overall , our findings provide a framework for extending cue-integration research to more natural contexts in which decision times are unconstrained and sensory cues vary substantially over time .
Traditional cue combination models predict that the discrimination threshold in the combined condition should be smaller than that of either unimodal condition ( Clark and Yuille , 1990 ) . With a fixed stimulus duration , this prediction has been shown to hold for visual/vestibular heading discrimination in both human and animal subjects ( Fetsch et al . , 2009 , 2011 ) , consistent with optimal cue combination . In contrast , the discrimination thresholds of subjects in our reaction-time task appear to be substantially sub-optimal . For the example subject of Figure 2A , psychometric functions in the combined condition lie between the visual and vestibular functions . Correspondingly , discrimination thresholds for the combined condition are intermediate between visual and vestibular thresholds for this subject , and for high coherences , are substantially greater than the optimal predictions ( Figure 2B ) . This pattern of results was consistent across subjects ( Figure 2C , Figure 2—figure supplement 1 ) . In no case did subjects feature a significantly lower discrimination threshold in the combined condition than the better of the two unimodal conditions ( p>0 . 57 , one-tailed , Supplementary file 2A ) . For the largest visual motion coherence ( 70% ) , all subjects except one showed thresholds in the combined condition that were significantly greater than visual thresholds and significant greater than optimal predictions of a conventional cue-integration scheme ( p<0 . 05 , Supplementary file 2A ) . These data lie in stark contrast to previous reports using fixed duration stimuli ( Fetsch et al . , 2009 , 2011 ) in which combined thresholds were generally found to improve compared to the unimodal conditions , as expected by standard optimal multisensory integration models . To summarize this contrast , we compare the ratio of observed to predicted thresholds in the combined condition for our subjects to human and monkey subjects performing a similar task in a fixed duration setting ( Fetsch et al . , 2009 ) . We found this ratio to be significantly greater for our subjects ( Figure 2C; two-sample t test , t ( 77 ) = 3 . 245 , p=0 . 0017 ) . This indicates that , with respect to predictions of standard multisensory integration models , our subjects performed significantly worse than those engaged in a similar fixed-duration task . A different picture emerges if we take not only discrimination thresholds but also reaction times into account . Short reaction times imply that subjects gather less information to make a decision , yielding greater discrimination thresholds . Longer reaction times may decrease thresholds , but at the cost of time . Consequently , if subjects decide more rapidly in the combined condition than the visual condition , they might feature higher discrimination thresholds in the combined condition even if they make optimal use of all available information . Thus , to assess if subjects perform optimal cue combination , we need to account for the timing of their decisions . Average reaction times depended on stimulus condition , motion coherence , and heading direction . In general , reaction times were faster for larger heading magnitudes , and reaction times in the vestibular condition were faster than those in the visual condition ( Figure 3 for subject D2 , Figure 3—figure supplement 1 for other subjects ) . In the combined condition , however , reaction times were much shorter than those seen for the visual condition and were comparable to those of the vestibular condition ( Figure 3 ) . Thus , subjects spent substantially more time integrating evidence in the visual condition , which boosted their discrimination performance when compared to the combined condition . Note also that discrimination thresholds in the combined condition were substantially smaller than vestibular thresholds , especially at 70% coherence ( Figures 2 and 3 ) . Thus , adding optic flow to a vestibular stimulus decreased the discrimination threshold with essentially no loss of speed . A similar overall pattern of results was observed for the other subjects ( Figure 3—figure supplement 1 ) . These data provide clear evidence that subjects made use of both visual and vestibular information to perform the reaction-time task , but the benefits of cue integration could not be appreciated by considering discrimination thresholds alone . 10 . 7554/eLife . 03005 . 006Figure 3 . Discrimination performance and reaction times for subject D2 . Behavioral data ( symbols with error bars ) and model fits ( lines ) are shown separately for each motion coherence . Top plot: reaction times as a function of heading; bottom plot: proportion of rightward choices as a function of heading . Mean reaction times are shown for correct trials , with error bars representing two SEM ( in some cases smaller than the symbols ) . Error bars on the proportion rightward choice data are 95% confidence intervals . Although reaction times are only shown for correct trials , the model is fit to data from both correct and incorrect trials . See Figure 3—figure supplement 1 for behavioral data and model fits for all subjects . Figure 3—figure supplement 2 shows the fitted model parameters per subject . DOI: http://dx . doi . org/10 . 7554/eLife . 03005 . 00610 . 7554/eLife . 03005 . 007Figure 3—figure supplement 1 . Psychometric functions , chronometric functions , and model fits for all subjects . Behavioral data ( symbols with error bars ) and model fits ( lines ) are , for clarity , shown separately for each different coherence of the visual motion stimulus . The reaction time shown is the mean reaction time for correct trials , with error bars showing two SEMs ( sometimes smaller than the symbols ) . Error bars on the proportion of rightward choices are 95% confidence intervals . Note that reaction times are shown only for correct trials , while the model is fit to both correct and incorrect trials . DOI: http://dx . doi . org/10 . 7554/eLife . 03005 . 00710 . 7554/eLife . 03005 . 008Figure 3—figure supplement 2 . Model parameters for fits of the optimal model and two alternative parameterizations . Based on the maximum likelihood parameters of full model fits for each subject , the four top plots show how drift rate and normalized bounds are assumed to depend on visual motion coherence . The solid lines show fits for the model described in the main text . The dashed lines show fits for an alternative parameterization with one additional parameter ( see Supplementary file 1 ) . The circles show the fits of a model that , instead of linking them by a parametric function , fits these drifts and bounds for each coherence separately . As can be seen , the parametric functions qualitatively match these independent fits . The bottom bar graphs show drift rate and bound for the vestibular modalities and fitted non-decision times for each subject , all for the model parameterization described in the text . All error bars show ±1 SD of the parameter posterior . Each color corresponds to a separate subject , with color scheme given by the bottom left bar graph . DOI: http://dx . doi . org/10 . 7554/eLife . 03005 . 008 To investigate whether subjects accumulate evidence optimally across both time and sensory modalities , we built a model that integrates visual and vestibular cues optimally to perform the heading discrimination task , and we compare predictions of the model to data from our human subjects . The model builds upon the structure of diffusion models ( DMs ) , which have previously been shown to account nicely for the tradeoff between speed and accuracy of decisions ( Ratcliff , 1978; Ratcliff and Smith , 2004; Palmer et al . , 2005 ) . Additionally , DMs are known to optimally integrate evidence over time ( Laming , 1968; Bogacz et al . , 2006 ) , given that the reliability of the evidence is time-invariant ( such that , at any point in time from stimulus onset , the stimulus provides the same amount of information about the task variable ) . However , DMs have neither been used to integrate evidence from several sources , nor to handle evidence whose reliability changes over time , both of which are required for our purposes . In the context of heading discrimination , a standard DM would operate as follows ( Figure 4A ) : consider a diffusing particle with dynamics given by x˙=ksin ( h ) +η ( t ) , where h is the heading direction , k is a positive constant relating particle drift to heading direction , and η ( t ) is unit variance Gaussian white noise . The particle starts at x ( 0 ) = 0 , drifts with an average slope given by ksin ( h ) , and diffuses until it hits either the upper bound θ or the lower bound −θ , corresponding to rightward and leftward choices , respectively . The decision time is determined by when the particle hits a bound . Larger |h|'s lead to shorter decision times and more correct decisions because the drift rate is greater . Lower bound levels , |θ| , also lead to shorter decision times but more incorrect decisions . Errors ( hitting bound θ when h < 0 , or hitting bound −θ when h > 0 ) can occur due to the stochasticity of particle motion , which is meant to capture the variability of the momentary sensory evidence . The Fisher information in x ( t ) regarding h , a measure of how much information x ( t ) provides for discriminating heading ( Papoulis , 1991 ) , is Ix ( sin ( h ) ) = k2 per second , showing that k is a measure of the subject's sensitivity to changes in heading direction . This sensitivity depends on the subject's effectiveness in estimating heading from the cue , which in turn is influenced by the reliability of the cue itself ( e . g . , coherence ) . 10 . 7554/eLife . 03005 . 009Figure 4 . Extended diffusion model ( DM ) for heading discrimination task . ( A ) A drifting particle diffuses until it hits the lower or upper bound , corresponding to choosing ‘left’ or ‘right’ respectively . The rate of drift ( black arrow ) is determined by heading direction . The time at which a bound is hit corresponds to the decision time . 10 particle traces are shown for the same drift rate , corresponding to one incorrect and nine correct decisions . ( B ) Despite time-varying cue sensitivity , optimal temporal integration of evidence in DMs is preserved by weighting the evidence by the momentary measure of its sensitivity . The DM representing the combined condition is formed by an optimal sensitivity-weighted combination of the DMs of the unimodal conditions . DOI: http://dx . doi . org/10 . 7554/eLife . 03005 . 009 Now consider both a visual ( vis ) and a vestibular ( vest ) source of evidence regarding h , x˙vis=kvis ( c ) sin ( h ) +ηvis ( t ) and x˙vest=kvestsin ( h ) +ηvest ( t ) , where kvis ( c ) indicates that the sensitivity to the cue in the visual modality depends on motion coherence , c . Combining these two sources of evidence by a simple sum , x˙vis+x˙vest , would amount to adding noise to x˙vest for low coherences ( kvis ( c ) ≈ 0 ) , which is clearly suboptimal . Rather , it can be shown that the two particle trajectories are combined optimally by weighting their rates of change in proportion to their relative sensitivities ( see Supplementary file 1 for derivation ) : ( 1 ) x˙comb=kvis2 ( c ) kvis2 ( c ) +kvest2x˙vis+kvest2kvis2 ( c ) +kvest2x˙vest . This allows us to model the combined condition by a single new DM , x˙comb=kcomb ( c ) sin ( h ) +ηcomb ( t ) , which is optimal because it preserves all information contained in both xvis and xvest ( Figure 4B; see ‘Materials and methods’ and Supplementary file 1 for a formal treatment ) . The sensitivity ( drift rate coefficient ) in the combined condition , ( 2 ) kcomb ( c ) =kvis2 ( c ) +kvest2 , is a combination of the sensitivities of the unimodal conditions and is therefore always greater than the largest unimodal sensitivity . So far we have assumed that the reliability of each cue is time-invariant . However , as the motion velocity changes over time , so does the amount of information about h provided by each cue , and with it the subject's sensitivity to changes in h . For the vestibular and visual conditions , motion acceleration a ( t ) and motion velocity v ( t ) , respectively , are assumed to be the physical quantities that modulate cue sensitivity ( ‘Materials and methods’ and ‘Discussion’ ) . To account for these dynamics , the DMs are modified to x˙vest=a ( t ) kvestsin ( h ) +ηvest ( t ) and x˙vis=v ( t ) kvis ( c ) sin ( h ) +ηvis ( t ) . Note that once the drift rate in a DM changes with time , it generally loses its property of integrating evidence optimally over time . For example , at the beginning of each trial when motion velocity is low , x˙vis is dominated by noise and integrating x˙vis is fruitless . Fortunately , weighting the momentary visual evidence , x˙vis , by the velocity profile recovers optimality of the DM ( ‘Materials and methods’ ) . This temporal weighting causes the visual evidence to contribute more at high velocities , while the noise is downweighted at low velocities . Similarly , vestibular evidence is weighted by the time course of acceleration . The new , weighted particle trajectories are described by the DMs X˙vis=v ( t ) x˙vis and X˙vest=a ( t ) x˙vest . The two unimodal DMs are combined as before , resulting in the combined DM given by X˙comb=d ( t ) x˙comb , where the sensitivity profile d ( t ) is a weighted combination of the unimodal sensitivity profiles , ( 3 ) d ( t ) =kvis2 ( c ) kcomb2 ( c ) v2 ( t ) +kvest2kcomb2 ( c ) a2 ( t ) . ( Figure 4B; see Supplementary file 1 for derivation ) . These modifications to the standard DM are sufficient to integrate evidence optimally across time and sensory modalities , even as the sensitivity to the evidence changes over time . The model assumes that subjects know their cue sensitivities , kvis ( c ) and kvest , as well as the temporal sensitivity profiles , a ( t ) and v ( t ) , of each stimulus . In this respect , our model provides an upper bound on performance , since subjects may not have perfect knowledge of these variables , especially since stimulus modalities and visual motion coherence values are randomized across trials ( ‘Discussion’ ) . We tested whether subjects combined evidence optimally across both time and cues by evaluating how well the model outlined above could explain the observed behavior . The bounds , θ , of the modified DM , and the sensitivity parameters ( kvis , kvest and kcomb ) , were allowed to vary between the visual , vestibular , and combined conditions . Varying the bound was essential to capture the deviation of the discrimination threshold in the combined condition from that predicted by traditional cue combination models ( Figure 2 ) . Indeed , this discrimination threshold is inversely proportional to bound and sensitivity ( see Supplementary file 1 ) . Since the sensitivity in the bimodal condition is not a free parameter ( it is determined by Equation 2 ) , the height of the bound is the only parameter that could modulate the discrimination thresholds . The noise terms ηvis and ηvest play crucial roles in the model , as they relate to the reliability of the momentary sensory evidence . To specify the manner in which such noise may depend on motion coherence , we relied on fundamental assumptions about how optic flow stimuli are represented by the brain . We assumed that heading is represented by a neural population code in which neurons have heading tuning curves that , within the range of heading tested in this experiment ( ±16° , Figure 5A ) , differ in their heading preferences but have similar shapes . This is broadly consistent with data from area MSTd ( Fetsch et al . , 2011 ) , but the exact location of such a code is not important for our argument . For low coherence , motion energy in the stimulus is almost uniform for all heading directions , such that all neurons in the population fire at approximately the same rate ( Figure 5A , dark blue curve ) . For high coherence , population neural activity is strongly peaked around the actual heading direction ( Figure 5A , cyan curve ) ( Morgan et al . , 2008; Fetsch et al . , 2011 ) . 10 . 7554/eLife . 03005 . 010Figure 5 . Scaling of momentary evidence statistics of the diffusion model ( DM ) with coherence . ( A ) Assumed neural population activity giving rise to the DM mean and variance of the momentary evidence , and their dependence on coherence . Each curve represents the activity of a population of neurons with a range of heading preferences , in response to optic flow with a particular coherence and a heading indicated by the dashed vertical line . ( B ) Expected pattern of reaction times if variance is independent of coherence . If neither the DM bound nor the DM variance depend on coherence , the DM predicts the same decision time for all small headings , regardless of coherence . This is due to the DM drift rate , kvis ( c ) sin ( h ) being close to 0 for small headings , h≈0 , independent of the DM sensitivity kvis ( c ) . ( C ) Expected pattern of reaction times when variance scales with coherence . If both DM sensitivity and DM variance scale with coherence while the bound remains constant , the DM predicts different decision times across coherences , even for small headings . Greater coherence causes an increase in variance , which in turn causes the bound to be reached more quickly for higher coherences , even if the heading , and thus the drift rate , is small . DOI: http://dx . doi . org/10 . 7554/eLife . 03005 . 010 Based on this representation , and assuming that the response variability of the neurons belongs to the exponential family with linear sufficient statistics ( Ma et al . , 2006 ) ( an assumption consistent with in vivo data [Graf et al . , 2011] ) , heading discrimination can be performed optimally by a weighted sum of the activity of all neurons , with weights monotonically related to the preferred heading of each neuron . For a straight forward heading , h = 0 , this sum should be 0 , and for h > 0 ( or h < 0 ) it should be positive ( or negative ) , thus sharing the basic properties of the momentary evidence , x˙ , in our DM . This allowed us to deduce the mean and variance of the momentary evidence driving x˙ , based on what we know about the neural responses . First , the sensitivity , kvis ( c ) , which determines how optic flow modulates the mean drift rate of x˙ , scales in proportion with the ‘peakedness’ of the neural activity , which in turn is proportional to coherence . We assumed a functional form of kvis ( c ) given by aviscγvis , where avis and γvis are positive parameters . Second , the variance of x˙ is assumed to be the sum of the variances of the neural responses . Since experimental data suggest that the variance of these responses is proportional to their firing rate ( Tolhurst et al . , 1983 ) , the sum of the variances is proportional to the area underneath the population activity profile ( Figure 5A ) . Based on the experimental data of Britten et al . ( Heuer and Britten , 2007 ) , this area was assumed to scale roughly linearly with coherence , such that the variance of x˙ is proportional to 1+bviscγvis with free parameters bvis and γvis , the latter of which captures possible deviations from linearity . We further assumed the DM bound to be independent of coherence , and given by θσ , vis . Thus , the effect of motion coherence on the momentary evidence in the DM was modeled by four parameters: avis , γvis , bvis , and θσ , vis . The above scaling of the diffusion variance by coherence , which is a consequence of the neural code for heading , makes an interesting prediction: reaction times for headings near straight ahead should be inversely proportional to coherence in the visual condition , even though the mean drift rate , kvis ( c ) sin ( h ) , is very close to 0 . This is indeed what we observed: subjects tended to decide faster for higher coherences even when h ≈ 0 ( Figure 3 , Figure 3—figure supplement 1 ) . This aspect of the data can only be captured by the model if the DM variance is allowed to change with coherence ( Figure 5B , C ) . To summarize , in the combined condition , the diffusion variance was assumed to be proportional to 1+bcombcγcomb , while the bound was fixed at θσ , comb . By contrast , the diffusion rate ( sensitivity ) cannot be modeled freely but rather needs to obey kcomb ( c ) =kvis2 ( c ) +kvest2 in order to ensure optimal cue combination . The sensitivity kvest and bound θσ , vest in the vestibular condition do not depend on motion coherence and were thus model parameters that were fitted directly . Observed reaction times were assumed to be composed of the decision time and some non-decision time . The decision time is the time from the start of integrating evidence until a decision is made , as predicted by the diffusion model . The non-decision time includes the motor latency and the time from stimulus onset to the start of integrating evidence . As the latter can vary between different modalities , we allowed it to differ between visual , vestibular , and combined conditions , but not for different coherences , thus introducing the model parameters tnd , vis , tnd , vest , and tnd , comb . Although the fitted non-decision times were similar across stimulus conditions for most subjects ( Figure 3—figure supplement 2 ) , a model assuming a single non-decision time resulted in a small but significant decrease in fit quality ( Figure 7—figure supplement 2A ) . Overall , 12 parameters were used to model cue sensitivities , bounds , variances , and non-decision times in all conditions , and these 12 parameters were used to fit 312 data points for subjects that were tested with 6 coherences ( 168 data points for the three-coherence version ) . An additional 14 parameters ( 8 parameters for the three-coherence version; one bias parameter per coherence/condition , one lapse parameter across all condition ) controlled for biases in the motion direction percept and for lapses of attention that were assumed to lead to random choices ( ‘Materials and methods’ ) . Although these additional parameters were necessary to achieve good model fits ( Figure 7—figure supplement 2A ) , it is critical to note that they could not account for differences in heading thresholds or reaction times across stimulus conditions . As such , the additional parameters play no role in determining whether subjects perform optimal multisensory integration . Alternative parameterizations of how drift rates and bounds depend on motion coherence yielded qualitatively similar results , but caused the model fits to worsen decisively ( Supplementary file 1; Figure 7—figure supplement 2A ) . Critically , our model predicts that the unimodal sensitivities kvis ( c ) and kvest relate to the combined value by kcombpredicted ( c ) =kvis ( c ) 2+kvest2 , if subjects accumulate evidence optimally across cues . To test this prediction , we fitted separately the unimodal and combined sensitivities , kvis ( c ) , kvest and kcomb to the complete data set from each individual subject using maximum likelihood optimization ( ‘Materials and methods’ ) , and then compared the fitted values of kcomb to the predicted values , kcombpredicted ( c ) . Predicted and observed sensitivities for the combined condition are virtually identical ( Figure 6 ) , providing strong support for near-optimal cue combination across both time and cues . Remarkably , for low coherences at which optic flow provides no useful heading information , the sensitivity in the combined condition was not significantly different from that of the vestibular condition ( Figure 6 ) . Thus , subjects were able to completely suppress noisy visual information and rely solely on vestibular input , as predicted by the model . 10 . 7554/eLife . 03005 . 011Figure 6 . Predicted and observed sensitivity in the combined condition . The sensitivity parameter measures how sensitive subjects are to a change of heading . The solid red line shows predicted sensitivity for the combined condition , as computed from the sensitivities of the unimodal conditions ( dashed lines ) . The combined sensitivity measured by fitting the model to each coherence separately ( red squares ) does not differ significantly from the optimal prediction , providing strong support to the hypothesis that subjects accumulate evidence near-optimally across time and cues . Data are averaged across datasets ( except 0% , 12% , 51% coherence: only datasets B2 , D2 , F2 ) , with shaded areas and error bars showing the 95% CIs . DOI: http://dx . doi . org/10 . 7554/eLife . 03005 . 011 Having established that cue sensitivities combine according to Equation 2 , the model was then fit to data from each individual subject under the assumption of optimal cue combination . Model fits are shown as solid curves for example subject D2 ( Figure 3 ) , as well as for all other subjects ( Figure 3—figure supplement 1 ) . Sensitivity parameters , bounds , and non-decision times resulting from the fits are also shown for each subject , condition , and coherence ( Figure 3—figure supplement 2 ) . For 8 of 10 datasets , the model explains more than 95% of the variance in the data ( adjusted R2 > 0 . 95 ) , providing additional evidence for near-optimal cue combination across both time and cues ( Figure 7A ) . The subjects associated with these datasets show a clear decrease in reaction times with larger |h| , and this effect is more pronounced in the visual condition than in the vestibular and combined conditions ( Figure 3 , Figure 3—figure supplement 1 ) . The remaining two subjects ( C and F ) feature qualitatively different behavior and lower R2 values of approximately 0 . 80 and 0 . 90 , respectively ( Figure 3—figure supplement 1 ) . These subjects showed little decline in reaction times with larger values of |h| , and their mean reaction times were more similar across the visual , vestibular and combined conditions . 10 . 7554/eLife . 03005 . 012Figure 7 . Model goodness-of-fit and comparison to alternative models . ( A ) Coefficient of determination ( adjusted R2 ) of the model fit for each of the ten datasets . ( B ) Bayes factor of alternative models compared to the optimal model . The abscissa shows the base-10 logarithm of the Bayes factor of the alterative models vs the optimal model ( negative values mean that the optimal model out-performs the alternative model ) . The gray vertical line close to the origin ( at a value of −2 on the abscissa ) marks the point at which the optimal model is 100 times more likely than each alternative , at which point the difference is considered ‘decisive’ ( Jeffreys , 1998 ) . Only the ‘separate k's‘ model has more parameters than the optimal model , but the Bayes factor indicates that the slight increase in goodness-of-fit does not justify the increased degrees of freedom . The ‘no cue weighting’ model assumes that visual and vestibular cues are weighted equally , independent of their sensitivities . The ‘weighting by acceleration’ and ‘weighting by velocity’ models assume that the momentary evidence of both cues is weighted by the acceleration and velocity profile of the stimulus , respectively . The ‘no temporal weighting’ model assumes that the evidence is not weighted over time according to its sensitivity . The ‘no cue/temporal weighting’ model lacks both weighting of cues by sensitivity and weighting by temporal profile . All of the tested alternative models explain the data decisively worse than the optimal model . Figure 7—figure supplement 1 shows how individual subjects contribute to this model comparison , and the results of a more conservative Bayesian random-effects model comparison that supports same conclusion . Figure 7—figure supplement 2 compares the proposed model to ones with alternative parameterizations . DOI: http://dx . doi . org/10 . 7554/eLife . 03005 . 01210 . 7554/eLife . 03005 . 013Figure 7—figure supplement 1 . Model comparison per subject , and random-effects model comparison . ( A ) Shows the contribution of each subject to the model comparison shown in Figure 7B . As in Figure 7B , the grey line shows the threshold above which the alternative models provide a decisively worse ( if negative ) or better ( if positive ) model fit . As can be seen , the model comparison is mostly consistent across subjects , except for models that weight both modalities either by acceleration or velocity only . Even in these cases , pooling across subjects leads to a decisively worse fit of the alternative model when compared to the optimal model ( Figure 7 ) . ( B ) and ( C ) Show the results of a random-effects Bayesian model comparison ( Stephan et al . , 2009 ) . This model comparison infers the probability of each model to have generated the behavior observed for each subject , and is less sensitive to model fit outliers than the fixed-effects comparison shown in Figure 7B ( e . g . , a single subject might strongly support an otherwise unsupported model , which could skew the overall comparison ) . ( B ) Shows the inferred distribution over all compared models , and supports the optimal model with exceedance probability p≈0 . 664 ( probability that the optimal model is more likely that any other model ) . This random-effects comparison causes models with very similar predictions to share some probability mass—in our case the optimal model and the model assuming evidence weighting by the velocity time-course . In ( C ) we perform the same comparison without the ‘weighting by velocity’ model , in which case the exceedance probability supporting the optimal model rises to p≈0 . 953 . DOI: http://dx . doi . org/10 . 7554/eLife . 03005 . 01310 . 7554/eLife . 03005 . 014Figure 7—figure supplement 2 . Model comparison for models with alternative parameterization . ( A ) Compares the optimal model as described in the main text to various alternative models . The first model changes how drifts and bounds relate to coherence ( see Supplementary file 1 ) , and introduces one additional parameter . The second model fits drifts and bounds separately for all coherences . The other models either use a single non-decision time ( instead of one or each modality ) , no heading biases , or a combination of both . The figure shows the Bayes factor , illustrating that in all cases the alternative models are decisively worse ( grey line close to origin indicating threshold ) than the original model . ( B and C ) Show the overall model goodness-of-fit ( left panels ) of two model that used an alternative parameterization of how drifts and bounds depend on coherence ( see ( A ) ) . Furthermore , it compares these models , which still perform optimal evidence accumulation across both time and cues , to sub-optimal models ( right panels ) that do not ( except ‘separate k's’ , which is potentially optimal ) . These figures are analogous to Figure 7 and show that neither change of parameterization qualitatively changes our conclusions . DOI: http://dx . doi . org/10 . 7554/eLife . 03005 . 014 Critically , the model nicely captures the observation that the psychophysical threshold in the combined condition is typically greater than that for the visual condition , despite near-optimal combination of momentary evidence from the visual and vestibular modalities ( e . g . , Figure 3 , 70% coherence , Figure 2—figure supplement 1 , Figure 3—figure supplement 1 ) . Thus , the model fits confirm quantitatively that apparent sub-optimality in psychophysical thresholds can arise even if subjects combine all cues in a statistically optimal manner , emphasizing the need for a computational framework that incorporates both decision accuracy and speed . To further assess and validate the critical design features of our modified DM , we evaluated six alternative ( mostly sub-optimal ) versions of the model to see if these variants are able to explain the data equally well . We compared these variants to the optimal model using Bayesian model comparison , which trades off fit quality with model complexity to determine whether additional parameters significantly improve the fit ( Goodman , 1999 ) . With regard to optimality of cue integration across modalities , we examined two model variants . The first variant ( also used to generate Figure 6 ) eliminates the relationship , kcomb ( c ) =kvis2 ( c ) +kvest2 ( Equation 2 ) , between the sensitivity parameters in the combined and single-cue conditions . Instead , this variant allows independent sensitivity parameters for the combined condition at each coherence , thus introducing one additional parameter per coherence . Since this variant is strictly more general than the optimal model , it must fit the data at least as well . However , if the subjects' behavior is near optimal , the additional degrees of freedom in this variant should not improve the fit enough to justify the addition of these parameters . This is indeed what we found by Bayesian model comparison ( Figure 7B , ‘separate k's’ ) , which shows the optimal model to be ∼1070 times more likely than the variant with independent values of kcomb ( c ) . This is well above the threshold value that is considered to provide ‘decisive’ evidence in favor of the optimal model ( we use Fisher's definition of decisive [Jeffreys , 1998] according to which a model is said to be decisively better if it is >100 times more likely to have generated the data ) . The second model variant had the same number of parameters as the optimal model , but assumed that the cues are always weighted equally . Evidence in the combined condition was given by the simple average , X˙comb=12 ( X˙vis+X˙vest ) , ignoring cue sensitivities . The resulting fits ( Figure 7B , ‘no cue weighting’ ) are also decisively worse than those of the optimal model . Together , these model variants strongly support the hypothesis that subjects weight cues according to their relative sensitivities , as given by Equation 2 . These effects were largely consistent across individual subjects ( Figure 7—figure supplement 1A ) . To test the other key assumption of our model—that subjects temporally weight incoming evidence according to the profile of stimulus information—we tested three model variants that modified how temporal weighting was performed without changing the number of parameters in the model . If we assumed that the temporal weighting of both modalities followed the acceleration profile of the stimulus while leaving the model otherwise unchanged , the model fit worsened decisively ( Figure 7B , ‘weighting by acceleration’ ) . Assuming that the weighting of both modalities followed the velocity profile of the stimulus also decisively reduced fit quality ( Figure 7B , ‘weighting by velocity’ ) , although this effect was not consistent across subjects ( Figure 7—figure supplement 1A ) . If we completely removed temporal weighting of cues from the model , fits were dramatically worse than the optimal model ( Figure 7B , ‘no temporal weighting’ ) . Finally , for completeness , we also tested a model variant that neither performs temporal weighting of cues nor considers the relative sensitivity to the cues . Again , this model variant fit the data decisively worse than the optimal model ( Figure 7B , ‘no cue/temporal weighting’ ) . Thus , subjects seem to be able to take into account their sensitivity to the evidence across time as well as across cues . All of these model comparisons received further support from a more conservative random-effects Bayesian model comparison , shown in Figure 7—figure supplement 1B , C . Finally , we also considered if a parallel race model could account for our data . The parallel race model ( Raab , 1962; Miller , 1982; Townsend and Wenger , 2004; Otto and Mamassian , 2012 ) postulates that the decision in the combined condition emerges from the faster of two independent races toward a bound , one for each sensory modality . Because it does not combine information across modalities , the parallel race model predicts that decisions in the combined condition are caused by the faster modality . Consequently , choices in the combined condition are unlikely to be more correct ( on average ) than those of the faster unimodal condition . For all but one subject , the vestibular modality is substantially faster , even when compared to the visual modality at high coherence and controlling for the effect of heading direction ( 2-way ANOVA , p<0 . 0001 for all subjects except C ) . Critically , all of these subjects feature significantly lower psychophysical thresholds in the combined condition than in the vestibular condition ( p<0 . 039 for all subjects except subject C , p=0 . 210 , Supplementary file 2A ) . Furthermore , we performed standard tests ( Miller's bound and Grice's bound ) that compare the observed distribution of reaction times with that predicted by the parallel race model ( Miller , 1982; Grice et al . , 1984 ) . These tests revealed that all but two subjects made significantly slower decisions than predicted by the parallel race model for most coherence/heading combinations ( p<0 . 05 for all subjects except subjects F and B2; Supplementary file 2B ) , and no subject was faster than predicted ( p>0 . 05 , all subjects; Supplementary file 2B ) . Based on these observations , we can reject the parallel race model as a viable hypothesis to explain the observed behavior .
We have shown that , when subjects are allowed to choose how long to accumulate evidence in a cue integration task , their behavior no longer follows the standard predictions of optimal cue integration theory that normally apply when stimulus presentation time is controlled by the experimenter . Particularly , they feature worse discrimination performance ( higher psychophysical thresholds ) in the combined condition than would be predicted from the unimodal conditions—in some cases even worse than the better of the two unimodal conditions . This occurs because subjects tend to decide more quickly in the combined condition than in the more sensitive unimodal condition and thus have less time to accumulate evidence . This indicates that a more general definition of optimal cue integration must incorporate reaction times . Indeed , subjects' behavior could be reproduced by an extended diffusion model that takes into account both speed and accuracy , thus suggesting that subjects accumulate evidence across both time and cues in a statistically near-optimal manner ( i . e . , with minimal information loss ) despite their reduced discrimination performance in the combined condition . Previous work on optimal cue integration ( e . g . , Ernst and Banks , 2002; Battaglia et al . , 2003; Knill and Saunders , 2003; Fetsch et al . , 2009 ) was based on experiments that employed fixed-duration stimuli and was thus able to ignore how subjects accumulate evidence over time . Moreover , previous work relied on the implicit assumption that subjects make use of all evidence throughout the duration of the stimulus . However , this assumption need not be true and has been shown to be violated even for short presentation durations ( Mazurek et al . , 2003; Kiani et al . , 2008 ) . Therefore , apparent sub-optimality in some previous studies of cue integration or in some individual subjects ( Battaglia et al . , 2003; Fetsch et al . , 2009 ) might be attributable to either truly sub-optimal cue combination , to subjects halting evidence accumulation before the end of the stimulus presentation period , or to the difficulty in estimating stimulus processing time ( Stanford et al . , 2010 ) . Unfortunately , these potential causes cannot be distinguished using a fixed-duration task . Allowing subjects to register their decisions at any time during the trial alleviates this potential confound . We model subjects' decision times by assuming an accumulation-to-bound process . In the multisensory context , this raises the question of whether evidence accumulation is bounded for each modality separately , as assumed by the parallel race model , or whether evidence is combined across modalities before being accumulated toward a single bound , as in co-activation models and our modified diffusion model . Based on our behavioral data , we can rule out parallel race models , as they cannot explain lower psychophysical thresholds ( better sensitivity ) in the combined condition relative to the faster vestibular condition . Further evidence against such models is provided by neurophysiological studies which demonstrate that visual and vestibular cues to heading converge in various cortical areas , including areas MSTd ( Gu et al . , 2006 ) , VIP ( Schlack et al . , 2005; Chen et al . , 2011b ) , and VPS ( Chen et al . , 2011a ) . Activity in area MSTd can account for sensitivity-based cue weighting in a fixed-duration task ( Fetsch et al . , 2011 ) , and MSTd activity is causally related to multi-modal heading judgments ( Britten and van Wezel , 1998 , 2002; Gu et al . , 2012 ) . These physiological studies strongly suggest that visual and vestibular signals are integrated in sensory representations prior to decision-making , inconsistent with parallel race models . Our model makes the assumption that sensory signals are integrated prior to decision-making and is in this sense similar to co-activation models that have been used previously to model reaction times in multimodal settings ( Miller , 1982; Corneil et al . , 2002; Townsend and Wenger , 2004 ) . However , it differs from these models in important aspects . First , co-activation models have been introduced to explain reaction times that are faster than those predicted by parallel race models ( Raab , 1962; Miller , 1982 ) . Our subjects , in contrast , feature reaction times that are slower than those of parallel race models in almost all conditions ( Supplementary file 2B ) . We capture this effect by an elevated effective bound in the combined condition as compared to the faster vestibular condition , such that cue combination remains optimal despite longer reaction times . Second , co-activation models usually combine inputs from the different modalities by a simple sum ( e . g . , Townsend and Wenger , 2004 ) . This entails adding noise to the combined signal if the sensitivity to one of the modalities is low , which is detrimental to discrimination performance . In contrast , we show that different cues need to be weighted according to their sensitivities to achieve statistically optimally integration of multisensory evidence at each moment in time ( Equation 2 ) . Another alternative to co-activation models are serial race models , which posit that the race corresponding to one cue needs to be completed before the other one starts ( e . g . , Townsend and Wenger , 2004 ) . These models can be ruled out by observing that they predict reaction times in the combined condition to be longer than those in the slower of the two unimodal conditions . This is clearly violated by the subjects' behavior . Optimal accumulation of evidence over time requires the momentary evidence to be weighted according to its associated sensitivity . For the vestibular modality , we assume that the temporal profile of sensitivity to the evidence follows acceleration . This may appear to conflict with data from multimodal areas MSTd , VIP , and VPS , where neural activity in response to self-motion reflects a mixture of velocity and acceleration components ( Fetsch et al . , 2010; Chen et al . , 2011a ) . Note , however , that the vestibular stimulus is initially encoded by otolith afferents in terms of acceleration ( Fernandez and Goldberg , 1976 ) . Thus , any neural representation of vestibular stimuli in terms of velocity requires a temporal integration of the acceleration signal , and this integration introduces temporal correlations into the signal . As a consequence , a neural response that is maximal at the time of peak stimulus velocity does not imply a simultaneous peak in the information coded about heading direction . Rather , information still follows the time course of its original encoding , which is in terms of acceleration . In contrast , the time course of the sensitivity to the visual stimulus is less clear . For our model we have intuitively assumed it to follow the velocity profile of the stimulus , as information per unit time about heading certainly increases with the velocity of the optic flow field , even when there is no acceleration . This assumption is supported by a decisively worse model fit if we set the weighting of the visual momentary evidence to follow the acceleration profile ( Figure 7B , ‘weighted by acceleration’ ) . Nonetheless , we cannot completely exclude any contribution of acceleration components to visual information ( Lisberger and Movshon , 1999; Price et al . , 2005 ) . In any case , our model fits make clear that temporal weighting of vestibular and visual inputs is necessary to predict behavior when stimuli are time-varying . The extended DM model described here makes the strong assumption that cue sensitivities are known before combining information from the two modalities , as these sensitivities need to be known in order to weight the cues appropriately . As only the sensitivity to the visual stimulus changes across trials in our experiment , it is possible that subjects can estimate their sensitivity ( as influenced by coherence ) during the initial low-velocity stimulus period ( Figure 1C ) in which heading information is minimal but motion coherence is salient . Thus , for our task , it is reasonable to assume that subjects can estimate their sensitivity to cues . We have recently begun to consider how sensitivity estimation and cue integration can be implemented neurally . The neural model ( Onken et al . , 2012 . Near optimal multisensory integration with nonlinear probabilistic population codes using divisive normalization . The Society for Neuroscience annual meeting 2012 ) estimates the sensitivity to the visual input from motion sensitive neurons and uses this estimate to perform near-optimal multisensory integration with generalized probabilistic population codes ( Ma et al . , 2006; Beck et al . , 2008 ) using divisive normalization . We intend to extend this model to the integration of evidence over time to predict neural responses ( e . g . , in area LIP ) that should roughly track the temporal evolution of the decision variable ( xcomb ( t ) , ‘Materials and methods’ ) in the DM model . This will make predictions for activity in decision-making areas that can be tested in future experiments . In closing , our findings establish that conventional definitions of optimality do not apply to cue integration tasks in which subjects’ decision times are unconstrained . We establish how sensory evidence should be weighted across modalities and time to achieve optimal performance in reaction-time tasks , and we show that human behavior is broadly consistent with these predictions but not with alternative models . These findings , and the extended diffusion model that we have developed , provide the foundation for building a general understanding of perceptual decision-making under more natural conditions in which multiple cues vary dynamically over time and subjects make rapid decisions when they have acquired sufficient evidence .
Seven subjects ( 3 males ) aged 23–38 years with normal or corrected-to-normal vision and no history of vestibular deficits participated in the experiments . All subjects but one were informed of the purposes of the study . Informed consent was obtained from all participants and all procedures were reviewed and approved by the Washington University Office of Human Research Protections ( OHRP ) , Institutional Review Board ( IRB; IRB ID# 201109183 ) . Consent to publish was not obtained in writing , as it was not required by the IRB , but all subjects were recruited for this purpose and approved verbally . Of these subjects , three ( subjects B , D , F; 1 male ) participated in a follow-up experiment roughly 2 years after the initial data collection , with six coherence levels instead of the original three . The six-coherence version of their data is referred to as B2 , D2 , and F2 . Procedures for the follow-up experiment were approved by the Institutional Review Board for Human Subject Research for Baylor College of Medicine and Affiliated Hospitals ( BCM IRB , ID# H-29411 ) and informed consent and consent to publish was given again by all three subjects . The apparatus , stimuli , and task design have been described in detail previously ( Fetsch et al . , 2009; Gu et al . , 2010 ) , and are briefly summarized here . Subjects were seated comfortably in a padded racing seat that was firmly attached to a 6-degree-of-freedom motion platform ( MOOG , Inc ) . A 3-chip DLP projector ( Galaxy 6; Barco , Kortrijk , Belgium ) was mounted on the motion platform behind the subject and front-projected images onto a large ( 149 × 127 cm ) projection screen via a mirror mounted above the subject’s head . The viewing distance to the projection screen was ∼70 cm , thus allowing for a field of view of ∼94° × 84° . Subjects were secured to the seat using a 5-point racing harness , and a custom-fitted plastic mask immobilized the head against a cushioned head mount . Seated subjects were enclosed in a black aluminum superstructure , such that only the display screen was visible in the darkened room . To render stimuli stereoscopically , subjects wore active stereo shutter glasses ( CrystalEyes 3; RealD , Beverly Hills , CA ) which restricted the field of view to ∼90° × 70° . Subjects were instructed to look at a centrally-located , head-fixed target throughout each trial . Sounds from the motion platform were masked by playing white noise through headphones . Behavioral task sequences and data acquisition were controlled by Matlab and responses were collected using a button box . Visual stimuli were generated by an OpenGL accelerator board ( nVidia Quadro FX1400 ) , and were plotted with sub-pixel accuracy using hardware anti-aliasing . In the visual and combined conditions , visual stimuli depicted self-translation through a 3D cloud of stars distributed uniformly within a virtual space 130 cm wide , 150 cm tall , and 75 cm deep . Star density was 0 . 01/cm3 , with each star being a 0 . 5 cm × 0 . 5 cm triangle . Motion coherence was manipulated by randomizing the three-dimensional location of a percentage of stars on each display update while the remaining stars moved according to the specified heading . The probability of a single star following the trajectory associated with a particular heading for N video updates is therefore ( c/100 ) N , where c denotes motion coherence ( ranging from 0–100% ) . At the largest coherence used here ( 70% ) , there is only a 3% probability that a particular star would follow the same trajectory for 10 display updates ( 0 . 17 s ) . Thus , it was practically not possible for subjects to track the trajectories of individual stars . This manipulation degraded optic flow as a heading cue and was used to manipulate visual cue reliability in the visual and combined conditions . ‘Zero’ coherence stimuli had c set to 0 . 1 , which was practically indistinguishable from c = 0 , but allowed us to maintain a precise definition of the correctness of the subject's choice . In all stimulus conditions , the task was a single-interval , two-alternative forced choice ( 2AFC ) heading discrimination task . In each trial , human subjects were presented with a translational motion stimulus in the horizontal plane ( Gaussian velocity profile; peak velocity , 0 . 403 m/s; peak acceleration , 0 . 822 m/s2; total displacement , 0 . 3 m; maximum duration , 2 s ) . Heading was varied in small steps around straight ahead ( ±0 . 686° , ±1 . 96° , ±5 . 6° , ±16° ) and subjects were instructed to report ( by a button press ) their perceived heading ( leftward or rightward relative to an internal standard of straight ahead ) as quickly and accurately as possible . In the visual and combined conditions , cue reliability was varied across trials by randomly choosing the motion coherence of the visual stimulus from among either a group of three values ( 25% , 37% , and 70% , subjects A–G ) or a group of six values ( 0% , 12% , 25% , 37% , 51% , and 70% , subjects B2 , D2 , F2 ) . A coherence of 25% means that 25% of the dots move in a direction consistent with the subject's heading , whereas the remaining 75% of the dots are relocated randomly within the dot cloud . In the combined condition , visual and vestibular stimuli always specified the same heading ( there was no cue conflict ) . During the main phase of data collection , subjects were not informed about the correctness of their choices ( no feedback ) . In the vestibular and combined conditions , platform motion was halted smoothly but rapidly immediately following registration of the decision , and the platform then returned to its original starting point . In the visual condition , the optic flow stimulus disappeared from the screen when a decision was made . In all conditions , 2 . 5 s after the decision , a sound informed the subjects that they could initiate the next trial by pushing a third button . Once a trial was initiated , the stimulus onset occurred following a randomized delay period ( truncated exponential; mean , 987 ms ) . Prior to data collection , subjects were introduced to the task for 1–2 week ‘training’ sessions , in which they were informed about the correctness of their choices by either a low-frequency ( incorrect ) or a high-frequency ( correct ) sound . The training period was terminated once their behavior stabilized across consecutive training sessions . During training , subjects were able to adjust their speed-accuracy trade-off based on feedback . During subsequent data collection , we did not observe any clear changes in the speed-accuracy trade-off exhibited by subjects . Analyses and statistical tests were performed using MATLAB R2013a ( The Mathworks , MA , USA ) . For each subject , discrimination thresholds were determined separately for each combination of stimulus modality ( visual-only , vestibular-only , combined ) and coherence ( 25% , 37% , and 70% for subjects A–G; 0% , 12% , 25% , 37% , 51% , and 70% for subjects B2 , D2 , F2 ) by plotting the proportion of rightward choices as a function of heading direction ( Figure 2A ) . The psychophysical discrimination threshold was taken as the standard deviation of a cumulative Gaussian function , fitted by maximum likelihood methods . We assumed a common lapse rate ( proportion of random choices ) across all stimulus conditions , but allowed for a separate bias parameter ( horizontal shift of the psychometric function ) for each modality/coherence . Confidence intervals for threshold estimates were obtained by taking 5000 parametric bootstrap samples ( Wichmann and Hill , 2001 ) . These samples also form the basis for statistical comparisons of discrimination thresholds: two thresholds were compared by computing the difference between their associated samples , leading to 5000 threshold difference samples . Subsequently , we determined the fraction of differences that were below or above zero , depending on the directionality of interest . This fraction determined the raw significance level for accepting the null hypothesis ( no difference ) . The reported significance levels are Bonferroni-corrected for multiple comparisons . All comparisons were one-tailed . Following traditional cue combination analyses ( Clark and Yuille , 1990 ) , the optimal threshold σpred , c in the combined condition for coherence c was predicted from the visual threshold σvis , c and the vestibular threshold σvest by σpred , c2=σvis , c2σvest2/ ( σvis , c2+σvest2 ) . Confidence intervals and statistical tests were again based on applying this formula to individual bootstrap samples of the unimodal threshold estimates . Supplementary file 2A reports the p-values for all subjects and all comparisons . For each dataset , we evaluated the absolute goodness-of-fit of the optimal model ( Figure 7A ) by finding the set of model parameters φ that maximized the likelihood of the observed choices and reaction times , and then computing the average coefficient of determination , R2 ( Dφ ) =12 ( Rpsych2 ( φ ) +Rchron2 ( φ ) ) . Here , Rpsych2 ( φ ) and Rchron2 ( φ ) denote the adjusted R2 values for the psychometric and chronometric functions , respectively , across all modalities/coherences . The value of Rpsych2 for the psychometric function was based on the probability of making a correct choice across all heading angles , coherences , and conditions , weighted by the number of observations , and adjusted for the number of model parameters . The same procedure , based on the mean reaction times , was used to find Rchron2 , but we additionally distinguished between mean reaction times for correct and incorrect choices , and fitted both weighted by their corresponding number of observations ( see SI for expressions for Rpsych2 ( φ ) and Rchron2 ( φ ) ) . We compared different variants of the full model ( Figure 7B ) by Bayesian model comparison based on Bayes factors , which were computed as follows . First , we found for each model M and subject s the set of parameters φ that maximized the likelihood , φs , M*=argmaxφp ( data of subj s|φ , M ) . Second , we approximated the Bayesian model evidence , measuring the model posterior probability while marginalizing over the parameters , up to a constant by the Bayesian information criterion , lnp ( M|s ) ≈−12BIC ( s , M ) with BIC ( s , M ) =−2lnp ( s|φs , M* , M ) +kMlnNs . Here , kM is the number of parameters of model M , and Ns is the number of trials for dataset s , respectively . Based on this , we computed the Bayes factor of model M vs the optimal model Mopt by pooling the model evidence over datasets , resulting in ∑s ( lnp ( M|s ) −lnp ( Mopt|s ) ) . These values , converted to a base-10 logarithm , are shown in Figure 7B . In this case , a negative log10-difference of 2 implies that the optimal model is 100 times more likely given the data than the alternative model , a difference that is considered decisive in favor of the optimal model ( Jeffreys , 1998 ) . To determine the faster stimulus modality for each subject , we compared reaction times for the vestibular condition with those for the visual condition at 70% coherence . We tested the difference in the logarithm of these reaction times by a 2-way ANOVA with stimulus modality and heading direction as the two factors , and we report the main effect of stimulus modality on reaction times . Although we performed a log-transform of the reaction times to ensure their normality , a Jarque–Bera test revealed that normality did not hold for some heading directions . Thus , we additionally performed a Friedman test on subsampled data ( to have the same number of trials per modality/heading ) which supported the ANOVA result at the same significance level . In the main text , we only report the main effect of stimulus modality on reaction time from the 2-way ANOVA . Detailed results of the 2-way ANOVA , the Jarque–Bera test , and the Friedman test are reported for each subject in Supplementary file 2C . Here we outline the critical extensions to the diffusion model . Detailed derivations and properties of the model are described in the Supplementary file 1 . Discretizing time into small steps of size Δ allows us to describe the particle trajectory x ( t ) in a DM by a random walk , x ( t ) =∑n∈1:tδxn , where each of the steps δxn ∼ ( ksin ( h ) Δ , Δ ) , called the momentary evidence , are normally distributed with mean ksin ( h ) Δ and variance Δ ( 1:t denotes the set of all steps up to time t ) . This representation is exact in the sense that it recovers the diffusion model , x˙=ksin ( h ) +η ( t ) , in the limit of Δ→0 . For the standard diffusion model , the posterior probability of sin ( h ) after observing the stimulus for t seconds , and under the assumption of a uniform prior , is given by Bayes rule ( 4 ) p ( sin ( h ) |δx1:t ) ∝∏n∈1:tp ( δxn|sin ( h ) ) ∝N ( sin ( h ) |x ( t ) kt , 1k2t ) , where δx1:t is the momentary evidence up to time t . From this we can derive the belief that heading is rightward , resulting in ( 5 ) p ( h>0|δx1:n ) =p ( sin ( h ) >0|δx1:t ) =∫0πp ( sin ( h ) |δx1:t ) dh=Φ ( x ( t ) t ) , where Φ ( · ) denotes the standard cumulative Gaussian function . This shows that both the posterior of the actual heading angle , as well as the belief about ‘rightward’ being the correct choice , only depend on x ( t ) rather than the whole trajectory δx1:t . The above formulation assumes that evidence is constant over time , which is not the case for our stimuli . Considering the visual cue and assuming that its associated sensitivity varies with velocity v ( t ) , the momentary evidence δxvis , n∼N ( vnkvis ( c ) sin ( h ) Δ , Δ ) is Gaussian with mean vnkvis ( c ) sin ( h ) Δ , where vn is the velocity at time step n , and variance Δ . Using Bayes rule again to find the posterior of sin ( h ) , it is easy to shown that xvis ( t ) is no longer sufficient to determine the posterior distribution . Rather , we need to perform a velocity-weighted accumulation , Xvis ( t ) =∑n∈1:tvnδxvis , n to replace xvis ( t ) , and replace time t with V ( t ) =∑n∈1:tvn2Δ , resulting in the following expression for the posterior ( 6 ) p ( sin ( h ) |δxvis , 1:t ) =p ( sin ( h ) |Xvis ( t ) , V ( t ) ) =N ( sin ( h ) |Xvis ( t ) kvis ( c ) V ( t ) , 1kvis2 ( c ) V ( t ) ) . Consequently , the belief about ‘rightward’ being correct can also be fully expressed by Xvis ( t ) and V ( t ) . This shows that optimal accumulation of evidence with a single-particle diffusion model with time-varying evidence sensitivity requires the momentary evidence to be weighted by its momentary sensitivity . A similar formulation holds for the posterior over heading based on the vestibular cue , however the vestibular cue is assumed to be weighted by the temporal profile of stimulus acceleration , instead of velocity . When combining multiple cues into a single DM , X˙comb=d ( t ) ( d ( t ) kcombsin ( h ) +ηcomb ( t ) ) , we aim to find expressions for kcomb and d ( t ) that keep the posterior over sin ( h ) unchanged , that is ( 7 ) p ( sin ( h ) |δxcomb , 1:t ) =p ( sin ( h ) |δxvis , 1:t , δxvest , 1:t ) . δxcomb , 1:t is the sequence of momentary evidence in the combined condition , following δxcomb , n∼N ( dnkcomb ( c ) sin ( h ) Δ , Δ ) . Expanding the probabilities reveals the equality to hold if the combined sensitivity is given by kcomb2 ( c ) =kvis2 ( c ) +kvest2 , and d ( t ) is expressed by Equation 3 , leading to Equation 1 for optimally combining the momentary evidence ( see Supplementary file 1 for derivation ) . The model used to fit the behavioral data is described in the main text . We never averaged data across subjects as they feature qualitatively different behavior , due to different speed-accuracy tradeoffs . Furthermore , for subjects performing both the three-coherence and the six-coherence version of the experiment , we treated either version as a separate data set . For each modality/coherence combination ( 7 combinations for 3 coherences , 13 combinations for 6 coherences ) we fitted one bias parameter that prevents behavioral biases from influencing model fits . The fact that performance of subjects often fails to reach 100% correct even for the highest coherences and largest heading angles was modeled by a lapse rate , which describes the frequency with which the subject makes a random choice rather than one based on accumulated evidence . This lapse rate was assumed to be independent of stimulus modality or coherence , and so a single lapse rate parameter is shared among all modality/coherence combinations . All model fits sought to find the model parameters φ that maximize the likelihood of the observed choices and reaction times for each dataset . As in Palmer et al . ( 2005 ) , we have assumed the likelihood of the choices to follow a binomial distribution , and the reaction times of correct and incorrect choices to follow different Gaussian distributions centered on the empirical means and spread according to the standard error . Model predictions for choice fractions and reaction times for correct and incorrect choices were computed from the solution to integral equations describing first-passage times of bounded diffusion processes ( Smith , 2000 ) . See Supplementary file 1 for the exact form of the likelihood function that was used . To avoid getting trapped in local maxima of this likelihood , we utilized a three-step maximization procedure . First , we found a ( possibly local ) maximum by pseudo-gradient ascent on the likelihood function . Starting from this maximum , we used a Markov Chain Monte Carlo procedure to draw 44 , 000 samples from the parameter posterior under the assumption of a uniform , bounded prior . After this , we used the highest-likelihood sample , which is expected to be close to the mode of this posterior , as a starting point to find the posterior mode by pseudo-gradient ascent . The resulting parameter vector is taken as the maximum-likelihood estimate . All pseudo-gradient ascent maximizations were performed with the Optimization Toolbox of Matlab R2013a ( Mathworks ) , using stringent stopping criteria ( TolFun = TolX = 10−20 ) to prevent premature convergence . | Imagine trying out a new roller-coaster ride and doing your best to figure out if you are being hurled to the left or to the right . You might think that this task would be easier if your eyes were open because you could rely on information from your eyes and also from the vestibular system in your ears . This is also what cue combination theory says—our ability to discriminate between two potential outcomes is enhanced when we can draw on more than one of the senses . However , previous tests of cue combination theory have been limited in that test subjects have been asked to respond after receiving information for a fixed period of time whereas , in real life , we tend to make a decision as soon as we have gathered sufficient information . Now , using data collected from seven human subjects in a simulator , Drugowitsch et al . have confirmed that test subjects do indeed give more correct answers in more realistic conditions when they have two sources of information to rely on , rather than only one . What makes this result surprising ? Traditional cue combination theories do not consider that slower decisions allow us to process more information and therefore tend to be more accurate . Drugowitsch et al . show that this shortcoming causes such theories to conclude that multiple information sources might lead to worse decisions . For example , some of their test subjects made less accurate choices when they were presented with both visual and vestibular information , compared to when only visual information was available , because they made these choices very rapidly . By developing a theory that takes into account both reaction times and choice accuracy , Drugowitsch et al . were able to show that , despite different trade-offs between speed and accuracy , test subjects still combined the information from their eyes and ears in a way that was close to ideal . As such the work offers a more thorough account of human decision making . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"neuroscience"
] | 2014 | Optimal multisensory decision-making in a reaction-time task |
Lung cancer and chronic lung diseases impose major disease burdens worldwide and are caused by inhaled noxious agents including tobacco smoke . The cellular origins of environmental-induced lung tumors and of the dysfunctional airway and alveolar epithelial turnover observed with chronic lung diseases are unknown . To address this , we combined mouse models of genetic labeling and ablation of airway ( club ) and alveolar cells with exposure to environmental noxious and carcinogenic agents . Club cells are shown to survive KRAS mutations and to form lung tumors after tobacco carcinogen exposure . Increasing numbers of club cells are found in the alveoli with aging and after lung injury , but go undetected since they express alveolar proteins . Ablation of club cells prevents chemical lung tumors and causes alveolar destruction in adult mice . Hence club cells are important in alveolar maintenance and carcinogenesis and may be a therapeutic target against premalignancy and chronic lung disease .
Chronic lung diseases present tremendous health burdens attributed to dysfunctional alveolar repair ( Barnes et al . , 2015; Lozano et al . , 2012; Spella et al . , 2017 ) . Lung adenocarcinoma ( LUAD ) , the leading cancer killer worldwide , is mainly caused by chemical carcinogens of tobacco smoke that induce mutations of the Kirsten rat sarcoma viral oncogene homologue ( KRAS ) in yet unidentified pulmonary cells ( Torre et al . , 2015; Forbes et al . , 2011; Hecht , 1999; Campbell et al . , 2016; Cancer Genome Atlas Research Network , 2014 ) . The discovery of the cellular lineages and the transcriptional programs that underlie lung regeneration and carcinogenesis is extremely important , since epithelial developmental pathways are intimately related with oncogenic signaling to jointly regulate stemness and drug resistance ( Barbie et al . , 2009; Seguin et al . , 2014 ) . To this end , lineage-specific genes encoding epithelial proteins that support the physiological functions of the lungs were recently shown to suffer non-coding insertions and deletions in LUAD , lending further support to the longstanding notion that epithelial cells that express lung-restricted proteins are the cellular sources of LUAD ( Imielinski et al . , 2017 ) . However , these cells of origin of LUAD remain only partially charted . Previous pulmonary lineage tracing studies that utilized noxious insults and ectopic expression of oncogenes in the respiratory epithelium incriminated both airway and alveolar cells as progenitors of newly formed alveoli and/or LUAD in adult mice ( Zuo et al . , 2015; Kim et al . , 2005; Cho et al . , 2011; Xu et al . , 2012; Sutherland et al . , 2014; Mainardi et al . , 2014; Desai et al . , 2014 ) . To this end , airway epithelial cells ( AEC ) line the bronchi and include ciliated , basal , goblet , and Clara or club cells; alveolar type II cells ( ATII ) and alveolar macrophages ( AMΦ ) are distributed across the distal lung parenchyma; and bronchoalveolar stem cells ( BASC ) with dual AEC/ATII properties are located at the bronchoalveolar junctions . Established markers currently used to label these pulmonary lineages include acetylated tubulin ( TUBA1A ) for ciliated cells , keratin 5 ( KRT5 ) for basal cells , forkhead box J1 ( FOXJ1 ) for goblet cells , Clara cell secretory protein ( CCSP ) for club cells , surfactant protein C ( SFTPC ) and lysozyme 2 ( LYZ2 ) for ATII cells , and LYZ2 for AMΦ , are summarized in Figure 1A and Figure 1—figure supplement 1 , and are extensively studied in Desai et al . ( 2014 ) and Treutlein et al . ( 2014 ) . However , existing mouse models for lineage tracing feature incomplete and/or promiscuous lung cell labeling , that is cellular markings fail to identify all cells of a target lineage ( false negative marking ) or wrongfully identify other cells outside of the target lineage ( false positive marking ) ( Zuo et al . , 2015; Kim et al . , 2005; Cho et al . , 2011; Xu et al . , 2012; Sutherland et al . , 2014; Mainardi et al . , 2014; Desai et al . , 2014 ) . In addition , all studies that attempted to address the cellular origins of LUAD to date employed overexpression of oncogenes such as KRASG12D in the lungs , to conclude that ATII cells or BASC are the most probable culprits of the disease ( Kim et al . , 2005; Cho et al . , 2011; Xu et al . , 2012; Sutherland et al . , 2014; Mainardi et al . , 2014; Desai et al . , 2014 ) . However , it was recently shown that oncogenic KRASG12D-driven mouse lung tumors do not imitate the mutational landscape of human LUAD as closely as tobacco carcinogen-induced LUAD do ( Campbell et al . , 2016; Cancer Genome Atlas Research Network , 2014; Westcott et al . , 2015 ) . Here we aimed at identifying the cell lineage ( s ) that give rise to human-relevant tobacco carcinogen-triggered LUAD in mice and that regenerate adult murine alveoli after injury . For this , we combined mouse models of genetic labeling and ablation of airway and alveolar epithelial cells with noxious and tumorigenic insults to the adult lung . To achieve this , we adapted multi-hit chemical carcinogen exposure protocols to the murine C57BL/6 strain that is resistant to chemical tumor induction ( Miller et al . , 2003; Malkinson et al . , 1997; Stathopoulos et al . , 2007 ) , and corroborated the findings with the FVB strain that is susceptible to single-hit carcinogenesis ( Westcott et al . , 2015; Stathopoulos et al . , 2007; Vreka et al . , 2018 ) . We show that aging , toxic , and carcinogen insults to the adult mouse lung cause expansion of airway-marked cells to the alveolar parenchyma , where they express the alveolar marker SFTPC and facilitate alveolar repair and carcinogenesis . In addition , we report how airway cells preferentially sustain chemical-induced KRAS mutations leading to LUAD that are spatially linked with neighboring bronchi . Moreover , genetic ablation of airway cells is shown to hinder alveolar maintenance and carcinogenesis in mice , indicating a central role for these cells in alveolar regeneration and LUAD triggered in response to environmental challenges .
To evaluate the contribution of different epithelial lung cell lineages to chemical-induced LUAD , we crossed a CRE-reporter strain that switches somatic cells from membranous tdTomato ( mT; hereafter TOMATO ) to membranous GFP ( mG; hereafter GFP ) fluorescence upon CRE-mediated recombination ( mT/mG; hereafter TOMATO mice ) ( Muzumdar et al . , 2007 ) to six different CRE-driver strains on the C57BL/6 background ( Desai et al . , 2014; Oikonomou et al . , 2012; Okubo et al . , 2005; Hayashi et al . , 2002; Ogilvy et al . , 1998; Tronche et al . , 1999 ) . This permitted the permanent genetic GFP-labeling of different lung cell lineages ( mouse strains are listed in Figure 1A and Figure 1—figure supplement 2 , and in Materials and methods and in Appendix 1 ) . Double heterozygote offspring at six postnatal weeks ( i . e . , after mouse lung development is complete [Zuo et al . , 2015; Desai et al . , 2014] ) were examined for GFP-labeling ( results are shown in Figure 1A , Figure 1—figure supplements 3 and 4 , and in Figure 1—figure supplement 4—source data 1 ) . This approach labeled permanently all AEC of GFP;CCSP . CRE mice , some AEC and all ATII of GFP;SFTPC . CRE mice , some ATII and all AMΦ of GFP;LYZ2 . CRE mice , and various other cells in the remaining intercrosses ( Figure 1A , Figure 1—figure supplements 3–5 , and Figure 1—figure supplement 5—source data 1 ) . Co-localization of GFP-labeling with lineage protein markers ( listed in Figure 1A and Figure 1—figure supplement 1 ) revealed that genetic GFP-labeling in GFP;CCSP . CRE mice marked all airway epithelial cells including club and ciliated cells , in GFP;SFTPC . CRE mice most airway and all alveolar epithelial type II cells , and in GFP;LYZ2 . CRE mice some alveolar epithelial type II cells and all alveolar macrophages ( Figure 1B , Figure 1—figure supplements 6–8 , Figure 1—figure supplement 7—source data 1 , Figure 1—figure supplement 8—source data 1 ) . These findings show precise airway epithelial lineage labeling in GFP;CCSP . CRE mice and non-specific airway/alveolar/myeloid lineage labeling in GFP;SFTPC . CRE and GFP;LYZ2 . CRE mice . We next triggered LUAD in GFP;CCSP . CRE , GFP;SFTPC . CRE , and GFP;LYZ2 . CRE mice on the C57BL/6 background using repetitive exposures to the tobacco carcinogens urethane ( ethyl carbamate , EC; stand-alone mutagen and tumor promoter ) ( Westcott et al . , 2015; Miller et al . , 2003; Stathopoulos et al . , 2007; Vreka et al . , 2018 ) or 3-methylcholanthrene followed by butylated hydroxytoluene ( MCA/BHT; a two-hit mutagen/tumor promoter regimen ) ( Malkinson et al . , 1997 ) ( Figure 1C , Figure 1—figure supplements 9 and 10 , and Figure 1—figure supplement 10—source data 1 ) . In both models , preneoplastic ( airway epithelial hyperplasias and atypical alveolar hyperplasias ) and neoplastic ( adenoma and LUAD ) lesions classified according to established guidelines ( Nikitin et al . , 2004 ) were located both in the airways and the alveolar regions . However , established lung tumors were most frequently located near or inside the airways ( Figure 1C and Figure 1—figure supplement 11 ) . All hyperplasias and tumors of GFP;SFTPC . CRE and some of GFP;LYZ2 . CRE mice were GFP-labeled , but this was not informative , since baseline marking of GFP;SFTPC . CRE and GFP;LYZ2 . CRE mice were non-specific . Interestingly , all hyperplasias and tumors of GFP;CCSP . CRE mice contained GFP-labeled airway cells that did not express the club cell marker CCSP anymore , but had acquired expression of the alveolar epithelial markers SFTPC with or without LYZ2 ( Figure 1D , Figure 1—figure supplements 12–15 , and Figure 1—figure supplement 13—source data 1 ) . Identical results were recapitulated using single urethane hits to GFP;CCSP . CRE , GFP;SFTPC . CRE , and GFP;LYZ2 . CRE mice backcrossed >F12 to the susceptible FVB strain , which result in human LUAD-like mutations including KrasQ61R ( Westcott et al . , 2015; Vreka et al . , 2018; Kanellakis et al . , 2019 ) ( Figure 1D and Figure 1—figure supplements 16–19 ) . Collectively , these data support that airway cells contribute to chemical-induced LUAD , shifting from airway to alveolar marker expression during carcinogenesis . We next used digital droplet PCR ( ddPCR ) to determine the lung lineages that suffer KrasQ61R driver mutations at early time-points after single urethane hits ( Westcott et al . , 2015; Vreka et al . , 2018; Kanellakis et al . , 2019 ) . For this , GFP;CCSP . CRE and GFP;LYZ2 . CRE mice backcrossed >F12 to the susceptible FVB strain received urethane and duplexed ddPCR designed to single-copy-co-amplify Kras and RosamT was performed one and two weeks later . Interestingly , GFP-labeled cells of both mouse strains had KrasQ61R mutations at one week post-urethane , but KrasQ61R mutations selectively persisted in GFP-labeled airway cells in the lungs of GFP;CCSP . CRE mice at two weeks ( Figure 2A , Figure 2—figure supplement 1 , and Figure 2—source data 1 ) . In addition , three-dimensional reconstruction of tumor-bearing lungs of FVB mice at 6 months post-urethane using high-resolution micro-computed tomography ( μCT ) revealed that most lung tumors were spatially linked with the airways , in accord with pathology results ( Figure 2B and C , and Figure 2—source data 2 ) . These results support the involvement of airway cells in chemical-induced lung adenocarcinoma formation in mice . Since airborne carcinogens act globally on the respiratory field ( Franklin et al . , 1997 ) , we examined non-neoplastic alveolar areas of carcinogen-treated GFP;CCSP . CRE mice , to discover markedly increased numbers of GFP-labeled cells in the alveoli of carcinogen-treated mice compared with saline-treated or naïve controls ( Figure 3A , Figure 3—figure supplements 1 and 2 , and Figure 3—figure supplement 2—source data 1 ) . Immunostaining revealed that juxtabronchial GFP-labeled cells still expressed CCSP , but lost CCSP and acquired SFTPC expression when located in alveoli and tumors ( Figure 3B and Figure 3—figure supplements 3 and 4 ) . The expansion of airway cells after urethane exposure was also documented using bioluminescent imaging of double heterozygote offspring of CCSP . CRE intercrosses with Luciferase-expressing ( LUC ) mice ( Safran et al . , 2003 ) , a strain emitting light specifically from airway epithelia ( Figure 3—figure supplement 5 , and Figure 3—figure supplement 5—source data 2 ) . In addition , co-staining of human LUAD ( Giopanou et al . , 2015 ) for the alveolar marker SFTPC and the airway markers CCSP and KRT5 showed co-localization of SFTPC with KRT5 but not with CCSP ( Figure 3C and Figure 3—figure supplement 6 ) . These results suggest that airway epithelial cells expand to alveolar regions during field cancerization by tobacco carcinogens , a process involving either direct alveolar cell recycling by airway epithelial cells or transient CCSP expression by alveolar cells during carcinogenesis . Moreover , that human and murine LUAD carry airway imprints although their location and protein expression suggests an alveolar origin ( Desai et al . , 2014; Aberle et al . , 2011; Mason et al . , 2000; Lindskog et al . , 2014; Sutherland and Berns , 2010 ) . We next examined the kinetics of lineage-labeled cells during aging , injury , and repair . While the number of GFP-labeled cells in the alveoli of aging GFP;SFTPC . CRE and GFP;LYZ2 . CRE mice was stable , GFP-labeled airway cells in the alveoli of aging GFP;CCSP . CRE mice progressively increased and expressed SFTPC protein ( Figure 4A and B and Figure 4—source data 1 ) . Bleomycin treatment , which depletes alveolar type II cells ( Lawson et al . , 2005 ) , accelerated the accumulation of GFP-labeled airway cells in the alveoli and in urethane-triggered LUAD ( Figure 4C and D , Figure 4—figure supplements 1 and 2 , Figure 4—source data 2 , and Figure 4—figure supplement 2—source data 1 ) . GFP-labeled airway cells expressing the alveolar marker SFTPC also increased in the alveoli of GFP;CCSP . CRE mice exposed to perinatal hyperoxia that damages forming alveoli ( Rawlins et al . , 2009 ) , and in the alveoli of GFP;CCSP . CRE mice treated with naphthalene that kills airway epithelial cells ( Sutherland and Berns , 2010; Rawlins et al . , 2009 ) , but were not identified within the airways of naphthalene-treated GFP;CCSP . CRE mice; these appeared to be repopulated by GFP-labeled airway cells that express the club cell marker CCSP ( Figure 4E–4H , Figure 4—figure supplements 3 and 4 , Figure 4—source datas 3 and 4 , and Figure 4—figure supplement 4—source data 2 ) . In line with the latter finding , no GFP-labeled alveolar cells were identified in the airways of GFP;LYZ2 . CRE mice recovering from naphthalene-induced injury ( Figure 4G and H ) . Taken together , the data indicate that airway-originated cells repopulate both the airways and the alveoli during aging and recovery from injury , while alveolar cells do not reconstitute the airways , in line with previous findings ( Desai et al . , 2014; Rawlins et al . , 2009 ) . The observed alveolar spread of airway-labeled cells was explained by either peripheral migration of airway cells or transient CCSP expression by regenerating alveolar cells . To further examine the role of airway and alveolar cells in alveolar homeostasis and lung carcinogenesis , we ablated them by crossing CCSP . CRE , SFTPC . CRE , and LYZ2 . CRE mice to mice expressing Diphtheria toxin in somatic cells upon CRE-mediated recombination ( DTA mice ) ( Voehringer et al . , 2008 ) . Triple transgenic GFP;DRIVER . CRE;DTA intercrosses were also generated to evaluate ablation efficiency . As expected , SFTPC . CRE;DTA and GFP;SFTPC . CRE;DTA mice were fetal lethal ( no double or triple heterozygote offspring was obtained by n > 3 intercrosses , >10 litters , and >60 off-springs for each genotype; p<0 . 0001 , Fischer’s exact test ) . However , all other ablated mice survived till adulthood . Airway epithelial ablation was complete in GFP;CCSP . CRE;DTA mice , while some GFP-labeled alveolar macrophages persisted in GFP;LYZ2 . CRE;DTA mice , presumably freshly recruited monocytes initiating LYZ2 expression . Immunostaining revealed that the denuded airway epithelium of 12-week-old GFP;CCSP . CRE;DTA mice contained few flat CCSP+SFTPC+LYZ2+ immunoreactive cells , while the apparently intact alveolar spaces of GFP;LYZ2 . CRE;DTA mice harbored only some CCSP-SFTPC-LYZ2+immunoreactive alveolar macrophages ( Figure 5A , Figure 5—figure supplements 1 and 2 , and Figure 5—figure supplement 2—source data 1 ) . Remarkably , morphometric and functional analyses of 12-week-old DTA control , CCSP . CRE;DTA , and LYZ2 . CRE;DTA mice showed that LYZ2 . CRE;DTA mice displayed normal airway caliper and mean linear intercept ( measures of airway and alveolar structure ) , normal number of CD45+ CD11b+ myeloid cells in bronchoalveolar lavage ( BAL; measure of airspace inflammation ) , and normal airways resistance and static compliance ( measures of airway and alveolar function ) compared with DTA controls . However , CCSP . CRE;DTA mice displayed widened airway and alveolar dimensions with inflammatory interalveolar septal destruction evident by increased mean linear intercept , CD45+ CD11b+ cells in BAL , and static compliance ( Figure 5B and C and Figure 5—source data 1 ) , mimicking human chronic obstructive pulmonary disease ( Barnes et al . , 2015 ) . Finally , we exposed control and ablated mice to ten consecutive weekly urethane exposures . All mice survived six months into carcinogen treatment , and CCSP . CRE;DTA and LYZ2 . CRE;DTA mice were equally protected from LUAD development compared with controls ( Figure 5D and E , and Figure 5—source data 2 ) . Taken together , these results show that the CCSP+ airway lineage maintains postnatal alveolar structure and function , and , together with the LYZ2+ alveolar lineage , are required for lung adenocarcinoma development . We subsequently examined the transcriptomes of cell lines isolated from urethane-induced LUAD ( Kanellakis et al . , 2019 ) and of murine lungs with those of murine AEC isolated from tracheal explants , of murine ATII cells ( Frank et al . , 2016 ) , and of murine bone-marrow-derived macrophages ( BMDM ) . The AEC transcriptome was specifically enriched in LUAD cells compared with whole lungs ( Figure 6A and B , Figure 6—figure supplement 1 , and Figure 6—source data 1 ) . LUAD cell lines lost expression of epithelial markers compared with their native lungs , but displayed up-regulated expression of LUAD markers ( i . e . , Krt18 and Krt20 ) , of epidermal growth factor receptor ligands ( Areg and Ereg ) , and of the Myc oncogene ( Figure 6—figure supplements 2–4 , and Figure 6—figure supplement 2—source data 1 ) . Similar analyses of the transcriptomes of human LUAD and corresponding healthy lungs ( Kabbout et al . , 2013 ) , and of primary human AEC , ATII , and AMΦ ( Clark et al . , 2015; Dancer et al . , 2015; Lee et al . , 2009 ) also disclosed that the AEC transcriptome was significantly enriched in LUAD compared with healthy lungs ( Figure 6C and D and Figure 6—source data 2 ) . Gene set enrichment analyses ( GSEA ) showed that the mouse AEC transcriptome predominated over ATII/BMDM transcriptomes in LUAD cells ( Figure 6E , Figure 6—figure supplement 5 , and Figure 6—source data 3 ) . In addition , the human AEC transcriptome was enriched equally with ATII/AMΦ transcriptomes in human LUAD compared with healthy lungs ( Figure 6F , Figure 6—figure supplement 6 , and Figure 6—source data 4 ) . These results showed the presence of an anticipated alveolar and an unexpected airway epithelial transcriptomic signature in tobacco carcinogen-induced LUAD of mice and men . The more pronounced results in mice were plausible by the early nature of the human surgical specimens examined compared with our murine cell lines that present advanced metastatic tumor cells .
We characterized the dynamics of respiratory epithelial cells in the postnatal mouse lung during aging and after challenge with noxious and carcinogenic insults . The contributions of airway cells to chemical-induced lung adenocarcinoma are described for the first time ( Figure 7A ) . Although the peripheral location and molecular phenotype of murine and human lung adenocarcinoma ( i . e . , the expression of the alveolar epithelial marker SFTPC ) suggest an alveolar origin , we show here that both airway and alveolar cells are found in environmental-induced lung adenocarcinoma and that , in fact , airway cells may play a more prominent role during the initial steps of carcinogenesis . Furthermore , airway cells are implicated in postnatal alveolar maintenance during aging and recovery from injury . Our analyses facilitate insights into the dynamics of epithelial lineages in the postnatal lung ( Figure 7B ) and indicate that airway cells are essential for the sustained structural and functional integrity of adult alveoli . Finally , mouse and human lung adenocarcinomas are shown to bare transcriptome markings of highly enriched airway signatures , rendering our findings plausible in both experimental and human lung adenocarcinoma . This study addresses the cellular and molecular signatures of chemical-induced lung adenocarcinoma . Lung tumors induced in two different mouse strains by two different chemical regimens contained in tobacco smoke are shown to contain airway epithelial markings . This is important because human lung adenocarcinoma is inflicted by chronic exposure to tobacco smoke and other environmental exposures ( Hecht , 1999; Campbell et al . , 2016; Cancer Genome Atlas Research Network , 2014; Westcott et al . , 2015; Miller et al . , 2003; Malkinson et al . , 1997; Alexandrov et al . , 2016; Castelletti et al . , 2019 ) . As such , the mutation profile of the human disease is more closely paralleled by chemical-induced murine lung tumors compared with lung cancers triggered by transgenic expression of KrasG12C or KrasG12D in the respiratory epithelium ( Westcott et al . , 2015 ) . Although the latter transgenic tumors have been extensively studied ( Kim et al . , 2005; Cho et al . , 2011; Xu et al . , 2012; Sutherland et al . , 2014; Mainardi et al . , 2014; Desai et al . , 2014 ) , chemical-induced lung adenocarcinomas have not been investigated . In all mouse models we studied , all tumors contained the airway genetic marking , in contrast with the LYZ2 alveolar genetic marking which was dispensable for lung adenocarcinoma development . Our observations support the multi-stage field concept of chemical carcinogenesis ( Franklin et al . , 1997 ) , according to which tumor-initiated cells undergo multiple steps of genomic evolution and phenotypic appearance that include an obligatory airway-like stage . In fact , the prevalence of a different Kras mutation in urethane-induced tumors ( KrasQ61R ) compared to KRASG12C/D mutations in the transgenic mouse models has led to the suggestion that chemical carcinogens introduce KRAS mutations in a different population of tumor-initiating cells than mouse models of genetic KRAS activation ( Westcott et al . , 2015 ) . Our findings of airway epithelial cells being more sensitive than alveolar type II cells to KrasQ61R mutations during the initial steps of urethane-induced carcinogenesis further supports this notion and render airway cells an attractive novel target for premalignancy . The consistent finding of CCSP genetic markings ( indicative of airway epithelial origin ) together with SFTPC and LYZ2 protein expression ( indicative of alveolar epithelial phenotype ) in chemical-triggered lung adenocarcinomas and their precursor lesions implies three different scenarios for lung adenocarcinoma formation: i ) airway epithelial cells colonize the distal lung during carcinogenesis thereby activating obligate ( SFTPC+ ) and dispensable ( LYZ2+ ) alveolar transcriptomes; ii ) alveolar cells transit through an obligate CCSP+ with or without a dispensable LYZ2+ stage during the process; or iii ) lung adenocarcinoma arises from multipotent progenitors that express multiple epithelial markers , such as those found during pulmonary embryogenesis , in human lung adenocarcinoma , and in other chronic lung diseases ( Desai et al . , 2014; Frank et al . , 2016; Xu et al . , 2016 ) . However , in our view , the propensity of airway cells to survive KRAS mutations during early carcinogenesis , the close airway-proximity of lung tumors revealed by μCT and histology , as well as the fact that CCSP-labeled cells did not express the CCSP marker anymore , support a bronchial origin of these tumors . This view is in line with recent evidence for tobacco smoke-induced epigenetic changes that sensitize human airway epithelial cells to a single KRAS mutation ( Vaz et al . , 2017 ) . Along these lines , the split genetic markings of chemical-induced lung adenocarcinomas of GFP;LYZ2 . CRE mice indicates that LYZ2-labeled alveolar cells are dispensable for environmental lung adenocarcinoma , as opposed to what was previously shown for genetically-triggered lung adenocarcinoma ( Desai et al . , 2014 ) . Our approach focused on the integral assessment of changes in lung epithelial kinetics and transcriptome signatures during aging , injury , and carcinogenesis . The perpetual cell labeling approach we adopted was preferred over pulsed lineage tracing models because of the unprecedented accuracy of our CCSP . CRE strain in exclusively and completely labeling airway epithelial cells at the conclusion of development , allowing tracking of subsequent changes in adulthood . The identification of transcriptional programs that are activated during lung repair and carcinogenesis are of great importance for lung biology and are likely to lead to therapeutic innovations ( Nagel et al . , 2016 ) . To this end , insertions and deletions in lineage-restricted genes were recently shown to occur in human lung adenocarcinoma ( Imielinski et al . , 2017 ) . Moreover , integrin β3 and TANK-binding kinase one partner with oncogenic KRAS signaling to mediate cancer stemness and drug resistance ( Barbie et al . , 2009; Seguin et al . , 2014 ) . Along these lines , our findings of the involvement of airway epithelial cells in lung maintenance , repair , and carcinogenesis imply that at least some of these cells present lung stem cells with regenerative and malignant potential and thus marked therapeutic targets . This was evident in our hands by the facts that airway epithelial cells could maintain adult injured alveoli and sustain KRAS mutations induced by urethane . In conclusion , airway cells contribute to alveolar maintenance and lung carcinogenesis in response to environmental challenges . Since defective epithelial repair underlies the pathogenesis of chronic lung diseases and since abundantly transcribed genes are central to the mutational processes that cause cancer , this finding is of potential therapeutic importance for chronic pulmonary diseases and lung cancer .
All raw data used to generate the main Figures and Figure Supplements are provided as * . xlsx Source Data files . All mice were bred at the Center for Animal Models of Disease of the University of Patras . Experiments were designed and approved a priori by the Veterinary Administration of the Prefecture of Western Greece ( approval numbers 3741/16 . 11 . 2010 , 60291/3035/19 . 03 . 2012 , and 118018/578/30 . 04 . 2014 ) and were conducted according to Directive 2010/63/EU ( http://eur-lex . europa . eu/legal-content/EN/TXT/ ? qid=1486710385917&uri=CELEX:32010L0063 ) . Male and female experimental mice were sex- , weight ( 20–25 g ) - , and age ( 6–12 week ) -matched . n = 588 experimental and n = 165 breeder mice were used for this report . Sample size was calculated using power analysis on G*power . Experiments were randomized across different cages and mouse lungs were always examined by two blinded researchers . Sample numbers are included in the figures and figure legends . Archival tissue samples of patients with LUAD ( Giopanou et al . , 2015 ) that underwent surgical resection with curative intent between 2001 and 2008 at the University Hospital of Patras were retrospectively enrolled . The observational protocol for these studies adhered to the Helsinki Declaration and was approved by the Ethics Committee of the University Hospital of Patras , and all patients gave written informed consent . Urethane , ethyl carbamate , EC , CAS# 51-79-6; 3-methylcholanthrene , 3-methyl-1 , 2-dyhydrobenzo[j]aceanthrylene , MCA , CAS# 56-49-5; butylated hydroxytoluene , 2 , 6-Di-tert-butyl-4-methylphenol , BHT , CAS# 128-37-0; naphthalene , CAS# 91-20-3 , and Hoechst33258 nuclear dye ( CAS# 23491-45-4 ) , were from Sigma-Aldrich ( St . Louis , MO ) . Bleomycin A2 , ( ( 3-{[ ( 2'-{ ( 5S , 8S , 9S , 10R , 13S ) −15-{6-amino-2- [ ( 1S ) −3-amino-1-{[ ( 2S ) −2 , 3-diamino-3-oxopropyl]amino}−3-oxopropyl] −5-methylpyrimidin-4-yl}−13-[{[ ( 2R , 3S , 4S , 5S , 6S ) −3-{[ ( 2R , 3S , 4S , 5R , 6R ) −4- ( carbamoyloxy ) −3 , 5-dihydroxy-6- ( hydroxymethyl ) tetrahydro-2H-pyran-2-yl]oxy} −4 , 5-dihydroxy-6- ( hydroxymethyl ) tetrahydro-2H-pyran-2-yl]oxy} ( 1H-imidazol-5-yl ) methyl]−9-hydroxy-5-[ ( 1R ) −1-hydroxyethyl]−8 , 10-dimethyl-4 , 7 , 12 , 15-tetraoxo-3 , 6 , 11 , 14-tetraazapentadec-1-yl}−2 , 4'-bi-1 , 3-thiazol-4-yl ) carbonyl]amino}propyl ) ( dimethyl ) sulfonium; CAS #9041-93-4 , was from Calbiochem ( Darmstadt , Germany ) . D-Luciferin potassium salt , ( 4S ) −2- ( 6-hydroxy-1 , 3-benzothiazol-2-yl ) −4 , 5-dihydrothiazole-4-carboxylic acid , CAS #2591-17-5 , was from Gold Biotechnology ( St . Louis , MO ) . C57BL/6J ( C57BL/6; #000664 ) , FVB/NJ ( FVB; #001800 ) , B6 . 129 ( Cg ) -Gt ( ROSA ) 26Sortm4 ( ACTB-tdTomato , -EGFP ) Luo/J [mT/mG; TOMATO; #007676; ( Muzumdar et al . , 2007 ) ] , FVB . 129S6 ( B6 ) -Gt ( ROSA ) 26Sortm1 ( Luc ) Kael/J [LUC; #005125; ( Safran et al . , 2003 ) ] , B6 . 129P2-Gt ( ROSA ) 26Sortm1 ( DTA ) Lky/J [DTA; #009669; ( Voehringer et al . , 2008 ) ] , B6 . 129P2-Lyz2tm1 ( cre ) Ifo/J [LYZ2 . CRE; #004781; ( Desai et al . , 2014 ) ] , B6 . Cg-Tg ( Sox2-cre ) 1Amc/J [SOX2 . CRE; #008454; ( Hayashi et al . , 2002 ) ] , B6 . Cg-Tg ( Vav1-icre ) A2Kio/J [VAV . CRE; #008610; ( Ogilvy et al . , 1998 ) ] , and B6 . Cg-Tg ( Nes-cre ) 1Kln/J [NES . CRE; #003771; ( Tronche et al . , 1999 ) ] mice were from Jackson Laboratories ( Bar Harbor , MN ) . B6;CBA-Tg ( Scgb1a1-cre ) 1Vart/Flmg ( CCSP . CRE; European Mouse Mutant Archive #EM:04965 ) mice are described elsewhere ( Oikonomou et al . , 2012 ) and Tg ( Sftpc-cre ) 1Blh ( SFTPC . CRE; Mouse Genome Informatics #MGI:3574949 ) mice were donated by their founder ( Okubo et al . , 2005 ) . Mice were bred >F12 to the FVB background at the University of Patras Center for Animal Models of Disease . Six-week-old mice on the C57BL/6 background received ten consecutive weekly intraperitoneal urethane injections ( 1 g/Kg in 100 μL saline ) and were sacrificed 6–7 months after the first injection , or four consecutive weekly intraperitoneal MCA ( 15 mg/Kg in 100 μL saline ) followed by eight consecutive weekly intraperitoneal BHT injections ( 200 mg/Kg in 100 μL corn oil ) and were sacrificed 6–7 months after the first injection . Six-week-old mice on the FVB background received one intraperitoneal urethane injection ( 1 g/Kg in 100 μL saline ) and were sacrificed 6–7 months later ( Westcott et al . , 2015; Miller et al . , 2003; Malkinson et al . , 1997; Stathopoulos et al . , 2007; Vreka et al . , 2018 ) . Six-week-old mice ( C57BL/6 background ) received intratracheal bleomycin A2 ( 0 . 08 units in 50 μL saline ) or intraperitoneal naphthalene ( 250 mg/Kg in 100 μL corn oil ) ( Lawson et al . , 2005; Rawlins et al . , 2009 ) . In addition , preterm mothers of the C57BL/6 background and their offspring were exposed to room air ( 21% oxygen; control ) or 98% oxygen for two days before and four days after birth ( Rawlins et al . , 2009; Yee et al . , 2009 ) . Oxygen levels were continuously monitored . The gas stream was humidified to 40–70% by a deionized water-jacketed Nafion membrane tubing and delivered through a 0 . 22 μm filter before passage into a sealed Lexan polycarbonate chamber measuring 40 × 25×25 cm and accommodating 25 L gas at a flow rate of 5 L/min , resulting in complete gas exchange every 5 min . Mothers were cycled between litters on 21% and 98% oxygen every 24 hr to prevent oxygen toxicity and to control for nutritional support of the pups . After perinatal hyperoxia , mice remained at room air till sacrificed at eight weeks of age . Lung tumors were dissected from surrounding healthy lung parenchyma under sterile conditions , minced into 1 mm pieces , and cultured at 37°C in 5% CO2–95% air using Dulbecco’s Modified Eagle Medium ( DMEM ) , 10% FBS , 2 mM L-glutamine , 1 mM pyruvate , 100 U/mL penicillin , and 100 U/mL streptomycin . All cell lines were immortal and indefinitely phenotypically stable over >18 months and/or 60 passages , and were tumorigenic and metastatic in C57BL/6 mice ( Kanellakis et al . , 2019 ) . Cell lines were cultured in DMEM supplemented with 10% FBS and 100 IU/mL penicillin/streptomycin and were maintained in humidified incubators at 37°C with 95% air–5% CO2 . Cell lines were authenticated annually using the short tandem repeat method and were tested negative for Mycoplasma Spp . biannually by MycoAlert Mycoplasma Detection Kit ( LONZA; Verviers , Belgium ) . Ten archival formalin-fixed , paraffin-embedded tissue samples of patients with LUAD that underwent surgical resection with curative intent between 2001 and 2008 at the University Hospital of Patras were retrospectively enrolled ( Giopanou et al . , 2015 ) . The observational protocol for these studies adhered to the Helsinki Declaration and was approved by the Ethics Committee of the University Hospital of Patras , and all patients gave written informed consent . Urethane or saline treated FVB mice were sacrificed six months post urethane/saline injection . Lungs were inflated and fixed with 10% neutral buffered formalin overnight . They were then dehydrated and chemically dried for μCT scanning using a method kindly provided by Jeroen Hostens ( Bruker; Kontich , Belgium ) . Briefly , a gradient ethanol dehydration protocol ( from 70–100% ) was applied , followed by 2 hr incubation in Hexamethyldisilazane ( HMDS; Sigma , St . Louis , MO ) and 2 hr air-drying . The dehydrated lungs were then scanned in a Bruker SkyScan 1172 scanner at 41kV without filtration and with 5 . 94 μm voxel resolution ( exposure: 440 ms ) . The X-ray projections were obtained at 0 . 35° intervals with a scanning angular rotation of 180° and two frames were averaged for each rotation under a mean of 10 frames per random movement . 3D reconstructions were performed using NRECON software ( Bruker ) . Regions of interest for the whole lung and peripheral lung tissue were defined in the CT analysis software ( CTan; Bruker ) , thresholds applied to detect tissue from background , and a 3D volume rendering of the lungs were performed using the CTVox software ( Bruker ) . Mouse lungs were recoded ( blinded ) by laboratory members not participating in these studies and were always examined by two independent blinded participants of this study . The results obtained by each investigator were compared , and lungs were re-evaluated if deviant by >20% . Lungs and lung tumors were initially inspected macroscopically under a Stemi DV4 stereoscope equipped with a micrometric scale incorporated into one eyepiece and an AxiocamERc 5 s camera ( Zeiss , Jena , Germany ) in trans-illumination mode , allowing for visualization of both superficial and deeply-located lung tumors ( Stathopoulos et al . , 2007; Vreka et al . , 2018 ) . Tumor location was charted and diameter ( δ ) was measured . Tumor number ( multiplicity ) per mouse was counted and mean tumor diameter per mouse was calculated as the average of individual diameters of all tumors found in a given mouse lung . Individual tumor volume was calculated as πδ3/6 . Mean tumor volume per mouse was calculated as the average of individual volumes of all tumors found in a given mouse lung , and total lung tumor burden per mouse as their sum . Following macroscopic mapping of lung and lung tumor morphology , lungs of fluorescent reporter mice were imaged on a Leica MZ16F fluorescent stereomicroscope equipped with GFP and RFP filters and a DFC 300FX camera ( Leica Microsystems , Heidelberg , Germany ) in order to determine their macroscopic fluorescent pattern . Lung volume was measured by saline immersion , and lungs were embedded in paraffin , randomly sampled by cutting 5 μm-thick lung sections ( n = 10/lung ) , mounted on glass slides , and stained with hematoxylin and eosin for morphometry and histologic typing of lung tumors . For this , a digital grid of 100 intersections of vertical lines ( points ) was superimposed on multiple digital images of all lung sections from lung tissue of a given mouse using Fiji academic freeware ( https://fiji . sc/ ) . Total lung tumor burden was determined by point counting of the ratio of the area occupied by neoplastic lesions versus total lung area and by extrapolating the average ratio per mouse to total lung volume ( Hsia et al . , 2010 ) . The results of this stereologic approach were compared with the macroscopic method , and were scrutinized if deviant by >20% . To evaluate alveolar structure and size , we calculated mean linear intercept using randomly sampled hematoxylin and eosin-stained lung sections , as described elsewhere ( Hsia et al . , 2010 ) . For this , a digital grid of twenty random horizontal lines was superimposed on multiple digital images of all lung sections from lung tissue of a given mouse using Fiji . Mean linear intercept was calculated by counting the intercepts of interalveolar septae with the lines and the formula: Σ{2 x ( length of line/number of intercepts ) }/total number of lines . All quantifications were done by counting at least five random non-overlapping fields of view of at least ten sections per lung . For histology , lungs were inflated to 20 cmH2O pressure that provides for a lung volume equivalent to the resting volume of the lungs ( a . k . a . functional residual capacity in humans ) and enables precise histologic observations on airway and alveolar structure avoiding false interpretations resulting from the study of compressed or over-inflated lungs ( Hsia et al . , 2010 ) . Subsequently , lungs were fixed with 10% formalin overnight and were embedded in paraffin . Five-μm-thick paraffin sections were then counterstained with hematoxylin and eosin ( Sigma , St . Louis , MO ) and mounted with Entellan New ( Merck Millipore , Darmstadt , Germany ) . For immunofluorescence , lungs were inflated with a 2:1 mixture of 4% paraformaldehyde:Tissue-Tek ( Sakura , Tokyo , Japan ) , fixed in 4% paraformaldehyde overnight at 4°C , cryoprotected with 30% sucrose , embedded in Tissue-Tek and stored at −80°C . Ten-μm cryosections were then post-fixed in 4% paraformaldehyde for 10 min , treated with 0 . 3% Triton X‐100 for 5 min , and incubated in blocking solution containing 10% fetal bovine serum ( FBS ) , 3% bovine serum albumin ( BSA ) , 0 . 1% polyoxyethylene ( 20 ) sorbitanmonolaurate ( Tween 20 ) in 1x phosphate-buffered saline ( PBS ) for 1 hr . Following labeling with the indicated primary antibodies overnight at 4°C , sections were incubated with fluorescent secondary antibodies , counterstained with Hoechst 33258 and mounted with Mowiol 4–88 ( Calbiochem , Darmstadt , Germany ) . The following primary antibodies were used: rabbit anti-proliferating cell nuclear antigen ( PCNA , 1:3000 dilution , ab2426 , Abcam , London , UK ) , rabbit anti-LYZ2 ( 1:50 dilution , ab108508 , Abcam ) , rabbit anti-KRT5 ( 1:200 dilution , ab53121 , Abcam ) , rabbit anti-SFTPC ( 1:200 dilution , sc-13979 , Santa Cruz , Dallas , TX ) , rabbit anti-CCSP ( 1:200 dilution , sc-25555 , Santa Cruz ) , goat anti-CCSP ( 1:1000 dilution , sc-9772 , Santa Cruz ) , mouse anti-acetylated α-tubulin ( 1:2000 dilution , T7451 , Sigma-Aldrich , St . Lewis , MO ) , rabbit anti-SFTPC ( 1:500 dilution , AB3786 , Merck-Millipore , Burlington , MA ) , and mouse anti-KRT5 ( 1:200 dilution , MA5-17057 , Thermo Fisher Scientific , Waltham , MA ) . Alexa Fluor donkey anti-rabbit 488 ( A21206 , Thermo Fisher Scientific ) , Alexa Fluor donkey anti-mouse 568 ( ab175700 , Abcam ) , Alexa Fluor donkey anti-goat 568 ( A11057 , Thermo Fisher Scientific ) , Alexa Fluor donkey anti-rabbit 647 ( A31573 , Thermo Fisher Scientific ) , and Alexa Fluor donkey anti-mouse 647 ( A31571 , Thermo Fisher Scientific ) secondary antibodies were used at 1:500 dilution . For isotype control , the primary antibody was omitted . Bright-field images were captured with an AxioLab . A1 microscope connected to an AxioCamERc 5 s camera ( Zeiss , Jena , Germany ) whereas fluorescent microscopy was carried out either on an Axio Observer D1 inverted fluorescent microscope ( Zeiss , Jena , Germany ) or a TCS SP5 confocal microscope ( Leica Microsystems , Wetzlar , Germany ) with 20x , 40x and 63x lenses . Digital images were processed with Fiji . All quantifications of cellular populations were obtained by counting at least five random non-overlapping bronchial- , alveolar- , hyperplasia- , or tumor- containing fields of view per section . Following anesthesia induced by intraperitoneal ketamine ( 100 mg/Kg ) and xylazine ( 10 mL/Kg ) and tracheostomy , mice were mechanically ventilated by a Flexivent rodent ventilator ( Scireq , Montreal , Ontario , Canada ) . The whole procedure , described elsewhere ( Manali et al . , 2011 ) , lasted 15 min . After a 3 min run-in period of ventilation with 21% oxygen , a tidal volume of 10 mL/Kg , a respiratory rate of 150 breaths/min , and a positive end-expiratory pressure of 3 cmH2O , paralysis was induced using 8 mg/Kg intraperitoneal succinyl choline , and total respiratory system impedance was obtained by applying an 8-sec-long pseudorandom frequency oscillation ( 0 . 5–19 . 75 Hz ) to the airway opening . Thirty seconds prior to initiation of measurements , lung volume history was once controlled by a 6-sec-long inflation to 30 cm H2O pressure . Measurements were repeated thrice at 60 s intervals and were averaged . Data were fit into the constant phase model in order to fractionate total respiratory input impedance into airways resistance ( Raw ) and tissue damping and elastance coefficients . To obtain pressure-volume ( PV ) curves , the respiratory system was incrementally inflated and deflated to 40 mL/Kg total volume at seven steps each and airway pressures were recorded on each volume change . The slope of the linear portion of expiratory PV curves , which represents static compliance ( Cst ) , a measure of airspace function , was calculated manually . Operators were blinded to animal genotype . TOMATO , GFP;CCSP . CRE , and GFP;LYZ2 . CRE mice ( FVB strain ) received one intraperitoneal injection of urethane ( 1 g/Kg ) and lungs were then harvested one and two weeks post-urethane , homogenized , and subjected to DNA extraction and purification using GenElute Mammalian Genomic DNA Miniprep Kit ( Sigma-Aldrich , St . Louis , MO ) . DNA concentration and quality were assessed using a Nanodrop 1000 spectrophotometer ( Thermo Fisher Scientific , Waltham , MA ) . DNA concentration was converted to number of diploid copies according to the formula: DNA ( ng/µL ) /weight of mouse diploid genome ( 3 . 9 pg ) . Digital droplet PCR protocol and analysis was performed as described previously using reagents , equipment and software from BioRad Laboratories Inc ( Hercules , CA ) ( Mazaika and Homsy , 2014 ) . In brief , 20000 genome copies were used . Samples were normalized internally according to the number of accepted droplets and inter-sample normalization was performed according to the formula [x-min ( x ) ]/[max ( x ) -min ( x ) ] , where x represents the actual , min ( x ) the minimum , and max ( x ) the maximum number of accepted droplets . The data were reported as % positive/accepted droplets . Sequences of KrasQ61R primers and probe were: KrasQ61R forward: ATCTGACGTGCTTTGCCTGT , KrasQ61R reverse: CCCTCCCCAGTTCTCATGTA , and KrasQ61R probe: GACACAGCAGGTCAAGAGGAGTACA . The RosamT assay is registered as dCNS685684912 ( Bio-Rad ) with MIQE context: seq1:195–315:+CCAGTTCATGTACGGCTCCAAGGCGTACGTGAAGCACCCCGCCGACATCCCCGATTACAAGAAGCTGTCCTTCCCCGAGGGCTTCAAGTGGGAGCGCGTGATGAACTTCGAGGACGGCGGTCT . Primers and fluorescently labeled probes were combined in a mixture containing 18 μM forward and reverse primers and 5 μM labeled probes ( 20x primer/Taqman probe mix ) . Reactions were assembled to contain 12 . 5 μL 2x ddPCR mix no-UTP , 1 . 25 μL 20x KrasQ61Rprimer/Taqman probe Mix , 1 . 25 μL 20x RosamT custom primer/Taqman probe Mix and 10 μL DNA diluted in nuclease-free water . The ddPCR protocol included a first denaturation step at 95°C for 10 min followed by 40 cycles of denaturation at 95°C for 30 s and 40 cycles of annealing at 62 . 5°C for 60 s , and was performed in a BioRad T100 Thermal cycler . Results were analyzed with a BioRad QX100 droplet reader using the QuantaSoft software . The amplitude gathering thresholds of positive droplets were set at 3500 for the RosamT and at 10000 for the KrasQ61R probe , according to the manufacturer’s instructions . BAL was performed using three sequential aliquots of 1000 µL sterile ice-cold phosphate-buffered saline ( PBS ) . Fluid was combined and centrifuged at 260 g for 10 min to separate cells from supernatant . The cell pellet was resuspended in 1 ml PBS containing 2% fetal bovine serum , and the total cell count was determined using a grid hemocytometer according to the Neubauer method . Cell differentials were obtained by counting 400 cells on May-Grünwald-Giemsa-stained cytocentrifugal specimens . Total BAL cell numbers were calculated by multiplying the percentage of each cell type by total BAL cell number ( Stathopoulos et al . , 2007; Vreka et al . , 2018 ) . LUC;CCSP . CRE mice , bioluminescent reporters of CCSP-labeled cell mass , received one intraperitoneal injection of saline ( 100 μL saline ) or urethane ( 1 g/Kg in 100 μL saline ) and were serially imaged before treatment start , and at 150 and 210 days into treatment . Imaging was done on a Xenogen Lumina II ( Perkin-Elmer , Waltham , MA ) 5–20 min after delivery of 1 mg D-Luciferin sodium in 100 μL of sterile water to the retro-orbital vein , and data were analyzed using Living Image v . 4 . 2 ( Perkin-Elmer , Waltham , MA ) ( Stathopoulos et al . , 2007; Vreka et al . , 2018 ) . Triplicate cultures of 106 LUAD cells , BMDM ( obtained by 1 week bone marrow incubation with 100 ng/mL M-CSF ) , and tracheal AEC ( obtained by 1 week incubation of stripped mouse tracheal epithelium in DMEM ) were subjected to RNA extraction using Trizol ( Thermo Fisher ) followed by column purification and DNA removal ( Qiagen , Hilden , Germany ) . Whole lungs were homogenized in Trizol followed by the same procedure . Pooled RNA ( 5 μg ) was quality tested ( ABI 2000 Bioanalyzer; Agilent Technologies , Sta . Clara , CA ) , labeled , and hybridized to GeneChip Mouse Gene 2 . 0 ST arrays ( Affymetrix , Sta . Clara , CA ) . All data were deposited at GEO ( http://www . ncbi . nlm . nih . gov/geo/; Accession ID: GSE94981 ) and were analyzed on the Affymetrix Expression and Transcriptome Analysis Consoles together with previously reported ( Frank et al . , 2016; Kabbout et al . , 2013; Clark et al . , 2015; Dancer et al . , 2015; Lee et al . , 2009 ) murine ATII and human AEC , ATII , AMΦ , non-smokers lung , and LUAD microarray data ( Accession IDs: GSE82154 , GSE55459 , GSE46749 , GSE18816 , GSE43458 ) . qPCR was performed using first strand synthesis with specific primers ( Scgb1a1: ATCACTGTGGTCATGCTGTCC and GCTTCAGGGATGCCACATAAC; Sftpc: TCGTTGTCGTGGTGATTGTAG and AGGTAGCGATGGTGTCTGCT; Gusb: TTACTTTAAGACGCTGATCACC and ACCTCCAAATGCCCATAGTC ) and SYBR FAST qPCR Kit ( Kapa Biosystems , Wilmington , MA ) in a StepOne cycler ( Applied Biosystems , Carlsbad , CA ) . Ct values from triplicate reactions were analyzed with the 2-ΔCT method relative to Gusb . BAL cells were suspended in 50 μL PBS with 2% FBS and 0 . 1% NaN3 , were stained with anti-CD45 ( #11-0451-85; eBioscience; Santa Clara , CA ) and anti-CD11b ( #12-0112-82; eBioscience; Santa Clara , CA ) primary antibodies for 20 min in the dark at 0 . 5 μL antibody per million cells , and were analyzed on a CyFlowML cytometer with a sorter module using FloMax Software ( Partec , Darmstadt , Germany ) or FlowJo software ( TreeStar , Ashland , OR ) , as described previously ( Kanellakis et al . , 2019 ) . Perfused lungs were digested in RPMI-1640 medium containing collagenase XI ( 0 . 7 mg/mL; Sigma , St . Louis , MO ) and type IV bovine pancreatic DNase ( 30 μg/mL; Sigma , St . Louis , MO ) to obtain single-cell suspensions . After treatment with red blood cell lysis buffer ( BioLegend; San Diego , CA ) , single-cell suspensions were analyzed on a LSR II flow cytometer ( BD Bioscience , San Diego , CA ) , and data were examined with FlowJo . Dead cells were excluded using 4 , 6-diamidino-2-phenylindole ( DAPI; Sigma , St . Louis , MO ) . GSEA was performed with the Broad Institute pre-ranked GSEA module software ( http://software . broadinstitute . org/gsea/index . jsp ) ( Subramanian et al . , 2005 ) . In detail , genes significantly expressed ( log2 normalized expression >8 ) in murine tracheal airway cells , ATII cells ( Frank et al . , 2016 ) , and BMDM were cross-examined against the murine lung and chemical-induced LUAD cell line transcriptomes . In addition , previously reported human AEC , ATII , and AMΦ cellular signatures ( Clark et al . , 2015; Dancer et al . , 2015; Lee et al . , 2009 ) were cross-examined against the previously described transcriptomes of human normal lung tissue from never-smokers and of LUAD ( Kabbout et al . , 2013 ) . Sample size was calculated using power analysis on G*power ( http://www . gpower . hhu . de/ ) , assuming α = 0 . 05 , β = 0 . 05 , and effect size d = 1 . 5 ( Faul et al . , 2007 ) . No data were excluded from analyses . Animals were allocated to treatments by alternation and transgenic animals were enrolled case-control-wise . Data were collected by at least two blinded investigators from samples coded by non-blinded investigators . All data were normally distributed by Kolmogorov-Smirnov test , are given as mean ± SD , and sample size ( n ) always refers to biological and not technical replicates . Differences in frequency were examined by Fischer’s exact and χ2 tests and in means by t-test or one-way ANOVA with Bonferroni post-tests . Changes over time and interaction between two variables were examined by two-way ANOVA with Bonferroni post-tests . All probability ( P ) values are two-tailed and were considered significant when p<0 . 05 . All analyses and plots were done on Prism v8 . 0 ( GraphPad , La Jolla , CA ) . All raw data produced in this study are provided as * . xlsx source data supplements . The microarray data produced by this study were deposited at GEO ( http://www . ncbi . nlm . nih . gov/geo/; Accession ID: GSE94981 ) . Previously reported murine ATII and human AEC , ATII , AMΦ , non-smokers lung , and LUAD microarray data are available at GEO using Accession IDs GSE82154 , GSE55459 , GSE46749 , GSE18816 , and GSE43458 ) . | The deadliest form of lung cancer is called lung adenocarcinoma , or LUAD . Tobacco chemicals often cause the disease by damaging the genetic information of lung cells . The damage leads to harmful changes in the DNA sequence which prompt the cells to form tumors . For instance , the most common of these changes takes place in a gene called KRAS . However , it is still unclear exactly which type of lung cells are more likely to develop into a tumor . In the lungs , airway epithelial cells cover the inside of the passages that bring the air inside little sacks called alveoli , which are lined by alveolar cells . Previous studies have used genetic methods to switch on the KRAS mutation in different compartments of the mouse lung . This showed that groups of airway cells , of alveolar cells , and of a class of cells located at the junction between airways and alveoli could all give rise to cancer . However , these experiments did not examine how tobacco chemicals could give rise to tumors in different groups of lung cells . Here , Spella et al . triggered LUAD in adult mice by exposing them to the toxic chemicals found in tobacco smoke , but without making any change to the KRAS gene . These mice also had genetically engineered reporters that could be used to deduce where the resulting tumors came from . DNA sequencing showed that the airway epithelial cells gained KRAS mutations after the chemical treatment . When the airway epithelial cells were experimentally removed before the treatments with tobacco chemicals , these mice did not get LUAD tumors . Spella et al . also observed that the tobacco-induced tumors came from the epithelial cells in the airways , and not from the cells in the alveoli . Moreover , when the lung was damaged , airway cells could move to the alveoli and start adopting the identity of alveolar cells , thereby replenishing this population . Together , these experiments imply that tobacco-induced LUAD starts in the airway epithelial cells . These findings suggest that airway epithelial cells could be targeted to stop lung cancer early on . Further studies should also examine how airway epithelial cells can transition to look more like alveolar cells when the lungs get harmed . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"cell",
"biology",
"cancer",
"biology"
] | 2019 | Club cells form lung adenocarcinomas and maintain the alveoli of adult mice |
Type II topoisomerases manage DNA supercoiling and aid chromosome segregation using a complex , ATP-dependent duplex strand passage mechanism . Type IIB topoisomerases and their homologs support both archaeal/plant viability and meiotic recombination . Topo VI , a prototypical type IIB topoisomerase , comprises two Top6A and two Top6B protomers; how these subunits cooperate to engage two DNA segments and link ATP turnover to DNA transport is poorly understood . Using multiple biochemical approaches , we show that Top6B , which harbors the ATPase activity of topo VI , recognizes and exploits the DNA crossings present in supercoiled DNA to stimulate subunit dimerization by ATP . Top6B self-association in turn induces extensive DNA bending , which is needed to support duplex cleavage by Top6A . Our observations explain how topo VI tightly coordinates DNA crossover recognition and ATP binding with strand scission , providing useful insights into the operation of type IIB topoisomerases and related meiotic recombination and GHKL ATPase machineries .
The appropriate control of transcription , DNA replication , and chromosome segregation are essential to cell proliferation . These three processes are antagonized , however , by the double helical structure of DNA , which supercoils in response to helicase and polymerase activity and which promotes chromosome interlinkage during replicative synthesis ( reviewed in [Vos et al . , 2011; Wang , 2002] ) . In all cells , enzymes known as topoisomerases are used to overcome the natural topological impediments arising from these physical transactions that impinge on DNA . Many different topoisomerase families exist to relieve super-helical tension and DNA entanglements , all of which transiently form either single strand or double strand breaks to manipulate DNA topology ( reviewed in [Chen et al . , 2013; Forterre and Gadelle , 2009; Schoeffler and Berger , 2008] ) . Type II topoisomerases introduce transient double strand breaks into DNA , and play a key role in unlinking catenated DNA molecules ( Holm et al . , 1985; Spell and Holm , 1994; Zechiedrich and Cozzarelli , 1995 ) . The type IIA subfamily of topoisomerases , which are principally used in bacteria and eukaryotes , utilize a so-called ‘two-gate’ mechanism ( Roca et al . , 1996; Roca and Wang , 1994; Roca and Wang , 1992 ) , in which one DNA duplex ( termed the transport- or ‘T’-segment ) is captured by one half of the enzyme , actively passed through a second , protein-bound DNA duplex ( the gate- or ‘G’-segment ) , and expelled through the other end of the enzyme . T-segment capture is regulated by the ATP-dependent closure of one subunit dimer interface ( Roca and Wang , 1992; Wigley et al . , 1991 ) , referred to as the ‘ATP-gate’ , while a pair of catalytic tyrosine residues responsible for G-segment cleavage and opening reside in a second , separable subunit-subunit contact point termed the ‘DNA-gate’ ( Berger et al . , 1996; Morais Cabral et al . , 1997; Morrison and Cozzarelli , 1979; Tse et al . , 1980 ) . In most instances , repeated cycles of ATP binding and hydrolysis allow for the processive removal of multiple DNA crossings . The use of ATP by type II topoisomerases in general has been proposed to serve as a mechanism for preventing the inappropriate formation of potentially cytotoxic DNA breaks ( Bates et al . , 2011 ) ; however , the molecular basis for the coupling between ATP turnover and DNA cleavage has remained enigmatic for the superfamily as a whole . The type IIB topoisomerases , which are exemplified by DNA topoisomerase VI ( topo VI ) ( Bergerat et al . , 1997; Bergerat et al . , 1994; Forterre et al . , 2007 ) , share evolutionarily conserved catalytic elements with their type IIA counterparts but are structurally distinct ( Corbett and Berger , 2003; Nichols et al . , 1999 ) . Topo VI comprises an A2B2 heterotetramer formed by two Top6A and two Top6B subunits: Top6A forms a ‘U’-shaped dimer that serves as the DNA-gate for G-segment cleavage and opening ( Bergerat et al . , 1997; Nichols et al . , 1999 ) , while Top6B constitutes the ATP-gate and dimerizes in response to nucleotide binding ( Corbett and Berger , 2003 ) . Topo VI is thought to serve as the primary topoisomerase for DNA decatenation and supercoil relaxation in archaea and is required for endoreduplication and cell growth in plants ( Bergerat et al . , 1997; Bergerat et al . , 1994; Hartung et al . , 2002; Sugimoto-Shirasu et al . , 2002; Yin et al . , 2002 ) . Topo VI is also found sporadically throughout the bacterial domain , and a single chain variant , topo VIII , is found in certain plasmid-based mobile elements as well ( Forterre et al . , 2007; Gadelle et al . , 2014 ) . Interestingly , the type IIB topoisomerase scaffold has been co-opted to serve as the machinery responsible for introducing double-strand DNA breaks to initiate meiotic recombination in eukaryotes ( Bergerat et al . , 1997; Keeney et al . , 1997; Robert et al . , 2016; Vrielynck et al . , 2016 ) . How topo VI and its cousins engage DNA segments has yet to be determined . The ATPase region is generally well-preserved between type IIA and type IIB topoisomerases , with the exception of an additional helix-two-turn-helix ( H2TH ) domain of unknown function found in topo VI and topo VIII ( Bergerat et al . , 1997; Corbett and Berger , 2003; Gadelle et al . , 2014; Wigley et al . , 1991 ) . By contrast , the catalytic domains that comprise the DNA breakage-reunion region of type IIA and IIB enzymes have been extensively shuffled . One consequence of this rearrangement is that Top6A lacks a third subunit-subunit interface present in the type IIA enzymes , the ‘C-gate’ dimerization domain , which is thought to help mitigate the risk of aberrant double-strand break formation ( Bates et al . , 2011; Berger et al . , 1996; Nichols et al . , 1999; Roca , 2004 ) . To compensate for the loss of this element , type IIB topoisomerases appear to have evolved a stringent mechanism for controlling strand scission by Top6A that represses transesterase activity until ATP productively binds to Top6B ( Buhler et al . , 1998; Buhler et al . , 2001 ) . How Top6B activates Top6A is unknown; however , given that Spo11 , a paralog of Top6A used in meiotic recombination , is also thought to require activation for DNA cleavage , aspects of this control mechanism may be broadly conserved ( Lam and Keeney , 2014 ) . To better understand how type IIB topoisomerases coordinate DNA cleavage , we performed a comprehensive biochemical investigation of Methanosarcina mazei topo VI , a model mesophilic type IIB topoisomerase . We find that topo VI discriminates between linear and supercoiled DNA using an extensive and unanticipated DNA binding interface that specifically recognizes DNA crossings . Both gate closure and ATP hydrolysis by Top6B as well as transesterase activity by Top6A require engagement along this entire interface . Site-directed mutagenesis studies show that three conserved , positively charged regions on Top6B sense both the DNA bends and crossings present in supercoiled substrates and further serve to couple the binding of DNA crossings to B-subunit dimerization , nucleotide turnover , and DNA strand scission . Our results explain why type IIB topoisomerases absolutely depend upon the ATPase activity of the B-subunit to generate double strand breaks . These observations in turn reinforce the functional importance for DNA bending and potential T-segment-sensing elements in the related type IIA topoisomerases , and also provide insights as to how recently discovered meiotic Top6B homologs might promote Spo11 mediated strand scission during meiotic recombination .
We began our investigations of type IIB topoisomerase mechanism by measuring the affinity of M . mazei topo VI for DNAs of varying length or topological status . The relative affinity of the holoenzyme for fluorescein-labeled duplex DNAs ranging from 20 bp to 70 bp in length was assessed using a fluorescence anisotropy-based approach ( the predicted G-segment binding channel of a Top6A dimer is ~16–20 bp in length [Nichols et al . , 1999] ) . The DNA sequence used for these oligomers was based on a previously determined cleavage hotspot for Sulfolobus shibatae topo VI ( Buhler et al . , 2001 ) ( Figure 1—source data 1 ) . These experiments showed that whereas a 20 bp duplex binds relatively weakly to topo VI , apparent affinity increases with length , plateauing between 40 and 70 bp ( Figure 1A , Figure 1—source datas 2–3 ) . As the binding isotherms did not show any sign of complex interactions ( such as cooperativity ) and could be fit well by a single-site binding model ( Heyduk and Lee , 1990 ) , this result provided the first clue that topo VI might have more extensive interactions with DNA than previously hypothesized . To determine whether topo VI displayed any preference for the topological status of DNA , the relative binding affinities of the holoenzyme were next assessed for supercoiled plasmid vs . sheared , linear salmon-sperm DNA using a competitive binding assay . Topo VI was incubated with the fluorescein-labeled 70 bp duplex DNA and varying amounts of unlabeled , supercoiled plasmid or sheared salmon-sperm DNA . The relative affinity of topo VI for each substrate was determined by monitoring how well the competitor DNAs interfered with binding of the labeled probe . The response to the titration of supercoiled DNA or sheared salmon-sperm DNA was fit to an explicit competitive binding model ( Wang , 1995 ) to indirectly estimate affinities for the unlabeled substrates . Based on these measurements , topo VI showed a ~60-fold preference for supercoiled DNA ( KI , app = 0 . 6 ± 0 . 3 nM ) compared to sheared salmon-sperm DNA ( KI , app = 39 . 3 ± 2 . 6 nM ) ( Figure 1B ) . The difference in affinity between supercoiled and linear DNA suggested that topo VI might preferentially engage supercoiled substrates by binding to DNA crossings , DNA bends , or both . To distinguish between these modes , we examined the time-dependent processivity of topo VI in relaxing negatively supercoiled DNA . For type II topoisomerases in general , processivity describes the ability of a single enzyme to remain bound to a G-segment DNA during multiple strand passage events . For topo VI , the progress of ATP-dependent supercoil relaxation was followed by native agarose-gel electrophoresis , using a slight molar excess of plasmid over enzyme to disfavor the binding of two topo VI molecules to a single DNA substrate ( Figure 1C ) . A highly processive topoisomerase , such as Saccharomyces cerevisiae topoisomerase II ( ScTop2 ) , removes the majority of supercoils on a closed circular DNA in a single enzyme-DNA encounter ( Figure 1—figure supplement 1 ) as evidenced by a paucity of intermediate DNA topoisomers between the supercoiled substrate and fully relaxed plasmid product . In contrast , topo VI produced a broad distribution of intermediate topoisomers that were gradually converted to the fully relaxed distribution , a behavior more consistent with low processivity . To more thoroughly investigate supercoil processing by topo VI , we followed plasmid relaxation using two differently sized plasmids in a chase experiment . Following pre-incubation of a defined amount of topo VI with a slight molar excess of a primary 2 . 9 kb plasmid , a second , larger plasmid ( 6 . 5 kb ) was added along with ATP to serve as a competing substrate for any dissociated enzymes ( Figure 1D ) . In the case of a processive enzyme such as ScTop2 , the competing plasmid does not alter the initial rate at which a fully relaxed topoisomer distribution of the primary plasmid is generated ( Figure 1—figure supplement 1 ) . By contrast , topo VI again displayed clearly distributive behavior , relaxing both plasmids more slowly and simultaneously . Although assay conditions can modulate whether a topoisomerase acts processively or distributively ( salt concentration in particular ) , both timecourse experiments were run under low-salt conditions where type IIA topoisomerases are primarily processive . Collectively , these findings demonstrate that topo VI operates by a principally distributive mechanism , whereby once a DNA crossing is resolved by strand passage , the enzyme will tend to dissociate from the substrate before acting on a new crossing and/or bent DNA segment . A defining characteristic of type II topoisomerases is the coupling of ATP turnover with efficient and rapid strand passage . In type IIA topoisomerases , DNA binding strongly stimulates ATPase activity ( Lee et al . , 2013; Lindsley and Wang , 1993; Liu et al . , 1979; Mizuuchi et al . , 1978; Osheroff et al . , 1983; Sugino and Cozzarelli , 1980 ) . However , the coupling of DNA topological state to the magnitude of the ATP hydrolysis stimulation varies between different type IIA homologs ( Anderson et al . , 1998; Gubaev and Klostermeier , 2011; Harkins and Lindsley , 1998; McClendon et al . , 2005; Osheroff et al . , 1983; Sugino and Cozzarelli , 1980; Vaughn et al . , 2005 ) . To determine whether the ATPase activity of type IIB topoisomerases is stimulated in a DNA topology-dependent manner , we examined nucleotide turnover by wild-type topo VI in the absence and presence of linear sheared salmon-sperm DNA and supercoiled plasmid DNA substrates at varying ATP concentrations using a coupled assay . Hydrolysis was also measured for an ATPase-deficient topo VI construct ( Top6BE44A ) to identify non-specific activity arising from contaminating ATPases ( Figure 2A , Figure 2—figure supplement 1 ) . Although topo VI likely hydrolyzes ATP cooperatively , the data conformed to apparent Michalis-Menten behavior and were fit to this model ( Figure 2—source datas 1–2 ) . Topo VI showed negligible basal ATPase activity , and required the addition of a DNA substrate to hydrolyze ATP . Incubation with supercoiled DNA produced the maximal observed rate of hydrolysis ( and decreased the Km , app for ATP ) , resulting in a ~5-fold increase in catalytic efficiency ( kcat , app/Km , app ) over that for sheared salmon-sperm DNA . The observation that supercoiled DNA is more effective than linear substrates in activating ATP turnover indicates that topo VI not only interrogates DNA for specific topological features , but that its activity is potentiated when such features are recognized . ATP binding and hydrolysis by topo VI and many other enzymes that share its GHKL ATPase fold ( e . g . type IIA topoisomerases , Hsp90 , MutL , and MORC ATPases ) rely on nucleotide-dependent dimerization of ATP-binding domains to elicit biological activity ( Ali et al . , 2006; Ban et al . , 1999; Ban and Yang , 1998; Dutta and Inouye , 2000; Li et al . , 2016; Shiau et al . , 2006; Wigley et al . , 1991 ) . A mechanism in which supercoiled DNA binding , in particular T-segment engagement , promotes Top6B dimerization could thus explain why supercoiled DNA stimulates ATP turnover . To test this idea , we developed a Förster Resonance Energy Transfer ( FRET ) assay to monitor ATPase domain dimerization in the context of the topo VI holoenzyme . We first identified and mutated surface cysteines to non-reactive residues to create a fully functional ‘cys-lite’ construct of the holoenzyme . Thr155 of Top6B was then substituted with cysteine ( Figure 2—figure supplement 2A ) . Dual labeling with donor ( Alexa 555-maleimide ) and acceptor ( Alexa 647-maleimide ) fluorophores yielded an enzyme population containing an expected labeled mixture of correctly labeled donor-acceptor enzymes ( 50% ) , and both acceptor-acceptor ( 25% ) and donor-donor ( 25% ) labeled enzymes ( Figure 2—figure supplement 2B; labeling efficiency was determined by spectral absorption ) . The labeled topo VI holoenzymes were able to fully relax DNA and showed only a slight impairment ( ~2 fold ) of overall specific activity compared to wild-type topo VI ( Figure 2—figure supplement 2C ) . Using the labeled enzyme , bulk FRET efficiencies in the absence and presence of either linear or supercoiled DNA were first measured by scanning the spectral emission of both donor and acceptor fluorophores under excitation at 530 nm . The conformational response of the enzyme to AMPPNP , a non-hydrolyzable ATP analog , was then assessed for the enzyme alone and in the presence of each substrate over time ( Figure 2B ) . The addition of sheared salmon-sperm DNA and to a greater extent supercoiled DNA , led to minor but reproducible increases in FRET efficiency ( Figures 2C , 0 min time-point ) , suggesting that DNA binding alone alters the conformation of Top6B in the holoenzyme . By comparison , the addition of AMPPNP led to larger FRET responses , and FRET efficiency increased much more rapidly with supercoiled DNA compared to linear sheared salmon-sperm DNA . AMPPNP alone produced detectable but minor FRET changes when DNA was omitted , indicating that duplex binding is needed for ATPase domain dimerization ( Figure 2D , 120 min time-point ) . In conjunction with the ATPase data , these observations show that – unlike type IIA topoisomerases , whose ATPase regions efficiently dimerize in the absence of DNA ( [Gubaev and Klostermeier , 2011; Roca and Wang , 1992] and Figure 2—figure supplement 3 ) – topo VI utilizes the DNA geometries presented by supercoiled substrates to help favor nucleotide-dependent conformational changes associated with strand passage . Based on the ability of topo VI to recognize and utilize topological features in supercoiled DNA to promote activity , we set out to identify the structural elements responsible for this coupling . Working from an assumption that topology-sensing elements might consist in part of positively charged residues on the B subunit , we mapped both amino acid conservation ( derived from a multiple sequence alignment of Top6B homologs ) and electrostatic surface potential onto the known structure of M . mazei Top6B using ConSurf and ABPS ( Figure 3A–C and ( Ashkenazy et al . , 2010; Baker et al . , 2001 ) ) . By comparing positively charged interfaces against sequence conservation , we identified three different regions as candidate DNA interaction sites . The first prospective locus consisted of a small loop of basic residues ( KGRR186-189 ) ( Figure 3D ) within the predicted T-segment storage cavity of topo VI . A second feature comprised a trio of conserved basic residues ( R457 , K399 and K401 ) that are found within two spatially adjacent structural elements ( Figure 3E ) : the C-terminal , α-helical stalk of Top6B ( which connects the so-called ‘transducer’ domain of this subunit to Top6A ) , and a loop in the transducer domain containing the so-called ‘WKxY motif , ’ which is conserved in both Top6B and many meiotic Top6B-like proteins ( Robert et al . , 2016 ) . The third area of note , the H2TH domain , is embedded between the topo VI GHKL and transducer regions . The function of the topo VI H2TH domain has not been established , but this type of fold serves as a general nucleic-acid-binding element in a diverse number of proteins , including FpG/Nei DNA glycosylases , s13 ribosomal proteins , and sIHF type nucleoid-associated proteins ( Brodersen et al . , 2002; Sugahara et al . , 2000; Swiercz et al . , 2013; Zharkov et al . , 2002 ) . Comparison of nucleic-acid-bound H2TH domain structures with Top6B ( Figure 3—figure supplement 1 ) highlighted R263 , K268 and K308 as candidate residues that might interact with DNA ( Figure 3F ) . Having identified three potential sites for supercoil sensing on the surface of Top6B , six constructs were generated to assess the functional attributes of each region . Selected constructs included triple-neutral and triple-acidic mutations to the basic storage-cavity loop ( KGRR→AGAA and EGEE , referred to as KGRRAAA and KGRREEE ) , the C-terminal stalk ( Stalk/WKxYAAA and Stalk/WKxYEEE ) , and the H2TH domain ( H2THAAA and H2THEEE ) . All six mutant topo VI holoenzymes were soluble upon expression , purified to homogeneity ( as judged by SDS-PAGE ) , and appeared well-behaved based on gel-filtration chromatography profiles as compared to the wild-type enzyme ( Figure 3—figure supplement 2 ) . To assess overall activity , we next looked at the supercoil relaxation activity of the mutant enzymes compared to wild-type topo VI as a function of enzyme concentration ( Figure 4A ) . Both sets of KGRR and Stalk/WKxY mutants ( neutral and acidic ) proved completely unable to relax supercoiled substrate . By contrast , both sets of mutations to the H2TH region led to enzymes that were able to relax supercoiled DNA , but with ~20–30 fold lower efficiency than native topo VI . The activity profiles seen in enzyme titration assays were corroborated by timecourse assays at a fixed enzyme concentration ( Figure 4—figure supplement 1 ) . In some of these experiments , the open-circle ( nicked ) plasmid species increased over time; however , this increase was independent of both nucleotide and topo VI , and thus does not reflect an elevated nicking activity of the mutants . Collectively , these findings show that the KGRR loop and the Stalk/WKxY region are essential components for topo VI function , but that the H2TH domain , while important , is not strictly required for strand passage . To further investigate the role of each DNA-binding interface in the topo VI reaction cycle , the stimulatory effect of sheared salmon-sperm DNA and supercoiled DNA upon ATP hydrolysis activity of the six mutants was compared to wild-type enzyme . ATP hydrolysis rates were again measured using a coupled assay; however , ATP was held at 2 mM for these experiments , while the concentration of DNA substrate was varied to characterize the stimulatory effects of each substrate on each enzyme ( Figure 4B , Figure 4—source datas 1–2 ) . No DNA-stimulated ATP turnover was observed for either of the Stalk/WKxY mutants . Interestingly , the H2THAAA and H2THEEE mutants , which exhibited large defects in strand passage , showed similar levels of ATP hydrolysis stimulation by both DNA substrates as compared to wildtype topo VI . Moreover , whereas no additional ATP turnover was observed for the KGRRAAA and KGRREEE mutants on sheared salmon-sperm DNA , both variants showed an increased maximal rate of DNA-stimulated ATP hydrolysis compared to the wild-type enzyme on supercoiled DNA ( albeit with a more weakly coupled response to DNA concentration than wildtype topo VI or the H2TH mutants as judged by Kstim , DNA ) . All six mutants exhibited basal hydrolysis rates similar to both wild-type topo VI and the ATPase-deficient Top6ABE44A construct ( Figure 4—figure supplement 2 ) , indicating that the DNA-stimulated responses of each topo VI mutant are directly attributable to the introduced alterations . Collectively , these data indicate that the abrogation of strand passage activity by the KGRR loop mutants stems in part from a loss of an essential DNA-sensing motif required to carry out strand passage . However , unlike the Stalk/WKxY mutants , the KGRR loop mutants retain some feature which allows supercoiled , but not short linear DNAs , to promote ATP hydrolysis . Mutations to the H2TH domain additionally appear to largely decouple strand passage from ATP hydrolysis , yet do not appreciably alter the DNA dependence of ATPase activity . This result implies a role for the H2TH domain in facilitating A- and B-subunit coordination to minimize futile cycling . Since all three interfaces identified affect strand passage activity and its coupling to ATP turnover , we next tested whether the observed differences result directly from weakened binding to duplex DNA . Using fluorescence anisotropy , the affinity of each mutant was assessed for a range of duplex lengths ( 30 , 40 , 60 , and 70 bp ) found to exhibit moderate-to-tight binding to wild-type topo VI ( Figure 5A and Figure 5—source datas 1 and 3 ) . As with native topo VI , a single-site binding model adequately described the DNA-binding isotherms for the mutant panel; the one exception was the data for the H2THEEE mutant , which fit better to a cooperative model . This result suggests that charge reversal in the H2TH region may alter how longer duplexes are bound by the enzyme—although the direct binding data suggest that H2THEEE binds longer DNAs better than H2THAAA , both mutants display similar affinities for a 60 and 70 bp duplex in competitive binding experiments ( Figure 5—figure supplement 1 ) , indicating that differences in the fluorophore environment may underlie the higher Kd , app values seen in the direct binding study . Both Stalk/WKxY mutants were compromised for DNA binding overall ( as judged by the maximum observed changes in anisotropy ) , with the magnitude of the binding defects proving more severe for the acidic substitutions . This finding highlights the Stalk/WKxY region of Top6B as an important DNA-binding interface , a finding that helps explain both why the binding affinity of topo VI is higher for DNAs whose length exceeds what is necessary to bind a Top6A dimer alone and why mutations in this region lead to defects in both strand passage and ATP hydrolysis . By comparison , the KGRR loop and the H2TH domain mutants showed either no change or only a moderate decrease ( for the 60 and 70 bp duplexes ) in DNA affinity compared to wildtype topo VI , suggesting that these regions potentially contribute a more peripheral or secondary site of DNA binding . Because the KGRR loop and H2TH domain mutants minimally impacted affinity for short duplex DNAs as compared to the Stalk/WKxY mutants , we wondered whether these motifs might instead contribute to the preferential binding of topo VI seen for supercoiled DNA ( Figure 1B ) . To this end , the relative affinities of supercoiled plasmid and linear , sheared salmon-sperm DNA were assessed for both sets of KGRR and H2TH mutants , using the fluorescence anisotropy-based competition assay described earlier ( Figure 5B , and Figure 5—source datas 2–3 ) . The H2THAAA substitution minimally affected supercoiled DNA binding , whereas the H2THEEE and both KGRR substitutions resulted in a ~10–20 fold decrease of the overall affinity of topo VI for supercoiled DNA , with KGRREEE showing a greater defect than KGRRAAA . Both KGRR substitutions adversely impacted the binding of random linear DNA compared to wild type as well , a result concordant with this mutant’s negligible ATPase activity on sheared salmon-sperm DNA and which further suggests that this set of substitutions may ablate a secondary DNA-binding site on the holoenzyme . Together , these data indicate that both the KGRR loop and H2TH domain contribute to the preferential binding of topo VI to supercoiled substrates as compared to sheared salmon-sperm DNA , but that neither is solely responsible for this discrimination . Rather than contributing to overall DNA affinity , the biochemical and biophysical activities of our topo VI mutants implicate the KGRR loop and H2TH domain in recognizing supercoiled DNA and in coupling ATP hydrolysis to strand passage . Although these two motifs might recognize either the DNA crossings or bends present in plectonemic substrates , we hypothesized that the KGRR element in particular might sense T-segment occupancy directly due to its physical location in the holoenzyme ( Figure 3 ) . To address this question , we designed a fluorescently-labeled , 20 bp by 16 bp Holliday junction substrate that can form a stacked-X structure ( Duckett et al . , 1988; Ortiz-Lombardía et al . , 1999 ) as a mimic of a prospective duplex DNA crossing ( Figure 6A–B ) . Using fluorescence anisotropy , wildtype topo VI was found to bind this substrate nearly 4-fold more tightly than a single 20 bp DNA duplex ( Figure 6C , Figure 6—source datas 1–2 ) . We next asked whether mutations to the KGRR loop or the H2TH domain interfered with binding to the stacked-junction substrate . Whereas both H2TH mutants showed similar increases in affinity for the stacked-junction DNA as seen with native topo VI , the KGRRAAA mutant showed a clear decrease in affinity for this substrate compared to a 20 bp duplex ( and little to no change in affinity for a 20 bp duplex alone , Figure 6C , Figure 6—source datas 1–2 ) . The KGRREEE mutant displayed an even more pronounced defect in junction binding . Collectively , this response implicates the KGRR loop in the binding of DNA crossings by topo VI , potentially as a T-segment-sensing element . We next considered whether the binding of DNA crossings facilitated by the KGRR loop might affect how supercoiled DNA promotes the ATP-dependent dimerization of Top6B ( as observed for native topo VI [Figure 2B] ) , or whether this activity might instead arise from an H2TH domain interaction with supercoiled DNA . To address this question , we added the KGRRAAA and H2THAAA mutations into the topo VI construct used to monitor the conformational status of the ATPase domain by FRET . Following purification and labeling , we measured the emission spectra of both mutants alone and bound to supercoiled DNA . Similar to the wild-type construct , both the KGRRAAA and H2THAAA mutants showed increased FRET efficiencies in the presence of supercoiled substrate , independent of nucleotide ( Figure 6—figure supplement 1 ) . Interestingly , the ‘enzyme alone’ spectra suggest that each mutant alters the resting conformational status of the Top6B dimer compared to wild type , with the H2THAAA mutant taking on a more open state , and the KGRRAAA mutant taking on a more closed state . Although the H2THAAA mutant displayed a lower FRET signal than wild type , both in the presence of supercoiled DNA alone and with supercoiled DNA and nucleotide , the addition of AMPPNP produced a rapid FRET increase in the H2THAAA mutant similar to that of native topo VI , indicating that the ATPase region of this mutant responds to supercoiled DNA and nucleotide in a wildtype-like manner . By contrast , the KGRRAAA mutant initially manifested a higher FRET signal than either wildtype topo VI or the H2THAAA mutant in the presence of supercoiled DNA alone; however , the addition of AMPPNP failed to elicit any further increase in FRET ( Figure 6D ) . Given that ATP binding and hydrolysis rely on Top6B dimerization , and that the maximum observed ATPase rate of the KGRRAAA construct is actually greater than wild-type topo VI in the presence of supercoiled DNA ( Figure 4B ) , the high initial FRET signal for this mutant suggests that its Top6B subunits can adopt a ‘pre-dimerized’ ATPase competent state in the presence of supercoiled DNA alone . As a consequence , the rapid ATP turnover by the KGRR mutants likely arises from the decoupling of ATP hydrolysis and product release from a slow conformational change necessary for strand passage ( i . e . those that drive G-segment opening and T-segment release ) . In this view , the KGRR loop would serve not only as a sensor of DNA crossings , but also as an element that delays ATP turnover until T-segment binding or strand passage has occurred . Since the H2TH domain does not appear to participate in T-segment sensing ( Figure 6D ) , yet is important for the strand passage activity of topo VI ( Figure 4A ) , we considered whether this element might instead interact with the G-segment . The H2TH domains reside far from the site of G-segment cleavage in the Top6A dimer ( Corbett et al . , 2007; Graille et al . , 2008 ) ; however , a prior AFM study has reported that topo VI can bend DNA by 100–140° ( Thomson et al . , 2015 ) . Modeling DNAs with varying bend angles into structures of S . shibatae topo VI , which was captured in a splayed-open B-subunit conformation ( Graille et al . , 2008 ) , suggested that a ~70 bp duplex with a ~100˚ bend could span both H2TH domains in a topo VI holoenzyme , running along the helical Stalk/WKxY region of the Top6B transducer domains and through the Top6A catalytic center ( Figure 7A ) . Based on this model , a ~30 bp duplex would fully engage the Top6A dimer and one Stalk/WKxY element , whereas a ~40 bp duplex would be sufficient to span both Stalk/WKxY elements in a Top6A/Top6B heterotetramer ( Figure 7B ) . An extended G-segment interface of this nature would not only provide a physical rationale for the marked increase in affinity of topo VI for duplex DNA as substrate length is increased from 20 to 30 bp ( Figure 1A ) , but also would account for the observed DNA binding deficiencies exhibited by the Stalk/WKxY mutants ( Figure 5A ) . Similarly , the impaired DNA binding of the KGRRAAA mutant ( Figure 5—source data 1 ) may reflect its apparent altered conformational state ( Figure 6—figure supplement 1 ) , which might misalign the G-segment-binding surfaces of the B and A subunits to lower the affinity of the enzyme for duplexes that are not already pre-bent . If the H2TH domains do engage G-segment DNAs , they do not contribute appreciably to DNA binding , at least as judged by the affinity of the 70 bp duplex for wildtype topo VI compared to shorter duplexes ( Figure 1A ) . We therefore considered whether the H2TH domains might instead help bend DNA , serve as sensors for pre-bent substrates , and/or help couple B-subunit dimerization to G-segment cleavage or strand passage . To test these ideas , we first assessed the minimal length of DNA required for nucleotide-dependent G-segment cleavage . Topo VI was incubated with 40 , 60 , or 70 bp long 5’-labeled duplexes in the absence of nucleotide , or with ATP or AMPPNP . Reactions were analyzed by denaturing urea-formamide PAGE to separate cleaved and uncleaved oligonucleotide products . Although the absence of a T-segment strongly inhibits G-segment scission , topo VI produced clear cleavage products in the presence of either ATP or AMPPNP on the 70 bp duplex . Faint cleavage products were also produced from the 60 bp duplex , but only in the presence of AMPPNP . The length of the cleavage products suggest M . mazei topo VI is cutting DNA slightly off-center from the preferred site identified for its S . shibatae homolog; these products are instead consistent with strand scission occurring at a secondary site six nucleotides upstream of this locus ( Buhler et al . , 2001 ) . No cleavage was seen for any condition on the 40 bp duplex ( Figure 7C ) . We next assessed whether the H2TH domains play a role in the observed length dependence of the G-segment cleavage reaction by measuring the nucleotide-dependent cleavage activity of our functional mutant panel on a 70 bp duplex ( Figure 7C ) . The KGRR mutants showed a slight decrease in AMPPNP-dependent cleavage , while the two Stalk/WKxY mutants displayed a greater decrease in this activity ( the triple glutamate substitution proved the most severely compromised ) . These results are consistent with the impaired affinities that these mutant enzymes show for the 70 bp substrate ( Figure 5A ) . By contrast , neither H2TH mutant proved capable of supporting short duplex cleavage . Collectively , these findings support the idea that for a G-segment to bind productively to the Top6A dimer , it ideally should be sufficiently long to engage both the stalk and H2TH regions of Top6B . The inability of a 40 bp duplex to support cleavage , even though this DNA binds with higher affinity than a 20 bp duplex and is long enough to reach both Stalk/WKxY regions , suggests G-segment DNAs must engage at least one H2TH domain before strand scission can be triggered . One implication of H2TH contacts with the distal arms of an associated G-segment is that ATP-binding and ATPase domain dimerization might in turn alter G-segment bending . To test this prediction , we labeled opposing ends of the 70 bp duplex with Cy5 and Cy5 . 5 and monitored changes in the end-to-end distance by FRET for native topo VI and our panel of mutants . Bulk FRET efficiencies in the absence and presence of enzyme were measured by exciting Cy5 at 630 nm and scanning the spectral emission of both the donor and acceptor fluorophores ( Figure 7D ) . The time-dependent conformational response to the addition of AMPPNP was also assessed . The addition of wild-type topo VI alone to the labeled DNA led to a modest FRET increase , a result indicative of G-segment bending that accords with prior AFM data ( Thomson et al . , 2015 ) . The KGRR mutants produced a similar FRET increase; however , both sets of Stalk/WKxY and H2TH mutants yielded only a minor nucleotide-independent response ( between that of wildtype topo VI and the free duplex ) . Upon adding AMPPNP , FRET efficiency rapidly increased further for the labeled DNA incubated with topo VI , or the KGRRAAA or KGRREEE mutants , indicating that nucleotide-driven dimerization of the ATPase regions leads to additional DNA bending . This FRET increase did not occur when Cy5 and Cy5 . 5 were placed on separate duplexes ( Figure 7—figure supplement 1 ) , allowing us to attribute the observed changes in FRET with the doubly labeled DNA to intramolecular bending , rather than the binding of two segments in trans . While this result for the KGRR constructs initially appeared to contradict the inability of nucleotide to alter Top6B conformation in the KGRRAAA mutant on supercoiled DNA ( Figure 6D ) , we note that the substrate differed between these two experiments . Performing the Top6B dimerization experiment with excess , unlabeled 70 bp duplex showed that , similar to wildtype topo VI bound to linear DNA , the KGRRAAA mutant adopts a more open conformation when bound to the 70 bp duplex than when bound to supercoiled DNA , and that the addition of nucleotide can shift the conformational equilibrium of the enzyme toward a closed state ( Figure 7—figure supplement 2 ) . For their part , both Stalk/WKxY mutants produced a FRET increase in the presence of AMPPNP , albeit with substantially slowed kinetics that likely account for their negligible ATPase activities ( Figure 4B ) . By contrast , the H2TH mutants did not support any nucleotide-dependent increase in FRET , indicating that Top6B dimerization in these mutants no longer introduces DNA bending to the distal ends of a bound G-segment . Together , these observations both suggest that topo VI engages a G-segment using an extended interface that runs from one H2TH domain to the other , and that strand scission is stimulated by bending induced by Top6B dimerization . The observation that the ATPase activity of the H2TH mutants is decoupled from strand passage , yet also substantially impaired when compared to the futile cycling of the KGRR mutants ( Figure 4 ) , further suggests that there is a feedback mechanism which couples nucleotide turnover to efficient G-segment deformation and cleavage .
Using a broad range of functional and reporter assays ( summarized in Tables 1–2 ) , we show here that type IIB topoisomerases preferentially engage the DNA crossings and bends of supercoiled substrates , and that binding to supercoiled DNA in turn stimulates the nucleotide-dependent dimerization of Top6B and couples this movement to DNA cleavage and strand passage at a distance in Top6A . To recognize and exploit distinguishing features of supercoiled substrates , topo VI uses several previously unidentified DNA-binding elements integrated into Top6B , including: ( 1 ) a basic interface formed along the subunit’s C-terminal stalk and a conserved WKxY motif that is important for robust G-segment binding ( Figure 3E , Figure 5A ) , ( 2 ) a basic ‘KGRR’ loop in the GHKL domain that aids DNA crossing recognition and links controlled ATP turnover to productive strand passage ( Figure 3D , Figure 4 , Figure 6 ) , and ( 3 ) an H2TH DNA-binding domain that promotes nucleotide-dependent G-segment bending and links ATP turnover to DNA cleavage and strand passage ( Figure 3F , Figure 7C–D ) . Collectively , our data highlight new intermediate steps in the topo VI catalytic cycle ( Figure 8 ) and provide a molecular rationale for the essential role of Top6B in driving transesterase activity by Top6A ( Buhler et al . , 1998; Buhler et al . , 2001 ) . By demonstrating that efficient and productive Top6B dimerization requires nucleotide , supercoiled DNA , and an intact KGRR loop , our findings also suggest that the previously visualized , inactive conformation of the Top6A dimer ( Corbett et al . , 2007; Graille et al . , 2008; Nichols et al . , 1999 ) may represent a cleavage-suppression mechanism that can only be overcome when the regions identified here are occupied by the binding of an extended DNA crossing and when nucleotide induces the dimerization of Top6B . Besides promoting Top6B closure , our data also imply that T-segment engagement may actively control both ATP turnover and DNA-gate opening to permit strand passage . As with wild-type topo VI , the binding of the KGRRAAA mutant to supercoiled DNA alone promotes Top6B dimerization ( Figure 6—figure supplement 1 ) and supports ATP hydrolysis ( Figure 4B ) ; however , disruption of this region impairs the binding of a DNA crossover ( Figure 6C ) , does not support additional conformational response to nucleotide ( Figure 6D ) , blocks ATP hydrolysis on short linear DNA ( Figure 4B ) , and abolishes strand passage overall ( Figure 4A ) . This behavior suggests the bends or pre-formed crossings present in supercoiled DNA help to promote B-subunit dimerization and ATP hydrolysis , and may partially compensate for a loss of the KGRR element ( Figure 4B , Figure 6D ) , but that the coupling of ATP hydrolysis to strand passage requires the productive binding of a T-segment DNA ( Figure 2B–D , Figure 6D ) . Given that the T-segment storage cavity appears to be too small to accommodate DNA when fully closed ( Corbett et al . , 2007; Corbett and Berger , 2005; Graille et al . , 2008 ) , it has been proposed that T-segment engagement may potentiate opening of the Top6A dimer and separation of a cleaved G-segment . Considering this idea in light of our present findings suggests that the binding of the KGRR loops to a stored T-segment helps to suppress premature release of ATP hydrolysis products , which is normally linked to a slow conformational change associated with G-segment separation and subsequent T-segment release . This scheme offers a simple explanation for why the KGRR mutants rapidly hydrolyze ATP when bound to supercoiled DNA: weakening of the T-segment interaction with the KGRR loop allows for early ATP turnover , yet by not resolving the DNA crossing , Top6B remains pre-dimerized and primed to bind ATP again , leading to futile cycling without strand passage . Certain biochemical properties of topo VI identified here are somewhat surprising when considering the demands placed on the cell by transcription and replication . For example , the highly distributive nature of supercoil relaxation observed for topo VI ( Figure 1C–D ) is at odds with a need to remove the continual local build-up of superhelical tension arising from RNA polymerase advancement or replication fork progression . The maximal observed ATP hydrolysis and strand passage rates for Mm topo VI in vitro ( Figure 1C , Figure 2A ) are also much slower ( ~50–100 fold ) than rates generally observed for type IIA topoisomerases ( Higgins et al . , 1978; Lindsley and Wang , 1993; Osheroff et al . , 1983; Sugino and Cozzarelli , 1980 ) . These enzymatic properties raise important questions as to when and in which context topo VI acts in the cell . For instance , using estimates based on the genome size and generation time of M . mazei ( Appendix 1 ) , topo VI would appear to require ~50-fold greater specific activity to keep up with gene expression and chromosome duplication , or else be present at extremely high cellular concentrations . Although the source of this discrepancy may be due to differences between in vitro vs . in vivo rates of strand passage , it could alternatively arise from a missing factor that enhances topo VI activity . This second explanation , if true , has intriguing ramifications . For example , if a secondary factor were to accelerate topo VI’s strand passage rate by increasing processivity , then the distributive action of topo VI might reflect an auto-inhibitory mechanism that is manifest until the enzyme is localized to the appropriate chromosomal context . Along this line , multiple protein factors have been identified to bind topo VI in Arabidopsis thaliana , and may be obligate components of the topo VI machinery in plants ( Breuer et al . , 2007; Forterre and Gadelle , 2009; Kirik et al . , 2007; Sugimoto-Shirasu et al . , 2005 ) . If analogous factors exist for archaeal topo VI , it may be that Top6A and Top6B actually constitute the core of a larger type IIB topoisomerase complex . Eukaryotic topo IIIα , a type IA topoisomerase , exemplifies such a strategy , interacting with a RecQ family helicase and SSB/RPA-like factors to channel DNA strand passage into efficient Holliday junction resolution ( Plank et al . , 2006; Raynard et al . , 2006; Singh et al . , 2008; Wu et al . , 2006; Xu et al . , 2008 ) . The picture of the type IIB topoisomerase strand passage mechanism developed here reveals a rich set of regulatory mechanisms both shared with and divergent from type IIA topoisomerases . Of these , a central feature is the requirement that topo VI must engage a DNA crossing ( as found in supercoiled or catenated DNA ) to access a stable dimerized B-subunit conformation and productively turn over ATP ( Figure 8 ) . Contrariwise , Top6B takes on a predominantly open conformation when bound only to a prospective G-segment ( or in the absence of DNA [Figure 2 , Corbett et al . , 2007; Graille et al . , 2008] ) , even when nucleotide is present . This tight control over Top6B dimerization , and its reciprocal coupling to G-segment cleavage , likely helps compensate for the missing safeguard of a third dimerization interface – the so-called ‘C-gate’ – found in type IIA topoisomerases ( Roca et al . , 1996; Roca and Wang , 1994; Williams and Maxwell , 1999 ) . For its part , the ATPase region of type IIA topoisomerases does possess potential T-segment-sensing elements ( Tingey and Maxwell , 1996 ) ; however , in contrast to topo VI , the ATPase domains of type IIA topoisomerase holoenzymes readily dimerize upon binding ATP , even when DNA is absent ( Gubaev and Klostermeier , 2011; Roca and Wang , 1992 ) . Interestingly , in requiring the binding of a substrate T-segment for stable ATPase domain dimerization , topo VI echoes the behavior of Hsp90 , whose related GHKL ATPase fold strongly depends on client protein or co-chaperone engagement to drive ATPase association ( Ali et al . , 2006; Hessling et al . , 2009; Wolmarans et al . , 2016 ) . This co-dependency raises the possibility that other GHKL ATPases , such as MutL and MORC proteins , may similarly rely on substrate/cofactor interactions as a checkpoint to license nucleotide-dependent dimerization . Following DNA crossover recognition , ATP binding by Top6B is needed to trigger G-segment scission by Top6A ( Buhler et al . , 1998 ) . However , closing of the ATP gate also serves to further bend the G-segment through contacts mediated by the H2TH domain ( Figure 7D ) . Although H2TH mutants are unable to cleave short duplex substrates ( Figure 7C ) , they support weak strand passage activity on supercoiled DNA ( Figure 4A ) , a substrate that constrains DNA bends independent of enzyme binding . This suggests that DNA bending itself , whether innate or H2TH-mediated , may help promote DNA breakage by Top6A . Although the H2TH domain is specific to type IIB topoisomerases , type IIA enzymes also bend G-segment DNAs to support cleavage ( Dong and Berger , 2007; Laponogov et al . , 2009; Lee et al . , 2013; Lee et al . , 2012; Wohlkonig et al . , 2010 ) . This dependency raises the possibility that other nucleases or transesterases that rely on the TOPRIM fold beside type II topoisomerases ( e . g . OLD family enzymes and Spo11 ( Aravind et al . , 1998 ) ) may similarly require DNA deformation to promote strand scission . The discovery that Spo11 was related to the DNA-cleaving Top6A subunit of archaeal topo VI was a critical development in understanding how DNA breaks are formed to initiate meiotic recombination ( Bergerat et al . , 1997; Keeney et al . , 1997 ) . The realization that Top6A requires Top6B for DNA cleavage ( Buhler et al . , 1998 ) has in turn raised the question of whether Spo11 might partner with a similar regulatory factor during meiotic recombination . Recently , structurally homologous counterparts to Top6B have been recognized across a wide range of eukaryotic species ( MTop6B in plants , Top6BL in mammals , Rec102 in S . cerevisiae , and Mei-P22 in Drosophila ) ( Robert et al . , 2016; Vrielynck et al . , 2016 ) . Interestingly , the WKxY motif implicated here in G-segment binding is conserved between Top6B and some of its meiotic homologs ( e . g . mammalian MTop6B and plant Top6BL ) ( Robert et al . , 2016 ) , suggesting that this region could assist Spo11 with DNA targeting , and contribute to the signals necessary to activate DNA cleavage during meiosis . In those Top6B homologs where the WKxY motif is poorly conserved , alternative features on the transducer stalk may participate in binding to DNA . For example , the prospective WKxY motif in budding yeast Rec102 is highly divergent in sequence ( WEEQ ) , yet Spo11 hotspots from this organism display a sequence bias that extends beyond the predicted footprint of the Spo11 dimer . Interestingly , this bias maps to a distance of ±11–16 bp from the dyad of Spo11 ( Pan et al . , 2011 ) , compared to the ~17–20 bp distance between the Top6B Stalk/WKxY region and the Top6A dyad , consistent with the notion that non-Spo11 DNA interaction sites may have shifted during evolution . In topo VI , we find that Top6B dimerization further bends DNA to potentiate cleavage by Top6A . Surprisingly , components critical for Top6B-mediated dimerization are either highly divergent or missing in meiotic Top6B homologs . For example , both Topo6BL and MTopo6B contain a highly degenerate GHKL domain that lacks essential elements required for ATP binding ( only purine-binding elements are conserved , see Figure 8—figure supplement 1 ) , and Rec102 and Mei-P22 lack a GHKL domain entirely ( Dutta and Inouye , 2000; Robert et al . , 2016; Vrielynck et al . , 2016 ) . Insofar as DNA bending , the meiotic Top6B-like factors identified thus far also lack an H2TH domain ( Robert et al . , 2016; Vrielynck et al . , 2016 ) . Should Spo11 , like Top6A , require both DNA bending and allosteric activation to achieve a cleavage-competent state , these differences indicate that it is not the newly identified Top6B-like subunits alone that are responsible for mediating this event . Candidate factors that might further regulate Spo11-dependent break formation include additional partner proteins , post-translational modifications , and tension on or deformation of the DNA itself by factors responsible for sister chromatid pairing ( Lam and Keeney , 2014 ) . Future studies focused on defining how topo VI and Spo11-type systems physically engage DNA strands , respond to possible partner factors , and switch between inactive and active DNA-cleavage states will be needed to help shed light on how these systems operate .
Cloning of the M . mazei Top6B gene in frame with an N-terminally fused His6-tobacco etch virus ( TEV ) protease-cleavable tag and the M . mazei Top6A gene into a polycistronic expression vector was previously described ( Corbett et al . , 2007 ) . Oligonucleotides used for site directed mutagenesis were obtained from Integrated DNA Technology ( IDT , Coralville , IA ) . Mutant constructs were generated either by PCR amplification of the expression vector using primers containing the desired point substitutions followed by blunt-end ligation , or by quick-change mutagenesis ( Agilent , Santa Clara , CA ) . The following mutations were added to generate the ‘Cys-lite’ construct: C267S , C278A , C316A and C550A , all in Top6B . Mutagenesis was verified by Sanger sequencing ( Genewiz LLC , South Plainfield , NJ ) . Topo VI and functional mutant variants were overexpressed in E . coli BL21 ( DE3 ) Codon +RIL cells ( QB3-Macrolab , University of California-Berkeley , CA ) grown in ZYM-5052 auto-induction media ( Studier , 2005 ) . Wild-type topo VI was expressed in cultures grown at 37°C , whereas cultures expressing functional mutant constructs were shifted to 25°C upon reaching an OD600 of 0 . 4–0 . 6 . The KGRRAAA FRET assay construct was grown at 37°C to an OD600 of 2–3 in M9ZB media ( Studier , 2005 ) , cooled to 18°C , and then induced with IPTG ( 250 μM final concentration ) and grown overnight . Cultures were harvested by centrifugation at 24 hr following inoculation , resuspended in buffer A [20 mM HEPES-KOH pH 7 . 5 , 800 mM NaCl , 20 mM Imidazole , 10% ( v/v ) glycerol , 1 μg/mL pepstatin A , 1 μg/mL leupeptin , 1 mM PMSF] , and frozen drop-wise into liquid nitrogen for storage at −80°C . Proteins were purified as previously described ( Corbett et al . , 2007 ) . Harvested cells were lysed by sonication , and lysate was clarified by centrifugation . Clarified lysate was applied to a 5 mL HiTrap Ni2+ column ( GE Healthcare Life Sciences , Marlborough , MA , USA ) and washed with buffer A [20 mM HEPES-KOH pH 7 . 5 , 800 mM NaCl , 20 mM Imidazole , 10% ( v/v ) glycerol , 1 μg/mL pepstatin A , 1 μg/mL leupeptin , 1 mM PMSF] . Following a subsequent wash with buffer B [20 mM HEPES-KOH pH 7 . 5 , 150 mM NaCl , 20 mM Imidazole , 10% ( v/v ) glycerol , 1 μg/mL pepstatin A , 1 μg/mL leupeptin , 1 mM PMSF] , bound proteins were eluted by a 15-column volume gradient from buffer B to buffer C [20 mM HEPES-KOH pH 7 . 5 , 150 mM NaCl , 20 mM Imidazole , 10% ( v/v ) glycerol , 1 μg/mL pepstatin A , 1 μg/mL leupeptin , 1 mM PMSF] . Fractions containing the topo VI heterotetramer were applied to a 5 mL HiTrap SP cation-exchange column and 5 mL HiTrap Q anion-exchange column ( GE Healthcare Life Sciences ) in series and washed with buffer B . The HiTrap SP column was removed , and protein bound to the HiTrap Q column was eluted with a 10-column volume gradient from buffer B to buffer A . Peak fractions were concentrated by centrifugation ( Millipore Amicon Ultra 30K MWCO ) and incubated with 1 . 5 mg of His6-TEV protease ( QB3-Macrolab , University of California , Berkeley ) overnight at 4°C to remove His6 tags . Uncleaved proteins and His6-TEV protease were removed by applying the protease cleavage reaction to a HiTrap Ni2+ column equilibrated in buffer B . Flow-through was concentrated and applied to an Sephacryl-300 HR gel filtration column ( GE Healthcare Life Sciences ) equilibrated and run in sizing buffer [20 mM HEPES-KOH pH 7 . 5 , 300 mM KCl , 10% ( v/v ) glycerol] and concentrated by centrifugation ( Millipore Amicon Ultra 10K MWCO ) . Purity of peak fractions was assessed by SDS-PAGE and coomassie blue staining , and the concentration of tetramer was determined by absorbance at 280 nm using extinction coefficients determined by the ExPASY ProtParam webserver ( Gasteiger et al . , 2005 ) . Proteins were flash frozen in a final storage buffer [20 mM HEPES-KOH pH 7 . 5 , 300 mM KCl , 30% ( v/v ) glycerol , 1 mM Trisphosphine hydrochloride ( TCEP ) ] and stored in aliquots at −80°C for use in subsequent biochemical and biophysical studies . DNA substrates were resuspended in ddH2O and annealed from single strand DNA oligomers of complementary sequence ( Figure 1—source data 1 ) obtained from IDT . Annealing of the stacked-junction DNA substrate followed published protocols ( Duckett et al . , 1988 ) with a few modifications . The junction was prepared in 25 mM Tris HCl pH 7 . 9 , 25 mM NaCl , 10 mM MgCl2 and annealed by heating at 70°C for 2 hr , followed by cooling at 0 . 5°C/min to 4°C . Annealing reaction products were loaded onto a 5 mL HiTrap-Q anion exchange column ( GE Healthcare Life Sciences ) equilibrated in stacked junction ( SJ ) buffer A [25 mM NaCl , 25 mM Tris-HCl pH 7 . 9 , 10 mM MgCl2] . Contaminants were removed by washing with 55%/45% mix of SJ buffer A to SJ Buffer B [1 M NaCl , 25 mM Tris 7 . 9 , 10 mM MgCl2] . Correctly annealed substrate was eluted with 45%/55% Buffer A/Buffer B , pooled and dialyzed back into SJ buffer A , and concentrated by centrifugation ( Amicon Ultra 3K MWCO , EMD Millipore , Billerica , MA ) . Proper annealing for all substrates was assessed by native 15% PAGE run in 0 . 5x Tris-Borate-EDTA ( TBE ) buffer at 4°C . DNA binding by topo VI and functional mutants was assessed using fluorescence anisotropy . Protein was serially diluted in two-fold steps in binding assay dilution buffer [250 mM potassium glutamate , 5% ( v/v ) glycerol , 50 mM HEPES-KOH pH 7 . 5 and 1 mM TCEP] and incubated with fluorescein-labeled DNA substrate in the dark and on ice for 5 min . Reactions were diluted to final binding assay conditions [27 μL , 0 , 0 . 3–4000 nM enzyme , 20 nM labeled duplex , 50 mM potassium glutamate , 5% ( v/v ) glycerol , 20 mM HEPES-KOH pH 7 . 5 , 1 mM TCEP , 10 mM MgCl2 and 0 . 1 mg/mL BSA] , and incubated on ice an additional 10 min . Fluorescence anisotropy was measured at ambient temperature using a Clairiostar microplate reader ( BMG Labtech GmbH , Ortenberg , Germany ) by exciting at 482 nm ( band pass 16 nm ) and measuring parallel and perpendicular emission intensity at 530 nm ( band pass 40 nm ) , with an inline 504 nm long pass dichroic filter . Data are the average of three independent experiments , with all points normalized to the DNA alone condition and fit to the following single-site binding model: ( 1 ) ∆FA=∆FAmax ( L+P+Kd , app- ( L+P+Kd , app ) 2-4[L][P]2[L] ) where ΔFAmax is the maximal specific change in anisotropy , [L] is DNA substrate concentration , [P] is the concentration of topo VI construct , and Kd , app is the apparent dissociation constant for DNA substrate and enzyme . To test for cooperativity , binding isotherms were also fit to a Hill equation-type model: ( 2 ) ∆FA=∆FAmax ( [P]hKd , apph+[P]h ) where ΔFAmax is the maximal specific change in anisotropy , [P] is the concentration of topo VI construct , h is the apparent Hill coefficient , and Kd , app is the apparent dissociation constant for DNA substrate and enzyme . Competition assays were carried out similarly to binding assays , with protein diluted in binding assay dilution buffer and incubated with the 70 bp fluorescein-labeled duplex and either negatively supercoiled pSG483 plasmid DNA ( pBluescript SK derivative , 2927 bp ) or linear sheared salmon-sperm DNA ( ThermoFisher Scientific , Waltham , MA ) . Reactions were diluted to final binding assay conditions , except enzyme concentration was set at 100 nM , and competitor concentration varied from 0 . 1 μM bp to 106 . 5 μM bp DNA . Anisotropy data were fit to an explicit competition model ( Wang , 1995 ) , which fits to the parameters: [A] , total concentration of the competitor DNA substrate; [B] , total concentration of the labeled DNA probe; [P] , total topo VI concentration; KA , dissociation constant of the competitor DNA substrate; KB , dissociation constant of the labeled DNA probe; and ΔFAmax , the maximal specific change in fluorescence anisotropy for the probe . Topo VI holoenzyme was thawed and diluted in series with relaxation assay dilution buffer [300 mM potassium glutamate , 10% ( v/v ) glycerol , 20 mM HEPES-KOH pH 7 . 5 and 1 mM TCEP] and incubated with negatively supercoiled pSG483 plasmid DNA for 5 min on ice before dilution into final relaxation assay conditions [30 μL reactions , 0 , 0 . 3–20 nM topo VI for titration , 2 . 5 nM topo VI for timecourses , 50 mM potassium glutamate , 10% ( v/v ) glycerol 20 mM bis-tris-propane-HCl ( BTP-HCl ) pH 7 . 5 , 2 mM HEPES pH 7 . 5 , 1 mM TCEP , 10 mM MgCl2 , 0 . 1 mg/mL BSA , 3 . 5 nM pSG483 ( 10 . 2 μM bp DNA ) , and 1 mM ATP] . Reactions were initiated by addition of ATP , incubated at 30°C , and quenched by addition of SDS and EDTA to final concentrations of 1% and 10 mM respectively . Glycerol-based loading dye was added to samples which were run on a 1% ( w/v ) TAE agarose gel ( 40 mM sodium acetate , 50 mM Tris-HCl , pH 7 . 9 and 1 mM EDTA , pH 8 . 0 ) for 15 hr at ~2 V/cm . For visualization , gels were stained for 30 min with 0 . 5 μg/mL ethidium bromide in running buffer , de-stained in running buffer for 30 min , and exposed to UV trans-illumination . Experiments were carried out similarly for the plasmid-chase experiments , except that a 6 . 5 kb chase plasmid ( p1C ) was added with ATP to a final concentration of 10 . 2 μM bp when initiating reactions . ATP hydrolysis was measured using an established NADH-coupled assay ( Morrical et al . , 1986; Tamura and Gellert , 1990 ) . Topo VI was thawed and diluted with 300 mM potassium glutamate , 10% ( v/v ) glycerol , 50 mM BTP-HCl pH 7 . 5 and 5 mM TCEP to 3 . 75 μM enzyme , mixed 1:2 with sheared salmon-sperm DNA , supercoiled pSG483 , or ddH2O , and incubated for 5 min on ice . Enzyme/substrate mixes were diluted with NADH-PK/LDH coupling mix to final ATP hydrolysis assay conditions [100 μL reactions , 3 . 75 mM phosphoenolpyruvate , 150 μM NADH , 24 U pyruvate kinase and 36 U lactate dehydrogenase ( PK/LDH from rabbit muscle in buffered , aqueous glycerol solution , Sigma Aldrich , St Louis , MO ) , 0 . 1 mg/mL BSA , 50 mM BTP-HCl , pH 7 . 5 , 50 mM potassium glutamate , 5 mM TCEP , 10 mM MgCl2 , 5% ( v/v ) glycerol , 500 nM topo VI holoenzyme] . ATP titration reactions contained either 400 μM bp sheared salmon-sperm DNA , 400 μM bp negatively supercoiled pSG483 or no DNA , and were initiated by addition of ATP to a final concentration of 0 mM or 62 . 5 μM-4 mM diluted in two-fold steps . DNA titrations containing 3 . 12–800 μM bp DNA diluted in two-fold steps were initiated by addition of ATP to a final concentration of 2 mM . Reactions were incubated at 30°C and followed in clear 96-well plates ( Corning Inc , Corning , NY ) by absorbance at 340 nm using a Clairiostar microplate reader . Raw absorbance values were converted to NADH molar concentrations based on measurements from NADH standards in the final ATP hydrolysis assay condition . ATP hydrolysis rates were determined by fitting to the linear portion of NADH consumption curves . Data representing three independent experiments were fit to a standard Michaelis-Menten model: ( 3 ) V0=kcat , appET[S]Km , app+[S]where V0 is the observed turnover rate , kcat , app is the maximum turnover rate , [Et] is the total topo VI holoenzyme concentration , [S] is the concentration of ATP , and Km , app is the Michaelis constant for ATP . For DNA titration experiments , [S] is the concentration of DNA and the kcat-stim , DNA and Kstim , DNA parameters substitute for kcat , app and Km , app . Following purification , topo VI FRET constructs were labeled by reacting enzyme with 5-fold molar excess to enzyme of both Alexa Fluor 555 C2 maleimide and Alexa Fluor C2 647 maleimide ( ThermoFisher Scientific ) in sizing buffer overnight at 4°C . TCEP was also added at 50-fold molar excess to enzyme . Reactions were quenched with 5 mM DTT and applied to a HiPrep 26/10 Desalting column ( GE Healthcare Life Sciences ) to separate protein from unreacted dye . Proper labeling was imaged by SDS PAGE using a Typhoon FLA 9500 laser scanner ( GE Healthcare Life Sciences ) . Labeling efficiencies were determined by comparing absorption at 280 nm for protein to absorption at 555 nm for Alexa555 and 650 nm for Alexa647 . Proteins were brought to storage buffer conditions , flash frozen as aliquots in liquid nitrogen and stored at −80°C . For gate closure assays , labeled protein was diluted in 250 mM potassium glutamate , 10% ( v/v ) glycerol and 20 mM HEPES-KOH pH 7 . 5 to 1 μM , mixed 1:1 with 500 μM bp DNA substrate or ddH2O , incubated on ice for 5 min , and diluted to final assay conditions [20 μL reactions , 200 nM topo VI , 0 or 100 μM bp DNA , 50 mM potassium glutamate , 1 mM TCEP , 10% ( v/v ) glycerol , 20 mM HEPES-KOH pH 7 . 5 , 10 mM MgCl2 and 0 . 1 mg/mL BSA] . Fluorescence emission spectra were measured by exciting samples at 530 nm and measuring emission from 545 nm to 700 nm using a Fluoromax Fluorometer 4 ( HORIBA Jobin Yvon , Edison , NJ ) . Adenylyl-imidodiphosphate ( AMPPNP ) was added to a final concentration of 1 mM and changes to emission spectra were measured over time . Spectra were normalized by total emission intensity . Plotted FRET efficiencies ( E ) were determined ratiometrically from donor ( ID ) and acceptor ( IA ) peak intensities: ( 4 ) E=IAID+IA Topo VI was diluted in [250 mM potassium glutamate , 10% ( v/v ) glycerol , 10 mM MgCl2 and 20 mM HEPES-KOH pH 7 . 5] to 1 μM , mixed 1:1 with 500 nM fluorescein-labeled duplex ( Figure 1—source data 1 ) and incubated 5 min on ice . Reactions were diluted to a final cleavage reaction condition [20 μL reactions , 200 nM topo VI construct , 100 nM FAM-labeled duplex , 50 mM potassium glutamate , 1 mM TCEP , 10% ( v/v ) glycerol , 16 mM BTP-HCl pH 7 . 5 , 4 mM HEPES-KOH pH 7 . 5 , 10 mM MgCl2 , 0 . 1 mg/mL BSA and 15% DMSO] . ATP , AMPPNP or ddH2O were added to initiate reactions . Reactions were incubated at 30°C for 2 hr then quenched with SDS to a final concentration of 1% . Proteinase K was added to reactions at a final concentration of 0 . 3 mg/mL and incubated at 45°C for 1 hr . Formamide was added 1:1 to samples and cleavage products were separated on 7 M Urea-Formamide 0 . 5x TBE 12% PAGE . Gels were visualized using a Typhoon FLA 9500 laser scanner . DNA bending experiments used the same 70 bp duplex sequence from binding and cleavage experiments , except the substrate was modified to have a Cy5 replace the 5’-fluorocein on strand one and Cy5 . 5 was added to the 5’ end of strand 2 ( Figure 1—source data 1 ) . Reactions were prepared exactly as described for the DNA cleavage assays . Fluorescence emission spectra were measured by exciting samples at 630 nm and measuring emission from 645 nm to 850 nm using a Fluoromax Fluorometer 4 . AMPPNP was added to a final concentration of 1 mM and changes to emission spectra were measured over time . Spectra were normalized by total spectral emission . Plotted FRET efficiencies were calculated as for the gate closure assays . A ScTop2 construct containing labeling sites on the ATP gate ( ScTop2ΔCTR-cyslite-180C ) was generated from a previously described ScTop2 construct with a C-terminal truncation ( coding for residues 1–1177 , ScTop2ΔCTR , [Schmidt et al . , 2012] ) cloned in frame with an N-terminally fused His6-TEV protease-cleavable tag by introducing the following mutations: C48A , C381A , C471A , C731A . Proteins were overexpressed and purified as previously described ( Schmidt et al . , 2012 ) . In brief , S . cerevisiae strain BCY123 was transformed with a GAL1 shuttle vector containing the ScTop2ΔCTR ORF and grown in CSM-Ura- media with a 2% lactic acid and 1 . 5% glycerol carbon source at 30°C . Overexpression was induced by the addition of 2% galactose at A600 = 0 . 8 . Six hours following induction , cells were centrifuged , resuspended in 1 mM EDTA and 250 mM NaCl ( 1 mL/L liquid culture ) , and flash frozen drop-wise in liquid nitrogen . For purification , frozen cells were first lysed under liquid nitrogen using an SPEX SamplePrep 6870 Freezer Mill ( SPEX SamplePrep , Metuchen , NJ ) , and resultant powder was thawed and re-suspended in Buffer A300 [20 mM Tris-HCl pH 8 . 5 , 300 mM KCl , 20 mM imidazole , and 10% ( v/v ) glycerol , 1 μg/mL pepstatin A , 1 μg/mL leupeptin , and 1 mM PMSF] . Lysate was clarified by centrifugation and applied to a 5 mL HiTrap Ni2+ column equilibrated in buffer A . Following washing with buffer A , protein was eluted with buffer B [20 mM Tris-HCl pH 8 . 5 , 100 mM KCl , 200 mM imidazole , and 10% ( v/v ) glycerol , 1 μg/mL pepstatin A , 1 μg/mL leupeptin , and 1 mM PMSF] , and applied to a 5 mL HiTrap SP cation-exchange column . Bound protein was eluted with buffer C [20 mM Tris-HCl pH 8 . 5 , 500 mM KCl , 10% ( v/v ) glycerol , 1 μg/mL pepstatin A , 1 μg/mL leupeptin and 1 mM PMSF] . Peak fractions were concentrated by centrifugation ( Millipore Amicon Ultra 30K MWCO ) and incubated with 1 . 5 mg of His6 TEV protease overnight at 4°C . Uncleaved proteins and TEV protease were removed by applying the protease reaction to a HiTrap Ni2+ column equilibrated in buffer A . Flow-through was concentrated and applied to an Sephacryl-300 HR gel filtration column ( GE ) equilibrated and run in ScTop2 sizing buffer [20 mM Tris-HCl pH 7 . 9 , 500 mM KCl , 10% ( v/v ) glycerol] . Peak fractions were collected and concentrated ( Millipore Amicon Ultra 30K MWCO ) . Purity was estimated by SDS-PAGE and concentration was determined by absorbance at 280 nm . ScTop2 was flash frozen in a final storage buffer containing [20 mM Tris-HCl pH 7 . 9 , 500 mM KCl , 30% ( v/v ) glycerol] and stored in aliquots at −80°C . Plasmid relaxation assays and chase assays with ScTop2ΔCTR were carried out as described for topo VI , except that ScTop2ΔCTR was diluted in [500 mM KCl , 10% ( v/v ) glycerol , 20 mM Tris-HCl pH 7 . 9] and final relaxation assay conditions were [30 mM Tris-HCl pH 7 . 9 , 10 mM MgCl2 , 0 . 05 mg/mL BSA , 0 . 5 mM TCEP , 100 mM KCl , 10% ( v/v ) glycerol , 1 mM ATP , 3 . 5 nM ( 10 . 2 μM bp DNA ) pSG483 , and 2 . 5 nM topo II] , with 10 . 2 μM bp of the 6 . 5 kb plasmid added with ATP to initiate reactions for chase experiments . The ATP gate of ScTop2ΔCTR-cyslite-180C was labeled on a native cysteine residue ( 180C ) with the Alexa Fluor 555 C2 maleimide and Alexa Fluor C2 647 maleimide FRET pair following the same procedure as for Top6B , except that the reaction was carried out in ScTop2 sizing buffer and samples were flash frozen in the topo II storage buffer conditions . Gate closure assays were performed similarly as with topo VI , except protein was diluted in 500 mM KCl , 10% ( v/v ) glycerol and 20 mM Tris-HCl pH 7 . 9 , and final assay conditions were 200 nM topo II , 0 or 100 μM bp DNA , 100 mM KCl , 2% ( v/v ) glycerol , 10 mM Tris-HCl pH 7 . 9 , 5 mM MgCl2 and either 0 mM or 1 mM AMPPNP . Fluorescence emission spectra were measured as with topo VI . All data were plotted and fit using Prism Version 7 ( RRID: SCR_015807 , ( GraphPad Software , La Jolla , CA ) ) . Mapping of sequence conservation in relation to tertiary structure was aided by the Consurf web server ( RRID: SCR_002320 , [Ashkenazy et al . , 2010] ) . Coordinates for bent DNA models were generated using the 3DNA web server ( Zheng et al . , 2009 ) . Pymol was used for structure visualization and comparison ( RRID: SCR_000305 , [The PyMOL Molecular Graphics System , Schrödinger , LLC] ) . | Each human cell contains genetic information stored on approximately two meters of DNA . Like holiday lights in a storage box , packing so much DNA into such a small space leads to its entanglement . This snarled DNA prevents the cell from properly accessing and copying its genes . Type II topoisomerases are a group of enzymes that remove DNA tangles . They attach to one segment of a DNA tangle , cut it in half , remove the knot , and then repair the broken DNA strand . The process requires the proteins to ‘burn’ chemical energy . If topoisomerases make mistakes when they cut and reseal DNA , they could damage genetic information and harm cells . It is still unclear how these proteins recognize DNA tangles and use energy to remove knots instead of adding them . Here , Wendorff and Berger use biochemical approaches to look into topo VI , a type II topoisomerase found in plants and certain single-celled organisms . When DNA is tangled , it forms sharp bends and crossings . Their experiments reveal that topo VI has certain ‘sensors’ that detect where DNA bends , and others that recognize the crossings . Only when both features are present does the enzyme start working and using energy . These sensors act as fail-safes to ensure that topo VI only breaks DNA when it encounters a proper knot , and is not ‘set loose’ on untangled DNA . Future work will look at topo VI at an atom-by-atom level to reveal how exactly the enzymes ‘see’ DNA bends and crossings , and how interactions with the correct type of DNA triggers energy use and DNA untangling . Knowing more about topo VI can help researchers to understand how human and bacterial topoisomerases work . These results could also be generalized to other enzymes , for example those that help the genetic processes at play when sperm and egg cells form . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"biochemistry",
"and",
"chemical",
"biology",
"structural",
"biology",
"and",
"molecular",
"biophysics"
] | 2018 | Topoisomerase VI senses and exploits both DNA crossings and bends to facilitate strand passage |
Neuronal circuits are known to integrate nutritional information , but the identity of the circuit components is not completely understood . Amino acids are a class of nutrients that are vital for the growth and function of an organism . Here , we report a neuronal circuit that allows Drosophila larvae to overcome amino acid deprivation and pupariate . We find that nutrient stress is sensed by the class IV multidendritic cholinergic neurons . Through live calcium imaging experiments , we show that these cholinergic stimuli are conveyed to glutamatergic neurons in the ventral ganglion through mAChR . We further show that IP3R-dependent calcium transients in the glutamatergic neurons convey this signal to downstream medial neurosecretory cells ( mNSCs ) . The circuit ultimately converges at the ring gland and regulates expression of ecdysteroid biosynthetic genes . Activity in this circuit is thus likely to be an adaptation that provides a layer of regulation to help surpass nutritional stress during development .
Animals frequently find themselves in situations of nutritional deprivation . To combat these lean periods , physiological mechanisms have evolved that are common to both vertebrates and invertebrates ( Waterson and Horvath , 2015; Zhang et al . , 2002 ) . Such mechanisms require the animal to integrate sensory perception of nutrient deprivation with appropriate metabolic changes . The nervous system plays a central role in this process , and communication between multiple neuronal cell types can regulate the necessary metabolic and hormonal changes required for coordinating an organismal response ( Chantranupong et al . , 2015; Waterson and Horvath , 2015 ) . The vertebrate hindbrain acts as a central regulator of energy balance . Nuclei of the solitary tract integrate energy status signals from relevant inputs such as blood-borne endocrine signals and synaptic signals from the gastrointestinal tract and peripheral neurons , to modulate appetite and feeding ( Grill and Hayes , 2012 ) . However , the specific identity and circuitry of neurons responsible for sensing and responding to nutritional cues is not completely understood . This is in part due to the complexity of the vertebrate brain , in which monitoring activity in specific neuronal subtypes is challenging . A less-complex nervous system , consisting of approximately 10 , 000 neurons ( Scott et al . , 2001 ) , compared with 70 million neurons in the mouse brain ( Economo et al . , 2016 ) , makes Drosophila larvae a powerful system to elucidate central brain circuitry underlying systemic responses to nutrient deprivation ( Bjordal et al . , 2014 ) . Intracellular signaling mechanisms shape neural responses across circuits and contribute greatly to systemic outcome . For example , ghrelin , a gut-derived orexigenic hormone , affects synaptic plasticity under conditions of nutrient deprivation through intracellular signaling involving calcium ( Yang et al . , 2011 ) . Nonetheless , intracellular signaling pathways responsible for synaptic plasticity in circuits that regulate organismal responses to an altered nutrient status need further elucidation . Intracellular calcium signaling evolved in parallel with multi-cellularity ( Cai , 2008 ) , and may therefore function in coordinating systemic metabolic responses ( Chantranupong et al . , 2015 ) of metazoans . A key component of intracellular calcium signaling is the Inositol 1 , 4 , 5-trisphosphate receptor ( IP3R ) . These are calcium channels that mediate intracellular calcium release from the endoplasmic reticulum ( ER ) in response to extracellular stimuli ( Streb et al . , 1983 ) . In vertebrates , calcium release through IP3R2 and IP3R3 is required in various classes of non-excitable cells for metabolic control ( Wang et al . , 2012 ) and exocrine secretion of insulin or amylase from the pancreas ( Berggren et al . , 2004; Futatsugi et al . , 2005 ) . IP3R1 is expressed in different classes of neurons where it regulates processes ranging from synaptic plasticity ( Nishiyama et al . , 2000 ) to axonal guidance ( Xiang et al . , 2002 ) . Due to the broad expression of most components of metazoan intracellular signaling , including the IP3R family , identifying cell-specific function in vivo can be challenging . Drosophila genetics provides the tools for such cell-specific analysis . In Drosophila , IP3R is encoded by the single itpr gene ( Hasan and Rosbash , 1992 ) . IP3R mutants exhibit delayed moulting ( Venkatesh and Hasan , 1997 ) recently attributed to release of the steroid hormone ecdysone from the prothoracic gland ( Yamanaka et al . , 2015 ) . While null alleles are lethal as second instar larvae , heteroallelic hypomorphs exhibit developmental and metabolic phenotypes . These range from lethality across larval stages to hyperphagic adults with altered lipid metabolism . The focus of adult metabolic defects observed in itpr mutants appears to be the central nervous system ( Subramanian et al . , 2013a , 2013b ) . For a better understanding of neuronal IP3R function in the context of metabolic regulation , we chose to study the Drosophila larval to pupal transition . This transition requires systemic integration of the nutritional state of late-stage larvae with release of hormones that drive pupariation ( Andersen et al . , 2013 ) . Here , we identify a neural circuit that allows Drosophila larvae to overcome chronic protein-deprivation and pupariate . We demonstrate that nutrient sensitive plasticity of this circuit requires intracellular calcium signaling in newly identified glutamatergic neurons of the ventral ganglion .
To assess the role of IP3R in nutrient stress , itpr mutants ( itprka1091/ug3 ) were transferred as early third-instar larvae from a normal diet ( ND ) to a protein-deprived diet ( PDD ) containing only sucrose ( Figure 1A and Figure 1—figure supplement 1A and B ) . At 120 hr post transfer , wild-type larvae exhibited complete pupariation on either ND or PDD , whereas itpr mutant larvae exhibited a decrease in pupariation on ND which was worsened significantly on PDD ( Figure 1A ) . Pupariation in wild-type larvae was delayed slightly when subjected to PDD , whereas itpr mutants , which barely pupate on PDD , were delayed considerably even on a normal diet ( Figure 1B , C , D and E ) . Protein is the likely nutritional cue necessary for pupariation by itpr mutants , because the extent of pupariation on a lipid-deprived diet was significantly higher than on PDD and closer to pupariation on a normal diet ( Figure 1—figure supplement 1C ) . Moreover , pupariation in itpr mutants was restored when the PDD was supplemented with amino acids and vitamins ( Figure 1—figure supplement 1D ) . Interestingly , itpr mutant larvae feed in excess of controls ( Figure 1—figure supplement 1E ) , but their body weights are similar to that of wild-type larvae at the time of transfer to the PDD , indicating that excess feeding may be an attempt to compensate for metabolic changes due to altered intracellular calcium signaling ( Figure 1—figure supplement 1F ) . Pan-neuronal restoration of itpr function in the mutant by expressing wild-type itpr cDNA ( itpr+ ) with elav-GAL4 rescued the pupariation deficit , whereas pan-neuronal knockdown of the IP3R mimicked the itpr mutant phenotype on PDD , while exhibiting complete pupariation on ND ( Figure 1A , B , C , D and E ) . These data demonstrate that pupariation in a protein-deprived condition requires intracellular calcium signaling through the IP3R in neurons . 10 . 7554/eLife . 17495 . 003Figure 1 . Pupariation in a protein-deprived environment requires intracellular calcium signaling in neurons . ( A ) Representative images of larvae and pupae of indicated genotypes subjected to either protein-deprived diet ( PDD ) or normal diet ( ND ) . Percentages refer to pupariation . B and D Percentage pupariation of indicated genotypes represented as mean ± SEM over hours after transfer to the indicated media at 80–88 hr after egg laying ( AEL ) . Dotted lines indicate 50% viability . C and E Bars represent mean percentage pupariation at 120 hr ( ± SEM ) . Larvae were transferred to the indicated diet at 80–88 hr AEL . All pupariation experiments were performed with N ≥ 6 batches , with 25 larvae in each batch . Bars with the same alphabet represent statistically indistinguishable groups ( one-way ANOVA with a post hoc Tukey’s test p<0 . 05 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 17495 . 00310 . 7554/eLife . 17495 . 004Figure 1—figure supplement 1 . Pupariation under protein-deprivation requires neuronal IP3R . ( A ) Schematic representation of the amino acid starvation paradigm ( B ) High-magnification images of the anterior region of 80–88 hr AEL larvae of indicated genotypes C and D Bars show indicated genotypes subjected to lipid-depleted food ( C ) and 100 mM sucrose ( PDD ) supplemented with 5x RPMI ( growth supplement ) ( D ) N ≥ 6 batches with 25 larvae each . ( E ) Bars represent normalized food intake as measured by amount of red dye fed mixed with food . No significant interaction was observed between genotype and diet ( p = 0 . 98 ) . N ≥ 6 batches with 25 larvae each . ( F ) Bars represent weight of indicated genotypes at 80–88 hr AEL . Bars with the same alphabet represent statistically indistinguishable groups ( one-way ANOVA with a post hoc Tukey’s test p<0 . 05 ) . For two-way ANOVA , numbers represent the variable genotype and alphabets represent diets ( p<0 . 05 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 17495 . 004 To identify neuronal subsets that require itpr function for pupariation in protein-deprived conditions , IP3R knockdown was performed in various neuronal subsets . Among the tested subsets , the strongest pupariation deficit was observed upon IP3R knockdown in glutamatergic neurons ( Figure 2—figure supplement 1A ) . Peptidergic neurons also showed a significant pupariation deficit , but this effect was not diet specific ( Figure 2—figure supplement 1A and Table 2 ) . Interestingly , when pan-neuronal knockdown of the IP3R was restricted to the brain lobes by introducing tsh-GAL80 , pupariation was restored from less than 25% ( Figure 1E ) to more than 75% ( Figure 2—figure supplement 1B ) . These data indicate that IP3R function is required to a greater extent in neurons located in the ventral ganglion for pupariation on PDD as compared with neurons located in the central brain lobes . We therefore used two independent GAL4 strains , vglutVGN6341 and vglutVGN9281 ( 2 ) ( subsequently referred to as VGN6341 and VGN9281-2 ) , with contrasting expression patterns within the glutamatergic population of the ventral ganglion ( Figure 2A ) , to delimit the subpopulation required for pupariation on PDD . The VGN6341 and VGN9281-2 strains differ in their expression patterns especially in the third thoracic and initial abdominal segments of the ventral ganglion ( Figure 2A ) . IP3R knockdown with VGN6341 resulted in significantly reduced pupariation on PDD , whereas larvae with IP3R knockdown in VGN9281-2-expressing cells pupariated similar to control larvae ( Figure 2B ) . Normal pupariation was observed when IP3R knockdown was restricted by introducing tsh-GAL80 in the background of VGN6341 ( referred to as restricted VGN6341; Figure 2B ) . tsh-GAL80 restricted expression of VGN6341 , to the central brain and a few neurons in the ventral ganglion . In a complementary experiment , pupariation in the itpr mutant was restored by expression of UAS-itpr+ in VGN6341 expressing neurons but not with the GAL80-restricted VGN6341 neurons ( Figure 2C ) . Comparison of VGN9281-2 and VGN6341 expression with and without tsh-GAL80 pointed to a region between the third thoracic and fifth abdominal segments in the ventral ganglion , subsequently designated as the mid-ventral ganglion or mVG , where IP3R function is required for pupariation on PDD ( Figure 2D ) . These VGN6341-expressing cells were confirmed as glutamatergic by co-immunostaining of VGN6341 driven GFP with DvGlut , an established marker of glutamatergic neurons ( Daniels et al . , 2004 ) ( Figure 2E , Figure 2—figure supplement 1C ) . Although a role for peripheral glutamatergic neurons of the mVG remains possible , we focussed on the central ones as they were more easily and repeatedly identifiable . 10 . 7554/eLife . 17495 . 005Figure 2 . Knockdown of the IP3R in glutamatergic neurons prevents pupariation upon PDD . ( A ) Expression patterns of GAL4 drivers used in ( B ) determined using UAS-eGFP and co-stained with anti-nc82 . ( B and C ) Bars show mean percentage pupariation ( ± SEM ) of the indicated genotypes on PDD . N ≥ 6 batches with 25 larvae each . ( D ) Images of selected substacks of the ventral ganglion of VGN6341-GAL4 , with and without tsh-GAL80 , expressing UAS-eGFP , double labelled with anti-nc82 . ( E ) Selected substacks showing overlap of all dvGlut-positive cells and GFP-positive cells marked by VGN6341-GAL4 in the ventral ganglion . Scale bars indicate 50 µm . Bars with the same alphabet represent statistically indistinguishable groups ( one-way ANOVA with a post hoc Tukey’s test p<0 . 05 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 17495 . 00510 . 7554/eLife . 17495 . 006Figure 2—figure supplement 1 . Knockdown of the IP3R in glutamatergic neurons prevents pupariation upon protein-deprivation . ( A and B ) Bars indicate mean percentage pupariation ( ± SEM ) upon itpr knockdown with different neuronal drivers on PDD . N ≥ 6 batches with 25 larvae each . ( C ) VGN6341-GAL4 driven GFP-positive cells overlap with dvGlut-positive cells in the larval ventral ganglion . Scale bar indicates 50µm . Bars with the same alphabet represent statistically indistinguishable groups ( one-way ANOVA with a post hoc Tukey’s test p<0 . 05 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 17495 . 00610 . 7554/eLife . 17495 . 007Table 1 . Validated Hits from the GPCR RNAi Screen . Percentage pupariation ( rounded to the nearest integer ) upon GPCR knockdown performed with VGN6341-GAL4 as well as their rescue by overexpression of dSTIM and constitutively active form of Gq on PDD . DOI: http://dx . doi . org/10 . 7554/eLife . 17495 . 007Sl . no . RNAi line ( CG ) Receptor categoryReceptorLarvae to pupae with RNAi ( % ) Larvae to pupae rescue with UAS-dSTIM ( % ) Larvae to pupae rescue with UAS-AcGq ( % ) 116785FrizzledFrizzled 30804428784NPDPyrokinin 2 receptor 116908437395NPDsNPF receptor16848048795NPDPyrokinin 2 receptor 220725652114NPDFmrf receptor208844614593NPDCCHamide-2 receptor20685674356AcetylcholineMuscarinic acetylcholine receptor at 60C20888086515NPDTachykinin-like receptor at 86C206856916766MonoaminesTyramine receptor II2492761015274 ( Earlier 33310 ) GABA , GlutamateMetabotropic GABA-B receptor subtype 12876721110823NPDSIFa receptor2878881210001NPDAllatostatin receptor44927610 . 7554/eLife . 17495 . 008Table 2 . Percentage pupariation on ND . Percentage pupariation ( rounded to the nearest integer ) on ND of indicated genotypes . DOI: http://dx . doi . org/10 . 7554/eLife . 17495 . 008Sl . no . GenotypePercentage pupariation on ND1OK371-GAL4>itpr IR972VGN6341-GAL4>itpr IR953ChAT-GAL4>itpr IR994Dimm-GAL4>itpr IR635Restricted dimm-GAL4>itpr IR866Ppk-GAL4>itpr IR1007Restricted VGN6341-GAL4>itpr IR978VGN6341-GAL4>mAChR IR999Restricted VGN6341-GAL4>mAChR IR10010Dimm-GAL4>mAChR IR9711Dimm-GAL4>mGluRA IR83 To identify GPCRs that function in glutamatergic neurons and stimulate the IP3R under protein-deficient conditions , a genetic RNAi ( IR ) screen was performed with VGN6341 and all publicly available RNAi strains for GPCRs in the Drosophila genome ( Figure 3—figure supplement 1 , Supplementary file 3 ) . Twelve GPCRs identified in the screen ( Table 1 ) were validated as functioning upstream of the IP3R , by rescue of pupariation with two components of the IP3 signaling pathway , a constitutively active Gαq transgene AcGq and dSTIM ( Table 1 ) . Among the identified GPCRs , we tested further the role of the muscarinic Acetylcholine Receptor ( Flybase mAChR-A , mAChR-60C , CG4356 and henceforth , mAChR ) in pupariation . The mAChR activates Gq/PLCβ signaling leading to IP3 production and IP3-mediated Ca2+ release in Drosophila cells ( Millar et al . , 1995; Srikanth et al . , 2006 ) . Knockdown of the mAChR in glutamatergic neurons of the ventral ganglion significantly reduced pupariation on protein-deficient media and expression of the mAChR+ transgene in the same neuronal subset in the itpr mutant rescued pupariation ( Figure 3A ) . These data support mAChR stimulation followed by Ca2+ release through the IP3R as a mechanism required by glutamatergic neurons of the mVG for pupariation under conditions of nutrient stress . 10 . 7554/eLife . 17495 . 009Figure 3 . Cholinergic inputs convey nutrient-stress signals to glutamatergic neurons of the ventral ganglion . ( A ) Bars indicate mean percentage pupariation ( ± SEM ) of indicated genotypes subjected to PDD . N ≥ 6 batches with 25 larvae each . ( B ) Schematic illustrating the setup used to image neurons of interest from the larval ventral ganglion . ( C ) Representative images showing calcium activity measured by GCaMP6m in the mVG neurons of indicated genotypes at indicated time points from a time series . ( D and E ) Traces represent time series of the mean normalized GCaMP6m responses ( ± SEM ) from the mVG neurons of the indicated genotypes upon stimulation with carbamylcholine ( CCh ) . ( F ) and ( G ) Box plots represent Area under the Curve ( F ) and Peak change in fluorescence ( G ) quantified from ( E ) . In box plots , center lines show the medians; box limits indicate the 25th and 75th percentiles , whiskers extend 1 . 5 times the interquartile range from the 25th and 75th percentiles , open circles represent each data point and numbers below represent total number of cells measured . ( H ) Bars indicate mean percentage pupariation ( ± SEM ) of the indicated genotypes on PDD . N ≥ 6 batches with 25 larvae each . ( I ) Image showing GRASP between pickpocket-GAL4 and VGN6341-LexA . J and K Traces represent time series of mean normalized GCaMP6m responses ( ± SEM ) from mVG glutamatergic cells upon optogenetic activation of either cholinergic ( J ) or aminergic ( K ) domains . Grey box indicates duration of optogenetic activation . Bars with the same alphabet represent statistically indistinguishable groups ( one-way ANOVA with a post hoc Tukey’s test p<0 . 05 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 17495 . 00910 . 7554/eLife . 17495 . 010Figure 3—figure supplement 1 . Identification of GPCRs that stimulate IP3-mediated calcium signaling in response to PDD . ( A ) Schematic represents the total number of genes tested in the forward genetic screen with RNAi lines for GPCRs and the modifier screen with AcGq and dSTIM yielding the validated hits . The hits are summarized in Table 1 N ≥ 3 batches with 25 larvae each . DOI: http://dx . doi . org/10 . 7554/eLife . 17495 . 01010 . 7554/eLife . 17495 . 011Figure 3—figure supplement 2 . Cholinergic neurons are important for responding to nutrient stress . A and B Traces represent time series of mean normalized GCaMP6m responses ( ± SEM ) from the glutamatergic cells in the mVG . ( A ) Stimulation with either saline or carbachol after incubation with a mAChR specific antagonist , atropine . ( B ) GCaMP6m and RFP channels upon CCh stimulation . ( C ) Area under the curve of indicated genotypes on ND calculated from time series curves in Figure 3D . D and E Area under the curve for indicated genotypes at either 2 hr or 18 hr on ND ( D ) and PDD ( E ) . The response in the mutant was reduced irrespective of media at 2 hr . On ND , the response was unchanged over time in both , control and itpr mutant . However , the response declined over time on PDD in both . Correspondingly , there was a significant interaction between the mutation and time on PDD ( p<0 . 001 ) but not on ND ( p = 0 . 91 ) by two-way ANOVA; same numbers represent the variable genotype , and same alphabets represent the variable time as statistically indistinguishable ( p<0 . 05 ) . F G and H Bars represent mean percentage pupariation ( ± SEM ) of the indicated genotypes when synaptic activity is blocked using UAS Shits . Animals were subjected to restrictive ( 29°C ) or permissive ( 22°C ) temperature for 48 hr from 80 to 88 hr AEL on indicated media . N > 5 batches of 25 larvae each . ( I ) Bars represent mean percentage pupariation ( ± SEM ) of the indicated genotypes on PDD . N > 4 batches of 25 larvae each . Bars with the same alphabet represent statistically indistinguishable groups ( one-way ANOVA with a post hoc Tukey’s test p<0 . 05 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 17495 . 01110 . 7554/eLife . 17495 . 012Figure 3—figure supplement 3 . ppk class IV multidendritic neurons activate VGN6341 marked glutamatergic interneurons . A and B Complete confocal stacks ( z-project ) of GFP and nc82 patterns obtained when the two GRASP constructs ( UAS-CD4GFP1-10; LexAop-CD4GFP11 ) are driven by either ppk-GAL4 ( A ) or VGN6341-LexA ( B ) . ( C ) Expression pattern of ppk-QF in the larval brain . ( D ) Traces represent time series of mean normalized jRCaMP1b responses ( ± SEM ) from mVG glutamatergic cells upon optogenetic activation of the ppk domain . Blue box indicates duration of optogenetic activation . DOI: http://dx . doi . org/10 . 7554/eLife . 17495 . 012 Next we assessed the response of VGN6341-expressing neurons in the mVG to carbamylcholine ( CCh ) , an mAChR agonist ( Offermanns et al . , 1994 ) . We measured CCh responses in ex vivo preparations of third-instar larval brains ( Figure 3B ) that were subjected to either ND or PDD for 18 hr , by VGN6341-driven expression of the genetically encoded calcium indicator GCaMP6m . Upon stimulation with CCh , calcium transients were observed in multiple VGN6341 marked cells from larvae fed on a normal diet ( Figure 3C , D and Figure 3—figure supplement 2C ) . The specificity of this response was ascertained by pre-incubation with atropine , an established and specific antagonist of the mAChR ( Fryer and Maclagan , 1984 ) , which abolished the response ( Figure 3—figure supplement 2A and B ) . Not surprisingly , the response to CCh was significantly attenuated in itpr mutants on a normal diet ( Figure 3C , D and Figure 3—figure supplement 2C ) . On PDD , the response to CCh was reduced in neurons from control larvae , and it was nearly absent in neurons from itpr mutants ( Figure 3C , E , F and G; Videos 1 and 2 ) . To assess whether there was a temporal component to these responses , we measured responses at either 2 hr or 18 hr on NDD and PDD . In control larvae , the CCh response of glutamatergic neurons reduced significantly from 2 hr to 18 hr on PDD ( Figure 3—figure supplement 2E ) . The strongly attenuated CCh response observed in itpr mutants at 18 hr on PDD was not evident at 2 hr on PDD ( Figure 3—figure supplement 2E ) . These data indicate that a physiological change in the mVG glutamatergic neurons happens on PDD over time . The change in response to CCh over time was not evident on ND in either control or itpr mutant neurons ( Figure 3—figure supplement 2D ) . Importantly , the response of glutamatergic neurons in itpr mutants could be rescued by supplementing PDD with essential amino acids ( EAA ) ( Figure 3E ) . Thus , loss of dietary EAA appears to be a cause for abrogation of the CCh response in itpr mutant glutamatergic neurons after 18 hr on PDD . The response to CCh was also absent in glutamatergic neurons of the mVG with knockdown of either mAChR or itpr on ND ( Figure 3D and Figure 3—figure supplement 2C ) and PDD ( Figure 3E , F , G ) . The absence of the CCh response on ND in itpr and mAChR knockdowns , even though validating the original response to be through the IP3R , did not correlate with a defect in pupariation ( Table 2 ) , suggesting that mAChR and IP3R function in mVG neurons is critically required on PDD but not relevant on ND ( see discussion ) . The response to CCh on PDD was rescued to a significant extent by over-expression of either mAChR+ or itpr+ in VGN6341-marked neurons of itpr mutants ( Figure 3E , F and G ) , and correlated with rescue of pupariation observed in these genotypes ( Figures 2C and 3A ) . These data suggest that cholinergic inputs convey protein-starvation to VGN6341-marked glutamatergic neurons possibly through acetylcholine , that activate mAChR and the IP3R for pupariation . 10 . 7554/eLife . 17495 . 013Video 1 . Response to CCh in VGN 6341 neurons of the mVG from control larvae on PDD . The green flash indicates point of point of addition of CCh . DOI: http://dx . doi . org/10 . 7554/eLife . 17495 . 01310 . 7554/eLife . 17495 . 014Video 2 . Response to CCh in VGN 6341 neurons of the mVG from itpr mutant larvae on PDD . Green flash indicates point of addition of CCh . DOI: http://dx . doi . org/10 . 7554/eLife . 17495 . 014 To test if cholinergic stimuli signal protein-deprivation , recycling of synaptic vesicles was blocked in cholinergic neurons by expression of the dynamin mutant transgene Shibirets ( UAS-Shits1 ) with the cholinergic driver ChaT-GAL4 . Pupariation was reduced significantly when cholinergic transmission was blocked , by transferring larvae ( 80–88 hr AEL ) to the restrictive temperature ( 29°C ) for 48 hr ( till 128–136 hr AEL ) , concurrent with protein-deprivation ( Figure 3—figure supplement 2G ) . The same experiment performed either with larvae on ND or with larvae on PDD but at the permissive temperature ( 22°C ) supported pupariation ( Figure 3—figure supplement 2F and H ) . When Shits expression was driven by a combination of ChaT-GAL4 and ppk-GAL80 ( blocking GAL4 activity in multidendritic sensory neurons ) , normal pupariation was observed on the protein-deficient diet ( Figure 3—figure supplement 2G and H ) , suggesting the requirement of neurons expressing the Pickpocket ( ppk ) channel ( Adams et al . , 1998 ) in sensing protein-deficient conditions prior to pupariation . This idea was tested directly by expression of Shits or a hyperpolarising potassium channel Kir2 . 1 in neurons expressing Pickpocket ( ppk-GAL4 ) , both of which caused a severe pupariation deficit ( Figure 3H , Figure 3—figure supplement 2G ) . Pupariation was restored in these animals by supplementing the protein-deficient diet with a mixture of EAA ( Figure 3H ) . Activity in ppk-GAL4-marked neurons is thus required for pupariation on PDD but not on ND . In contrast , pupariation was unaffected when activity was inhibited in cholinergic neurons marked by 19-12-GAL4 expression , which does not overlap with ppk-GAL4 ( Figure 3—figure supplement 2G and H; Yan et al . , 2013 ) . The presence of direct cholinergic inputs to VGN6341-marked glutamatergic neurons was tested next by performing an experiment for genetic reconstitution across synaptic partners ( GRASP ) ( Feinberg et al . , 2008 ) . GRASP signals between VGN6341- and ppk-marked neurons were detected in the neuropil of the ventral ganglion ( Figure 3I ) , which is a synaptically dense region along the midline . GRASP constructs expressing the split GFP components individually with either ppk-GAL4 or VGN6341-LexA had no GFP expression ( Figure 3—figure supplement 3A and B ) . To test if the observed connections are functional , cholinergic neurons were marked by ChaT-LexA and optogenetically activated with a red-shifted channelrhodopsin variant , LexAop-CsChrimson . Calcium transients were observed in glutamatergic cells of the mVG simultaneously with optogenetic activation of cholinergic neurons on either ND or PDD ( Figure 3J; Video 3 ) . On PDD , interestingly , the transients appeared to oscillate . As a control , we tested optogenetic activation of aminergic neurons ( marked by HL9-LexA ) . This did not elicit a response in glutamatergic neurons of the mVG ( Figure 3K ) . To confirm that activation of VGN6341 neurons was through inputs from ppk neurons , we performed an optogenetic activation experiment with TrpA1-QF which marks class IV multidendritic neurons ( Petersen and Stowers , 2011; subsequently referred to as ppk-QF ) . A robust activation of VGN6341-marked cells expressing the red shifted calcium indicator jRCaMP1b was observed upon optogenetic activation of ppk neurons expressing QUAS-ChR2 ( Figure 3—figure supplement 3 ) . Knockdown of the IP3R in cholinergic neurons did not change pupariation on PDD ( Figure 2—figure supplement 1A ) , and restoring IP3R function in cholinergic neurons failed to rescue the itpr mutant ( Figure 3—figure supplement 2I ) . As normal pupariation was observed in animals with cholinergic knockdown of the IP3R on PDD , we did not test cholinergic stimulation of glutamatergic neurons in IP3R mutants . Instead , to understand the basis of the pupariation defect on PDD , we investigated next the postsynaptic partners of glutamatergic neurons marked by VGN6341-GAL4 . 10 . 7554/eLife . 17495 . 015Video 3 . Calcium transients observed in glutamatergic neurons of the mVG as a result of optogenetic activation of cholinergic neurons . DOI: http://dx . doi . org/10 . 7554/eLife . 17495 . 015 Neuropeptides are known to modulate organismal responses to changes in diet in vertebrates as well as insects ( Morton et al . , 2014; Nässel and Winther , 2010 ) . To test , whether the mVG interneurons synapse on to peptidergic cells , we imaged the two domains at high resolution . Anterior glutamatergic projections arising from near the central mVG interneurons ( Figure 4A and B , arrow heads ) reach posterior projections of the mNSCs ( Figure 4A and B , asterisks ) . These anterior glutamatergic projections appear to originate from the central mVG neurons , extend laterally toward the midline for a short distance , and then project to the anterior ( Figure 4D and video 4 ) . They do not arise from peripheral mVG neurons that appear to project solely to the midline neuropil of the VG . Similar connectivity was observed upon marking the mVG glutamatergic neurons and specifically the mNSCs ( Figure 4C ) . 10 . 7554/eLife . 17495 . 016Figure 4 . Glutamatergic neurons in the larval ventral ganglion project to peptidergic neurons in the mNSC . A and B Selected confocal stacks showing the neurites marked by VGN6341-GAL4 driven UAS-eGFP ( green ) and their merged patterns with dimm-LexA-driven expression of LexAop-mCherry ( red ) . The boxed area in A is shown in B as a high-magnification image . Arrow heads indicate VGN6341-GAL4 expressing neurites projecting toward the mNSCs . Asterisks mark dimm-LexA labelled projections . ( C ) Neurites marked by VGN6341-GAL4-driven eGFP ( arrow heads ) overlap with projections of the mNSCs marked by Dilp2mCherry ( asterisks ) . ( D ) Selected high-magnification confocal images of VGN6341-GAL4 driven UAS-eGFP with an anterior projecting neurite from a midline mVG neuron . The white arrow head marks the same co-ordinates in all three images . The yellow arrow head shows the ascending projections . Scale bars represent 50 µm in A and B and 10 µm in C and D . DOI: http://dx . doi . org/10 . 7554/eLife . 17495 . 01610 . 7554/eLife . 17495 . 017Video 4 . Projections from VGN 6341-marked glutamatergic neurons to the mNSCs . DOI: http://dx . doi . org/10 . 7554/eLife . 17495 . 017 Additionally , a GRASP experiment was performed between VGN6341-LexA-marked neurons and peptidergic neurons marked by dimm-GAL4 ( Figure 5A ) . The ensuing GRASP presented a complex pattern comprising medial Neurosecretory Cells ( mNSCs ) in the central brain , multiple projections between the mVGs and the mNSCs , and a few cell bodies and projections in the periphery of the ventral ganglion ( Figure 5A and Figure 5—figure supplement 1A ) . Controls with either GAL4 or LexA driving individual GRASP constructs elicited no GFP immunostaining ( Figure 5—figure supplement 2A and B ) . To test the functional significance of this apparent connectivity between peptidergic and glutamatergic cells , we measured pupariation on PDD after knockdown of various glutamatergic receptors in dimm-positive neurons . Amongst the five receptors tested , pupariation was reduced significantly in animals with knockdown of the metabotropic glutamate receptor A ( mGluRA; Figure 5B and Figure 5—figure supplement 1B ) . Knockdown of mGluRA with two independent RNAi strains showed differing but significant deficits in pupariation ( 10% and 40%; Figure 5B , Figure 5—figure supplement 1B ) . We attribute this difference to the strength of the RNAi knockdown in peptidergic cells from the two different RNAi strains . Knockdown of other glutamate receptor classes resulted in normal pupariation ( Figure 5—figure supplement 1B ) . Restriction of mGluRA knockdown in peptidergic neurons of the central brain and in Dilp2-positive mNSCs ( Dilp2-GAL4 ) also resulted in pupariation deficits on PDD ( Figure 5B ) . In contrast , normal pupariation was observed by restricting knockdown of IP3R to VGN6341-marked glutamatergic neurons of the central brain ( Figure 2B ) . Taken together these data support the innervation of central brain peptidergic neurons , specifically the mNSCs , by glutamatergic neurons of the ventral ganglion . This innervation appears relevant for pupariation on PDD . 10 . 7554/eLife . 17495 . 018Figure 5 . Glutamatergic neurons in the larval ventral ganglion convey signals to peptidergic neurons of the mNSC . ( A ) High-magnification images of the mNSC area in a GRASP experiment between the peptidergic and the glutamatergic domains stained for GFP and Dilp2 . ( B ) Bars represent mean percentage pupariation ( ± SEM ) of larvae subjected to mGluRA knockdown using indicated GAL4 drivers on PDD . N ≥ 6 batches with 25 larvae each . ( C ) A z-project of selected substacks at higher magnification showing the mNSC region from ( A ) . Scale bar indicates 10 µm . Arrow heads point to weakly stained cells . ( D ) Confocal images showing the mNSC region of the Dilp2-GAL4 simultaneously driving an axonal and dendritic marker ( Dilp2>UAS-DenMark , UAS-SyteGFP ) . E and F Traces represent time series of mean normalized GCaMP responses ( ± SEM ) from peptidergic cells in the mNSC of the indicated genotypes on ND ( E ) or PDD ( F ) . G and H Quantification of area under the curve from ( E ) and ( F ) . Box plots and symbols are as described for Figure 3F . ( I ) Bars represent mean percentage pupariation ( ± SEM ) of indicated genotypes subjected to PDD . N ≥ 6 batches with 25 larvae each . Scale bars indicate 50µm unless specified otherwise Bars with the same alphabet represent statistically indistinguishable groups ( one-way ANOVA with a post hoc Tukey’s test p<0 . 05 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 17495 . 01810 . 7554/eLife . 17495 . 019Figure 5—figure supplement 1 . IP3R signaling in glutamatergic neurons is required for activation of medial neurosecretory cells . ( A ) Confocal image showing GRASP between the peptidergic ( dimm-GAL4 ) and the glutamatergic ( VGN6341-LexA ) domains stained for GFP and Dilp2 . ( B ) Bars represent mean percentage pupariation ( ± SEM ) of indicated genotypes on PDD . N ≥ 6 batches with 25 larvae each . ( C ) and ( D ) Representative images of the mNSC showing GCaMP6m response upon thermogenic activation by VGN6341-GAL4 . ( E ) Traces represent time series of mean normalized GCaMP6m responses ( ± SEM ) from mNSCs of the indicated genotypes upon thermogenic activation using VGN6341-GAL4 . The grey box indicates duration of thermogenic activation . ( F ) Traces represent time series of mean normalized GCaMP6m responses ( ± SEM ) from the VGN6341 neurons upon self-optogenetic activation with CsChrimson . The red box indicates duration of optogenetic activation . ( G ) Area under the curves of GCaMP6m responses from the VGN6341 neurons upon self-optogenetic activation . ( H ) Traces represent time series of mean GCaMP6m responses ( ± SEM ) from the mNSC in itpr mutants upon activation of glutamatergic cells either acutely or chronically . ( I ) Traces represent time series traces of mean GCaMP6m responses ( ± SEM ) of oscillating cells in the mNSC ( from E ) observed upon thermogenic activation of VGN6341 neurons in control larvae ( 4/33 on ND and 17/36 on PDD ) . Bars with the same alphabet represent statistically indistinguishable groups ( one-way ANOVA with a post hoc Tukey’s test p<0 . 05 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 17495 . 01910 . 7554/eLife . 17495 . 020Figure 5—figure supplement 2 . Neurons of the mVG activate peptidergic neurons in the mNSCs . A and B A z-project image of GFP and the nc82 patterns obtained when either dimm-GAL4 ( A ) or VGN6341-LexA ( B ) drive the complete GRASP constructs ( UAS-CD4GFP1-10; LexAop-CD4GFP11 ) . The VGN-6341-LexA image is the same as that shown in Figure 3—figure supplement 2 . ( C ) Traces represent time series of mean normalized GCaMP6m responses ( ± SEM ) from the peptidergic mNSCs upon optogenetic activation . Red boxes indicate duration of optogenetic activation . The VGN6341-GAL4 activation trace is the same as that shown in Figure 5F . DOI: http://dx . doi . org/10 . 7554/eLife . 17495 . 020 mNSCs in Drosophila central brain secrete multiple neuropeptides , including insulin-like peptides ( Dilps ) , which regulate the response of an organism to dietary changes ( Dus et al . , 2015 ) . Upon co-immunostaining with an antibody against Dilp2 ( Géminard et al . , 2009 ) , five of the seven Dilp2-positive cells in the mNSCs were 'GRASPed' by VGN6341 ( Figure 5C ) as suggested earlier by immunostaining of mVG cells and the mNSCs ( Figure 4C ) . The presence of dendritic fields close to the mNSC soma has been described earlier ( Nässel et al . , 2008; Vallejo et al . , 2015 ) and was confirmed by simultaneous expression of an axonal and a dendritic marker with Dilp2-GAL4 ( Figure 5D ) . Thus , axonal projections from VGN6341-marked glutamatergic neurons probably stimulate a subset of mNSCs . The presence of functional synaptic connections between VGN6341-GAL4 and mNSCs was tested next . CsChrimson-expressing VGN6341 cells were optogenetically activated in ex vivo preparations in which GCaMP6m was expressed in peptidergic cells , including the mNSCs , marked by dimm-LexA::p65 . Upon optogenetic stimulation of CsChrimson in VGN6341 marked neurons calcium transients were observed in the mNSCs of larvae from both normal and protein-deficient diets ( Figure 5E , F , G and H; Video 5 ) . Similar activation of the mNSCs was obtained when VGN6341 marked neurons were thermogenically activated with dTrpA1 , a temperature-activated cation channel ( Figure 5—figure supplement 1C and E ) . Activation of VGN6341 marked neurons in the itpr mutant , however , did not evoke transients either on ND ( Figure 5E and G ) or on PDD ( Figure 5F and H ) . The underlying basis for this defect appears to be an inability of the VGN6341-marked neurons to stimulate the mNSCs in itpr mutants , because their optogenetic self-activation evoked robust calcium transients in itpr mutants on ND and PDD ( Figure 5—figure supplement F and G ) . Activation of the Restricted VGN6341-GAL4 , where GAL4 expression is absent from the mVG region , did not elicit a signal from the mNSCs ( Figure 5—figure supplement 2C ) . 10 . 7554/eLife . 17495 . 021Video 5 . Calcium transients observed in peptidergic neurons in the mNSCs of control larvae on PDD upon optogenetic activation of VGN 6341 neurons . DOI: http://dx . doi . org/10 . 7554/eLife . 17495 . 021 Next , we tested if artificial activation of glutamatergic neurons marked by VGN6341 compensated for reduced itpr function in the context of pupariation on PDD . NaChBac is a bacterial sodium channel that increases excitability of Drosophila neurons ( Nitabach et al . , 2006 ) . Expression of either dTrpA1 , NaChBac , or CsChrimson in glutamatergic neurons marked by VGN6341 indeed rescued the pupariation defect of the itpr mutant on PDD ( Figure 5I ) . For rescue experiments with either dTrpA1 or CsChrimson , activation was for 48 hr and this period was concurrent with transfer to PDD from 80–88 hr to 128–136 hr AEL . Taken together , these results suggest that glutamatergic interneurons in the larval mVG receive cholinergic signals indicating absence of dietary amino acids , process these signals in an IP3R-dependent manner , and convey this information to peptidergic cells in the mNSCs for mounting a suitable physiological response to enable pupariation . The effect of CCh-induced calcium transients on neuropeptide release in the mNSCs was tested next . For this purpose , an ex vivo preparation was taken in which cells marked by dimm-LexA::p65 expressed the rat atrial natriuretic peptide fused with GFP ( ANF::GFP ) under LexAop control . Peptide release in Drosophila neurons has been studied previously by measuring release of ANF::GFP ( Shakiryanova et al . , 2006 ) . Upon stimulation with 50 µM CCh , a decay in GFP fluorescence was observed in the mNSCs ( Figure 6A; Video 6 ) . Application of saline did not show a change in fluorescence ( Figure 6A ) . The extent of release observed was higher in brains from larvae on PDD as compared to larvae on ND ( Figure 6A ) . Release of ANF::GFP was significantly attenuated in mNSCs of the itpr mutant and importantly , this could be rescued by expressing mAChR+ in glutamatergic neurons marked by VGN6341 ( Figure 6B , C and F ) . Moreover , peptide release from the mNSCs was reduced significantly upon knockdown of mAChR in VGN6341-marked glutamatergic neurons and not by direct knockdown of mAChR in peptidergic neurons with dimm-GAL4 ( Figure 6G ) . Thus , cholinergic stimulation of glutamatergic neurons in the mVG appears to regulate peptide release from the mNSCs by activating mGluRA ( Figure 6G ) . Further support for the ability of glutamatergic neurons to stimulate peptidergic release from the mNSCs comes from optogenetic stimulation of VGN6341-marked glutamatergic neurons ( Figure 6D ) . Enhanced release of ANF::GFP was observed during the period of optogenetic stimulation ( Figure 6H ) . Conversely , acute optogenetic inhibition of VGN6341-marked glutamatergic neurons using halorhodopsin , a light-activated chloride pump known to hyperpolarize neurons ( Berni et al . , 2012; Inada et al . , 2011 ) , during CCh stimulation inhibited peptide release from mNSCs ( Figure 6E and I ) . 10 . 7554/eLife . 17495 . 022Figure 6 . mAChR stimulation in glutamatergic neurons modulates enhanced peptide release from the mNSCs upon protein-deprivation . ( A–C ) Traces represent a time series of mean normalized peptide release ( ANF::GFP; ± SEM ) from mNSCs of the indicated genotypes upon Carbachol ( CCh ) stimulation . D and E Traces represent a time series of mean normalized peptide release ( ANF::GFP; ± SEM ) on PDD from the mNSCs upon optogenetic activation of VGN6341-GAL4 ( E ) and using CCh under acute inhibition from the VGN6341-GAL4 ( F ) . Red and green boxes indicate duration of activation and inhibition , respectively . ( F–I ) Box plots of CCh stimulated peptide release ( ANF::GFP ) quantified by area under the curve from the mNSCs of the indicated genotypes on PDD . Box plots and symbols are as described for Figure 3F . Bars with the same alphabet represent statistically indistinguishable groups ( one-way ANOVA with a post hoc Tukey’s test p<0 . 05 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 17495 . 02210 . 7554/eLife . 17495 . 023Video 6 . Peptide release from mNSC of control larvae on PDD as observed by the decrease in ANF::GFP intensity upon addition of CCh . Green flash indicates point of addition of CCh . DOI: http://dx . doi . org/10 . 7554/eLife . 17495 . 023 Although somatic peptide release is known ( De-Miguel and Nicholls , 2015; Trueta and De-Miguel , 2012 ) , we also tested if peptide release was affected more specifically in varicosities of the mNSCs that project to the ring gland . In the itpr mutant , ANF::GFP release was significantly reduced upon stimulation with 50 µm carbachol ( Figure 7A , C and E ) . The extent of release could be rescued by expression of mAChR+ in glutamatergic neurons marked by VGN6341 ( Figure 7D and E ) . Similar results were observed when expression of ANF::GFP was restricted to a subset of mNSCs marked by dilp2-GAL4 ( Figure 7F–H ) . 10 . 7554/eLife . 17495 . 024Figure 7 . mAChR stimulation of glutamatergic neurons modulates peptide release from varicosities at the ring gland upon protein-deprivation . ( A ) Time series of ANF::GFP release from varicosities in the ring gland of the indicated genotypes at the indicated time intervals , after stimulation by Carbachol ( CCh ) . ( B–D ) Traces represent a time series of mean normalized peptide ( ± SEM ) release from varicosities at the ring glands of the indicated genotypes after Carbachol ( CCh ) stimulation . ( E ) Box plots representing CCh-stimulated peptide release with ANF::GFP quantified by area under the curve of the indicated genotypes on PDD from ( B–D ) . Box plots and symbols are as described for Figure 3F . F and G Traces represent a time series of mean normalized peptide release ( ANF::GFP; ± SEM ) from varicosities at the ring glands of indicated genotypes upon Carbachol ( CCh ) stimulation . ( H ) Box plots representing CCh-stimulated peptide release with ANF::GFP quantified by area under the curve of the indicated genotypes on PDD from ( F and G . ) . Box plots and symbols are as described for Figure 3F . Bars with the same alphabet represent statistically indistinguishable groups ( one-way ANOVA with a post hoc Tukey’s test p<0 . 05 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 17495 . 024 Among the peptides released from the mNSCs , Dilp2 has been implicated in the response to protein deficiency ( Géminard et al . , 2009; Sano et al . , 2015 ) . Therefore , we measured Dilp2 levels in the mNSCs of larvae subjected to either normal or protein-deficient diets before and after subjecting them to CCh stimulation for 30 min . Dilp2 levels were assessed by immunohistochemistry using a previously validated antibody ( Géminard et al . , 2009 ) . Significant release of Dilp2 was observed post-CCh stimulation , as interpreted from the reduced staining observed in wild-type mNSCs of larvae on ND and PDD ( Figure 6A and B ) . In the itpr mutant , however , CCh-stimulated Dilp2 release was significantly lower , irrespective of the diets ( Figure 8A and B ) . CCh stimulation of brains with itpr knockdown in neurons marked by VGN6341 also induced weaker release of Dilp2 ( Figure 8C ) . Taking these observations forward , when Dilp2+was over-expressed using Dilp2-GAL4 , or when Dilp2 positive neurons were maintained in an excitable state by expression of NaChBac , a partial rescue of pupariation in the itpr mutant was observed on PDD ( Figure 8D ) . Taken together , these results suggest that pupariation on a protein-deficient diet requires mVG-mediated neuropeptide release from the mNSCs . Alternately , or in parallel , pupariation on PDD might require up-regulation of Dilp2 synthesis in the mNSCs , triggered by mGluRA signaling . However this seems unlikely because Dilp2 mRNA levels are reduced to equal extents upon starvation in both wild-type and itpr mutant brains ( Figure 8—figure supplement 1 ) . 10 . 7554/eLife . 17495 . 025Figure 8 . Glutamatergic neurons regulate Dilp2 release upon protein-starvation . ( A ) Dilp2 staining in larval brains from the indicated genotypes before and after stimulation with 50µM CCh for 30 min . ( B ) and ( C ) Box plots representing percentage release of Dilp2 from the respective genotypes subjected to ND or PDD . Box plots and symbols are as described for Figure 2F . A significant interaction was observed between genotype and diet ( p<0 . 001 ) . ( D ) Bars represent percentage pupariation as mean ± SEM of indicated genotypes on PDD . N ≥ 6 batches with 25 larvae each . Bars with the same alphabet represent statistically indistinguishable groups ( one-way ANOVA with a post hoc Tukey’s test p<0 . 05 ) . For two-way ANOVA , numbers represent the variable genotype and alphabets represent diets ( p<0 . 05 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 17495 . 02510 . 7554/eLife . 17495 . 026Figure 8—figure supplement 1 . dilp2 mRNA levels are not altered in the itpr mutant . ( A ) Bar graph of the mean fold change ( ± SEM ) of dilp2 mRNA levels in the larval CNS . All larvae were fed normal food until 82–86 hr AEL and then transferred to the respective media for 18 hr before dissection at 100–104 hr AEL . Two way ANOVA revealed a significant effect of diet ( p = 0 . 00018 ) but not of the itpr mutation ( p = 0 . 804 ) on dilp2 mRNA levels . No significant interaction between diet , and the mutation was observed ( p = 0 . 706 ) . Same numbers represent the variable genotype , and same alphabets represent the variable time as statistically indistinguishable ( p<0 . 05 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 17495 . 026 A peak of ecdysone release late in the wandering third-instar triggers larval pupariation on ND ( Warren et al . , 2006 ) . Expression levels of genes encoding enzymes of the ecdysteroid biosynthetic pathway have been studied ( McBrayer et al . , 2007; Shimada-Niwa and Niwa , 2014 ) , in larvae on a normal diet . Specifically , transcripts of shadow ( sad ) , spookier ( spok ) , phantom ( phm ) , neverland ( nvd ) and disembodied ( dib ) are up-regulated dramatically before pupariation ( McBrayer et al . , 2007; Shimada-Niwa and Niwa , 2014; Warren et al . , 2006 ) . It is therefore likely that the neural circuit identified for pupariation on PDD affects ecdysone synthesis by regulating ecdysteroid biosynthesis . Levels of sad , spok , phm , nvd and dib transcripts over time were characterized by analysis of RNA isolated from wild-type prothoracic glands from larvae on PDD . Transcripts for all these genes peaked approximately 42 hr post protein-deprivation in wild-type larvae aged 80–88 hr AEL ( Figure 9A ) . The expression levels of these genes were lower in the itpr mutant at 42 hr and 66 hr and could be significantly rescued by overexpression of mAChR+in the neurons marked by VGN6341 , especially at 66 hr ( Figure 9B ) . Their up-regulation corresponded to a rescue in pupariation as well ( Figure 3A ) . Taken together these data indicate that glutamatergic neurons of the mVG regulate expression of genes required for the ecdysone peak for pupariation on PDD in a mAChR- and itpr-dependent manner ( Figure 9C and D ) . Ecdysteroid synthesis and release is regulated by neuropeptides , predominantly prothoracicotropic hormone ( PTTH ) ( McBrayer et al . , 2007; Mirth et al . , 2005 ) , as well as through insulin signaling ( Colombani et al . , 2005; Mirth et al . , 2005 ) . We propose that this regulation of ecdysone synthesis by the mVG is through neuropeptide release from the mNSCs . 10 . 7554/eLife . 17495 . 027Figure 9 . IP3R signaling in glutamatergic neurons regulates the expression of ecdysone biosynthetic genes during protein-deprivation . ( A ) Normalized fold changes in the mRNA levels of the indicated genes represented as means ± SEM at indicated time points after 83–85 hr AEL on PDD from wild-type ring glands ( n ≥ 3 ) . ( B ) Bars represent mean fold changes ( ± SEM ) of expression levels of respective ecdysteroid-synthesizing genes as shown in ( A ) from the ring glands of indicated genotypes at indicated time points ( n ≥ 5 ) . Bars with the same alphabet represent statistically indistinguishable groups ( one-way ANOVA with a post hoc Tukey’s test p<0 . 05 ) . ( C ) Schematics of the neuronal circuit required for pupariation under protein-deprivation in early third instar larvae ( 80–88 hr AEL ) . Upon amino acid deprivation , glutamatergic neurons of the mVG are activated by ppk inputs . These glutamatergic neurons activate peptidergic cells in the mNSC to release peptides to further modulate ecdysteroid gene expression . In itpr mutants upon amino acid deprivation , glutamatergic inputs from the mVG to the mNSCs remain silent . ( D ) Schematic model of the signaling mechanisms observed in the circuit for pupariation under protein-deprived conditions . DOI: http://dx . doi . org/10 . 7554/eLife . 17495 . 027
Similar to other insect larvae , the larval phase in Drosophila is dedicated to feeding and attaining a nutritional state ready for metamorphosis ( Nijhout , 2003 ) . Feeding and nutrient sensing thus constitute important aspects of larval life . The glutamatergic interneurons identified here exhibit diet-induced plasticity in their response to carbachol , an acetylcholine mimic . Their response appears integral to circuit function , and very likely drives enhanced neuropeptide secretion from the mNSCs and modulates ecdysteroid synthesis in the ring gland . Unlike the central neurons described recently ( Bjordal et al . , 2014 ) , the glutamatergic interneurons do not sense amino acid levels directly . Instead , they reside in the ventral ganglion and receive inputs from sensory neurons of the class-IV multidendritic type marked by the ppk-GAL4 . The precise identity of ppk-positive cells that sense the lack of amino acids needs to be determined . Vertebrate neurosecretory cells located in the hypothalamus are integral to nutrient-sensing and energy homeostasis ( Sternson , 2013 ) and have been equated to the mNSCs ( Hartenstein , 2006 ) . A recent report identified the importance of peripheral inputs to the central brain for maintaining nutritional homeostasis ( Zeng et al . , 2015 ) . There is thus a remarkable similarity in the nutrient-sensing circuit we describe here to the circuits proposed in mammals . In vertebrates , neuronal circuit perturbation involving peripheral and central circuits remains challenging . The glutamatergic neurons identified here appear to regulate neuropeptide release from the mNSCs and might be equivalent to neuronal regulators of hypothalamic neurons . Intracellular Ca2+ release through the IP3R has been implicated in the regulation of lipid and carbohydrate metabolism in vertebrates and invertebrates ( Agrawal et al . , 2009; Ozcan et al . , 2012; Wang et al . , 2012 ) . Vertebrate IP3R is encoded by three genes ( Furuichi et al . , 1989; Yamamoto-Hino et al . , 1994 ) , thus creating the possibility for layered , complex regulation . Mice null for IP3R2 and IP3R3 exhibit digestive defects associated with loss of exocrine secretion from pancreatic β cells and the salivary gland ( Futatsugi et al . , 2005 ) . Intracellular calcium signaling in the context of neuronal regulation of a systemic response to nutritional cues as described here , remains to be tested in vertebrates . We speculate that IP3R-mediated calcium release in glutamatergic interneurons stimulates glutamate release onto presynaptic terminals of neurosecretory cells . We propose that reducing IP3R function in glutamatergic neurons in the mVG alters their connectivity with peptidergic neurons . Synapse formation induced by glutamate release has been reported ( Kwon and Sabatini , 2011 ) . Our data do not have the resolution to distinguish between physical and molecular bases of synaptic dysfunction . Activity-dependent rescue ( observed in Figure 4I ) restores the connectivity ( Figure 5—figure supplement 1H ) , suggesting that these molecular changes , at least in part affect excitability . In addition to mAChR , our screen identified eleven other GPCRs that signal through intracellular Ca2+ and whose knockdown in glutamatergic neurons resulted in loss of pupariation on PDD . Although each GPCR leads to the same systemic phenotype , the mechanisms in each case might differ , and each requires further study . Neuropeptide receptors constitute 8 out of 12 of these receptors . Interestingly , the neuropeptide receptors we report here , except SIFaR , have all been implicated in regulating feeding in vertebrates and insects ( Caers et al . , 2012; Hentze et al . , 2015; Sano et al . , 2015 ) . The calcium transients observed in glutamatergic neurons in the mVG upon optogenetic activation of cholinergic inputs were markedly different in larvae on ND and PDD . On PDD , the transients were oscillating , whereas on ND oscillations were either dampened or absent . Moreover , oscillations in the glutamatergic neurons appear to drive oscillations in the mNSCs of larvae on PDD as seen by the increase in neurons that oscillate ( Figure 5—figure supplement 1I ) . Such oscillations are known to cause exocytosis of neuropeptides ( Tse et al . , 1993 ) and may underlie enhanced neuropeptide secretion required for pupariation on PDD . The molecular basis of such diet-induced plasticity leading to robust oscillatory activity in the glutamatergic neurons of the mVG needs investigation . In a recent report , dopaminergic neurons in the larval brain of Drosophila were demonstrated as directly sensing amino acids leading to their activation and consequent changes in food-intake behaviour on an amino acid deficient medium ( Bjordal et al . , 2014 ) . These findings , however , do not explain the full range of responses to amino acid deprivation . Our results describe a response mechanism that organisms employ to overcome the developmental consequences of protein-deprivation ( Chen et al . , 2015 ) . Higher levels of Dilp2 were observed in the larval mNSCs of IP3R mutants , as well as in larvae with IP3R knockdown in glutamatergic neurons of the ventral ganglion . We attribute the excess Dilp2 to reduced release of Dilp2 from the mNSCs . The absence of up-regulation of dilp2 transcripts in IP3R mutants supports this idea ( Figure 8—figure supplement 1 ) . This is , to our knowledge , the first report of Dilp2 regulation by neurons in the ventral ganglion . Environmental nutrients regulating Dilp function has been well documented ( Géminard et al . , 2009; Kim and Neufeld , 2015; Rajan and Perrimon , 2012 ) . Most of these studies report remote sensing attributed to the fat body ( Géminard et al . , 2009; Koyama and Mirth , 2016; Rajan and Perrimon , 2012; Sano et al . , 2015 ) . The regulation we report here is another layer in the overall regulation of Dilps in Drosophila that seems to be particularly important upon protein-deprivation . Such complex layers of modulation are not surprising given that insulin signaling is important in different aspects of development and growth in Drosophila ( Colombani et al . , 2005 ) . The Dilps have been speculated as regulators of pupariation ( Koyama et al . , 2014 ) , and Dilp neurons project to the ring gland where ecdysone production occurs ( Cao and Brown , 2001 ) . Further work will be required to understand the regulation of Dilp secretion as a balance between neuronal and fat body signaling . Ultimately , all these layers of regulation seem to depend on the environmental context and developmental stage . In insects , steroid hormones control developmental transitions ( Thummel et al . , 1990 ) , including larval moults and metamorphosis . In third-instar larvae , there are ecdysone peaks ranging from several small ones to a bigger commitment peak prior to pupariation ( Warren et al . , 2006 ) . Ecdysone is the switch for developmental change , hence needs to be tightly regulated through transcriptional control of ecdysteroid genes ( McBrayer et al . , 2007 ) and ecdysteroid biosynthesis is reported to be influenced by environmental conditions ( Shimada-Niwa and Niwa , 2014 ) . Serotonergic regulation of ecdysone release by changing neurite projections in a nutrient-dependent manner has been reported recently ( Shimada-Niwa and Niwa , 2014 ) . Our results do not rule out connectivity between mVG neurons and the supra-oesophageal ganglion in the central brain that could additionally stimulate the ring gland . We propose that the circuit identified here functions as a further layer of regulation , required during either sudden starvation or nutrient deprivation . It is required to make the key developmental decision of whether and when to pupariate . Energy homeostasis at a systemic level involves integrating environmental cues with internal states . The circuit we describe is such an integrator .
Drosophila strains were grown on cornmeal medium supplemented with yeast ( ND ) at 25°C unless otherwise noted . The protein-deprived diet ( PDD ) contained 100 mM sucrose with 1% agar . For optogenetic experiments , egg laying was carried out in cornmeal medium supplemented with 200 µM all-trans-retinal ( ATR ) , and larvae were transferred at 84 ± 4 hr onto ND or PDD with 400 µM ATR . Canton S was used as wild-type ( WT ) throughout . A table of all stocks used is appended as Supplementary file 1 . The itpr IR was used with UAS-dicer in all experiments . For the GPCR RNAi screen , RNAi lines were obtained from either VDRC or NIG fly stock centres . Larvae at 84 ± 4 hr post egg laying were transferred to PDD or ND in batches of 25 and were scored for pupariation . At least six independent batches were performed for each genotype on each media . These are reported as percentage pupariation . For experiments involving diet-based rescues , PDD was supplemented either EAA ( 1x MEM EAA , GIBCO ) or growth supplements ( 5x RPMI 1640 Amino acid solution , Sigma ) . For rate of pupariation , all genotypes in Figure 1 were monitored every 12 hr after transfer . Immunostaining of larval Drosophila brains was performed as described previously ( Daniels et al . , 2008 ) . Briefly , larval brains were dissected in 1x phosphate buffered saline ( PBS ) and fixed with 4% Paraformaldehyde or Bouin’s fixative for dvGlut staining . They were washed three to four times with 0 . 2% phosphate buffer , pH 7 . 2 containing 0 . 2% Triton-X 100 ( PTX ) and blocked with 0 . 2% PTX containing 5% normal goat serum ( NGS ) for four hours at 4°C . Respective primary antibodies were incubated overnight ( 14–16 hr ) at 4°C . For dvGlut staining , the brains were incubated for 60–72 hr at 4°C . After washing three to four times with 0 . 2% PTX at room temperature , they were incubated in the respective secondary antibodies for 2 hr at room temperature . The following primary antibodies were used: rabbit anti-GFP antibody ( 1:10 , 000; A6455 , Life Technologies , RRID:AB_221570 ) , mouse anti-GFP antibody ( 1:50; Santa Cruz Biotechnology , RRID:AB_627695 ) , rabbit anti-dsRed ( 1:500; 632496 , Clontech , RRID:AB_10015246 ) mouse anti-nc82 ( anti-brp ) antibody ( 1:50; a kind gift from Erich Buchner , RRID:AB_2314869 , ) , rabbit anti-dvGlut ( 1:1000; a kind gift from Aaron DiAntonio , RRID:AB_2314346 ) , rat anti-dilp2 ( 1:400; a kind gift from Pierre Leopold ) . Secondary antibodies were used at a dilution of 1:400 as follows: anti-rabbit Alexa Fluor 488 ( #A11008 , Life Technologies , RRID:AB_143165 ) , anti-mouse Alexa Fluor 488 ( #A11001 , Life Technologies , RRID:AB_141367 ) , anti-mouse Alexa Fluor 568 ( #A11004 , Life Technologies , RRID:AB_141371 ) , anti-rabbit Alexa Fluor 594 ( #A11037 , Life Technologies , RRID:AB_10561549 ) and anti-rat Alexa Fluor 633 ( #A21094 , Life Technologies , RRID:AB_10561523 ) . Confocal images were obtained on the Olympus Confocal FV1000 microscope ( Olympus ) with a 40x , 1 . 3 NA objective or with a 60x , 1 . 4 NA objective . Images were visualized using either the FV10-ASW 4 . 0 viewer ( Olympus ) or Fiji ( RRID:SCR_002285 ) ( Schindelin et al . , 2012 ) . Larval brains were dissected in hemolymph-like saline ( HL3 ) ( 70 mM NaCl , 5 mM KCl , 20 mM MgCl2 , 10 mM NaHCO3 , 5 mM trehalose , 115 mM sucrose , 5 mM HEPES , 1 . 5 mM Ca2+ , pH 7 . 2 ) , embedded in 0 . 2% low-melt agarose ( Invitrogen ) , and bathed in HL3 . GCaMP6m was used as the genetically encoded calcium sensor . ANF::GFP was expressed genetically to quantify vesicular release . Images were taken as a time series on an XY plane at an interval of 4 s using a 20x objective with an NA of 0 . 7 on an Olympus FV1000 inverted confocal microscope ( Olympus Corp . , Japan ) . For thermogenetic experiments , a heated stage was used to shift the temperature to 30°C to activate TrpA1 . For optogenetic stimulation , a 633-nm laser line was used for activation of CsChrimson while simultaneously acquiring images with the 488 nm laser line , and the images were acquired every 1 . 5 s . For channelrhodopsin activation 488 nm laser line was used . For optogenetic inhibition experiments a green 543-nm laser line was driven simultaneously with image acquisition using the 488 nm laser line . All live imaging experiments were performed with at least five independent brain preparations and the exact number of cells for each experiment are indicated in the figures . The raw images were extracted using Image J1 . 48 and regions of interest ( ROI ) selected using the Time Series Analyser plugin . ∆F/F was calculated using the formula ∆F/F = ( Ft-F0 ) /F0 , where Ft is the fluorescence at time t and F0 is baseline fluorescence corresponding to the average fluorescence over the first ten frames . Area under the curve was calculated from the point of stimulation which was considered as 0th second for stimulation up to 300 s using Microsoft Excel ( Microsoft ) and plotted using BoxPlotR ( Spitzer et al . , 2014 ) . ∆F/F Release was calculated as ( F0-Ft ) /F0 where Ft is the fluorescence at time t and F0 is baseline fluorescence corresponding to the average fluorescence over the first 10 frames . Area under the curve was calculated from 0s to 600s using Microsoft Excel ( Microsoft ) , and box plots were plotted using BoxPlotR ( Spitzer et al . , 2014 ) . For experiments with UAS-Shits , a heated microscopic stage was used . Larval brains on ND or PDD were dissected and stained for Dilp2 . Image acquisition was performed using similar acquisition settings and were processed using Fiji . Cells and background were marked using an ROI selection plugin and then the intensity across the stacks was measured . Brightest values were obtained using a Max function in Microsoft Excel . Intensity values were calculated for each cell in brains irrespective of ND or PDD , by subtracting background . To then obtain percentage release upon stimulation , intensity of cells were normalized to average intensity on the same diet . At least five independent brain preparations per genotype per condition were used and the exact number of cells are indicated in the figure . Flies were transferred every 2 hr to obtain larvae that were very tightly staged . Ring glands from larvae of the appropriate genotype and age were dissected in phosphate buffer saline prepared in double distilled water treated with diethyl pyrocarbonate ( Sigma ) . Each sample consisted of five Ring glands or 5 CNS and these were homogenized in 500 µl TRIzol per sample by vortexing immediately after dissection . At least three biological replicate samples were made for each genotype . After homogenization the sample was kept on ice and processed within 30 min or stored at −80°C until processing for up to 4 weeks . RNA was isolated by following manufacturer’s protocol for TRIzol ( Ambion , ThermoFischer Scientific ) . Purity of the isolated RNA was estimated by NanoDrop spectrophotometer ( Thermo Scientific ) and integrity was checked by running it on a 1% Tris-EDTA agarose gel . Approximately 100 ng of total RNA was used per sample for cDNA synthesis . DNAse treatment and first strand synthesis were performed as described previously ( Pathak et al . , 2015 ) . Quantitative real time PCRs ( qPCRs ) were performed in a total volume of 10 µl with Kapa SYBR Fast qPCR kit ( KAPA Biosystems , Wilmington , MA ) on an ABI 7500 fast machine operated with ABI 7500 software ( Applied Biosystems ) . Technical duplicates were performed for each qPCR reaction . A melt analysis was performed at the end of the reaction to ensure the specificity of the product . The fold change of gene expression in any experimental condition relative to wild-type was calculated as 2−△△Ct where △△Ct = ( Ct ( target gene ) –Ct ( rp49 ) ) Expt . - ( Ct ( target gene ) – Ct ( rp49 ) ) Control . rp49 was used as the internal control and the primer sequences used are provided in Supplementary file 2 . Primers for genes of the ecdysteroid biosynthetic pathway have been described previously ( Shimada-Niwa and Niwa , 2014 ) . The dimmed-LexA::p65 construct was created using recombineering techniques based on those of ( Warming et al . , 2005 ) . 5′ and 3′ homology arms comprising about 200 bases each of NLS::LexA and the HSP70 terminator were amplified from pBPnlsLexA::p65Uw , a gift of Gerald Rubin ( Pfeiffer et al . , 2010 ) and inserted into pSK+-rpsL-kana ( Wang et al . , 2009 ) to create a selectable generic landing cassette . Primers carrying gene-specific homology arms to target the cassette to the first coding exon of dimmed were used to PCR-amplify this cassette , and the resulting dimmed-flanked marker cassette was recombined into P[acman] BAC clone CH321-46B06 ( Venken et al . , 2009 ) ( obtained from Children’s Hospital Oakland Research Institute , Oakland , CA ) , replacing the coding portion of the first dimmed coding exon while leaving intact the endogenous 5′ UTR as well as the following introns and exons ( although these are presumably no longer transcribed because of the inserted terminator sequences ) . The landing-site cassette was then replaced via a second recombination with full-length LexA::p65-HSP70 , also amplified from pBPnlsLexA::p65Uw . The recombined regions of the BAC were sequence-verified , and the finished BAC was integrated into attP site VK00033 ( Venken et al . , 2006 ) on chromosome arm 3L by Genetic Services , Inc . ( Cambridge , MA ) . For the VGN6341-LexA , the 505 bp fragment of the dvGlut gene enhancer as in VGN6341-GAL4 ( Syed et al . , 2016 ) was PCR amplified from wild-type DNA , the sequence was verified and cloned into pDONR-221-p1-p5r ( Invitrogen ) to get an entry clone by performing the BP reaction . This entry clone along with the LexA entry clone and the destination vector were combined in an LR reaction to generate VGN6341-LexA . For the LexAop-ANF::GFP clone , the sequence of ANF:GFP was PCR amplified from DNA isolated from UAS-ANF::GFP flies and cloned into pDONR-221-p5-p2 ( Invitrogen ) to get an entry clone by performing the BP reaction . This entry clone along with the LexAop entry clone and the destination vector were combined in an LR reaction to generate LexAop-ANF::GFP . The entry clones pENTR L5-LexAp65-L2 ( 41437 ) and pENTR L1-13XLexAop2-R5 ( 41433 ) and the destination vector pDESTsvaw ( 32318 ) were obtained from Addgene . The BP and LR reactions were performed using the Multisite Gateway Pro cloning kit ( Invitrogen , 12537–102 ) following the half volume protocol described in ( Petersen and Stowers , 2011 ) . All primer sequences are listed in Supplementary file 2 . All statistical tests are mentioned in the figure legends and were performed using Origin 8 . 0 . Supplementary file 4 has all statistical tests and their p-values . | Insect larvae must feed voraciously to accumulate enough nutrients to tide them over the pupal stage of their lifecycle . Unlike larvae , pupae do not feed but instead use their stored energy reserves to fuel their metamorphosis into adults . To maximise their chances of survival , insect larvae must carefully time their transformation into pupae based on both the availability of nutrients in the environment and their own energy stores . The circuit of neurons within the larval nervous system that detects external nutrient levels , and then relays that information to the insect’s metabolic system , remains unknown . This circuit is also of interest because many animal species are thought to use it to slow down their metabolism during periods of food deprivation . Jayakumar et al . therefore set out to identify this circuit by studying how genetically modified fruit fly larvae transform into pupae when nutrients are in short supply . The experiments show that mutant larvae that lack a protein called IP3R struggle to turn into pupae when fed a diet deficient in proteins . IP3R proteins are ion channels that control the release of calcium ions from stores within the cells . Jayakumar et al . showed that food that is deficient in nutrients triggers some larval neurons to release a chemical called acetylcholine , which in turn activates receptors on certain other neurons that communicate using the signalling molecule glutamate . In normal insects , this causes the glutamate-producing neurons to release calcium ions through their IP3R channels . The calcium ions then activate a chain of events that ultimately causes other cells to produce a hormone called ecdysone , which drives the transformation from larva to pupa . In IP3R mutants , by contrast , the absence of calcium ion release keeps the insect in the larval stage . This circuit helps to explain how some insects and other animals are able to survive being deprived of food for extended periods . Further work will be required to understand how a lack of protein in the diet changes the signalling properties of cells in various parts of the circuit . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"neuroscience"
] | 2016 | Drosophila larval to pupal switch under nutrient stress requires IP3R/Ca2+ signalling in glutamatergic interneurons |
Fatty acyl reductases ( FARs ) are involved in the biosynthesis of fatty alcohols that serve a range of biological roles . Insects typically harbor numerous FAR gene family members . While some FARs are involved in pheromone biosynthesis , the biological significance of the large number of FARs in insect genomes remains unclear . Using bumble bee ( Bombini ) FAR expression analysis and functional characterization , hymenopteran FAR gene tree reconstruction , and inspection of transposable elements ( TEs ) in the genomic environment of FARs , we uncovered a massive expansion of the FAR gene family in Hymenoptera , presumably facilitated by TEs . The expansion occurred in the common ancestor of bumble bees and stingless bees ( Meliponini ) . We found that bumble bee FARs from the expanded FAR-A ortholog group contribute to the species-specific pheromone composition . Our results indicate that expansion and functional diversification of the FAR gene family played a key role in the evolution of pheromone communication in Hymenoptera .
Accumulation of DNA sequencing data is greatly outpacing our ability to experimentally assess the function of the sequenced genes , and most of these genes are expected to never be functionally characterized ( Koonin , 2005 ) . Important insights into the evolutionary processes shaping the genomes of individual species or lineages can be gathered from predictions of gene families , gene ortholog groups and gene function . However , direct experimental evidence of the function of gene family members is often unavailable or limited ( Lespinet et al . , 2002; Demuth et al . , 2006; Rispe et al . , 2008; Cortesi et al . , 2015; Niimura and Nei , 2006 ) . Gene duplication is hypothesized to be among the major genetic mechanisms of evolution ( Ohno , 1970; Zhang , 2003 ) . Although the most probable evolutionary fate of duplicated genes is the loss of one copy , the temporary redundancy accelerates gene sequence divergence and can result in gene subfunctionalization or neofunctionalization—acquisition of slightly different or completely novel functions in one copy of the gene ( Innan and Kondrashov , 2010; Lynch and Conery , 2000 ) . The alcohol-forming fatty acyl-CoA reductases ( FARs , EC 1 . 2 . 1 . 84 ) belong to a multigene family that underwent a series of gene duplications and subsequent gene losses , pseudogenizations and possibly functional diversification of some of the maintained copies , following the birth-and-death model of gene family evolution ( Eirín-López et al . , 2012 ) . FARs exhibit notable trends in gene numbers across organism lineages; there are very few FAR genes in fungi , vertebrates and non-insect invertebrates such as Caenorhabditis elegans , whereas plant and insect genomes typically harbor an extensive number of FAR gene family members ( Eirín-López et al . , 2012 ) . FARs are critical for production of primary fatty alcohols , which are naturally abundant fatty acid ( FA ) derivatives with a wide variety of biological roles . Fatty alcohols are precursors of waxes and other lipids that serve as surface-protective or hydrophobic coatings in plants , insects and other animals ( Wang et al . , 2017; Cheng and Russell , 2004; Jaspers et al . , 2014 ) ; precursors of energy-storing waxes ( Metz et al . , 2000; Teerawanichpan and Qiu , 2010a; Teerawanichpan and Qiu , 2012 ) ; and components of ether lipids abundant in the cell membranes of cardiac , nervous and immunological tissues ( Nagan and Zoeller , 2001 ) . Additionally , in some insect lineages , FARs were recruited for yet another task—biosynthesis of fatty alcohols that serve as pheromones or pheromone precursors . Moths ( Lepidoptera ) are the most well-studied model of insect pheromone biosynthesis and have been the subject of substantial research effort related to FARs . Variation in FAR enzymatic specificities is a source of sex pheromone signal diversity among moths in the genus Ostrinia ( Lassance et al . , 2013 ) and is also responsible for the distinct pheromone composition in two reproductively isolated races of the European corn borer Ostrinia nubilalis ( Lassance et al . , 2010 ) . Divergence in pheromone biosynthesis can potentially install or strengthen reproductive barriers , ultimately leading to speciation ( Smadja and Butlin , 2009 ) . However , the biological significance of a large number of insect FAR paralogs remains unclear , as all FARs implicated in moth and butterfly sex pheromone biosynthesis are restricted to a single clade , indicating that one FAR group was exclusively recruited for pheromone biosynthesis ( Lassance et al . , 2010; Liénard et al . , 2010; Liénard et al . , 2014; Antony et al . , 2009 ) . While more than 20 FARs have been experimentally characterized from 23 moth and butterfly ( Lepidoptera ) species ( Tupec et al . , 2017 ) , FARs from other insect orders have received far less attention . Single FAR genes have been isolated and experimentally characterized from Drosophila ( Diptera ) ( Jaspers et al . , 2014 ) , the European honey bee ( Hymenoptera ) ( Teerawanichpan et al . , 2010b ) and the scale insect Ericeus pela ( Hemiptera ) ( Hu et al . , 2018 ) . Our limited knowledge of FAR function prevents us from drawing inferences about the biological significance of the FAR gene family expansion in insects . Bumble bees ( Hymenoptera: Apidae ) are a convenient experimental model to study insect FAR evolution because the majority of bumble bee species produces fatty alcohols as species-specific components of male marking pheromones ( MMPs ) ( Ayasse and Jarau , 2014 ) , which are presumed to be biosynthesized by some of the numerous bumble bee FAR gene family members ( Buček et al . , 2016 ) . Bumble bee males employ MMPs to attract conspecific virgin queens ( Goulson , 2010 ) . In addition to fatty alcohols , MMPs generally contain other FA derivatives and terpenoid compounds . MMP fatty alcohols consist predominantly of saturated , mono-unsaturated and poly-unsaturated fatty alcohols with 16–18 carbon atoms ( Ayasse and Jarau , 2014 ) . Pheromone mixtures from three common European bumble bee species , B . terrestris , B . lucorum and B . lapidarius , are representative of the known MMP chemical diversity . Fatty alcohols are the major compounds in MMPs of B . lapidarius ( hexadecanol and Z9-hexadecenol ) and accompany electroantennogram-active compounds in B . terrestris ( hexadecanol , octadecatrienol , octadecenol ) and B . lucorum ( hexadecanol , Z9 , Z12-octadecadienol , Z9 , Z12 , Z15-octadecatrienol , octadecanol ) ( Bergström et al . , 1973; Kullenberg et al . , 1973; Kullenberg et al . , 1970; Urbanová et al . , 2001; Sobotník et al . , 2008; Luxová et al . , 2003; Zácek et al . , 2009 ) . In our previous investigation of the molecular basis of pheromone diversity in bumble bees , we found that the substrate specificities of fatty acyl desaturases ( FADs ) , enzymes presumably acting upstream of FARs in pheromone biosynthesis ( Tillman et al . , 1999 ) , are conserved across species despite differences in the compositions of their unsaturated FA-derived pheromone components ( Buček et al . , 2013 ) . These findings suggest that the substrate specificity of FADs contributes only partially to the species-specific composition of FA-derived MMPs ( Buček et al . , 2013 ) . The fatty alcohol content in bumble bee MMPs is therefore presumably co-determined by the enzymatic specificity of other pheromone biosynthetic steps , such as FA biosynthesis/transport or FA reduction . Analysis of the B . terrestris male labial gland ( LG ) transcriptome uncovered a remarkably high number of putative FAR paralogs , including apparently expressed pseudogenes , strongly indicating dynamic evolution of the FAR gene family and the contribution of FARs to the LG-localized MMP biosynthesis ( Buček et al . , 2016 ) . Here , we aimed to determine how the members of the large FAR gene family in the bumble bee lineage contribute to MMP biosynthesis . We sequenced B . lapidarius male LG and fat body ( FB ) transcriptomes and functionally characterized the FAR enzymes , along with FAR candidates from B . terrestris and B . lucorum , in a yeast expression system . We combined experimental information about FAR enzymatic specificities with quantitative information about bumble bee FAR expression patterns , as well as comprehensive gas chromatography ( GC ) analysis of MMPs and their FA precursors in the bumble bee male LG , with inference of the hymenopteran FAR gene tree . In addition , we investigated the content of transposable elements ( TEs ) in the genomic environment of FAR genes in genomes of two bumble bee species , B . terrestris and B . impatiens . We conclude that a dramatic TE-mediated expansion of the FAR gene family started in the common ancestor of the bumble bee ( Bombini: Bombus ) and stingless bee ( Meliponini ) lineages , which presumably shaped the pheromone communication in these lineages .
We sequenced , assembled and annotated male LG and FB transcriptomes of the bumble bee species B . lapidarius . The LG is the MMP-producing organ and is markedly enlarged in males , while the FB was used as a reference tissue not directly involved in MMP biosynthesis ( Žáček et al . , 2015 ) . Searches for FAR-coding transcripts in the LG and FB transcriptomes of B . lapidarius and the previously sequenced FB and LG transcriptomes of B . lucorum and B . terrestris ( Buček et al . , 2013; Prchalová et al . , 2016 ) yielded 12 , 26 and 16 expressed FAR homologs in B . lapidarius , B . terrestris and B . lucorum , respectively ( Figure 1—figure supplement 1 ) . To gain insight into the evolution of FAR gene family in Hymenoptera , we reconstructed a FAR gene tree using predicted FARs from species representing ants , Vespid wasps , parasitoid wasps and several bee lineages ( Figure 1 ) . We assigned the names FAR-A to FAR-K to 11 FAR ortholog groups that were retrieved as branches with high bootstrap support in the FAR gene tree . These ortholog groups typically encompass one or more FARs from each of the hymenopteran species used in the tree inference , with the exception of apparent species-specific FAR duplications or losses ( Figure 1 ) . Notably , we identified a massive expansion of the FAR-A ortholog group in the bumble bee and stingless bee ( subfamily Meliponini ) sister lineages ( Figure 1 , Figure 2 ) . The number of FAR homologs is inflated by a large number of predicted FAR genes and FAR transcripts with incomplete protein coding sequences lacking catalytically critical regions , such as the putative active site , NAD ( P ) + binding site or substrate binding site ( Figure 1 , Figure 1—source data 1 ) . The FAR gene tree also indicates expansion of the FAR-A ortholog group in the ant Camponotus floridanus and the mining bee Andrena vaga . However , this expansion is not present in two other ant species ( Acromyrmex echinatior and Harpegnathos saltator ) and two other mining bee species ( Andrenidae: Camptopoeum sacrum and Panurgus dentipes ) ( Figure 1 , Figure 2 ) . Several additional expansions of FAR families can be inferred from the FAR gene tree , including extensive FAR-B gene expansion in ants ( Formicoidea ) , along with many more lineage-specific FAR gene duplications and minor expansions . We also reconstructed a FAR gene tree encompassing FARs from three representatives of non-hymenopteran insect orders—the beetle Tribolium castaneum , the moth Bombyx mori and the fly Drosophila melanogaster ( Figure 1—figure supplement 2 ) . The only functionally characterized FAR from D . melanogaster—Waterproof ( NP_651652 . 2 ) , which is involved in the biosynthesis of a protective wax layer ( Jaspers et al . , 2014 ) —was placed in the FAR-J ortholog group ( Figure 1—figure supplement 2 ) . The FAR-G ortholog group includes a FAR gene from Apis mellifera with unclear biological function ( Teerawanichpan et al . , 2010b ) and a sex pheromone-biosynthetic FAR from B . mori ( Moto et al . , 2003 ) ( Figure 1—figure supplement 2 ) . In the gene tree , the majority of FAR ortholog groups contain predicted FARs from both hymenopteran and non-hymenopteran insect species , although the bootstrap support is <80% for some FAR groups . The presence of these FAR ortholog groups in representatives of Hymenoptera , Coleoptera , Lepidoptera and Diptera indicates that these groups are ancestral to holometabolous insects . FAR-D is the only FAR group that does not include any non-hymenopteran FARs from our dataset ( Figure 1—figure supplement 2 ) and thus presumably represents Hymenoptera-specific FAR gene family expansions . FAR-Ds , however , do not contain the complete set of three catalytically critical regions ( i . e . the putative active site , NAD ( P ) + binding site and substrate binding site ) and their enzymatic role is therefore unclear . To uncover the details of genetic organization of FAR-A genes , we attempted to analyze the shared synteny of FAR genes in the genomes of B . terrestris and A . mellifera ( Stolle et al . , 2011 ) . We aligned the A . melifera and B . terrestris genomes , but we were not able to identify any positional A . mellifera homologs of B . terrestris FAR-A genes ( data not shown ) . While the majority of FAR genes belonging to the non-FAR-A gene ortholog group localize to the B . terrestris genome assembled to linkage groups , most of the B . terrestris FAR-A genes localize to unlinked short scaffolds ( Supplementary file 1 ) . Some of the FAR-A genes in the B . terrestris genome are arranged in clusters ( Figure 1—figure supplement 3 ) . Unfortunately , the genome assembly of B . impatiens is not mapped to chromosomes to allow similar analysis . A genome assembly consisting of short scaffolds is often indicative of a repetitive structure in the assembled genomic region . Our analysis of the distribution of TEs in the vicinity of FAR genes in the B . terrestris and B . impatiens genomes confirmed that TEs are significantly enriched around FAR-A genes compared to the genome-wide average around randomly selected genes ( Figure 3AC; B . terrestris: p < 0 . 0001 , B . impatiens: p < 0 . 0001 ) . On average , more than 50% of the 10 kb regions surrounding FAR-A genes are formed by TEs , compared to an average of 10% around randomly selected B . terrestris genes . In contrast , the densities of TEs in the vicinity of FAR genes not belonging to the FAR-A group do not differ from the genome-wide average ( Figure 3AC; B . terrestris: p = 0 . 793; B . impatiens: p = 0 . 880 ) . Also , there is a statistically significant difference in TE densities between FAR-A genes and non-FAR-A genes ( Man-Whitney U-test: B . terrestris: U = 167 , p < 0 . 0001; B . impatiens: U = 78 , p = 0 . 0002 ) . Although all major known TE families are statistically enriched in the neighborhood of the FAR-A genes ( Figure 3BD ) , the Class I comprising retroid elements contributes considerably to the elevated repeat content around FAR-A genes ( Figure 3 , Figure 3—source data 1 ) . We selected 10 promising MMP-biosynthetic FAR candidates that were ( 1 ) among the 100 most abundant transcripts in the LG and were substantially more abundant in LG than in FB based on RNA-Seq-derived normalized expression values ( Figure 1—figure supplement 1 and ref . ( Buček et al . , 2016 ) ) and ( 2 ) included all the predicted catalytically critical regions of FARs—the putative active site , NAD ( P ) + binding site and substrate binding site in the protein coding sequence ( Figure 1—source data 1 ) . By employing reverse transcription-quantitative PCR ( RT-qPCR ) on an expanded set of bumble bee tissues , we confirmed that the FAR candidates follow a general trend of overexpression in male LG compared to FB , flight muscle and gut ( all from male bumble bees ) and virgin queen LG ( Figure 4 , Figure 4—figure supplement 1 , p < 0 . 05 ) . Notably , B . lapidarius FAR-A1 ( BlapFAR-A1 ) and B . terrestris FAR-J ( BterFAR-J ) transcripts are also abundant in virgin queen LG , where they are expressed at levels comparable to those in male LG ( Figure 4—figure supplement 1 ) . We performed a detailed analysis of transesterifiable fatty acyls ( free FAs and fatty acyls bound in esters ) and fatty alcohols in LGs and FBs of 3-day-old B . lapidarius , B . terrestris and B . lucorum males to identify the products and predict the potential FAR substrates in the male LG . In the LGs , we detected 4 , 14 and 19 individual fatty alcohol compounds in B . lapidarius , B . lucorum and B . terrestris , respectively ( Figure 5 ) . A limited number of fatty alcohols ( mainly 16:OH , Z9 , Z12-18:OH and Z9 , Z12 , Z15-18:OH ) also were detected in FBs of B . lucorum and B . terrestris , but at substantially lower abundance than in LGs ( Figure 5—source data 1 ) . To assess the apparent in vivo specificity of all FARs expressed in LGs and FBs , we calculated the fatty alcohol ratios ( see Equation 1 in Materials and methods ) , that is the ratios of the quantity of particular fatty alcohol to the quantity of its hypothetical fatty acyl precursors ( Figure 5—figure supplement 1 ) . These ratios are greater than 50% for most of the fatty alcohols in LGs and even approach 100% for some of the monounsaturated >C20 fatty alcohols , suggesting high overall conversion rates of acyl substrates to alcohols . The full-length coding regions of the FAR candidates were isolated from male LG cDNA libraries using gene-specific PCR primers ( Supplementary file 2 ) . In general , the FAR candidates share high to very high protein sequence similarity within each ortholog group ( Figure 1—figure supplement 4 , Figure 1—figure supplement 5 ) . FARs from three bumble bee species belonging to the FAR-J ortholog group are nearly identical , sharing 97 . 2–99 . 7% protein sequence identity; BlucFAR-A1 and BterFAR-A1 share 99 . 4% protein sequence identity with each other and 60 . 9–61 . 1% with BlapFAR-A1 . BlucFAR-A2 and BterFAR-A2 share 94 . 8% protein sequence identity ( Figure 1—figure supplement 4 ) . BlucFAR-J was not cloned because of its very high similarity to BterFAR-J ( 99 . 7% sequence identity , two amino acid differences ) . We cloned two versions of BlapFAR-A1: one that was custom-synthesized based on the predicted full-length coding sequence assembled from RNA-Seq data and one called BlapFAR-A1-short that we consistently PCR-amplified from B . lapidarius male LG cDNA . BlapFAR-A1-short has an in-frame internal 66 bp deletion in the coding region that does not disrupt the predicted active site , putative NAD ( P ) + binding site or putative substrate binding site ( Figure 6—figure supplement 1C ) . Using RT-qPCR with specific primers for each variant , we confirmed that both BlapFAR-A1 and BlapFAR-A1-short are expressed in the B . lapidarius male LG and virgin queen LG ( Figure 4—figure supplement 2 ) . To test whether the MMP-biosynthetic FAR candidates code for enzymes with fatty acyl reductase activity and to uncover their substrate specificities , we cloned the candidate FAR coding regions into yeast expression plasmids , heterologously expressed the FARs in Saccharomyces cerevisiae and assayed the fatty alcohol production by GC ( Figure 6—figure supplement 2 , Figure 6—figure supplement 3 ) . His-tagged FARs were detected in all yeast strains transformed with plasmids bearing FARs ( Figure 6—figure supplement 4 , Figure 6—figure supplement 1A ) , while no His-tagged proteins were detected in the negative control ( yeasts transformed with an empty plasmid ) . In addition to the major protein bands corresponding to the theoretical FAR molecular weight , we typically observed protein bands with lower and/or higher molecular weight ( Figure 6—figure supplement 4 ) . The synthetic BlucFAR-A1-opt and BlucFAR-A2-opt coding regions with codon usage optimized for S . cerevisiae showed a single major western blot signal corresponding to the position of the predicted full-length protein ( Figure 6—figure supplement 4 ) . The shortened heterologously expressed proteins thus presumably represent incompletely transcribed versions of full-length FARs resulting from ribosome stalling ( Angov , 2011 ) , while the higher molecular weight bands might correspond to aggregates of full-length and incompletely translated FARs . Because the codon-optimized BlucFAR-A1-opt and BlucFAR-A2-opt exhibit the same overall specificity in yeast expression system as the respective non-codon-optimized FARs ( Figure 6—figure supplement 5 ) , we employed non-codon-optimized FARs for further functional characterization . We only used the codon-optimized versions of BlucFAR-A1 and BlucFAR-A2 in experiments with exogenously supplemented substrates to increase the possibility of product detection , as the optimized FARs produce overall higher quantities of fatty alcohols ( Figure 6—figure supplement 5 , Figure 6—source data 1 ) . Characterization of FAR enzymatic activities involved identification of numerous individual FA derivatives , denoted using the length of the carbon chain ( e . g . 20: for 20-carbon chain ) , the number of double bonds ( either as the position/configuration of the double bond ( s ) , if known , for example Z9 , or by ‘:X’ , for example :1 and :2 for monounsaturated and diunsaturated FAs , respectively ) and the C1 moiety ( COOH for acid , OH for alcohol , Me for methyl ester , CoA for CoA-thioester ) . Functional characterization of FARs from B . terrestris and B . lucorum in yeast indicated that saturated C16 to C26 fatty alcohols are produced by both BterFAR-A1 and BlucFAR-A1 and BterFAR-J enzymes ( Figure 6A , Figure 6—figure supplement 3 ) ; Bter/BlucFAR-A1 prefers C22 substrates , whereas BterFAR-J has an optimal substrate preference slightly shifted to C24 . Unlike any of the other characterized FARs , BterFAR-A1 and BlucFAR-A1 are also capable of reducing supplemented monounsaturated Z15-20: acyl to the corresponding alcohol ( Figure 6B , Figure 6—figure supplement 6C ) . Both BterFAR-A2 and BlucFAR-A2 reduce only 16: and 18: acyls ( Figure 6A , Figure 6—figure supplement 7 ) . Characterization of B . lapidarius FARs showed that BlapFAR-A1 , in contrast to BterFAR-A1 and BlucFAR-A1 , produces Z9-16:OH and Z9-18:OH ( Figure 6A , Figure 6—figure supplement 2 ) . BlapFAR-A4 produces 16:OH and Z9-16:OH , together with lower quantities of 14:OH and Z9-18:OH ( Figure 6A , Figure 6—figure supplement 8A ) . BlapFAR-A5 produces 16:OH as a major product and lower amounts of 14:OH , Z9-16:OH , 18:OH and Z9-18:OH ( Figure 6A , Figure 6—figure supplement 8A ) . In addition , both BlapFAR-A1 and BlapFAR-A4 are capable of reducing supplemented polyunsaturated fatty acyls ( Z9 , Z12-18: and Z9 , Z12 , Z15-18: ) to their respective alcohols ( Figure 6B , Figure 6—figure supplement 6AB ) . Similarly to BterFAR-J , BlapFAR-J also reduces saturated C16 to C26 acyls ( Figure 6A , Figure 6—figure supplement 3 ) . No fatty alcohols were detected in the negative control , that is the yeasts transformed with empty plasmid ( Figure 6—figure supplement 2 , Figure 6—figure supplement 8B ) . The total quantities of fatty alcohols accumulated in FAR-expressing yeasts range from few milligrams to tens of milligrams per liter of culture ( Figure 6—figure supplement 8B ) , the exceptions being BterFAR-A2 and BlucFAR-A2 which both produce sub-milligram quantities of fatty alcohols . We did not detect the formation of fatty aldehydes in any of the yeast cultures ( data not shown ) , confirming that the studied FARs are strictly alcohol-forming fatty acyl-CoA reductases . In contrast to BlapFAR-A1 , BlapFAR-A1-short does not produce detectable amounts of any fatty alcohol ( Figure 6—figure supplement 1B ) , suggesting that the missing 22-amino acid region ( Figure 6—figure supplement 1C ) is necessary for the retention of FAR activity . Overall , the FAR specificities determined in yeast correlate well with the composition of LG fatty alcohols and fatty acyls ( Figure 5 ) . The exceptions are Z9 , Z12-18:OH and Z9 , Z12 , Z15-18:OH , as none of the studied FARs from B . lucorum or B . terrestris reduce the corresponding acyls .
In the present work , we substantially broadened our limited knowledge of the function of FARs in Hymenoptera , one of the largest insect orders . The experimentally determined reductase specificity of FARs that are abundantly expressed in bumble bee male LGs is consistent with their role in MMP biosynthesis . The majority of these MMP-biosynthetic FARs belong to the FAR-A ortholog group . Reconstruction of the FAR gene family evolution indicates the onset of FAR-A gene expansion in the common ancestor of bumble bee and stingless bee lineages after their divergence from honey bee lineage . We therefore propose that the strategy of bumble bees and stingless bees to employ fatty alcohols as marking pheromones was shaped by FAR gene family expansion . Our analysis of TE distribution in the B . terrestris genome indicates that TEs enriched in the vicinity of FAR-A genes might have substantially contributed to the dramatic expansion of the FAR-A gene group . In the future , the increasing availability of annotated Hymenopteran genome assemblies should enable us to more precisely delineate the taxonomic extent and evolutionary timing of the massive FAR gene family expansion and assess in detail the role of TEs in the process .
Specimens of Bombus lucorum and Bombus lapidarius were obtained from laboratory colonies established from naturally overwintering bumble bee queens . The Bombus terrestris specimens originated from laboratory colonies obtained from a bumble bee rearing facility in Troubsko , Czech Republic . LG and FB samples used for transcriptome sequencing were prepared from 3-day-old B . lapidarius males by pooling tissues from three specimens from the same colony . The cephalic part of the LG and a section of the abdominal peripheral FB were dissected , transferred immediately to TRIzol ( Invitrogen ) , then flash-frozen at −80°C and stored at this temperature prior to RNA isolation . For cloning of FARs and RT-qPCR analysis of tissue-specific gene expression , RNA was isolated from individual bumble bee tissues by guanidinium thiocyanate-phenol-chloroform extraction followed by RQ1 DNase ( Promega ) treatment and RNA purification using the RNeasy Mini Kit ( Qiagen ) . The tissue sample for RNA isolation from virgin queen LGs consisted of pooled glands from two specimens . A nanodrop ND-1000 spectrophotometer ( Thermo Fisher ) was employed to determine the isolated RNA concentration . The obtained RNA was kept at −80°C until further use . The cDNA libraries of LGs from 3-day-old bumble bee males were constructed from 0 . 50 μg total RNA using the SMART cDNA Library Construction Kit ( Clontech ) with either Superscript III ( Invitrogen ) or M-MuLV ( New England Biolabs ) reverse transcriptase . The male LG and FB transcriptomes of B . lapidarius were sequenced and assembled as previously described for the transcriptomes of male LGs and FBs of B . lucorum and B . terrestris ( Buček et al . , 2013; Prchalová et al . , 2016 ) . Briefly , total RNA was isolated from the LGs and FBs of three 3-day-old B . lapidarius males and pooled into a one FB and one LG sample . Total RNA ( 5 µg ) from each of the samples was used as starting material . Random primed cDNA libraries were prepared using poly ( A ) + enriched mRNA and standard Illumina TrueSeq protocols ( Illumina ) . The resulting cDNA was fragmented to an average of 150 bp . RNA-Seq was carried out by Fasteris ( Fasteris ) and was performed using an Illumina HiSeq 2500 Sequencing System . Quality control , including filtering high-quality reads based on the fastq score and trimming the read lengths , was carried out using CLC Genomics Workbench software v . 7 . 0 . 1 ( http://www . clcbio . com ) . The complete transcriptome libraries were assembled de novo using CLC Genomics Workbench software . FAR expression values were calculated by mapping Illumina reads against the predicted coding regions of FAR sequences using bowtie2 v2 . 2 . 6 ( Langmead and Salzberg , 2012 ) and counting the mapped raw reads using ht-seq v0 . 9 . 1 ( Anders et al . , 2015 ) . The raw read counts were normalized for the FAR coding region length and the total number of reads in the sequenced library , yielding reads per kilobase of transcript per million mapped reads ( RPKM ) values ( Mortazavi et al . , 2008 ) . A constant value of 1 was added to each RPKM value and subsequently log2-transformed and visualized as heatmaps using the ggplot2 package in R ( Core Team R , 2016 ) . Complete short read ( Illumina HiSeq2500 ) data for FB and LG libraries from B . lapidarius and previously sequenced B . lucorum ( Buček et al . , 2013 ) were deposited in the Sequence Read Archive ( https://www . ncbi . nlm . nih . gov/sra ) with BioSample accession numbers SAMN08625119 , SAMN08625120 , SAMN08625121 , and SAMN08625122 under BioProject ID PRJNA436452 . The FARs of B . lucorum and B . lapidarius were predicted based on Blast2GO transcriptome annotation and their high protein sequence similarity to previously characterized FARs from the European honey bee Apis mellifera ( Teerawanichpan et al . , 2010b ) and the silk moth Bombyx mori ( Moto et al . , 2003 ) . FAR sequences from species across the Hymenoptera phylogeny were retrieved from publicly available resources . When available , genome assembly-derived FAR sequences were used instead of transcriptome assembly-derived sequences to minimize the impact of misidentification of alternative splice variants as distinct genes on inference of FAR gene expansion . However , in the Euglossa dilemma genome-derived proteome , we failed to identify a FAR-A gene , but we did detect FAR-As in the transcriptome sequencing-derived dataset . We therefore used the Euglossa dilemma transcriptome rather than genome for downstream analyses . FARs from annotated genomes ( Bombus impatiens ( Sadd et al . , 2015 ) , Bombus terrestris ( Sadd et al . , 2015 ) , Apis mellifera ( Janoušek et al . , 2016 ) , Camponotus floridanus ( Bonasio et al . , 2010 ) , Acromyrmex echinatior ( Nygaard et al . , 2011 ) , Harpegnathos saltator ( Bonasio et al . , 2010 ) , Nasonia vitripenis ( Werren et al . , 2010 ) , Polistes canadensis ( Patalano et al . , 2015 ) , Dufourea novaeangliae ( Kapheim et al . , 2015 ) , Ceratina calcarata ( Rehan et al . , 2016 ) , Melipona quadrifasciata ( Woodard et al . , 2011 ) and Megachile rotundata ( Woodard et al . , 2011 ) ) of other hymenopteran species were retrieved by blastp ( Altschul et al . , 1990 ) searches ( E-value cutoff 10−5 ) of the species-specific NCBI RefSeq protein database or UniProt protein database using predicted protein sequences of B . lucorum , B . lapidarius and B . terrestris FARs ( accessed February 2017 ) . An additional round of blastp searches using FARs found in the first blastp search round did not yield any additional significant ( E-value <10−5 ) blastp hits , indicating that all FAR homologs were found in the first round of blastp searches ( data not shown ) . FARs from non-annotated transcriptomes ( Bombus rupestris ( Peters et al . , 2017 ) , Tetragonula carbonaria ( Peters et al . , 2017 ) , Euglossa dilemma ( Peters et al . , 2017 ) , Epeolus variegatus ( Peters et al . , 2017 ) , Colletes cunicularius ( Peters et al . , 2017 ) , Melitta haemorrhoidalis ( Peters et al . , 2017 ) , Camptopoeum sacrum ( Peters et al . , 2017 ) , Panurgus dentipes ( Peters et al . , 2017 ) , and Andrena vaga ( Peters et al . , 2017 ) ) were retrieved via local tblastn search ( E-value cutoff 10−5 ) of the publicly available contig sequences ( BioProjects PRJNA252240 , PRJNA252285 , PRJNA252310 , PRJNA252262 , PRJNA252324 , PRJNA252208 , PRJNA252153 , PRJNA252205 , and PRJNA252325 ) using Bombus FARs as a query . The longest translated ORFs were used as a query in tblastn searches against NCBI non-redundant nucleotide database ( nr/nt ) and ORFs not yielding highly scoring blast hits annotated as FARs were rejected . For FARs with multiple splice variants predicted from the genome sequence , only the longest protein was used for gene tree reconstruction . In the case of transcriptome assembly-derived FARs , we predicted as alternative splice variants those FAR transcripts that were truncated but otherwise identical in sequence to another FAR transcript in the transcriptome . These FARs were not included in gene tree reconstruction . The active site , conserved Rossmann fold NAD ( P ) + binding domain ( NABD ) ( Rossmann et al . , 1974 ) and a putative substrate binding site in FAR coding sequences were predicted using Batch conserved domain search ( Marchler-Bauer et al . , 2015 ) . The matrix of protein identities was calculated using Clustal Omega with default parameters ( https://www . ebi . ac . uk/Tools/msa/clustalo/ accessed February 2018 ) . The protein sequences of predicted hymenopteran FARs were aligned using mafft v7 . 305 . The unrooted gene tree was inferred in IQTREE v1 . 5 . 5 with 1000 ultrafast bootstrap approximation replicates ( Minh et al . , 2013 ) , and with a model of amino acid substitution determined by ModelFinder ( Kalyaanamoorthy et al . , 2017 ) implemented in IQTREE . The tree was visualized and annotated using the ggtree package ( Yu et al . , 2016 ) in R programming language . The genomes of A . mellifera and B . terrestris were aligned using MAUVE 2 . 4 . 0 ( Darling et al . , 2004 ) . The genomic position of predicted B . terrestris FAR genes was visually inspected using the NCBI Graphical sequence viewer ( accessed January 2018 at Nucleotide Entrez Database ) . TE-enrichment analysis in the vicinity of FAR genes in the B . terrestris and B . impatiens genomes was carried out to explore the impact of TEs in extensive expansion of FAR-A genes . TE annotation using the NCBI RepeatMasker provided insufficient detail . Thus , TE annotations for B . terrestris and B . impatiens were obtained from the Human Genome Sequencing Center FTP ( ftp://ftp . hgsc . bcm . edu/; accessed October 2018 ) . We used the approach described by Sadd et al . to identify different types of TEs ( Sadd et al . , 2015 ) . For B . terrestris , genome version 2 . 1 was used , and for B . impatiens version 1 . 0 was used . TE density around FAR genes was calculated 10 kb upstream and downstream of each FAR gene , separately for FAR-A genes and non-FAR-A genes . Statistical significance was assessed by permutation test . We compared FAR-A/non-FAR-A gene set average TE density to the null distribution of the average TE densities around B . terrestris and B . impatiens genes built from 10 , 000 randomly sampled gene sets with size corresponding to that of the FAR-A/non-FAR-A gene set from the publicly available RefSeq gene set downloaded for the respective genome versions from the NCBI FTP . TE densities were analyzed for a pooled set of all TEs and separately for each TE class and major TE family ( Class I: LINE , LTR , LARD , DIRS; Class II: DNA , TIR , MITE , TRIM , Maveric , Helitron ) using custom shell scripts and bedtools ( Supplementary file 3 ) , a suite of Unix genomic tools ( Quinlan and Hall , 2010 ) . R programming language was used for statistical analysis . First-strand cDNA was synthesized from 0 . 30 μg total RNA using oligo ( dT ) 12-18 primers and Superscript III reverse transcriptase . The resulting cDNA samples were diluted 5-fold with water prior to RT-qPCR . The primers used for the assay ( Supplementary file 2 ) were designed with Primer-BLAST ( https://www . ncbi . nlm . nih . gov/tools/primer-blast/ ) ( Ye et al . , 2012 ) and tested for amplification efficiency and specificity by employing amplicon melting curve analysis on dilution series of pooled cDNAs from each species . The reaction mixtures were prepared in a total volume of 20 μL consisting of 2 μL sample and 500 nM of each primer using LightCycler 480 SYBR Green I Master kit ( Roche ) . The reactions were run in technical duplicates for each sample . RT-qPCR was performed on a LightCycler 480 Instrument II ( Roche ) in 96-well plates under the following conditions: 95°C for 60 s , then 45 cycles of 95°C for 30 s , 55°C for 30 s and 72°C for 30 s followed by a final step at 72°C for 2 min . The acquired data were processed with LightCycler 480 Software 1 . 5 ( Roche ) and further analyzed with MS Excel ( Microsoft Corporation ) . FAR transcript abundances were normalized to the reference genes phospholipase A2 ( PLA2 ) and elongation factor 1α ( eEF1α ) as described ( Hornáková et al . , 2010 ) . The predicted coding regions of FARs from B . lucorum , B . lapidarius and B . terrestris were amplified by PCR from LG cDNA libraries using gene-specific primers ( Supplementary file 2 ) and Phusion HF DNA polymerase ( New England Biolabs ) . Parts of the full-length coding sequence of BlapFAR-A5 were obtained by RACE procedure using SMART cDNA Library Construction Kit . The PCR-amplified sequences containing the 5' and 3' ends of BlapFAR-A5 were inserted into pCRII-TOPO vector using TOPO TA Cloning kit ( Invitrogen ) and sequenced by Sanger method . The resulting sequences overlapped with contig sequences retrieved from the B . lapidarius transcriptome . The full-length BlapFAR-A5-coding region was subsequently isolated using gene-specific PCR primers . The sequence of BlapFAR-A1 and yeast codon-optimized sequences of BlucFAR-A1-opt and BlucFAR-A2-opt were obtained by custom gene synthesis ( GenScript ) ; see Supplementary file 2 for synthetic sequences . The individual FAR coding regions were then inserted into linearized pYEXTHS-BN vector ( Holz et al . , 2002 ) using the following restriction sites: Bter/BlucFAR-A1 and BlapFAR-J at SphI-NotI sites; Bter/BlucFAR2 , BlapFAR-A1 , BlapFAR-A1-short and BlapFAR-A5 at BamHI-NotI sites; and BlucFAR-A1-opt/FAR-A2-opt and BlapFAR-A4 at BamHI-EcoRI sites . In the case of BterFAR-J , the Taq DNA polymerase ( New England Biolabs ) -amplified sequence was first inserted into pCRII-TOPO vector and then subcloned into pYEXTHS-BN via BamHI-EcoRI sites using the In-Fusion HD Cloning kit ( Clontech ) . The resulting vectors containing FAR sequences N-terminally fused with 6×His tag were subsequently transformed into E . coli DH5α cells ( Invitrogen ) . The plasmids were isolated from bacteria with Zyppy Plasmid Miniprep kit ( Zymo Research ) and Sanger sequenced prior to transformation into yeast . The protein-coding sequences of all studied FARs were deposited to GenBank ( see Key Resources Table for accession numbers ) . Expression vectors carrying FAR-coding sequences were transformed into Saccharomyces cerevisiae strain BY4741 ( MATa his3Δ1 leu2Δ0 met15Δ0 ura3Δ0 ) ( Brachmann et al . , 1998 ) using S . c . EasyComp Transformation Kit ( Invitrogen ) . To test FAR specificity , yeasts were cultured for 3 days in 20 mL synthetic complete medium lacking uracil ( SC−U ) supplemented with 0 . 5 mM Cu2+ ( inducer of heterologous gene expression ) , 0 . 2% peptone and 0 . 1% yeast extract . The yeast cultures were then washed with water and the cell pellets lyophilized before proceeding with lipid extraction . FAR specificities were determined with the FARs acting on natural substrates present in yeast cells and with individual fatty acyls added to the cultivation media , with the respective fatty alcohols present in the LGs of studied bumble bees . For this purpose , yeast cultures were supplemented with the following fatty acyls: 0 . 1 mM Z9 , Z12-18:COOH ( linoleic acid , Sigma-Aldrich ) , Z9 , Z12 , Z15-18:COOH ( α-linolenic acid , Sigma-Aldrich ) or Z15-20:Me solubilized with 0 . 05% tergitol . We chose Z15-20: as a representative monounsaturated >C20 fatty acyl substrate because Z15-20:OH is the most abundant monounsaturated fatty alcohol in B . terrestris LG ( Figure 5 ) . The level of heterologous expression of bumble bee FARs was assayed by western blot analysis of the whole-cell extracts ( obtained via sonication ) using anti-6×His tag antibody-HRP conjugate ( Sigma-Aldrich ) and SuperSignal West Femto Maximum Sensitivity Substrate kit ( Thermo Fisher Scientific ) . Lipids were extracted from bumble bee tissue samples under vigorous shaking using a 1:1 mixture of CH2Cl2/MeOH , followed by addition of an equal amount of hexane and sonication . The extracts were kept at −20°C prior to GC analysis . Base-catalyzed transesterification was performed as described previously ( Matousková et al . , 2008 ) with modifications: the sample was shaken vigorously with 1 . 2 mL CH2Cl2/MeOH 2:1 and glass beads ( 0 . 5 mm ) for 1 hr . After brief centrifugation to remove particulate debris , 1 mL supernatant was evaporated under nitrogen , and the residue was dissolved using 0 . 2 mL 0 . 5 M KOH in methanol . The mixture was shaken for 0 . 5 hr and then neutralized by adding 0 . 2 mL solution of Na2HPO4 and KH2PO4 ( 0 . 25 M each ) and 35 μL 4 M HCl . The obtained FAMEs were extracted with 600 μL hexane and analyzed by gas chromatography . For quantification purposes , either 1-bromodecane ( 10:Br ) or 1-bromoeicosane ( 20:Br ) were added to the extracts as internal standards . Standards of Z9 , Z12 , Z15-18:OH and Z15-20:OH were prepared from their corresponding acids/FAMEs by reduction with LiAlH4 . The Z9-18:Me standard was prepared by reacting oleoyl chloride with methanol . Other FAME and fatty alcohol standards were obtained from Nu-Chek Prep and Sigma-Aldrich . The FA-derived compounds in extracts were identified based on the comparison of their retention times with the standards and comparison of measured MS spectra with those from spectral libraries . Double bond positions were assigned after derivatization with dimethyl disulfide ( Carlson et al . , 1989 ) . The fatty alcohol ratio is calculated according to Equation 1 , ( 1 ) Fattyalcoholratio=n ( FattyalcoholX ) n ( FattyalcoholX ) +n ( FattyacylX ) 100%where n is the amount in moles and X is the fatty chain structure of certain length , degree of unsaturation and double bond position/configuration . The fatty acyl term in Equation 1 stands for all transesterifiable fatty acyls present in the sample , for example free FAs , fatty acyl-CoAs , and triacylglycerols , containing the same fatty chain structure . The fatty alcohol ratio thus represents the hypothetical degree of conversion of total fatty acyls ( as if they were available as FAR substrates , that is fatty acyl-CoAs ) to the respective fatty alcohol and reflects the apparent FAR specificity in the investigated bumble bee tissue or yeast cell . The Z15-20:CoA precursor , methyl Z15-eicosenoate ( Z15-20:Me , 4 ) , was synthesized by a new and efficient four-step procedure , starting from inexpensive and easily available cyclopentadecanone . The C1–C15 part of the molecule was obtained by Baeyer-Villiger oxidation of cyclopentadecanone , followed by subsequent methanolysis of the resulting lactone 1 and Swern oxidation of the terminal alcohol group of 2; the C16–C20 fragment was then connected to the aldehyde 3 by Wittig olefination . All reactions were conducted in flame- or oven-dried glassware under an atmosphere of dry nitrogen . THF , CH2Cl2 and MeOH were dried following standard methods under a nitrogen or argon atmosphere . Petroleum ether ( PE , 40–65°C boiling range ) was used for chromatographic separations . TLC plates ( silica gel with fluorescent indicator 254 nm , Fluka or Macherey-Nagel ) were used for reaction monitoring . Flash column chromatographic separations were performed on silica gel 60 ( 230–400 mesh , Merck or Acros ) . IR spectra were taken on an ALPHA spectrometer ( Bruker ) as neat samples using an ATR device . 1H and 13C NMR spectra were recorded in CDCl3 on an AV III 400 HD spectrometer ( Bruker ) equipped with a cryo-probe or an AV III 400 spectrometer ( Bruker ) equipped with an inverse broad-band probe at 400 MHz for 1H and 100 MHz for 13C . 1H NMR chemical shifts were provided in ppm using TMS as external standard; 13C NMR chemical shifts were referenced against the residual solvent peak . The connectivity was determined by 1H-1H COSY experiments . GC-MS ( EI ) measurements were performed on an Agilent 5975B MSD coupled to a 6890N gas chromatograph ( Agilent Technologies ) . High-resolution MS ( HRMS ) spectra were measured on a Q-Tof micro spectrometer ( resolution 100000 ( ESI ) , Waters ) or GCT Premier orthogonal acceleration TOF mass spectrometer ( EI and CI , Waters ) . All lipid quantifications in yeast and in bumble bee LGs and FBs , and RT-qPCR transcript quantifications in bumble bee tissues were performed using three biological replicates ( in addition , technical duplicates were used for RT-qPCR ) ; the number of biological replicates is indicated as N in figures and tables . The results are reported as mean value ±S . D . Significant differences were determined by one-way analysis of variance ( ANOVA ) followed by post-hoc Tukey’s honestly significant difference ( HSD ) test or by a two-tailed t-test as indicated in figures and in Results section . | Many insects release chemical signals , known as sex pheromones , to attract mates over long distances . The pheromones of male bumble bees , for example , contain chemicals called fatty alcohols . Each species of bumble bee releases a different blend of these chemicals , and even species that are closely related may produce very different ‘cocktails’ of pheromones . The enzymes that make fatty alcohols are called fatty acyl reductases ( or FARs for short ) . Any change to a gene that encodes one of these enzymes could change the final mix of pheromones produced . This in turn could have far-reaching effects for the insect , and in particular its mating success . Over time , these changes could even result in new species . Yet no one has previously looked into how the genes for FAR enzymes have evolved in bumble bees , or how these genes might have shaped the evolution of this important group of insects . Tupec , Buček et al . set out to learn what genetic changes led the males of three common species of bumble bees to make dramatically different mixes of pheromones . Comparing the genetic information of bumble bees with that of other insects showed that the bumble bees and their close relatives , stingless bees , often had extra copies of genes for certain FAR enzymes . Inserting some of these genes into yeast cells caused the yeast to make the correct blend of bumble bee pheromones , confirming that these genes did indeed produce the mixture of chemicals in these signals . Further , detailed analysis of the bumble bees’ genetic information revealed many genetic sequences , called transposable elements , close to the genes for the FAR enzymes . Transposable elements make the genetic material less stable; they can be ‘cut’ or ‘copied and pasted’ in multiple locations and often cause other genes to be duplicated or lost . Tupec et al . concluded that these transposable elements led to a dramatic increase in the number of genes for FAR enzymes in a common ancestor of bumble bees and stingless bees , ultimately allowing a new pheromone ‘language’ to evolve in these insects . These results add to our understanding of the chemical and genetic events that influence what chemicals insects use to communicate with each other . Tupec , Buček et al . also hope that a better knowledge of the enzymes that insects use to make pheromones could have wide applications . Other insects – including pest moths – use a similar mixture of fatty alcohols as pheromones . Artificially produced enzymes , such as FAR enzymes , could thus be used to mass-produce pheromones that may control insect pests . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"evolutionary",
"biology"
] | 2019 | Expansion of the fatty acyl reductase gene family shaped pheromone communication in Hymenoptera |
Communication between neighboring tissues plays a central role in guiding organ morphogenesis . During heart tube assembly , interactions with the adjacent endoderm control the medial movement of cardiomyocytes , a process referred to as cardiac fusion . However , the molecular underpinnings of this endodermal-myocardial relationship remain unclear . Here , we show an essential role for platelet-derived growth factor receptor alpha ( Pdgfra ) in directing cardiac fusion . Mutation of pdgfra disrupts heart tube assembly in both zebrafish and mouse . Timelapse analysis of individual cardiomyocyte trajectories reveals misdirected cells in zebrafish pdgfra mutants , suggesting that PDGF signaling steers cardiomyocytes toward the midline during cardiac fusion . Intriguingly , the ligand pdgfaa is expressed in the endoderm medial to the pdgfra-expressing myocardial precursors . Ectopic expression of pdgfaa interferes with cardiac fusion , consistent with an instructive role for PDGF signaling . Together , these data uncover a novel mechanism through which endodermal-myocardial communication can guide the cell movements that initiate cardiac morphogenesis .
Organogenesis relies upon the coordinated regulation of precisely defined patterns of cell movement . Multiple precursor cell populations must convene at the appropriate location and organize into the correct configuration in order to insure proper organ function . Differential adhesion and paracrine signaling between neighboring tissues often influence the specific routes traveled by precursor cells during morphogenesis ( Scarpa and Mayor , 2016 ) . However , the molecular mechanisms through which tissue interactions guide organ assembly remain poorly understood . Heart formation requires the coordinated movement of myocardial precursor cells from their bilateral origins toward the embryonic midline , where they meet and merge through a process called cardiac fusion ( Evans et al . , 2010 ) . Cardiac fusion is essential for the construction of the heart tube , which provides a fundamental foundation for subsequent steps in cardiac morphogenesis . During cardiac fusion , the medial movement of the myocardium is considered to be a collective cell behavior: the cardiomyocytes travel along relatively parallel paths with very little neighbor exchange ( Holtzman et al . , 2007 ) and simultaneously form intercellular junctions and create a primitive epithelial sheet ( Linask , 1992; Manasek , 1968; Stainier et al . , 1993; Trinh and Stainier , 2004; Ye et al . , 2015 ) . Whether these coherent patterns of myocardial movement reflect active migration or passive morphogenesis is not yet resolved ( Aleksandrova et al . , 2015; Dehaan , 1963; Varner and Taber , 2012; Xie et al . , 2016; Ye et al . , 2015 ) . In either case , it is important to elucidate the specific signals that dictate the medial direction of myocardial trajectories during cardiac fusion . Several lines of evidence indicate that cardiac fusion is mediated by interactions between the myocardium and the adjacent anterior endoderm . In both mouse and zebrafish , mutations that block endoderm formation or disrupt endoderm integrity also inhibit cardiac fusion ( Alexander et al . , 1999; Holtzman et al . , 2007; Kawahara et al . , 2009; Kikuchi et al . , 2001; Kupperman et al . , 2000; Li et al . , 2004; Mendelson et al . , 2015; Molkentin et al . , 1997; Osborne et al . , 2008; Ragkousi et al . , 2011; Roebroek et al . , 1998; Ye and Lin , 2013; Yelon et al . , 1999 ) . Studies tracking both endodermal and myocardial movement in chick have suggested that endodermal contraction provides a physical force that pulls the myocardium toward the midline ( Aleksandrova et al . , 2015; Cui et al . , 2009; Varner and Taber , 2012 ) . However , while endodermal forces may influence initial phases of cardiac fusion , the observed patterns of endoderm behavior seem insufficient to account for the entire path traversed by the moving cardiomyocytes ( Aleksandrova et al . , 2015; Cui et al . , 2009; Varner and Taber , 2012; Xie et al . , 2016; Ye et al . , 2015 ) . Moreover , observations of myocardial cell protrusions have suggested that these cells may actively migrate in response to endodermal cues ( Dehaan , 1963; Haack et al . , 2014; Ye et al . , 2015 ) . While it is clear that the endoderm plays an important role in facilitating cardiac fusion , the molecular underpinnings of the endodermal-myocardial relationship are still unknown . Here , we reveal a novel connection between the endoderm and myocardium by discovering a new role for platelet-derived growth factor ( PDGF ) signaling . PDGFs signal through receptor tyrosine kinases and are well known for their mitogenic activity ( Andrae et al . , 2008 ) , as well as for their role in guiding the migration of mesenchymal cells ( Ataliotis et al . , 1995; Yang et al . , 2008 ) . However , PDGF signaling has not been previously implicated in heart tube assembly , even though it is known to be important for later aspects of heart development , such as the contribution of cardiac neural crest cells to the outflow tract ( Morrison-Graham et al . , 1992; Schatteman et al . , 1995; Tallquist and Soriano , 2003 ) , the formation of the inflow tract ( Bleyl et al . , 2010 ) , and the formation of epicardial derivatives ( Smith et al . , 2011 ) . Our analysis of early morphogenetic defects caused by mutation of the gene encoding PDGF receptor alpha ( Pdgfra ) uncovers an essential function for Pdgfra during cardiac fusion in both zebrafish and mouse . Notably , through live imaging of individual cell movements in zebrafish mutants , we find that pdgfra is crucial for guiding cardiomyocyte movement toward the midline . Furthermore , our studies suggest that expression of PDGF ligands by the anterior endoderm could facilitate interaction of this tissue with the pdgfra-expressing myocardial precursors . Thus , our work supports a model in which PDGF signaling underlies communication between the endoderm and myocardium and thereby directs the cell movements that initiate heart tube assembly . These insights into the regulation of cardiomyocyte behavior provide new ideas regarding the etiology of diseases associated with aberrant cell movement ( Friedl and Gilmour , 2009 ) , including congenital heart diseases ( CHDs ) caused by defective myocardial morphogenesis ( Bleyl et al . , 2010; Briggs et al . , 2012; Neeb et al . , 2013; Samsa et al . , 2013 ) .
In a screen for ethylnitrosourea-induced mutations that disrupt cardiac morphogenesis in zebrafish ( Auman et al . , 2007 ) , we identified a recessive lethal mutation , refuse-to-fuse ( ref ) , that causes abnormal cardiac chamber morphology . Instead of the normal curvatures of the wild-type ventricle ( Figure 1A ) , ref mutants often displayed a bifurcated ventricle at 48 hours post-fertilization ( hpf ) ( Figure 1B ) . This phenotype was the most common among a range of cardiac defects in ref mutants ( Figure 1—figure supplement 1; Table 1 ) . On rare occasions , we found ref mutants with cardia bifida , a condition in which two separate hearts form in lateral positions ( Table 1 ) . The observed bifurcated ventricle and cardia bifida phenotypes led us to hypothesize that the ref mutation might interfere with cardiac fusion . In wild-type embryos , cardiac fusion results in the formation of a ring of cardiomyocytes at the midline by the 22 somite stage ( Figure 1C ) . In contrast , ref mutant cardiomyocytes failed to fuse into a ring and instead remained in separate bilateral domains ( Figure 1D; Table 1 ) or fused only in posterior positions , creating a horseshoe shape ( Figure 1E; Table 1 ) . Similar fusion defects were also observed when examining a broader portion of the anterior lateral plate mesoderm ( ALPM ) encompassing the heart fields ( Figure 1F , G and Figure 5A–F ) . 10 . 7554/eLife . 21172 . 003Figure 1 . Cardiac fusion defects in refuse-to-fuse ( ref ) mutants . ( A , B ) Three-dimensional reconstructions depict wild-type ( wt ) and ref mutant hearts expressing the myocardial reporter transgene Tg ( myl7:egfp ) at 48 hpf . In contrast to the normal contours of the wt heart ( A ) , the ref mutant heart ( B ) often displays a bifurcated , two-lobed ventricle and a misshapen atrium . See Figure 1—figure supplement 1 and Table 1 for more information on the range of cardiac phenotypes observed in ref mutants at 48 hpf . A: atrium; V: ventricle . ( C–G ) Dorsal ( C–E ) and ventral ( F , G ) views , anterior to the top , of wt ( C , F ) and ref ( D , E , G ) mutant embryos displaying the expression of myl7 at the 22 somite stage ( 22 s; C–E ) and the localization of ZO-1 at the 18 somite stage ( 18 s; F , G ) . ( C–E ) In ref mutants at this stage , cardiomyocytes typically fail to fuse at the midline ( D ) or only fuse posteriorly ( E ) . Note that the ref mutation is incompletely penetrant , although its penetrance is more evident at 20 s than at 48 hpf ( Table 2 ) , suggesting that some ref mutants recover as development proceeds . ( F , G ) ZO-1 localization highlights junctions forming within the maturing epithelium of the ALPM in both wt and ref mutant embryos . The ventral portion of the neural tube located at the midline is also visible . By 18 s , the wt ALPM ( F ) has initiated fusion at the midline , whereas the two sides of the ref mutant ALPM are still separate ( G ) . Scale bars: 60 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 21172 . 00310 . 7554/eLife . 21172 . 004Figure 1—figure supplement 1 . ref mutant embryos display a range of cardiac defects . ( A , B ) Lateral views of live wild-type ( wt , A ) and ref mutant ( B ) embryos at 48 hpf . ref mutants exhibit a mild pericardial edema . ( C–F ) Lateral views from the right side of wt ( C ) and ref mutant ( D–F ) hearts reveal a variety of abnormal cardiac morphologies in ref mutants . The most commonly observed phenotype was a bifurcated ventricle ( D ) , some other embryos exhibited abnormal looping ( E ) or a severely shrunken heart ( F ) , and a few embryos had cardia bifida ( not shown ) . See Tables 1 and 2 for more information regarding the incomplete penetrance and variable expressivity of the ref mutant phenotype . ( C'–F' ) Cartoons outline the cardiac morphologies shown in C–F , with the ventricle in magenta and the atrium in green . DOI: http://dx . doi . org/10 . 7554/eLife . 21172 . 00410 . 7554/eLife . 21172 . 005Figure 1—figure supplement 2 . The anterior endoderm forms normally in ref mutant embryos . ( A–D ) Lateral views , dorsal to the right , depict the expression of axial at 8 . 5 hpf ( A , B ) or sox17 at 8 hpf ( C , D ) . The number and distribution of endoderm precursor cells is similar in wt ( A , C ) and ref mutant ( B , n = 8/8; D , n = 14/14 ) embryos during gastrulation stages . ( E–H ) Dorsal views , anterior to the top , depict the anterior endoderm , visualized with the expression of axial at 24 hpf ( E , F ) or the endodermal reporter transgene Tg ( sox17:egfp ) at 18 s ( G , H ) . The width and morphology of the anterior endoderm is similar in wt ( E , G ) and ref mutant ( F , n = 13/13; H , n = 7/7 ) embryos during the stages when cardiac fusion takes place . Scale bars: 60 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 21172 . 00510 . 7554/eLife . 21172 . 006Figure 1—figure supplement 3 . Endocardial cells move to the midline in ref mutant embryos . ( A , B ) Dorsal views , anterior to the top , depict the expression of the endothelial reporter transgene Tg ( fli1a:negfp ) in the endocardial precursor population at 18 s ( n = 8 ) . In both wt ( A ) and ref mutant ( B ) embryos , the endocardial precursors move to the midline during the process of cardiac fusion . Scale bar: 60 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 21172 . 00610 . 7554/eLife . 21172 . 007Table 1 . Variable expressivity of cardiac phenotype in ref mutant embryos . DOI: http://dx . doi . org/10 . 7554/eLife . 21172 . 007StagePhenotypeNumber of embryos22 s∗ No fusion8Only posterior fusion748 hpf†Bifurcated ventricle229Shrunken heart147Abnormal looping74Cardia bifida4∗ Tabulated from 15 embryos with morphologically evident phenotypes from clutch depicted in Figure 1C–E and in Table 2 . See Figure 2—figure supplement 2 for additional data . † Tabulated from 454 embryos with morphologically evident phenotypes collected from multiple clutches . See Figure 1—figure supplement 1 for representative images . 10 . 7554/eLife . 21172 . 008Table 2 . Penetrance of cardiac phenotype in ref mutant embryos . DOI: http://dx . doi . org/10 . 7554/eLife . 21172 . 008StageTotal # embryos# +/+ embryos# +/− embryos# −/− embryos# with evident cardiac defectsApproximate penetrance22 s 611427201575% 48 hpf522NG∗NG∗NG∗5844%† ∗ NG=not genotyped . † Calculated with the assumption that 25% of embryos are -/- . Since prior studies in zebrafish have shown that defects in endoderm specification or morphogenesis can inhibit cardiac fusion ( Alexander et al . , 1999; Holtzman et al . , 2007; Kawahara et al . , 2009; Kikuchi et al . , 2001; Kupperman et al . , 2000; Mendelson et al . , 2015; Osborne et al . , 2008; Ye and Lin , 2013; Yelon et al . , 1999 ) , we examined the status of the endoderm in ref mutants . During gastrulation stages , the specification and movement of endodermal cells appeared normal in ref mutants ( Figure 1—figure supplement 2A–D ) . In addition , the differentiation and morphology of the anterior endoderm in ref mutants appeared intact during the stages when cardiac fusion takes place ( Figure 1—figure supplement 2E–H ) . The normal appearance of the ref mutant endoderm was consistent with the unaltered progress of the endocardial precursor cells in ref mutants: endocardial cells require interactions with the anterior endoderm for their medial movement during cardiac fusion ( Holtzman et al . , 2007; Wong et al . , 2012; Xie et al . , 2016 ) , and the ref mutant endocardium seemed to reach the midline normally ( Figure 1—figure supplement 3 ) . Taken together , our data suggest that defects in myocardial movement , as opposed to defects in the endoderm , cause the bifurcated cardiac morphology in ref mutants . In order to identify the genomic lesion responsible for the ref mutant phenotype , we mapped the ref locus to a <0 . 1 cM region on linkage group 20 containing six annotated genes ( Figure 2A ) . Our examination of ref mutant cDNA revealed missplicing in one of these genes , platelet-derived growth factor receptor alpha ( pdgfra ) ( Figure 2B ) . Specifically , we noted that exon 14 was omitted or truncated in the pdgfra messages detected in ref mutant cDNA . Furthermore , ref mutant genomic DNA contained a G-to-A mutation in the first nucleotide of intron 15 of pdgfra ( Figure 2C , D ) . Since a G at the exon/intron boundary is an essential conserved feature of splice sites , we infer that this mutation would disrupt pdgfra splicing . The misspliced pdgfra messages found in ref mutants cause a frameshift in the coding sequence , resulting in a premature truncation prior to the transmembrane domain of Pdgfra ( Figure 2D ) . In concordance with the concept that premature stop codons often lead to nonsense-mediated decay , we detected a global reduction of pdgfra mRNA in ref mutants ( Figure 2E , F ) . 10 . 7554/eLife . 21172 . 009Figure 2 . ref is a loss-of-function mutation in pdgfra . ( A ) Polymorphic markers ( z21170 , kdr_e28 , 5’scfd2 , z14614 ) were used to map meiotic recombination events , narrowing the region containing the ref mutation to <0 . 1 cM on linkage group ( LG ) 20 . ( See also Table 3 ) . Fractions indicate frequencies of proximal ( magenta ) and distal ( green ) recombination between markers and ref . Six annotated genes are present in this region ( GRCv10 ) ; sequence analysis of kdr , kita , gsx2 , lnx1 , and fip1l1a in ref mutants revealed only missense mutations that led to conserved amino acid changes . ( B ) RT-PCR spanning exons 13–19 of pdgfra generates a single , properly spliced product from homozygous wt embryos and multiple , smaller products from ref mutant embryos . Sequencing revealed that exon 14 was either omitted or truncated in these smaller products; in all cases , the observed missplicing would result in a frameshift followed by a premature stop codon . Although we did not detect any normally spliced pdgfra products in ref mutants , we cannot rule out the presence of low levels of wild-type mRNA . RT-PCR of gapdh demonstrates use of comparable amounts of template . ( C , D ) Sequencing the e14i15 exon-intron boundary of pdgfra revealed that ref mutant genomic DNA contains a G-to-A mutation in a conserved intronic nucleotide required for proper splicing . Chromatograms ( C ) and sequence alignment ( D ) show position of the mutation relative to reference sequences . Schematics ( D ) depict the proteins predicted to result from the wt , ref , and b1059 alleles of pdgfra; immunoglobulin ( magenta ) , transmembrane ( green ) , and tyrosine kinase ( blue ) domains are shown . ( E , F ) Lateral views depict expression of pdgfra at 11 s . Expression levels are lower in ref mutants ( F; n = 5/5 ) than in wt ( E ) . ( G–J ) Dorsal views , anterior to the top , of myl7 expression at 22 s . In contrast to wt ( G ) , cardiac fusion defects are evident in embryos injected with a pdgfra morpholino ( MO ) ( H ) , b1059 homozygous mutant embryos ( I ) , and ref/b1059 transheterozygous mutant embryos ( J ) . See Figure 2—figure supplement 2 for additional information on the prevalence of each of these phenotypes . Scale bars: 60 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 21172 . 00910 . 7554/eLife . 21172 . 010Figure 2—figure supplement 1 . ref mutants display craniofacial defects . ( A–F ) Lateral ( A–C ) and ventral ( D–F ) views of embryos at 6 days post-fertilization ( dpf ) , stained with Alcian blue to show the dorsal and ventral cartilage in the head . Embryos homozygous for ref ( B , E; n = 9/9 ) or transheterozgous for ref and b1059 ( C , F; n = 5/6 ) display similar dorsal cartilage formation defects . Specifically , the dorsal ethmoid plate is diminished or absent in these mutants . Scale bars: 60 μm . ( D'–F' ) Cartoons illustrate the dorsal cartilage structures shown in D–F . DOI: http://dx . doi . org/10 . 7554/eLife . 21172 . 01010 . 7554/eLife . 21172 . 011Figure 2—figure supplement 2 . Comparison of cardiac fusion phenotypes resulting from alteration of PDGF signaling . Bar graph compares severity of cardiac fusion phenotypes observed at 22 s in three sets of experiments: one set of experiments comparing ref homozygous mutant embryos , b1059 homozygous mutant embryos , ref/b1059 transheterozygous mutant embryos , pdgfra MO-injected embryos , and wt sibling embryos ( as in Figure 2G–J ) ; one set of experiments comparing embryos treated with DMSO and embryos treated with Pdgfr inhibitor V ( as in Figure 5—figure supplement 1 ) ; and one set of experiments comparing wt embryos and Tg ( hsp70l:pdgfaa-2A-mCherry ) embryos , following heat shock at the tailbud stage ( as in Figure 7Q–S ) . Phenotypes were classified as fused ( blue; as in Figure 2G ) , almost fused ( light blue; no representative image shown ) , only posterior fusion ( yellow; as in Figure 2H ) , or no fusion ( red; as in Figure 2J ) . The n for each data set is presented under its corresponding bar . The severity of the cardiac fusion defects appears similar in ref homozygous mutant embryos , b1059 homozygous mutant embryos , ref/b1059 transheterozygous mutant embryos , and embryos treated with Pdgfr inhibitor V . In contrast , the phenotypes observed in pdgfra MO-injected embryos are less severe , possibly reflecting the efficacy of this MO . In addition , we note that the effects caused by mutation of pdgfra and overexpression of pdgfaa are relatively similar . The similarity in phenotypes caused by loss and gain of PDGF signaling is reminiscent of results obtained in studies of the role of PDGF signaling during Xenopus gastrulation ( Damm and Winklbauer , 2011; Nagel et al . , 2004 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 21172 . 011 We next compared the ref mutant phenotype to the effects of another mutation in pdgfra , b1059 . The b1059 allele is a missense mutation that disrupts a conserved residue within the tyrosine kinase domain of Pdgfra ( Eberhart et al . , 2008 ) ( Figure 2D ) . Previous studies of b1059 mutant embryos focused on their dorsal jaw defects ( Eberhart et al . , 2008 ) ; our analysis also uncovered dorsal jaw defects in ref mutants ( Figure 2—figure supplement 1B , E ) , as well as cardiac fusion defects in b1059 mutants ( Figure 2G , I; Figure 2—figure supplement 2 ) . Through complementation testing , we found that ref and b1059 fail to complement each other; transheterozygotes displayed defects in both cardiac fusion ( Figure 2J; Figure 2—figure supplement 2 ) and dorsal jaw formation ( Figure 2—figure supplement 1C , F ) . Finally , we found that injection of a morpholino targeting pdgfra also interfered with cardiac fusion ( Figure 2H; Figure 2—figure supplement 2 ) . Together , our mapping , sequencing , complementation testing , and morpholino data support the conclusion that the ref mutation causes inappropriate splicing of pdgfra , resulting in diminished pdgfra function and cardiac fusion defects . Although prior work in mouse has revealed functions for PDGF signaling during later stages of heart development ( Grüneberg and Truslove , 1960; Richarte et al . , 2007; Schatteman et al . , 1995 ) , these studies did not report an earlier role for PDGFRα during cardiac fusion or heart tube assembly . In contrast , analysis of the Patch ( Ph ) mutant , carrying a chromosome deletion including Pdgfra , did reveal early cardiac phenotypes , as well as yolk sac defects ( Orr-Urtreger et al . , 1992 ) . Exploration of the early functions of PDGFRα has been complicated by the variability of Pdgfra mutant phenotypes , due in part to genetic background ( Grüneberg and Truslove , 1960; Orr-Urtreger and Lonai , 1992; Schatteman et al . , 1995; Soriano , 1997; Tallquist and Soriano , 2003 ) . Since the C57BL/6 background was reported to generate more severe phenotypes in the Ph mutant ( Orr-Urtreger et al . , 1992 ) , we chose to analyze mouse embryos carrying Pdgfra null alleles on a C57BL/6J background at E9 . 5 , using expression of Nkx2-5 to highlight heart morphology ( Figure 3A–E' ) . We note that we encountered Pdgfra null mutants on this background at E9 . 5 less often than predicted ( Table 4 ) , potentially because they fail to survive until this stage . Although the cause of this loss remains to be identified , it is consistent with the expression of Pdgfra during early embryogenesis ( e . g . Artus et al . , 2013; Palmieri et al . , 1992; Schatteman et al . , 1992 ) , and with studies in other vertebrates revealing a role for Pdgfra during gastrulation ( Nagel et al . , 2004; Yang et al . , 2008 ) . 10 . 7554/eLife . 21172 . 012Figure 3 . Pdgfra mouse mutants display defects in heart tube assembly . ( A–E' ) Ventral views ( A , B , D , E ) , lateral views ( A' , B' , C ) , and transverse sections ( A'' , E' ) depict Nkx2-5 expression in wt ( A–A'' ) and Pdgfra homozygous mutant ( B–E' ) mice on a C57BL/6J background at E9 . 5 . Some Pdgfra mutants display relatively mild cardiac defects ( B , B' ) , including defects in heart tube rotation . Other Pdgfra mutants display severe cardiac defects that could result from hindered cardiac fusion ( C–E' ) . These defects include incomplete heart tube assembly ( C , D ) with two inflow/common atrial regions ( red arrows ) and a single ventricle ( black arrowhead ) or with cardia bifida ( E , E'; arrows indicate unfused ventricles and atria ) . See Tables 4 and 5 for additional information on the prevalence of mutant phenotypes . A: atrium; AVC: atrioventricular canal; LA: left atrium; LV: left ventricle; OFT: outflow tract; RA: right atrium; RV: right ventricle; V: ventricle . Scale bars: 100 μm . ( F–I ) Ventral views depict localization of Pdgfra mRNA ( F–H ) and PDGFRα protein ( I ) in wt embryos from E7 . 5 to E8 . 5 . Pdgfra expression is seen in anterior lateral plate mesoderm ( ALPM ) at E7 . 5 ( F ) and persists in the caudal region of the forming heart at E8 . 0 ( G; arrows ) and in the inflow tract at E8 . 5 ( H; arrows ) . Pdgfra is also expressed in somites ( S ) and caudal lateral plate mesoderm ( LPM ) at E8 . 0-E8 . 5 . Although mRNA levels are diminished , PDGFRα protein localization is maintained throughout the forming heart at E8 . 0 ( I ) . FH: forming heart; H: heart . ( J , J' ) Ventral view ( J ) and transverse section ( J' ) depict the localization of Pdgfa mRNA at E8 . 0 . Black line in J indicates plane of transverse section shown in J’ . At this stage , Pdgfa expression is seen in the foregut endoderm ( Fg ) and lateral ectoderm . DOI: http://dx . doi . org/10 . 7554/eLife . 21172 . 01210 . 7554/eLife . 21172 . 013Table 3 . Primers used to map recombinants . DOI: http://dx . doi . org/10 . 7554/eLife . 21172 . 013MarkerForward primerReverse primer5’SCFD2CGCGTTACCAGAGAGACACATTCTCGGCAGGATAAATTGGZ14614AAACACATGCACAATGGTAGAAACAGCAAGTTCAGCCAAAACAZ21170AAACATTGCTTTTGGCCACTCTCACTCCCCCACACTGTTTkdr_e28TATGATAACGCTCCGCCTCTCAGGGGAATGTCCACAAAAC10 . 7554/eLife . 21172 . 014Table 4 . Genotypes encountered in progeny from intercrosses of Pdgfra heterozygotes at E9 . 5 . DOI: http://dx . doi . org/10 . 7554/eLife . 21172 . 014 TotalWild-typeHeterozygous mutantHomozygous mutantNumber of embryos10830∗64∗14†Observed ratio0 . 92 . 00 . 4‡Expected ratio1 . 02 . 01 . 0∗ No observed phenotype . † See Table 5 for detail on observed phenotypes . ‡ Chi-squared test p<0 . 05 compared to expected . Our analysis revealed a range of early defects in cardiac morphogenesis in homozygous Pdgfra mutants at E9 . 5 ( Figure 3A–E'; Table 5 ) . Wild-type hearts had undergone looping and exhibited distinct left and right atrial and ventricular chambers ( Figure 3A–A'' ) . Some Pdgfra homozygous mutant hearts displayed relatively mild defects in heart looping as well as in the size and shape of the cardiac chambers and their inflow and outflow tracts ( Figure 3B , B' ) . Other Pdgfra mutant hearts displayed more severe disruptions that could be the consequence of abnormal cardiac fusion ( Figure 3C–E' ) : the most prominent were embryos with a split inflow/common atrial region connected to a single ventricle ( Figure 3C , D ) , and we also observed a single embryo with near total cardia bifida ( Figure 3E , E' ) . Pdgfra mutants were often smaller than wild-type littermates , consistent with previous observations ( Orr-Urtreger et al . , 1992 ) ; however , the observed severe cardiac defects ( C-E' ) are not likely a result of general developmental delay , as these mutant phenotypes do not resemble wild-type cardiac morphology at younger stages . Severely affected embryos had not turned , as previously observed ( Soriano , 1997 ) . We did not observe omphalocele as reported ( Soriano , 1997 ) , although these previous observations were made at later time points than examined here . The majority of Pdgfra mutants died by E10 . 5 , slightly earlier than reported for the majority of Ph mutants ( Orr-Urtreger et al . , 1992 ) . Altogether , our data uncover a previously unappreciated influence of Pdgfra on the early stages of cardiac morphogenesis in mice . In combination with the phenotype of ref mutants , these studies suggest that Pdgfra plays a conserved role in regulating heart tube assembly . 10 . 7554/eLife . 21172 . 015Table 5 . Cardiac phenotypes observed in Pdgfra mutants at E9 . 5 . DOI: http://dx . doi . org/10 . 7554/eLife . 21172 . 015Number of embryosTotal14Normal phenotype2Abnormal looping5Split inflow/common atrial region6Cardia bifida1 To further elucidate how Pdgfra influences heart tube assembly , we next examined the expression pattern of pdgfra during cardiac fusion in zebrafish . We found robust expression of pdgfra within the ALPM and in migrating neural crest cells ( Figure 4A–D ) . The domains of pdgfra expression in the ALPM matched those of hand2 ( Figure 4E–J ) , which is expressed in the territories that contain myocardial precursor cells and is excluded from the territories containing endocardial precursors ( Schoenebeck et al . , 2007 ) . As cardiac fusion proceeds , hand2 continues to be expressed in the cardiomyocytes that reach the midline ( Figure 5A–C ) , while pdgfra expression appears to be absent from these cells ( Figure 4D ) . 10 . 7554/eLife . 21172 . 016Figure 4 . pdgfra is expressed within the ALPM while cardiac fusion is underway . ( A–D ) Dorsal views , anterior to the top , depict pdgfra expression in wt embryos at 10 s ( A ) , 12 s ( B ) , 14 s ( C ) , and 18 s ( D ) . Arrows ( A ) indicate pdgfra expression in the ALPM and the neural crest ( NC ) . Asterisk ( D ) denotes position of the myocardium by 18 s; although pdgfra is expressed in the myocardial precursors within the ALPM at earlier stages , its expression in these cells appears to be gone by this time point . ( E–J ) Comparison of hand2 ( green ) and pdgfra ( red ) expression patterns demonstrates their overlap in the wt ALPM at 10 s . ( E–G ) Three-dimensional confocal reconstructions of dorsal views , anterior to the top , focused on the left side of the ALPM ( area outlined by a dashed box in ( A ) . Arrows ( E ) indicate pdgfra expression in the ALPM and the NC . ( H–J ) Single transverse ( XZ ) sections from ( E–G ) , respectively . Yellow arrow ( H ) indicates overlap of hand2 and pdgfra expression in the ALPM . ( K–P ) Comparison of axial ( green ) and pdgfra ( red ) expression patterns demonstrates lack of pdgfra expression in the axial-expressing anterior endoderm in wt embryos at 12 s . ( K–M ) Three-dimensional confocal reconstructions of dorsal views , anterior to the top; arrows ( K ) indicate pdgfra expression in the ALPM and the neural crest . ( N–P ) Single transverse ( XZ ) sections from ( K–M ) , respectively . Arrowheads ( N ) indicate axial expression in the anterior endoderm , adjacent to , but not overlapping with , pdgfra expression in the ALPM . Scale bars: 60 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 21172 . 016 Similarly , we found that mouse Pdgfra is expressed in the ALPM at E7 . 5 ( Figure 3F ) and later becomes confined to the caudal aspect of the forming heart tube by E8 . 0 ( Figure 3G ) and to the inflow tract of the looping heart ( Figure 3H ) as well as the dorsal mesocardium ( Prall et al . , 2007 ) by E8 . 5 . In more mature hearts , Pdgfra is expressed in the atrioventricular valves and epicardium ( Chong et al . , 2011; Orr-Urtreger et al . , 1992 ) . Even though Pdgfra mRNA levels had declined in the anterior cardiac mesoderm by the beginning of heart tube formation ( Figure 3G ) , we found persistent PDGFRα protein expression in the forming heart at this stage ( Figure 3I ) . PDGFRα was also found in the more caudal domains defined by Pdgfra mRNA expression , including the caudal aspect of the forming heart corresponding to its future inflow tract and coelomic mesothelium ( Figure 3I ) ( Bax et al . , 2010 ) . We did not observe pdgfra expression within the anterior endoderm during cardiac fusion in either zebrafish or mouse ( Figures 3G and 4K–P; [Prall et al . , 2007] ) . In zebrafish , comparison of axial and pdgfra expression demonstrated mutually exclusive expression domains ( Figure 4K–P ) . Lack of pdgfra expression in the anterior endoderm is also consistent with previous expression analysis in mouse ( Orr-Urtreger and Lonai , 1992 ) , as well as with the lack of anterior endoderm defects in ref mutant embryos ( Figure 1—figure supplement 2 ) . Altogether , the pdgfra expression patterns in both zebrafish and mouse indicate that pdgfra could act within the ALPM to regulate the progression of cardiac fusion . Although our analysis pointed toward a role for pdgfra within the ALPM during cardiac fusion , we also considered the possibility that pdgfra expression in the early embryo ( Ataliotis et al . , 1995; Liu et al . , 2002; Mercola et al . , 1990; Yang et al . , 2008 ) could indirectly affect cardiac fusion by influencing processes such as mesoderm involution during gastrulation ( Yang et al . , 2008 ) . However , we did not observe any defects in the size , shape , or bilateral spacing of the ALPM in ref mutants at the 8–12 somite stages ( Figure 5A , D , G ) , indicating that early ALPM morphogenesis is intact in these embryos . Moreover , we found that pharmacological inhibition of Pdgfr activity at the tailbud stage can disrupt cardiac fusion ( Figure 5—figure supplement 1; Figure 2—figure supplement 2 ) , further supporting the conclusion that pdgfra activity influences cardiac fusion after gastrulation is complete . 10 . 7554/eLife . 21172 . 017Figure 5 . pdgfra influences the movement of the ALPM after the 15 somite stage . ( A–F ) Dorsal views , anterior to the top , depict expression of hand2 in the wt ( A–C ) and ref mutant ( D–F ) ALPM from 12 to 20 s . The morphology and position of the ALPM are indistinguishable in wt ( A ) and ref mutant ( D ) embryos at 12 s . After 15 s ( B , C , E , F ) , disrupted movement of the ALPM is evident in ref mutants . Scale bar: 60 μm . ( G ) Graph illustrates the average distance between the two sides of the ALPM in wt and ref mutant embryos from 8 to 20 s . In each embryo , the distance between the sides of the ALPM was calculated by measuring the distance between the medial edges of the hand2-expressing domains at three equidistant points ( 200 μm apart ) along the anterior-posterior axis . The largest of those three measurements was selected as representative of the maximum distance between the bilateral ALPM domains for that embryo . Dots represent the selected measurements from individual embryos . The distance between the bilateral sheets in ref mutant embryos begins to diverge significantly from wt after 15 s . Error bars represent the standard error . Asterisks indicate p<0 . 05 ( Student’s t-test ) : p=0 . 99 at 8 s; p=0 . 58 at 12 s; p=0 . 30 at 14 s; p=0 . 053 at 15 s; p=0 . 012 at 16 s; and p=0 . 00012 at 20 s . DOI: http://dx . doi . org/10 . 7554/eLife . 21172 . 01710 . 7554/eLife . 21172 . 018Figure 5—figure supplement 1 . PDGF signaling is required after gastrulation for proper cardiac fusion . ( A , B ) Dorsal views , anterior to the top , display myl7 expression at 22 s in representative embryos treated with DMSO ( A ) or Pdgfr inhibitor V ( B ) from the tailbud stage until 22 s . DMSO-treated control embryos exhibited normal cardiac fusion , whereas treatment with Pdgfr inhibitor V disrupted cardiac fusion . See Figure 2—figure supplement 2 for additional information on the prevalence of these phenotypes . Scale bar: 60 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 21172 . 01810 . 7554/eLife . 21172 . 019Figure 5—figure supplement 2 . Overexpression of pdgfaa influences ALPM movement by the 15 somite stage . ( A–C ) Dorsal views , anterior to the top , depict expression of hand2 in the ALPM of Tg ( hsp70l:pdgfaa-2A-mCherry ) embryos from 12 to 20 s , following heat shock at the tailbud stage . See Figure 5A–C for images of representative wt embryos at the same stages . Overexpression of pdgfaa at the tailbud stage does not seem to affect the morphology and position of the ALPM at 12 s ( A ) , but it does disrupt ALPM movement by 16 s ( B ) and 20 s ( C ) . Scale bar: 60 μm . ( D ) Graph illustrates the average distance between the two sides of the ALPM from 12 to 20 s in Tg ( hsp70l:pdgfaa-2A-mCherry ) embryos , following heat shock at the tailbud stage , shown together with the data presented in Figure 5G . The distance between the sides of the ALPM domains in transgenic embryos was measured as described for Figure 5G . Dots represent measurements from individual embryos . The distance between the bilateral sheets of ALPM in Tg ( hsp70l:pdgfaa-2A-mCherry ) embryos begins to diverge significantly from wt by 15 s . Error bars represent the standard error . Asterisks indicate p<0 . 05 in comparison to wt embryos ( Student’s T-test ) : p=0 . 47 at 12 s; p=0 . 12 at 14 s; p=0 . 024 at 15 s; p=0 . 090 at 16 s; and p=0 . 000027 at 20 s . DOI: http://dx . doi . org/10 . 7554/eLife . 21172 . 019 To determine when cardiac fusion first goes awry in ref mutants , we began by comparing the distance between the left and right sides of the ALPM in wild-type and ref mutant embryos . Until the 15 somite stage , the spacing between the bilateral domains of the ALPM was normal in ref mutants ( Figure 5G ) . After the 15 somite stage , the ref mutants began to display an evident phenotype: whereas the two sides of the wild-type ALPM continued to move toward each other , the sides of the ref mutant ALPM stayed apart ( Figure 5B , C , E , F , G ) . Thus , although the initial stages of ALPM convergence are unaffected in ref mutants , the ref mutant ALPM is unable to approach the midline normally during cardiac fusion . We next sought to elucidate the cellular defects responsible for the inhibition of cardiac fusion in ref mutants . Do ref mutant cardiomyocytes move at a sluggish rate or are they misdirected ? Previous studies have shown that VEGF signaling can regulate the speed of endocardial precursor movement during cardiac fusion ( Fish et al . , 2011 ) , suggesting the possibility that PDGF signaling might set the pace of myocardial precursor movement . Alternatively , PDGF signaling has been shown to control the direction of mesodermal movement during gastrulation ( Damm and Winklbauer , 2011; Nagel et al . , 2004 ) , suggesting that it could also guide the route taken by myocardial cells during cardiac fusion . To test these hypotheses , we tracked individual cell movements over time , using the myocardial reporter transgene Tg ( myl7:egfp ) ( Holtzman et al . , 2007; Huang et al . , 2003 ) to follow the patterns of cardiomyocyte behavior in live embryos ( Figure 6 ) . 10 . 7554/eLife . 21172 . 020Figure 6 . pdgfra regulates the directionality of cardiomyocyte movement . ( A–F ) Representative timelapse experiments indicate patterns of cell movement in wt ( A , B ) and ref mutant ( C–F ) embryos carrying the Tg ( myl7:egfp ) transgene . ( A , C , E ) Three-dimensional confocal reconstructions of select timepoints within each timelapse depict the typical changes in myocardial morphology seen over time in wt ( A ) , mildly affected ref mutant ( C ) and severely affected ref mutant ( E ) embryos . ( B , D , F ) Tracks show the movements of the innermost cardiomyocytes in these embryos over the course of a ~2 hr timelapse . Cell tracks are colored from blue to red , indicating the location of each cell from the beginning to the end of its trajectory . See also Videos 1–3 . Scale bar: 60 μm . ( G–L ) Quantitative analysis of cardiomyocyte movements . 168 and 137 cells were analyzed from eight wt and six ref mutant embryos , respectively . Graphs depict the average speed of individual cells ( G , distance/time ) , the average efficiency index of individual cells ( H , displacement/distance ) , the average velocity ( displacement/time ) of individual cells along the anterior-posterior axis ( I ) and along the medial-lateral axis ( J ) , the average medial-lateral velocity per embryo ( K ) , and the direction of the overall trajectory of individual cells ( L ) . Dots in ( J ) are colored to depict the embryo to which they belong , and the same color scheme is used in ( K ) . In ( L ) , individual cells are grouped into 10 bins based on their net direction of movement; length of each radial bar represents the number of cells in each bin . The velocity along the medial-lateral axis ( J , K ) and the direction of cell trajectories ( L ) were significantly altered in ref mutants , indicating the misdirection of ref mutant cardiomyocytes and implicating pdgfra in steering the medial direction of cardiomyocyte movement . Error bars represent the standard deviation; p values were determined using Student’s T-test . DOI: http://dx . doi . org/10 . 7554/eLife . 21172 . 020 We initiated our timelapse analysis at the 16 somite stage , the earliest timepoint when we could robustly detect Tg ( myl7:egfp ) expression . Consistent with our analysis of ALPM position ( Figure 5 ) , the bilateral populations of cardiomyocytes in ref mutants were already slightly farther apart than their wild-type counterparts were at the 16 somite stage ( Figure 6A , C , E ) . By following the movements of these cells during cardiac fusion , we found that wild-type cardiomyocytes display a coherent pattern of collective movement without significant neighbor exchange ( Figure 6A , B ) , consistent with our prior work ( Holtzman et al . , 2007 ) . Cardiomyocytes in ref mutants exhibited similar patterns of coherent movement ( Figure 6C–F ) . However , while wild-type cardiomyocytes moved progressively toward the midline ( Figure 6A , B; Video 1 ) , the medial movement of ref mutant cardiomyocytes seemed severely diminished , even though these cells still appeared to be in motion ( Figure 6C–F; Videos 2–3 ) . In ref mutants with a relatively mild phenotype , a posterior subset of cardiomyocytes still exhibited sufficient medial movement to fuse at the midline ( Figure 6C , D; Video 2 ) . In more severely affected ref mutants , medial movement appeared lost along the entire anterior-posterior extent of the myocardium ( Figure 6E , F; Video 3 ) . 10 . 7554/eLife . 21172 . 021Video 1 . Cardiomyocytes in a wild-type embryo undergo medially directed movement during cardiac fusion . ( A , B ) Representative timelapse movie ( A ) and associated tracks ( B ) of cardiomyocyte movement occurring during cardiac fusion in a wild-type embryo carrying the Tg ( myl7:egfp ) transgene . ( A ) Movie of drift-corrected three-dimensional reconstructions of 30 confocal slices taken at ~4 min intervals for ~2 hr , starting when eGFP could first be detected in the ALPM . ( B ) The movements of individual cardiomyocytes at the innermost region of the ALPM were tracked ( dots , B ) at each time point . Their positions over the previous 80 min are depicted as connected colored tracks ( blue-to-red , beginning-to-end ) . Blank frames indicate brief pauses in acquisition for refocusing . Arrows indicate initial starting position of cardiomyocytes . Asterisks indicate GFP+ cells that are not cardiomyocytes . Scale bar: 40 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 21172 . 02110 . 7554/eLife . 21172 . 022Video 2 . Not all cardiomyocytes in a mildly affected ref mutant embryo undergo medially directed movement during cardiac fusion . ( A , B ) Representative timelapse movie ( A ) and associated tracks ( B ) of cardiomyocyte movement in a mildly affected ref mutant embryo carrying the Tg ( myl7:egfp ) transgene . Images were acquired as described for Video 1; however , drift correction was not applied to this movie and thus its tracks were not included in further quantitative analysis . In mildly affected ref mutant embryos , posterior cardiomyocytes display sufficient medial movement to fuse at the midline , while anterior cardiomyocytes do not . Arrows indicate initial starting position of cardiomyocytes . Asterisks indicate GFP+ cells that are not cardiomyocytes . Scale bar: 40 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 21172 . 02210 . 7554/eLife . 21172 . 023Video 3 . Cardiomyocytes in a severely affected ref mutant embryo fail to display medially directed movement during cardiac fusion . ( A , B ) Representative timelapse movie ( A ) and associated tracks ( B ) of cardiomyocyte movement in a severely affected ref mutant embryo carrying the Tg ( myl7:egfp ) transgene . Images were acquired as described for Video 1 , with drift correction . In severely affected ref mutant embryos , none of the cardiomyocytes display measurable medial movement . Arrows indicate initial starting position of cardiomyocytes . Blank frames indicate brief pauses in acquisition for refocusing . Asterisks indicate GFP+ cells that are not cardiomyocytes . Scale bar: 40 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 21172 . 023 Lack of medial movement could be the result of defects in several aspects of cell behavior including speed , efficiency , and directionality . To distinguish between these possibilities , we performed quantitative analysis of individual cardiomyocyte trajectories . Compared to wild-type cardiomyocytes , ref mutant cardiomyocytes moved at a slightly slower average speed ( distance/time ) ( Figure 6G ) and with a slightly reduced efficiency ( displacement/distance ) ( Figure 6H ) . When examining velocity ( displacement/time ) along particular axes , we found no difference between the velocities of wild-type and ref mutant cardiomyocyte movement along the anterior-posterior axis ( Figure 6I ) . However , there was a substantial difference between the velocities of wild-type and ref mutant cardiomyocyte movement along the medial-lateral axis: the average velocity along the medial-lateral axis was 0 . 19 micron/min for wild-type cardiomyocytes , but was only 0 . 016 micron/min for ref mutant cardiomyocytes ( Figure 6J ) . This difference in cell behavior becomes even more striking when considering the variability in the ref mutant phenotype . Two of the six ref mutant embryos examined had a relatively mild phenotype , and the cardiomyocytes in these embryos exhibited an average medial-lateral velocity similar to that seen in wild-type embryos ( Figure 6K ) . In contrast , the other four ref mutant embryos displayed a more severe phenotype , and the cardiomyocytes in these embryos had an average medial-lateral velocity near or below zero ( Figure 6K ) . Further examination of the vectors of cell movement revealed that these deficiencies in medial-lateral velocity reflect the misdirection of ref mutant cardiomyocytes . In our wild-type timelapse data , almost all cardiomyocytes move in the medial direction , whereas over half of the cardiomyocytes in our ref mutant timelapse data show no medial movement , with many of these cells moving away from the midline ( Figure 6L ) . Together , these data reveal that pdgfra plays an important role in steering cardiomyocyte movement toward the midline during cardiac fusion . We next evaluated whether the expression patterns of genes encoding Pdgfra ligands could provide insight into how PDGF signaling confers directionality to cardiomyocyte movement . Initial examination demonstrated that both pdgfaa and pdgfab , but not pdgfc , are expressed in bilateral medial stripes within the anterior portion of the embryo ( Figure 7—figure supplement 1 ) . Deeper analysis of pdgfaa expression revealed that it is expressed in bilateral domains within the anterior endoderm between the 10 and 16 somite stages , positioned near the lateral edges of this tissue ( Figure 7A–J ) . The expression of pdgfaa within the anterior endoderm in zebrafish is grossly consistent with prior studies demonstrating expression of Pdgfa in the mouse foregut ( Palmieri et al . , 1992 ) . We readdressed this issue in mouse and found Pdgfa expression in endoderm at the rim of the foregut pocket ( Figure 3J ) as well as in the pharyngeal floor and pharyngeal pouches ( Figure 3J' ) , regions closely associated with the forming heart tube at E8 . 0 and earlier stages . Moreover , the pdgfaa expression domains in zebrafish are medially adjacent to the positions of the myocardial precursors within the ALPM ( Figure 7K–P ) , suggesting the possibility of a paracrine relationship between Pdgfa ligands in the endoderm and Pdgfra in the ALPM . 10 . 7554/eLife . 21172 . 024Figure 7 . pdgfaa is expressed in the anterior endoderm , medially adjacent to the ALPM . ( A–G ) Fluorescent in situ hybridization and immunofluorescence compare the expression of pdgfaa ( magenta ) and Tg ( sox17:egfp ) ( green ) in wt embryos at 13 s . ( A–C ) Dorsal views , anterior to the top , of a three-dimensional reconstruction show that pdgfaa is expressed in bilateral domains of the anterior endoderm , near the lateral edges of the endodermal sheet . Expression of pdgfaa is also evident in cranial rhombomeres . ( D–F ) A coronal ( XY ) slice of the same embryo demonstrates the overlap between pdgfaa and Tg ( sox17:egfp ) expression . ( G ) A sagittal ( ZY ) slice of the same embryo provides a lateral view . ( H–J ) Transverse cryosections compare the expression of pdgfaa ( blue ) and the expression of Tg ( sox17:egfp ) ( green ) in wt embryos at 13 s , showing that pdgfaa is expressed in lateral domains of the endodermal sheet . ( K–P ) Comparison of hand2 ( green ) and pdgfaa ( red ) expression patterns demonstrates that pdgfaa is expressed medially adjacent to the domains of hand2 expression in the ALPM . ( K–M ) Three-dimensional confocal reconstructions of dorsal views , anterior to the top . ( N–P ) Single transverse ( XZ ) slices from ( K–M ) , respectively . ( Q–S ) Dorsal views , anterior to the top , display the expression of myl7 at 22 s in nontransgenic ( Q ) or Tg ( hsp70l:pdgfaa-2A-mCherry ) ( R , S ) embryos , following heat shock at the tailbud stage . Ectopic expression of pdgfaa causes cardiac fusion defects . See Figure 2—figure supplement 2 and Figure 5—figure supplement 2 for additional information regarding this phenotype . Scale bars: 60 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 21172 . 02410 . 7554/eLife . 21172 . 025Figure 7—figure supplement 1 . Expression of genes encoding Pdgfra ligands . ( A–C ) Dorsal views , anterior to the top , display expression of pdgfaa ( A ) , pdgfab ( B ) , and pdgfc ( C ) in wt embryos at 14 s . Since Pdgfa homodimers and Pdgfc homodimers have been shown to interact with Pdgfra ( Andrae et al . , 2008 ) , we chose to examine the expression of the two zebrafish Pdgfa homologs , pdgfaa and pdgfab , as well as the expression of pdgfc ( Eberhart et al . , 2008 ) . Both pdgfaa ( A ) and pdgfab ( B ) are expressed in bilateral medial stripes ( arrows , A ) , consistent with the location of the anterior endoderm . We detected pdgfc ( C ) expression in the otic vesicles and somites at 14 s , but we did not detect expression over background levels within the anterior endoderm . Scale bar: 60 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 21172 . 025 To investigate whether the spatially restricted expression of pdgfa genes is important for the regulation of cardiac fusion , we induced pdgfaa expression throughout the embryo using the heat-inducible transgene Tg ( hsp70l:pdgfaa-2A-mCherry ) . Following heat shock at the tailbud stage , transgenic embryos displayed cardiac fusion defects similar to those observed in ref mutants ( Figure 7Q–S; Figure 2—figure supplement 2; Figure 5—figure supplement 2 ) . The ability of ectopic pdgfaa expression to disrupt cardiac fusion indicates that PDGF signaling has the potential to serve as an instructive influence in directing cardiomyocytes toward the midline .
Taken together , our studies point to a model in which the PDGF signaling pathway facilitates communication between the endoderm and the myocardium and thereby directs cardiomyocytes toward the midline during cardiac fusion . We propose that Pdgfa ligands , produced by the anterior endoderm , signal through the Pdgfra receptor in the ALPM in order to control the directionality of cardiomyocyte behavior . This connection parallels other examples in which PDGF ligand-receptor pairs in adjacent tissues influence cell movements ( e . g . in the kidney , neural crest , and gastrulating mesoderm [Eberhart et al . , 2008; Lindahl et al . , 1998; Yang et al . , 2008] ) , highlighting a paradigm for how tissue-tissue interactions establish the landscape of organogenesis ( Andrae et al . , 2008; Hoch and Soriano , 2003; Orr-Urtreger and Lonai , 1992 ) . Importantly , our demonstration of this new function for PDGF signaling represents the first insight into a pathway that guides the medial direction of cardiomyocyte movement . Moreover , our findings suggest a previously unappreciated molecular basis for the interactions between the endoderm and the myocardium that govern cardiac fusion . Further elucidation of the paracrine relationship between Pdgfa ligands in the endoderm and the Pdgfra receptor in the ALPM awaits the development of appropriate tools for the tissue-specific inactivation of these players . How might PDGF signaling confer directionality on the collective behavior of the cardiomyocytes ? In the absence of pdgfra function , myocardial movement is no longer directed toward the midline , implicating PDGF signaling in the arrangement of the forces that steer this epithelial tissue . One intriguing possibility is that PDGF signaling could direct medial movement via polarized Pdgfra activation that controls oriented formation of active protrusions , akin to the activity of the PDGF/VEGF receptor Pvr in Drosophila , which directs the collective movement of the epithelial border cells during oogenesis and the epidermal cells during dorsal closure ( Duchek et al . , 2001; Garlena et al . , 2015 ) . Alternatively , PDGF signaling could promote other types of epithelial reorganization that could facilitate directional movement , such as the rearrangement of adherens junctions or extracellular matrix at the medial edge of the myocardium , causing epithelial deformations that drive movement forward ( Xu et al . , 2005; Yang et al . , 2008 ) . Future examination of the relationship of PDGF signaling to the subcellular characteristics of the myocardium during cardiac fusion will help to elucidate the precise morphogenetic consequences of Pdgfra activity . It is likely that the PDGF signaling pathway works in parallel with additional pathways to influence cardiac cell movement during cardiac fusion . Although ref mutants fail to undergo proper cardiac fusion , they do not fully phenocopy mutants with primary endoderm defects ( e . g . casanova ( cas; sox32 ) or miles apart ( mil; s1pr2 ) [Alexander et al . , 1999; Kikuchi et al . , 2001; Kupperman et al . , 2000; Ye and Lin , 2013; Yelon et al . , 1999] ) . In cas and mil mutants , both the endocardial and myocardial precursors fail to move to the midline ( Holtzman et al . , 2007; Wong et al . , 2012; Xie et al . , 2016 ) ; moreover , their myocardial movement defects can be detected prior to the eight somite stage ( Ye et al . , 2015 ) . In contrast , the endocardial precursors seem to reach the midline normally in ref mutants , and the ref myocardial movement defects emerge only after the 15 somite stage . Most likely , other factors , such as VEGF signaling to the endocardium ( Fish et al . , 2011 ) or mechanical forces from the endoderm ( Aleksandrova et al . , 2015; Varner and Taber , 2012 ) , collaborate with PDGF signaling to control distinct aspects of endocardial and myocardial cell behavior during earlier and later phases of cardiac fusion . The cardiomyocyte movements that occur during cardiac fusion , guided by interactions with the endoderm , establish a foundation of proper tissue orientation and morphology upon which to assemble the initial heart tube . Our studies in zebrafish and mouse reveal a conserved influence of PDGF signaling on heart tube assembly . This influence is likely to be relevant to CHD in humans , since defects in cardiac morphology can originate in the cardiac precursor populations involved in cardiac fusion and heart tube assembly ( Prall et al . , 2007; Vincent and Buckingham , 2010 ) . More generally , the molecular mechanisms that control the direction of cardiomyocyte movement are likely to be relevant to the etiology of disorders that are caused by anomalous cell movement , potentially including ventricular septal defects , atrial septal defects , outflow tract defects , and trabeculation abnormalities ( Bax et al . , 2010; Bruneau , 2008; Ding et al . , 2004; Neeb et al . , 2013; Samsa et al . , 2013 ) , as well as inflow tract defects known to be associated with mutations in PDGFRA ( Bleyl et al . , 2010 ) . PDGFRA is deleted in a number of human families showing total anomalous pulmonary venous return ( TAPVR ) , in which the pulmonary arteries connect with the systemic venous system instead of the left atrium , a defect replicated in mouse and chick loss-of-function models ( Bleyl et al . , 2010 ) . TAPVR occurs in 1 in 15 , 000 live births and , while life-threatening , is at the mild end of the spectrum of morphogenetic defects that we have observed in Pdgfra knockout mice . Thus , our studies suggest the possibility of a broader involvement of PDGFRA mutations in CHD , specifically through their effects on heart tube assembly , and more globally as part of the spectrum of diseases associated with aberrant cardiac cell movements .
We used the following transgenic and mutant strains of zebrafish: Tg ( myl7:egfp ) twu34 ( Huang et al . , 2003 ) ( RRID:ZFIN_ZDB-GENO-050809-10 ) , Tg ( fli1a:negfp ) y7 ( Roman et al . , 2002 ) ( RRID:ZFIN_ZDB-GENO-060821-2 ) , Tg ( sox17:egfp ) ha01 ( Mizoguchi et al . , 2008 ) ( RRID:ZFIN_ZDB-GENO-080714-2 ) , pdgfrab1059 ( Eberhart et al . , 2008 ) ( RRID:ZFIN_ZDB-GENO-081008-1 ) , and ref ( pdgfrask16; this paper ) . The Tg ( hsp70l:pdgfaa-2A-mCherry ) sd44 transgene was assembled using the D-Topo vector ( Invitrogen , Carlsbad , CA ) with a pdgfaa cDNA lacking the stop codon ( Eberhart et al . , 2008 ) , in combination with established Gateway cloning vectors ( Kwan et al . , 2007 ) . The final destination vector was created by inserting a Cryaa:CFP cassette ( Hesselson et al . , 2009 ) into the pDestTol2pA4 vector ( gift from K . Kwan ) . Transgenic founders were established using standard techniques for Tol2-mediated transgenesis ( Fisher et al . , 2006 ) . We analyzed F2 embryos from four separate transgenic lines to evaluate the effect of pdgfaa overexpression on cardiac fusion . Embryos were heat shocked at 38°C for 45 min beginning at the tailbud stage and were then returned to 28°C . Transgenic and nontransgenic sibling embryos were distinguished based on their expression of mCherry following heat shock . All zebrafish work followed protocols approved by the UCSD IACUC . Pdgfra null embryos were generated by intercrossing heterozygous Pdgfratm11 ( EGFP ) Sor ( Hamilton et al . , 2003 ) ( RRID:MGI:5519063 ) mutant mice on a co-isogenic C57BL/6J background . In situ hybridization and immunohistochemistry were performed using standard protocols ( Prall et al . , 2007 ) , and genotyping was performed as described for Jax stock #007669 ( https://www . jax . org/strain/007669 ) . Images were captured using a Leica M125 microscope outfitted with a Leica DFC295 camera and processed using Adobe Photoshop . All mouse experiments were overseen by the Garvan Institute of Medical Research/St . Vincent's Hospital Animal Ethics Committee . Meiotic recombinants were mapped using polymorphic SSLP and SNP markers to identify a small critical interval on linkage group 20; PCR primers used for mapping are provided in Table 3 . Sequence analysis of candidate genes was performed on cDNA from homozygous wild-type and ref mutant embryos . PCR genotyping was used to identify ref mutant embryos following phenotypic analysis . The primer pair 5'-GTAGGTAAAAGTAAAGCTGGTA-3' and 5'-CAAGGGTGTGTTGAACCTGA-3' amplifies a 136 bp PCR product flanking the e14i15 boundary in the pdgfra locus and creates a KpnI restriction site within the wild-type allele , but not within the ref mutant allele . Digestion of the wild-type PCR product with KpnI creates fragments of 113 and 23 bp . A pdgfra morpholino ( 5'-CACTCGCAAATCAGACCCTCCTGAT-3' ) was designed to disrupt the splicing of exon 11 and thereby lead to premature truncation of Pdgfra prior to its kinase domain . We injected 12 ng of morpholino at the one-cell stage; this dose did not induce visible toxicity . Furthermore , injection of this morpholino into ref mutants did not increase the frequency or severity of their cardiac fusion defects . For pharmacological inhibition of PDGF signaling ( Kim et al . , 2010 ) , we incubated embryos in Pdgfr inhibitor V ( Calbiochem 521234 , Temecula , CA ) from the tailbud stage until the 22 somite stage . Three separate experiments were performed , using doses of 0 . 25–0 . 4 μM . The following probes and antibodies were used: myl7 ( ZDB-GENE-991019–3 ) , axial/foxa2 ( ZDB-GENE-980526–404 , sox17 ( ZDB-GENE-991213–1 ) , hand2 ( ZDB-GENE-000511–1 ) , pdgfra ( ZDB-GENE-990415–208 ) , pdgfaa ( ZDB-GENE-030918–2 ) , pdgfab ( ZDB-GENE-060929–124 ) , pdgfc ( ZDB-GENE-071217–2 ) , anti-GFP ( RRID:AB_300798; Abcam ab13970; 1:1000 ) , anti-ZO-1 ( RRID:AB_2533147; Zymed 33–9100; 1:200 ) , and donkey anti-mouse Alexa 488 ( RRID:AB_141607; Invitrogen; 1:300 ) . Standard in situ hybridization , fluorescent in situ hybridization , and immunofluorescence were performed using established protocols ( Alexander et al . , 1998; Brend and Holley , 2009; Yelon et al . , 1999 ) . Fluorescent in situ hybridization was combined with immunofluorescence as previously described ( Zeng and Yelon , 2014 ) . Standard in situ hybridization was combined with visualization of transgene expression by creating transverse sections following in situ hybridization , using standard cryoprotection , embedding , and sectioning techniques ( Garavito-Aguilar et al . , 2010 ) and then performing standard immunofluorescence for GFP on sections . Alcian blue staining was performed as previously described ( Kimmel et al . , 1998 ) . Trunks were removed for genotyping prior to Alcian staining . Images were captured using Zeiss M2Bio , AxioZoom and AxioImager microscopes outfitted with Axiocam cameras and processed with Adobe Photoshop . Confocal stacks were collected using a Leica SP5 confocal laser-scanning microscope and processed using Imaris ( Bitplane , Belfast , Ireland ) . Tg ( myl7:egfp ) embryos at the 14 somite stage were mounted head down in 0 . 8% low-melt agarose and placed on a coverslip bottom dish in wells made from a layer of 3% agarose . Timelapse images were collected using a Leica SP5 confocal microscope with a 20X objective , in a chamber heated to 28°C . Confocal stacks of GFP and brightfield images were collected every 4 min for 2–3 hr , starting around the 16 somite stage . In each stack , 30 confocal slices spanning the expression of Tg ( myl7:egfp ) were collected at ~3 μm intervals . Embryos were retained after completion of imaging , and we only analyzed data from embryos that appeared healthy for 24 hr following the timelapse . Image processing and cell tracking was performed on three-dimensional reconstructions generated with Imaris , using the semi-automated cell tracking module . In each embryo , we tracked 20–30 cells from the two most medial columns of cardiomyocytes on each side . Only tracks in which a cell position could be determined for each timepoint were used for further analysis . We also tracked the tip of the notochord in brightfield images at each timepoint . Although we observed a slight posterior retraction of the notochord over the course of our timelapse analysis , we found that this was the most consistent landmark to use as a reference point to correct for drift that occurred during imaging . Thus , the movement of the tracked notochord tip was subtracted from the movement of each tracked cardiomyocyte . Our wild-type tracking data were largely consistent with our prior studies ( Holtzman et al . , 2007 ) , including the velocity of movement , coherence of movement , lack of cell movement in the Z-axis , and direction of wild-type cardiomyocyte trajectories . Subtle differences between these two data sets are likely due to our current use of the notochord as a reference point and the slightly later stage at which we initiated these timelapse experiments . For quantitative analysis of cardiomyocyte movement , we extracted the X and Y position of each cell at each timepoint along its track , as previously described ( Holtzman et al . , 2007 ) . Cell movement properties , including overall speed ( distance/time ) , efficiency ( displacement/distance ) , velocity ( displacement/time ) , and direction , were then calculated for each individual cardiomyocyte . Velocity measurements were split into their X ( medial-lateral ) and Y ( anterior-posterior ) components . Cells along the anterior-posterior axis were further divided into top , middle , and bottom subsets , as in our prior work ( Holtzman et al . , 2007 ) . Direction was calculated as arctan[abs ( y-displacement ) / ( x-displacement ) ] , after aligning movement between the left and right sides . Graphs were made using Matlab ( Mathworks , Natick , MA ) and Prism ( Graphpad , La Jolla , CA ) software . All statistical analyses were performed using a two-tailed unpaired Student’s t-test . No statistical methods were used to predetermine sample sizes . Instead , sample sizes were determined based on prior experience with relevant phenotypes and standards within the zebrafish and mouse communities . All results were collected from at least two independent experiments ( technical replicates ) in which multiple embryos , from multiple independent matings , were analyzed ( biological replicates ) . | In the growing embryo , the heart initially develops in the form of a simple tube . Its outer layer is made up of muscular cells , called myocardial cells , that pump blood through the tube . Before the heart tube develops , two groups of myocardial cells exist – one on each side of the embryo . To assemble the heart , these two populations of cells must move as a group to the middle of the embryo , where they meet and merge through a process called cardiac fusion . This movement of myocardial cells toward the middle of the embryo depends upon interactions with a neighboring tissue called the endoderm . How the endoderm directs the movement of the myocardial cells was not well understood . The PDGF signaling pathway guides the movement of several different types of cells in the body , but it had not been previously linked to the early stages of heart tube assembly . In this pathway , a molecule called platelet-derived growth factor ( PDGF ) binds to PDGF receptors that sit on the surface of cells . Using microscopy and genetic analysis to study zebrafish and mouse embryos , Bloomekatz et al . now show that embryos that carry mutations in a gene that encodes a PDGF receptor suffer from defects in heart tube assembly . Further examination of the mutant zebrafish embryos revealed that the myocardial cells were not properly directed toward the middle of the embryo . In fact , many of these cells appeared to move away from the midline . Bloomekatz et al . also observed that , in normal embryos , the endoderm cells that lie adjacent to the myocardial cells produce PDGF . Therefore , it appears that PDGF produced by the endoderm could interact with PDGF receptors on the myocardial cells to direct these cells toward the middle of the embryo . The next step will be to figure out how this signaling influences the machinery inside the myocardial cells that controls their movement . Ultimately , this knowledge could lead to new ways to identify and treat congenital heart diseases . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"developmental",
"biology"
] | 2017 | Platelet-derived growth factor (PDGF) signaling directs cardiomyocyte movement toward the midline during heart tube assembly |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.